src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256i          vfitab;
 109     __m128i          vfitab_lo,vfitab_hi;
 110     __m128i          ifour       = _mm_set1_epi32(4);
 111     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 112     real             *vftab;
 113     __m256           dummy_mask,cutoff_mask;
 114     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 115     __m256           one     = _mm256_set1_ps(1.0);
 116     __m256           two     = _mm256_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_elec_vdw->data;
 135     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 140     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 141     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm256_setzero_ps();
 182         fiy0             = _mm256_setzero_ps();
 183         fiz0             = _mm256_setzero_ps();
 184         fix1             = _mm256_setzero_ps();
 185         fiy1             = _mm256_setzero_ps();
 186         fiz1             = _mm256_setzero_ps();
 187         fix2             = _mm256_setzero_ps();
 188         fiy2             = _mm256_setzero_ps();
 189         fiz2             = _mm256_setzero_ps();
 190         fix3             = _mm256_setzero_ps();
 191         fiy3             = _mm256_setzero_ps();
 192         fiz3             = _mm256_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_ps();
 196         vvdwsum          = _mm256_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             jnrE             = jjnr[jidx+4];
 208             jnrF             = jjnr[jidx+5];
 209             jnrG             = jjnr[jidx+6];
 210             jnrH             = jjnr[jidx+7];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215             j_coord_offsetE  = DIM*jnrE;
 216             j_coord_offsetF  = DIM*jnrF;
 217             j_coord_offsetG  = DIM*jnrG;
 218             j_coord_offsetH  = DIM*jnrH;
 219
 220             /* load j atom coordinates */
 221             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 223                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 224                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 225                                                  &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm256_sub_ps(ix0,jx0);
 229             dy00             = _mm256_sub_ps(iy0,jy0);
 230             dz00             = _mm256_sub_ps(iz0,jz0);
 231             dx10             = _mm256_sub_ps(ix1,jx0);
 232             dy10             = _mm256_sub_ps(iy1,jy0);
 233             dz10             = _mm256_sub_ps(iz1,jz0);
 234             dx20             = _mm256_sub_ps(ix2,jx0);
 235             dy20             = _mm256_sub_ps(iy2,jy0);
 236             dz20             = _mm256_sub_ps(iz2,jz0);
 237             dx30             = _mm256_sub_ps(ix3,jx0);
 238             dy30             = _mm256_sub_ps(iy3,jy0);
 239             dz30             = _mm256_sub_ps(iz3,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 243             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 244             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 245             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 246
 247             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 248             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 249             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 250             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 251
 252             /* Load parameters for j particles */
 253             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 254                                                                  charge+jnrC+0,charge+jnrD+0,
 255                                                                  charge+jnrE+0,charge+jnrF+0,
 256                                                                  charge+jnrG+0,charge+jnrH+0);
 257             vdwjidx0A        = 2*vdwtype[jnrA+0];
 258             vdwjidx0B        = 2*vdwtype[jnrB+0];
 259             vdwjidx0C        = 2*vdwtype[jnrC+0];
 260             vdwjidx0D        = 2*vdwtype[jnrD+0];
 261             vdwjidx0E        = 2*vdwtype[jnrE+0];
 262             vdwjidx0F        = 2*vdwtype[jnrF+0];
 263             vdwjidx0G        = 2*vdwtype[jnrG+0];
 264             vdwjidx0H        = 2*vdwtype[jnrH+0];
 265
 266             fjx0             = _mm256_setzero_ps();
 267             fjy0             = _mm256_setzero_ps();
 268             fjz0             = _mm256_setzero_ps();
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r00              = _mm256_mul_ps(rsq00,rinv00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 278                                             vdwioffsetptr0+vdwjidx0B,
 279                                             vdwioffsetptr0+vdwjidx0C,
 280                                             vdwioffsetptr0+vdwjidx0D,
 281                                             vdwioffsetptr0+vdwjidx0E,
 282                                             vdwioffsetptr0+vdwjidx0F,
 283                                             vdwioffsetptr0+vdwjidx0G,
 284                                             vdwioffsetptr0+vdwjidx0H,
 285                                             &c6_00,&c12_00);
 286
 287             /* Calculate table index by multiplying r with table scale and truncate to integer */
 288             rt               = _mm256_mul_ps(r00,vftabscale);
 289             vfitab           = _mm256_cvttps_epi32(rt);
 290             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 291             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 292             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 293             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 294             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 295             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 296
 297             /* CUBIC SPLINE TABLE DISPERSION */
 298             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 299             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 300             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 301                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 302             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 303                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 304             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 305                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 306             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 308             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 309             Heps             = _mm256_mul_ps(vfeps,H);
 310             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 311             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 312             vvdw6            = _mm256_mul_ps(c6_00,VV);
 313             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 314             fvdw6            = _mm256_mul_ps(c6_00,FF);
 315
 316             /* CUBIC SPLINE TABLE REPULSION */
 317             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 318             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 319             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 320                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 321             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 323             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 325             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 327             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 328             Heps             = _mm256_mul_ps(vfeps,H);
 329             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 330             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 331             vvdw12           = _mm256_mul_ps(c12_00,VV);
 332             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 333             fvdw12           = _mm256_mul_ps(c12_00,FF);
 334             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 335             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 339
 340             fscal            = fvdw;
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm256_mul_ps(fscal,dx00);
 344             ty               = _mm256_mul_ps(fscal,dy00);
 345             tz               = _mm256_mul_ps(fscal,dz00);
 346
 347             /* Update vectorial force */
 348             fix0             = _mm256_add_ps(fix0,tx);
 349             fiy0             = _mm256_add_ps(fiy0,ty);
 350             fiz0             = _mm256_add_ps(fiz0,tz);
 351
 352             fjx0             = _mm256_add_ps(fjx0,tx);
 353             fjy0             = _mm256_add_ps(fjy0,ty);
 354             fjz0             = _mm256_add_ps(fjz0,tz);
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             r10              = _mm256_mul_ps(rsq10,rinv10);
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq10             = _mm256_mul_ps(iq1,jq0);
 364
 365             /* Calculate table index by multiplying r with table scale and truncate to integer */
