src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  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           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_elec_vdw->data;
 134     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+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     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149     j_coord_offsetE = 0;
 150     j_coord_offsetF = 0;
 151     j_coord_offsetG = 0;
 152     j_coord_offsetH = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183         fix1             = _mm256_setzero_ps();
 184         fiy1             = _mm256_setzero_ps();
 185         fiz1             = _mm256_setzero_ps();
 186         fix2             = _mm256_setzero_ps();
 187         fiy2             = _mm256_setzero_ps();
 188         fiz2             = _mm256_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_ps();
 192         vvdwsum          = _mm256_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             jnrE             = jjnr[jidx+4];
 204             jnrF             = jjnr[jidx+5];
 205             jnrG             = jjnr[jidx+6];
 206             jnrH             = jjnr[jidx+7];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211             j_coord_offsetE  = DIM*jnrE;
 212             j_coord_offsetF  = DIM*jnrF;
 213             j_coord_offsetG  = DIM*jnrG;
 214             j_coord_offsetH  = DIM*jnrH;
 215
 216             /* load j atom coordinates */
 217             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 219                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 220                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 221                                                  &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm256_sub_ps(ix0,jx0);
 225             dy00             = _mm256_sub_ps(iy0,jy0);
 226             dz00             = _mm256_sub_ps(iz0,jz0);
 227             dx10             = _mm256_sub_ps(ix1,jx0);
 228             dy10             = _mm256_sub_ps(iy1,jy0);
 229             dz10             = _mm256_sub_ps(iz1,jz0);
 230             dx20             = _mm256_sub_ps(ix2,jx0);
 231             dy20             = _mm256_sub_ps(iy2,jy0);
 232             dz20             = _mm256_sub_ps(iz2,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 236             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 237             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 238
 239             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 240             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 241             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 245                                                                  charge+jnrC+0,charge+jnrD+0,
 246                                                                  charge+jnrE+0,charge+jnrF+0,
 247                                                                  charge+jnrG+0,charge+jnrH+0);
 248             vdwjidx0A        = 2*vdwtype[jnrA+0];
 249             vdwjidx0B        = 2*vdwtype[jnrB+0];
 250             vdwjidx0C        = 2*vdwtype[jnrC+0];
 251             vdwjidx0D        = 2*vdwtype[jnrD+0];
 252             vdwjidx0E        = 2*vdwtype[jnrE+0];
 253             vdwjidx0F        = 2*vdwtype[jnrF+0];
 254             vdwjidx0G        = 2*vdwtype[jnrG+0];
 255             vdwjidx0H        = 2*vdwtype[jnrH+0];
 256
 257             fjx0             = _mm256_setzero_ps();
 258             fjy0             = _mm256_setzero_ps();
 259             fjz0             = _mm256_setzero_ps();
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             r00              = _mm256_mul_ps(rsq00,rinv00);
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq00             = _mm256_mul_ps(iq0,jq0);
 269             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 270                                             vdwioffsetptr0+vdwjidx0B,
 271                                             vdwioffsetptr0+vdwjidx0C,
 272                                             vdwioffsetptr0+vdwjidx0D,
 273                                             vdwioffsetptr0+vdwjidx0E,
 274                                             vdwioffsetptr0+vdwjidx0F,
 275                                             vdwioffsetptr0+vdwjidx0G,
 276                                             vdwioffsetptr0+vdwjidx0H,
 277                                             &c6_00,&c12_00);
 278
 279             /* Calculate table index by multiplying r with table scale and truncate to integer */
 280             rt               = _mm256_mul_ps(r00,vftabscale);
 281             vfitab           = _mm256_cvttps_epi32(rt);
 282             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 283             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 284             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 285             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 286             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 287             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 288
 289             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 290             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 292             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 294             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 296             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 298             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 299             Heps             = _mm256_mul_ps(vfeps,H);
 300             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 301             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 302             velec            = _mm256_mul_ps(qq00,VV);
 303             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 304             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 305
 306             /* CUBIC SPLINE TABLE DISPERSION */
 307             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 308             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 309             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 311             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 313             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 314                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 315             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 316                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 317             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 318             Heps             = _mm256_mul_ps(vfeps,H);
 319             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 320             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 321             vvdw6            = _mm256_mul_ps(c6_00,VV);
 322             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 323             fvdw6            = _mm256_mul_ps(c6_00,FF);
 324
 325             /* CUBIC SPLINE TABLE REPULSION */
 326             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 327             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 328             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 330             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 332             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 334             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 335                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 336             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 337             Heps             = _mm256_mul_ps(vfeps,H);