 366             rt               = _mm256_mul_ps(r10,vftabscale);
 367             vfitab           = _mm256_cvttps_epi32(rt);
 368             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 369             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 370             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 371             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 372             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 373             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 374
 375             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 376             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 377                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 378             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 379                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 380             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 381                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 382             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 383                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 384             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 385             Heps             = _mm256_mul_ps(vfeps,H);
 386             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 387             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 388             velec            = _mm256_mul_ps(qq10,VV);
 389             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 390             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm256_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm256_mul_ps(fscal,dx10);
 399             ty               = _mm256_mul_ps(fscal,dy10);
 400             tz               = _mm256_mul_ps(fscal,dz10);
 401
 402             /* Update vectorial force */
 403             fix1             = _mm256_add_ps(fix1,tx);
 404             fiy1             = _mm256_add_ps(fiy1,ty);
 405             fiz1             = _mm256_add_ps(fiz1,tz);
 406
 407             fjx0             = _mm256_add_ps(fjx0,tx);
 408             fjy0             = _mm256_add_ps(fjy0,ty);
 409             fjz0             = _mm256_add_ps(fjz0,tz);
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             r20              = _mm256_mul_ps(rsq20,rinv20);
 416
 417             /* Compute parameters for interactions between i and j atoms */
 418             qq20             = _mm256_mul_ps(iq2,jq0);
 419
 420             /* Calculate table index by multiplying r with table scale and truncate to integer */
 421             rt               = _mm256_mul_ps(r20,vftabscale);
 422             vfitab           = _mm256_cvttps_epi32(rt);
 423             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 424             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 425             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 426             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 427             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 428             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 429
 430             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 431             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 432                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 433             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 434                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 435             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 436                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 437             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 438                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 439             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 440             Heps             = _mm256_mul_ps(vfeps,H);
 441             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 442             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 443             velec            = _mm256_mul_ps(qq20,VV);
 444             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 445             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 446
 447             /* Update potential sum for this i atom from the interaction with this j atom. */
 448             velecsum         = _mm256_add_ps(velecsum,velec);
 449
 450             fscal            = felec;
 451
 452             /* Calculate temporary vectorial force */
 453             tx               = _mm256_mul_ps(fscal,dx20);
 454             ty               = _mm256_mul_ps(fscal,dy20);
 455             tz               = _mm256_mul_ps(fscal,dz20);
 456
 457             /* Update vectorial force */
 458             fix2             = _mm256_add_ps(fix2,tx);
 459             fiy2             = _mm256_add_ps(fiy2,ty);
 460             fiz2             = _mm256_add_ps(fiz2,tz);
 461
 462             fjx0             = _mm256_add_ps(fjx0,tx);
 463             fjy0             = _mm256_add_ps(fjy0,ty);
 464             fjz0             = _mm256_add_ps(fjz0,tz);
 465
 466             /**************************
 467              * CALCULATE INTERACTIONS *
 468              **************************/
 469
 470             r30              = _mm256_mul_ps(rsq30,rinv30);
 471
 472             /* Compute parameters for interactions between i and j atoms */
 473             qq30             = _mm256_mul_ps(iq3,jq0);
 474
 475             /* Calculate table index by multiplying r with table scale and truncate to integer */
 476             rt               = _mm256_mul_ps(r30,vftabscale);
 477             vfitab           = _mm256_cvttps_epi32(rt);
 478             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 479             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 480             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 481             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 482             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 483             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 484
 485             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 486             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 487                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 488             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 489                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 490             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 491                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 492             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 493                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 494             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 495             Heps             = _mm256_mul_ps(vfeps,H);
 496             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 497             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 498             velec            = _mm256_mul_ps(qq30,VV);
 499             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 500             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velecsum         = _mm256_add_ps(velecsum,velec);
 504
 505             fscal            = felec;
 506
 507             /* Calculate temporary vectorial force */
 508             tx               = _mm256_mul_ps(fscal,dx30);
 509             ty               = _mm256_mul_ps(fscal,dy30);
 510             tz               = _mm256_mul_ps(fscal,dz30);
 511
 512             /* Update vectorial force */
 513             fix3             = _mm256_add_ps(fix3,tx);
 514             fiy3             = _mm256_add_ps(fiy3,ty);
 515             fiz3             = _mm256_add_ps(fiz3,tz);
 516
 517             fjx0             = _mm256_add_ps(fjx0,tx);
 518             fjy0             = _mm256_add_ps(fjy0,ty);
 519             fjz0             = _mm256_add_ps(fjz0,tz);
 520
 521             fjptrA             = f+j_coord_offsetA;
 522             fjptrB             = f+j_coord_offsetB;
 523             fjptrC             = f+j_coord_offsetC;
 524             fjptrD             = f+j_coord_offsetD;
 525             fjptrE             = f+j_coord_offsetE;
 526             fjptrF             = f+j_coord_offsetF;
 527             fjptrG             = f+j_coord_offsetG;
 528             fjptrH             = f+j_coord_offsetH;
 529
 530             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 531
 532             /* Inner loop uses 188 flops */
 533         }
 534
 535         if(jidx<j_index_end)
 536         {
 537
 538             /* Get j neighbor index, and coordinate index */
 539             jnrlistA         = jjnr[jidx];
 540             jnrlistB         = jjnr[jidx+1];
 541             jnrlistC         = jjnr[jidx+2];
 542             jnrlistD         = jjnr[jidx+3];
 543             jnrlistE         = jjnr[jidx+4];
 544             jnrlistF         = jjnr[jidx+5];
 545             jnrlistG         = jjnr[jidx+6];
 546             jnrlistH         = jjnr[jidx+7];
 547             /* Sign of each element will be negative for non-real atoms.
 548              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 549              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 550              */
 551             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 552                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 553
 554             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 555             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 556             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 557             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 558             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 559             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 560             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 561             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 562             j_coord_offsetA  = DIM*jnrA;
 563             j_coord_offsetB  = DIM*jnrB;
 564             j_coord_offsetC  = DIM*jnrC;
 565             j_coord_offsetD  = DIM*jnrD;
 566             j_coord_offsetE  = DIM*jnrE;
 567             j_coord_offsetF  = DIM*jnrF;
 568             j_coord_offsetG  = DIM*jnrG;
 569             j_coord_offsetH  = DIM*jnrH;
 570
 571             /* load j atom coordinates */
 572             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 573                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 574                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 575                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 576                                                  &jx0,&jy0,&jz0);
 577
 578             /* Calculate displacement vector */
 579             dx00             = _mm256_sub_ps(ix0,jx0);
 580             dy00             = _mm256_sub_ps(iy0,jy0);
 581             dz00             = _mm256_sub_ps(iz0,jz0);
 582             dx10             = _mm256_sub_ps(ix1,jx0);
 583             dy10             = _mm256_sub_ps(iy1,jy0);
 584             dz10             = _mm256_sub_ps(iz1,jz0);
 585             dx20             = _mm256_sub_ps(ix2,jx0);
 586             dy20             = _mm256_sub_ps(iy2,jy0);
 587             dz20             = _mm256_sub_ps(iz2,jz0);
 588             dx30             = _mm256_sub_ps(ix3,jx0);
 589             dy30             = _mm256_sub_ps(iy3,jy0);
 590             dz30             = _mm256_sub_ps(iz3,jz0);
 591
 592             /* Calculate squared distance and things based on it */
 593             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 594             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 595             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 596             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 597
 598             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 599             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 600             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 601             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 602
 603             /* Load parameters for j particles */
 604             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 605                                                                  charge+jnrC+0,charge+jnrD+0,
 606                                                                  charge+jnrE+0,charge+jnrF+0,
 607                                                                  charge+jnrG+0,charge+jnrH+0);
 608             vdwjidx0A        = 2*vdwtype[jnrA+0];
 609             vdwjidx0B        = 2*vdwtype[jnrB+0];
 610             vdwjidx0C        = 2*vdwtype[jnrC+0];
 611             vdwjidx0D        = 2*vdwtype[jnrD+0];
 612             vdwjidx0E        = 2*vdwtype[jnrE+0];
 613             vdwjidx0F        = 2*vdwtype[jnrF+0];
 614             vdwjidx0G        = 2*vdwtype[jnrG+0];
 615             vdwjidx0H        = 2*vdwtype[jnrH+0];
 616
 617             fjx0             = _mm256_setzero_ps();
 618             fjy0             = _mm256_setzero_ps();
 619             fjz0             = _mm256_setzero_ps();
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             r00              = _mm256_mul_ps(rsq00,rinv00);
 626             r00              = _mm256_andnot_ps(dummy_mask,r00);
 627
 628             /* Compute parameters for interactions between i and j atoms */
 629             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 630                                             vdwioffsetptr0+vdwjidx0B,
 631                                             vdwioffsetptr0+vdwjidx0C,
 632                                             vdwioffsetptr0+vdwjidx0D,
 633                                             vdwioffsetptr0+vdwjidx0E,
 634                                             vdwioffsetptr0+vdwjidx0F,
 635                                             vdwioffsetptr0+vdwjidx0G,
 636                                             vdwioffsetptr0+vdwjidx0H,
 637                                             &c6_00,&c12_00);
 638
 639             /* Calculate table index by multiplying r with table scale and truncate to integer */
 640             rt               = _mm256_mul_ps(r00,vftabscale);
 641             vfitab           = _mm256_cvttps_epi32(rt);
 642             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 643             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 644             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 645             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 646             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 647             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 648
 649             /* CUBIC SPLINE TABLE DISPERSION */
 650             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 651             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 652             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 653                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 654             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 655                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 656             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 657                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 658             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 659                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 660             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 661             Heps             = _mm256_mul_ps(vfeps,H);
 662             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 663             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 664             vvdw6            = _mm256_mul_ps(c6_00,VV);
 665             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 666             fvdw6            = _mm256_mul_ps(c6_00,FF);
 667
 668             /* CUBIC SPLINE TABLE REPULSION */
 669             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 670             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 671             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 672                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 673             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 674                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 675             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 676                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 677             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 678                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 679             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 680             Heps             = _mm256_mul_ps(vfeps,H);
 681             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 682             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 683             vvdw12           = _mm256_mul_ps(c12_00,VV);
 684             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 685             fvdw12           = _mm256_mul_ps(c12_00,FF);
 686             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 687             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 688
 689             /* Update potential sum for this i atom from the interaction with this j atom. */
 690             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 691             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 692
 693             fscal            = fvdw;
 694
 695             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 696
 697             /* Calculate temporary vectorial force */
 698             tx               = _mm256_mul_ps(fscal,dx00);
 699             ty               = _mm256_mul_ps(fscal,dy00);
 700             tz               = _mm256_mul_ps(fscal,dz00);
 701
 702             /* Update vectorial force */
 703             fix0             = _mm256_add_ps(fix0,tx);
 704             fiy0             = _mm256_add_ps(fiy0,ty);
 705             fiz0             = _mm256_add_ps(fiz0,tz);
 706
 707             fjx0             = _mm256_add_ps(fjx0,tx);
 708             fjy0             = _mm256_add_ps(fjy0,ty);
 709             fjz0             = _mm256_add_ps(fjz0,tz);
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             r10              = _mm256_mul_ps(rsq10,rinv10);
 716             r10              = _mm256_andnot_ps(dummy_mask,r10);
 717
 718             /* Compute parameters for interactions between i and j atoms */
 719             qq10             = _mm256_mul_ps(iq1,jq0);
 720
 721             /* Calculate table index by multiplying r with table scale and truncate to integer */
 722             rt               = _mm256_mul_ps(r10,vftabscale);
 723             vfitab           = _mm256_cvttps_epi32(rt);
 724             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 725             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 726             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 727             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 728             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 729             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 730
 731             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 732             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 733                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 734             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 735                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 736             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 737                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 738             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 739                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 740             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 741             Heps             = _mm256_mul_ps(vfeps,H);
 742             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 743             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 744             velec            = _mm256_mul_ps(qq10,VV);
 745             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 746             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 747
 748             /* Update potential sum for this i atom from the interaction with this j atom. */
 749             velec            = _mm256_andnot_ps(dummy_mask,velec);
 750             velecsum         = _mm256_add_ps(velecsum,velec);
 751
 752             fscal            = felec;
 753
 754             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 755
 756             /* Calculate temporary vectorial force */
 757             tx               = _mm256_mul_ps(fscal,dx10);
 758             ty               = _mm256_mul_ps(fscal,dy10);
 759             tz               = _mm256_mul_ps(fscal,dz10);
 760
 761             /* Update vectorial force */
 762             fix1             = _mm256_add_ps(fix1,tx);
 763             fiy1             = _mm256_add_ps(fiy1,ty);
 764             fiz1             = _mm256_add_ps(fiz1,tz);
 765
 766             fjx0             = _mm256_add_ps(fjx0,tx);
 767             fjy0             = _mm256_add_ps(fjy0,ty);
 768             fjz0             = _mm256_add_ps(fjz0,tz);
 769
 770             /**************************
 771              * CALCULATE INTERACTIONS *
 772              **************************/
 773
 774             r20              = _mm256_mul_ps(rsq20,rinv20);
 775             r20              = _mm256_andnot_ps(dummy_mask,r20);
 776
 777             /* Compute parameters for interactions between i and j atoms */
 778             qq20             = _mm256_mul_ps(iq2,jq0);
 779
 780             /* Calculate table index by multiplying r with table scale and truncate to integer */
 781             rt               = _mm256_mul_ps(r20,vftabscale);
 782             vfitab           = _mm256_cvttps_epi32(rt);
 783             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 784             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 785             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 786             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 787             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 788             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 789
 790             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 791             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 792                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 793             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 794                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 795             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 796                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 797             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 798                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 799             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 800             Heps             = _mm256_mul_ps(vfeps,H);
 801             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 802             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 803             velec            = _mm256_mul_ps(qq20,VV);
 804             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 805             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 806
 807             /* Update potential sum for this i atom from the interaction with this j atom. */
 808             velec            = _mm256_andnot_ps(dummy_mask,velec);
 809             velecsum         = _mm256_add_ps(velecsum,velec);
 810
 811             fscal            = felec;
 812
 813             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 814
 815             /* Calculate temporary vectorial force */
 816             tx               = _mm256_mul_ps(fscal,dx20);
 817             ty               = _mm256_mul_ps(fscal,dy20);
 818             tz               = _mm256_mul_ps(fscal,dz20);
 819
 820             /* Update vectorial force */
 821             fix2             = _mm256_add_ps(fix2,tx);
 822             fiy2             = _mm256_add_ps(fiy2,ty);
 823             fiz2             = _mm256_add_ps(fiz2,tz);
 824
 825             fjx0             = _mm256_add_ps(fjx0,tx);
 826             fjy0             = _mm256_add_ps(fjy0,ty);
 827             fjz0             = _mm256_add_ps(fjz0,tz);
 828
 829             /**************************
 830              * CALCULATE INTERACTIONS *
 831              **************************/
 832
 833             r30              = _mm256_mul_ps(rsq30,rinv30);
 834             r30              = _mm256_andnot_ps(dummy_mask,r30);
 835
 836             /* Compute parameters for interactions between i and j atoms */
 837             qq30             = _mm256_mul_ps(iq3,jq0);
 838
 839             /* Calculate table index by multiplying r with table scale and truncate to integer */
 840             rt               = _mm256_mul_ps(r30,vftabscale);
 841             vfitab           = _mm256_cvttps_epi32(rt);
 842             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 843             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 844             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 845             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 846             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 847             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 848
 849             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 850             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 851                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 852             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 853                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 854             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 855                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 856             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 857                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 858             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 859             Heps             = _mm256_mul_ps(vfeps,H);
 860             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 861             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 862             velec            = _mm256_mul_ps(qq30,VV);
 863             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 864             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 865
 866             /* Update potential sum for this i atom from the interaction with this j atom. */
 867             velec            = _mm256_andnot_ps(dummy_mask,velec);
 868             velecsum         = _mm256_add_ps(velecsum,velec);
 869
 870             fscal            = felec;
 871
 872             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 873
 874             /* Calculate temporary vectorial force */
 875             tx               = _mm256_mul_ps(fscal,dx30);
 876             ty               = _mm256_mul_ps(fscal,dy30);
 877             tz               = _mm256_mul_ps(fscal,dz30);
 878
 879             /* Update vectorial force */
 880             fix3             = _mm256_add_ps(fix3,tx);
 881             fiy3             = _mm256_add_ps(fiy3,ty);
 882             fiz3             = _mm256_add_ps(fiz3,tz);
 883
 884             fjx0             = _mm256_add_ps(fjx0,tx);
 885             fjy0             = _mm256_add_ps(fjy0,ty);
 886             fjz0             = _mm256_add_ps(fjz0,tz);
 887
 888             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 889             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 890             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 891             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 892             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 893             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 894             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 895             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 896
 897             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 898
 899             /* Inner loop uses 192 flops */
 900         }
 901
 902         /* End of innermost loop */
 903
 904         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 905                                                  f+i_coord_offset,fshift+i_shift_offset);
 906
 907         ggid                        = gid[iidx];
 908         /* Update potential energies */
 909         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 910         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 911
 912         /* Increment number of inner iterations */
 913         inneriter                  += j_index_end - j_index_start;
 914
 915         /* Outer loop uses 26 flops */
 916     }
 917
 918     /* Increment number of outer iterations */
 919     outeriter        += nri;
 920
 921     /* Update outer/inner flops */
 922
 923     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*192);
 924 }
 925 /*
 926  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_single
 927  * Electrostatics interaction: CubicSplineTable
 928  * VdW interaction:            CubicSplineTable
 929  * Geometry:                   Water4-Particle
 930  * Calculate force/pot:        Force
 931  */
 932 void
 933 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_single
 934                     (t_nblist                    * gmx_restrict       nlist,
 935                      rvec                        * gmx_restrict          xx,
 936                      rvec                        * gmx_restrict          ff,
 937                      t_forcerec                  * gmx_restrict          fr,
 938                      t_mdatoms                   * gmx_restrict     mdatoms,
 939                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 940                      t_nrnb                      * gmx_restrict        nrnb)
 941 {
 942     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 943      * just 0 for non-waters.
 944      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 945      * jnr indices corresponding to data put in the four positions in the SIMD register.
 946      */
 947     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 948     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 949     int              jnrA,jnrB,jnrC,jnrD;
 950     int              jnrE,jnrF,jnrG,jnrH;
 951     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 952     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 953     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 954     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 955     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 956     real             rcutoff_scalar;
 957     real             *shiftvec,*fshift,*x,*f;
 958     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 959     real             scratch[4*DIM];
 960     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 961     real *           vdwioffsetptr0;
 962     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 963     real *           vdwioffsetptr1;
 964     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 965     real *           vdwioffsetptr2;
 966     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 967     real *           vdwioffsetptr3;
 968     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 969     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 970     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 971     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 972     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 973     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 974     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 975     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 976     real             *charge;
 977     int              nvdwtype;
 978     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 979     int              *vdwtype;
 980     real             *vdwparam;
 981     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 982     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 983     __m256i          vfitab;
 984     __m128i          vfitab_lo,vfitab_hi;
 985     __m128i          ifour       = _mm_set1_epi32(4);
 986     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 987     real             *vftab;
 988     __m256           dummy_mask,cutoff_mask;
 989     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 990     __m256           one     = _mm256_set1_ps(1.0);
 991     __m256           two     = _mm256_set1_ps(2.0);
 992     x                = xx[0];
 993     f                = ff[0];
 994
 995     nri              = nlist->nri;
 996     iinr             = nlist->iinr;
 997     jindex           = nlist->jindex;
 998     jjnr             = nlist->jjnr;
 999     shiftidx         = nlist->shift;
1000     gid              = nlist->gid;
1001     shiftvec         = fr->shift_vec[0];
1002     fshift           = fr->fshift[0];
1003     facel            = _mm256_set1_ps(fr->epsfac);
1004     charge           = mdatoms->chargeA;
1005     nvdwtype         = fr->ntype;
1006     vdwparam         = fr->nbfp;
1007     vdwtype          = mdatoms->typeA;
1008
1009     vftab            = kernel_data->table_elec_vdw->data;
1010     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
1011
1012     /* Setup water-specific parameters */
1013     inr              = nlist->iinr[0];
1014     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1015     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1016     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1017     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1018
1019     /* Avoid stupid compiler warnings */
1020     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1021     j_coord_offsetA = 0;
1022     j_coord_offsetB = 0;
1023     j_coord_offsetC = 0;
1024     j_coord_offsetD = 0;
1025     j_coord_offsetE = 0;
1026     j_coord_offsetF = 0;
1027     j_coord_offsetG = 0;
1028     j_coord_offsetH = 0;
1029
1030     outeriter        = 0;
1031     inneriter        = 0;
1032
1033     for(iidx=0;iidx<4*DIM;iidx++)
1034     {
1035         scratch[iidx] = 0.0;
1036     }
1037
1038     /* Start outer loop over neighborlists */
1039     for(iidx=0; iidx<nri; iidx++)
1040     {
1041         /* Load shift vector for this list */
1042         i_shift_offset   = DIM*shiftidx[iidx];
1043
1044         /* Load limits for loop over neighbors */
1045         j_index_start    = jindex[iidx];
1046         j_index_end      = jindex[iidx+1];
1047
1048         /* Get outer coordinate index */
1049         inr              = iinr[iidx];
1050         i_coord_offset   = DIM*inr;
1051
1052         /* Load i particle coords and add shift vector */
1053         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1054                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1055
1056         fix0             = _mm256_setzero_ps();
1057         fiy0             = _mm256_setzero_ps();
1058         fiz0             = _mm256_setzero_ps();
1059         fix1             = _mm256_setzero_ps();
1060         fiy1             = _mm256_setzero_ps();
1061         fiz1             = _mm256_setzero_ps();
1062         fix2             = _mm256_setzero_ps();
1063         fiy2             = _mm256_setzero_ps();
1064         fiz2             = _mm256_setzero_ps();
1065         fix3             = _mm256_setzero_ps();
1066         fiy3             = _mm256_setzero_ps();
1067         fiz3             = _mm256_setzero_ps();
1068
1069         /* Start inner kernel loop */
1070         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1071         {
1072
1073             /* Get j neighbor index, and coordinate index */
1074             jnrA             = jjnr[jidx];
1075             jnrB             = jjnr[jidx+1];
1076             jnrC             = jjnr[jidx+2];
1077             jnrD             = jjnr[jidx+3];
1078             jnrE             = jjnr[jidx+4];
1079             jnrF             = jjnr[jidx+5];
1080             jnrG             = jjnr[jidx+6];
1081             jnrH             = jjnr[jidx+7];
1082             j_coord_offsetA  = DIM*jnrA;
1083             j_coord_offsetB  = DIM*jnrB;
1084             j_coord_offsetC  = DIM*jnrC;
1085             j_coord_offsetD  = DIM*jnrD;
1086             j_coord_offsetE  = DIM*jnrE;
1087             j_coord_offsetF  = DIM*jnrF;
1088             j_coord_offsetG  = DIM*jnrG;
1089             j_coord_offsetH  = DIM*jnrH;
1090
1091             /* load j atom coordinates */
1092             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1093                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1094                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1095                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1096                                                  &jx0,&jy0,&jz0);
1097
1098             /* Calculate displacement vector */
1099             dx00             = _mm256_sub_ps(ix0,jx0);
1100             dy00             = _mm256_sub_ps(iy0,jy0);
1101             dz00             = _mm256_sub_ps(iz0,jz0);
1102             dx10             = _mm256_sub_ps(ix1,jx0);
1103             dy10             = _mm256_sub_ps(iy1,jy0);
1104             dz10             = _mm256_sub_ps(iz1,jz0);
1105             dx20             = _mm256_sub_ps(ix2,jx0);
1106             dy20             = _mm256_sub_ps(iy2,jy0);
1107             dz20             = _mm256_sub_ps(iz2,jz0);
1108             dx30             = _mm256_sub_ps(ix3,jx0);
1109             dy30             = _mm256_sub_ps(iy3,jy0);
1110             dz30             = _mm256_sub_ps(iz3,jz0);
1111
1112             /* Calculate squared distance and things based on it */
1113             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1114             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1115             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1116             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1117
1118             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1119             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1120             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1121             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1122
1123             /* Load parameters for j particles */
1124             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1125                                                                  charge+jnrC+0,charge+jnrD+0,
1126                                                                  charge+jnrE+0,charge+jnrF+0,
1127                                                                  charge+jnrG+0,charge+jnrH+0);
1128             vdwjidx0A        = 2*vdwtype[jnrA+0];
1129             vdwjidx0B        = 2*vdwtype[jnrB+0];
1130             vdwjidx0C        = 2*vdwtype[jnrC+0];
1131             vdwjidx0D        = 2*vdwtype[jnrD+0];
1132             vdwjidx0E        = 2*vdwtype[jnrE+0];
1133             vdwjidx0F        = 2*vdwtype[jnrF+0];
1134             vdwjidx0G        = 2*vdwtype[jnrG+0];
1135             vdwjidx0H        = 2*vdwtype[jnrH+0];
1136
1137             fjx0             = _mm256_setzero_ps();
1138             fjy0             = _mm256_setzero_ps();
1139             fjz0             = _mm256_setzero_ps();
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             r00              = _mm256_mul_ps(rsq00,rinv00);
1146
1147             /* Compute parameters for interactions between i and j atoms */
1148             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1149                                             vdwioffsetptr0+vdwjidx0B,
1150                                             vdwioffsetptr0+vdwjidx0C,
1151                                             vdwioffsetptr0+vdwjidx0D,
1152                                             vdwioffsetptr0+vdwjidx0E,
1153                                             vdwioffsetptr0+vdwjidx0F,
1154                                             vdwioffsetptr0+vdwjidx0G,
1155                                             vdwioffsetptr0+vdwjidx0H,
1156                                             &c6_00,&c12_00);
1157
1158             /* Calculate table index by multiplying r with table scale and truncate to integer */
1159             rt               = _mm256_mul_ps(r00,vftabscale);
1160             vfitab           = _mm256_cvttps_epi32(rt);
1161             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1162             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1163             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1164             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1165             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1166             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1167
1168             /* CUBIC SPLINE TABLE DISPERSION */
1169             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1170             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1171             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1172                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1173             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1174                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1175             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1176                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1177             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1178                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1179             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1180             Heps             = _mm256_mul_ps(vfeps,H);
1181             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1182             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1183             fvdw6            = _mm256_mul_ps(c6_00,FF);
1184
1185             /* CUBIC SPLINE TABLE REPULSION */
1186             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1187             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1188             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1189                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1190             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1191                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1192             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1193                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1194             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1195                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1196             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1197             Heps             = _mm256_mul_ps(vfeps,H);
1198             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1199             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1200             fvdw12           = _mm256_mul_ps(c12_00,FF);
1201             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1202
1203             fscal            = fvdw;
1204
1205             /* Calculate temporary vectorial force */
1206             tx               = _mm256_mul_ps(fscal,dx00);
1207             ty               = _mm256_mul_ps(fscal,dy00);
1208             tz               = _mm256_mul_ps(fscal,dz00);
1209
1210             /* Update vectorial force */
1211             fix0             = _mm256_add_ps(fix0,tx);
1212             fiy0             = _mm256_add_ps(fiy0,ty);
1213             fiz0             = _mm256_add_ps(fiz0,tz);
1214
1215             fjx0             = _mm256_add_ps(fjx0,tx);
1216             fjy0             = _mm256_add_ps(fjy0,ty);
1217             fjz0             = _mm256_add_ps(fjz0,tz);
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             r10              = _mm256_mul_ps(rsq10,rinv10);
1224
1225             /* Compute parameters for interactions between i and j atoms */
1226             qq10             = _mm256_mul_ps(iq1,jq0);
1227
1228             /* Calculate table index by multiplying r with table scale and truncate to integer */
1229             rt               = _mm256_mul_ps(r10,vftabscale);
1230             vfitab           = _mm256_cvttps_epi32(rt);
1231             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1232             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1233             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1234             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1235             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1236             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1237
1238             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1239             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1240                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1241             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1242                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1243             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1244                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1245             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1246                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1247             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1248             Heps             = _mm256_mul_ps(vfeps,H);
1249             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1250             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1251             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1252
1253             fscal            = felec;
1254
1255             /* Calculate temporary vectorial force */
1256             tx               = _mm256_mul_ps(fscal,dx10);
1257             ty               = _mm256_mul_ps(fscal,dy10);
1258             tz               = _mm256_mul_ps(fscal,dz10);
1259
1260             /* Update vectorial force */
1261             fix1             = _mm256_add_ps(fix1,tx);
1262             fiy1             = _mm256_add_ps(fiy1,ty);
1263             fiz1             = _mm256_add_ps(fiz1,tz);
1264
1265             fjx0             = _mm256_add_ps(fjx0,tx);
1266             fjy0             = _mm256_add_ps(fjy0,ty);
1267             fjz0             = _mm256_add_ps(fjz0,tz);
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             r20              = _mm256_mul_ps(rsq20,rinv20);
1274
1275             /* Compute parameters for interactions between i and j atoms */
1276             qq20             = _mm256_mul_ps(iq2,jq0);
1277
1278             /* Calculate table index by multiplying r with table scale and truncate to integer */
1279             rt               = _mm256_mul_ps(r20,vftabscale);
1280             vfitab           = _mm256_cvttps_epi32(rt);
1281             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1282             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1283             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1284             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1285             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1286             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1287
1288             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1289             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1291             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1293             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1295             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1297             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1298             Heps             = _mm256_mul_ps(vfeps,H);
1299             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1300             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1301             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1302
1303             fscal            = felec;
1304
1305             /* Calculate temporary vectorial force */
1306             tx               = _mm256_mul_ps(fscal,dx20);
1307             ty               = _mm256_mul_ps(fscal,dy20);
1308             tz               = _mm256_mul_ps(fscal,dz20);
1309
1310             /* Update vectorial force */
1311             fix2             = _mm256_add_ps(fix2,tx);
1312             fiy2             = _mm256_add_ps(fiy2,ty);
1313             fiz2             = _mm256_add_ps(fiz2,tz);
1314
1315             fjx0             = _mm256_add_ps(fjx0,tx);
1316             fjy0             = _mm256_add_ps(fjy0,ty);
1317             fjz0             = _mm256_add_ps(fjz0,tz);
1318
1319             /**************************
1320              * CALCULATE INTERACTIONS *
1321              **************************/
1322
1323             r30              = _mm256_mul_ps(rsq30,rinv30);
1324
1325             /* Compute parameters for interactions between i and j atoms */
1326             qq30             = _mm256_mul_ps(iq3,jq0);
1327
1328             /* Calculate table index by multiplying r with table scale and truncate to integer */
1329             rt               = _mm256_mul_ps(r30,vftabscale);
1330             vfitab           = _mm256_cvttps_epi32(rt);
1331             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1332             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1333             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1334             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1335             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1336             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1337
1338             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1339             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1340                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1341             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1342                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1343             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1344                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1345             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1346                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1347             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1348             Heps             = _mm256_mul_ps(vfeps,H);
1349             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1350             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1351             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1352
1353             fscal            = felec;
1354
1355             /* Calculate temporary vectorial force */
1356             tx               = _mm256_mul_ps(fscal,dx30);
1357             ty               = _mm256_mul_ps(fscal,dy30);
1358             tz               = _mm256_mul_ps(fscal,dz30);
1359
1360             /* Update vectorial force */
1361             fix3             = _mm256_add_ps(fix3,tx);
1362             fiy3             = _mm256_add_ps(fiy3,ty);
1363             fiz3             = _mm256_add_ps(fiz3,tz);
1364
1365             fjx0             = _mm256_add_ps(fjx0,tx);
1366             fjy0             = _mm256_add_ps(fjy0,ty);
1367             fjz0             = _mm256_add_ps(fjz0,tz);
1368
1369             fjptrA             = f+j_coord_offsetA;
1370             fjptrB             = f+j_coord_offsetB;
1371             fjptrC             = f+j_coord_offsetC;
1372             fjptrD             = f+j_coord_offsetD;
1373             fjptrE             = f+j_coord_offsetE;
1374             fjptrF             = f+j_coord_offsetF;
1375             fjptrG             = f+j_coord_offsetG;
1376             fjptrH             = f+j_coord_offsetH;
1377
1378             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1379
1380             /* Inner loop uses 168 flops */
1381         }
1382
1383         if(jidx<j_index_end)
1384         {
1385
1386             /* Get j neighbor index, and coordinate index */
1387             jnrlistA         = jjnr[jidx];
1388             jnrlistB         = jjnr[jidx+1];
1389             jnrlistC         = jjnr[jidx+2];
1390             jnrlistD         = jjnr[jidx+3];
1391             jnrlistE         = jjnr[jidx+4];
1392             jnrlistF         = jjnr[jidx+5];
1393             jnrlistG         = jjnr[jidx+6];
1394             jnrlistH         = jjnr[jidx+7];
1395             /* Sign of each element will be negative for non-real atoms.
1396              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1397              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1398              */
1399             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1400                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1401
1402             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1403             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1404             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1405             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1406             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1407             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1408             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1409             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1410             j_coord_offsetA  = DIM*jnrA;
1411             j_coord_offsetB  = DIM*jnrB;
1412             j_coord_offsetC  = DIM*jnrC;
1413             j_coord_offsetD  = DIM*jnrD;
1414             j_coord_offsetE  = DIM*jnrE;
1415             j_coord_offsetF  = DIM*jnrF;
1416             j_coord_offsetG  = DIM*jnrG;
1417             j_coord_offsetH  = DIM*jnrH;
1418
1419             /* load j atom coordinates */
1420             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1421                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1422                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1423                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1424                                                  &jx0,&jy0,&jz0);
1425
1426             /* Calculate displacement vector */
1427             dx00             = _mm256_sub_ps(ix0,jx0);
1428             dy00             = _mm256_sub_ps(iy0,jy0);
1429             dz00             = _mm256_sub_ps(iz0,jz0);
1430             dx10             = _mm256_sub_ps(ix1,jx0);
1431             dy10             = _mm256_sub_ps(iy1,jy0);
1432             dz10             = _mm256_sub_ps(iz1,jz0);
1433             dx20             = _mm256_sub_ps(ix2,jx0);
1434             dy20             = _mm256_sub_ps(iy2,jy0);
1435             dz20             = _mm256_sub_ps(iz2,jz0);
1436             dx30             = _mm256_sub_ps(ix3,jx0);
1437             dy30             = _mm256_sub_ps(iy3,jy0);
1438             dz30             = _mm256_sub_ps(iz3,jz0);
1439
1440             /* Calculate squared distance and things based on it */
1441             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1442             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1443             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1444             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1445
1446             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1447             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1448             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1449             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1450
1451             /* Load parameters for j particles */
1452             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1453                                                                  charge+jnrC+0,charge+jnrD+0,
1454                                                                  charge+jnrE+0,charge+jnrF+0,
1455                                                                  charge+jnrG+0,charge+jnrH+0);
1456             vdwjidx0A        = 2*vdwtype[jnrA+0];
1457             vdwjidx0B        = 2*vdwtype[jnrB+0];
1458             vdwjidx0C        = 2*vdwtype[jnrC+0];
1459             vdwjidx0D        = 2*vdwtype[jnrD+0];
1460             vdwjidx0E        = 2*vdwtype[jnrE+0];
1461             vdwjidx0F        = 2*vdwtype[jnrF+0];
1462             vdwjidx0G        = 2*vdwtype[jnrG+0];
1463             vdwjidx0H        = 2*vdwtype[jnrH+0];
1464
1465             fjx0             = _mm256_setzero_ps();
1466             fjy0             = _mm256_setzero_ps();
1467             fjz0             = _mm256_setzero_ps();
1468
1469             /**************************
1470              * CALCULATE INTERACTIONS *
1471              **************************/
1472
1473             r00              = _mm256_mul_ps(rsq00,rinv00);
1474             r00              = _mm256_andnot_ps(dummy_mask,r00);
1475
1476             /* Compute parameters for interactions between i and j atoms */
1477             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1478                                             vdwioffsetptr0+vdwjidx0B,
1479                                             vdwioffsetptr0+vdwjidx0C,
1480                                             vdwioffsetptr0+vdwjidx0D,
1481                                             vdwioffsetptr0+vdwjidx0E,
1482                                             vdwioffsetptr0+vdwjidx0F,
1483                                             vdwioffsetptr0+vdwjidx0G,
1484                                             vdwioffsetptr0+vdwjidx0H,
1485                                             &c6_00,&c12_00);
1486
1487             /* Calculate table index by multiplying r with table scale and truncate to integer */
1488             rt               = _mm256_mul_ps(r00,vftabscale);
1489             vfitab           = _mm256_cvttps_epi32(rt);
1490             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1491             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1492             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1493             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1494             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1495             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1496
1497             /* CUBIC SPLINE TABLE DISPERSION */
1498             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1499             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1500             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1501                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1502             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1503                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1504             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1505                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1506             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1507                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1508             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1509             Heps             = _mm256_mul_ps(vfeps,H);
1510             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1511             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1512             fvdw6            = _mm256_mul_ps(c6_00,FF);
1513
1514             /* CUBIC SPLINE TABLE REPULSION */
1515             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1516             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1517             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1518                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1519             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1520                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1521             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1522                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1523             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1524                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1525             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1526             Heps             = _mm256_mul_ps(vfeps,H);
1527             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1528             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1529             fvdw12           = _mm256_mul_ps(c12_00,FF);
1530             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1531
1532             fscal            = fvdw;
1533
1534             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1535
1536             /* Calculate temporary vectorial force */
1537             tx               = _mm256_mul_ps(fscal,dx00);
1538             ty               = _mm256_mul_ps(fscal,dy00);
1539             tz               = _mm256_mul_ps(fscal,dz00);
1540
1541             /* Update vectorial force */
1542             fix0             = _mm256_add_ps(fix0,tx);
1543             fiy0             = _mm256_add_ps(fiy0,ty);
1544             fiz0             = _mm256_add_ps(fiz0,tz);
1545
1546             fjx0             = _mm256_add_ps(fjx0,tx);
1547             fjy0             = _mm256_add_ps(fjy0,ty);
1548             fjz0             = _mm256_add_ps(fjz0,tz);
1549
1550             /**************************
1551              * CALCULATE INTERACTIONS *
1552              **************************/
1553
1554             r10              = _mm256_mul_ps(rsq10,rinv10);
1555             r10              = _mm256_andnot_ps(dummy_mask,r10);
1556
1557             /* Compute parameters for interactions between i and j atoms */
1558             qq10             = _mm256_mul_ps(iq1,jq0);
1559
1560             /* Calculate table index by multiplying r with table scale and truncate to integer */
1561             rt               = _mm256_mul_ps(r10,vftabscale);
1562             vfitab           = _mm256_cvttps_epi32(rt);
1563             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1564             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1565             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1566             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1567             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1568             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1569
1570             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1571             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1572                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1573             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1574                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1575             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1576                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1577             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1578                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1579             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1580             Heps             = _mm256_mul_ps(vfeps,H);
1581             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1582             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1583             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1584
1585             fscal            = felec;
1586
1587             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1588
1589             /* Calculate temporary vectorial force */
1590             tx               = _mm256_mul_ps(fscal,dx10);
1591             ty               = _mm256_mul_ps(fscal,dy10);
1592             tz               = _mm256_mul_ps(fscal,dz10);
1593
1594             /* Update vectorial force */
1595             fix1             = _mm256_add_ps(fix1,tx);
1596             fiy1             = _mm256_add_ps(fiy1,ty);
1597             fiz1             = _mm256_add_ps(fiz1,tz);
1598
1599             fjx0             = _mm256_add_ps(fjx0,tx);
1600             fjy0             = _mm256_add_ps(fjy0,ty);
1601             fjz0             = _mm256_add_ps(fjz0,tz);
1602
1603             /**************************
1604              * CALCULATE INTERACTIONS *
1605              **************************/
1606
1607             r20              = _mm256_mul_ps(rsq20,rinv20);
1608             r20              = _mm256_andnot_ps(dummy_mask,r20);
1609
1610             /* Compute parameters for interactions between i and j atoms */
1611             qq20             = _mm256_mul_ps(iq2,jq0);
1612
1613             /* Calculate table index by multiplying r with table scale and truncate to integer */
1614             rt               = _mm256_mul_ps(r20,vftabscale);
1615             vfitab           = _mm256_cvttps_epi32(rt);
1616             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1617             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1618             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1619             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1620             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1621             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1622
1623             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1624             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1625                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1626             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1627                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1628             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1629                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1630             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1631                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1632             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1633             Heps             = _mm256_mul_ps(vfeps,H);
1634             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1635             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1636             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1637
1638             fscal            = felec;
1639
1640             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1641
1642             /* Calculate temporary vectorial force */
1643             tx               = _mm256_mul_ps(fscal,dx20);
1644             ty               = _mm256_mul_ps(fscal,dy20);
1645             tz               = _mm256_mul_ps(fscal,dz20);
1646
1647             /* Update vectorial force */
1648             fix2             = _mm256_add_ps(fix2,tx);
1649             fiy2             = _mm256_add_ps(fiy2,ty);
1650             fiz2             = _mm256_add_ps(fiz2,tz);
1651
1652             fjx0             = _mm256_add_ps(fjx0,tx);
1653             fjy0             = _mm256_add_ps(fjy0,ty);
1654             fjz0             = _mm256_add_ps(fjz0,tz);
1655
1656             /**************************
1657              * CALCULATE INTERACTIONS *
1658              **************************/
1659
1660             r30              = _mm256_mul_ps(rsq30,rinv30);
1661             r30              = _mm256_andnot_ps(dummy_mask,r30);
1662
1663             /* Compute parameters for interactions between i and j atoms */
1664             qq30             = _mm256_mul_ps(iq3,jq0);
1665
1666             /* Calculate table index by multiplying r with table scale and truncate to integer */
1667             rt               = _mm256_mul_ps(r30,vftabscale);
1668             vfitab           = _mm256_cvttps_epi32(rt);
1669             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1670             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1671             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1672             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1673             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1674             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1675
1676             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1677             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1678                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1679             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1680                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1681             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1682                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1683             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1684                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1685             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1686             Heps             = _mm256_mul_ps(vfeps,H);
1687             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1688             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1689             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1690
1691             fscal            = felec;
1692
1693             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1694
1695             /* Calculate temporary vectorial force */
1696             tx               = _mm256_mul_ps(fscal,dx30);
1697             ty               = _mm256_mul_ps(fscal,dy30);
1698             tz               = _mm256_mul_ps(fscal,dz30);
1699
1700             /* Update vectorial force */
1701             fix3             = _mm256_add_ps(fix3,tx);
1702             fiy3             = _mm256_add_ps(fiy3,ty);
1703             fiz3             = _mm256_add_ps(fiz3,tz);
1704
1705             fjx0             = _mm256_add_ps(fjx0,tx);
1706             fjy0             = _mm256_add_ps(fjy0,ty);
1707             fjz0             = _mm256_add_ps(fjz0,tz);
1708
1709             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1710             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1711             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1712             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1713             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1714             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1715             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1716             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1717
1718             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1719
1720             /* Inner loop uses 172 flops */
1721         }
1722
1723         /* End of innermost loop */
1724
1725         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1726                                                  f+i_coord_offset,fshift+i_shift_offset);
1727
1728         /* Increment number of inner iterations */
1729         inneriter                  += j_index_end - j_index_start;
1730
1731         /* Outer loop uses 24 flops */
1732     }
1733
1734     /* Increment number of outer iterations */
1735     outeriter        += nri;
1736
1737     /* Update outer/inner flops */
1738
1739     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*172);
1740 }