 338             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 339             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 340             vvdw12           = _mm256_mul_ps(c12_00,VV);
 341             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 342             fvdw12           = _mm256_mul_ps(c12_00,FF);
 343             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 344             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm256_add_ps(velecsum,velec);
 348             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 349
 350             fscal            = _mm256_add_ps(felec,fvdw);
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_ps(fscal,dx00);
 354             ty               = _mm256_mul_ps(fscal,dy00);
 355             tz               = _mm256_mul_ps(fscal,dz00);
 356
 357             /* Update vectorial force */
 358             fix0             = _mm256_add_ps(fix0,tx);
 359             fiy0             = _mm256_add_ps(fiy0,ty);
 360             fiz0             = _mm256_add_ps(fiz0,tz);
 361
 362             fjx0             = _mm256_add_ps(fjx0,tx);
 363             fjy0             = _mm256_add_ps(fjy0,ty);
 364             fjz0             = _mm256_add_ps(fjz0,tz);
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             r10              = _mm256_mul_ps(rsq10,rinv10);
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq10             = _mm256_mul_ps(iq1,jq0);
 374
 375             /* Calculate table index by multiplying r with table scale and truncate to integer */
 376             rt               = _mm256_mul_ps(r10,vftabscale);
 377             vfitab           = _mm256_cvttps_epi32(rt);
 378             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 379             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 380             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 381             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 382             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 383             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 384
 385             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 386             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 388             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 390             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 392             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 394             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 395             Heps             = _mm256_mul_ps(vfeps,H);
 396             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 397             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 398             velec            = _mm256_mul_ps(qq10,VV);
 399             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 400             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velecsum         = _mm256_add_ps(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             /* Calculate temporary vectorial force */
 408             tx               = _mm256_mul_ps(fscal,dx10);
 409             ty               = _mm256_mul_ps(fscal,dy10);
 410             tz               = _mm256_mul_ps(fscal,dz10);
 411
 412             /* Update vectorial force */
 413             fix1             = _mm256_add_ps(fix1,tx);
 414             fiy1             = _mm256_add_ps(fiy1,ty);
 415             fiz1             = _mm256_add_ps(fiz1,tz);
 416
 417             fjx0             = _mm256_add_ps(fjx0,tx);
 418             fjy0             = _mm256_add_ps(fjy0,ty);
 419             fjz0             = _mm256_add_ps(fjz0,tz);
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             r20              = _mm256_mul_ps(rsq20,rinv20);
 426
 427             /* Compute parameters for interactions between i and j atoms */
 428             qq20             = _mm256_mul_ps(iq2,jq0);
 429
 430             /* Calculate table index by multiplying r with table scale and truncate to integer */
 431             rt               = _mm256_mul_ps(r20,vftabscale);
 432             vfitab           = _mm256_cvttps_epi32(rt);
 433             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 434             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 435             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 436             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 437             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 438             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 439
 440             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 441             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 442                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 443             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 444                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 445             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 446                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 447             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 448                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 449             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 450             Heps             = _mm256_mul_ps(vfeps,H);
 451             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 452             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 453             velec            = _mm256_mul_ps(qq20,VV);
 454             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 455             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velecsum         = _mm256_add_ps(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             /* Calculate temporary vectorial force */
 463             tx               = _mm256_mul_ps(fscal,dx20);
 464             ty               = _mm256_mul_ps(fscal,dy20);
 465             tz               = _mm256_mul_ps(fscal,dz20);
 466
 467             /* Update vectorial force */
 468             fix2             = _mm256_add_ps(fix2,tx);
 469             fiy2             = _mm256_add_ps(fiy2,ty);
 470             fiz2             = _mm256_add_ps(fiz2,tz);
 471
 472             fjx0             = _mm256_add_ps(fjx0,tx);
 473             fjy0             = _mm256_add_ps(fjy0,ty);
 474             fjz0             = _mm256_add_ps(fjz0,tz);
 475
 476             fjptrA             = f+j_coord_offsetA;
 477             fjptrB             = f+j_coord_offsetB;
 478             fjptrC             = f+j_coord_offsetC;
 479             fjptrD             = f+j_coord_offsetD;
 480             fjptrE             = f+j_coord_offsetE;
 481             fjptrF             = f+j_coord_offsetF;
 482             fjptrG             = f+j_coord_offsetG;
 483             fjptrH             = f+j_coord_offsetH;
 484
 485             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 486
 487             /* Inner loop uses 162 flops */
 488         }
 489
 490         if(jidx<j_index_end)
 491         {
 492
 493             /* Get j neighbor index, and coordinate index */
 494             jnrlistA         = jjnr[jidx];
 495             jnrlistB         = jjnr[jidx+1];
 496             jnrlistC         = jjnr[jidx+2];
 497             jnrlistD         = jjnr[jidx+3];
 498             jnrlistE         = jjnr[jidx+4];
 499             jnrlistF         = jjnr[jidx+5];
 500             jnrlistG         = jjnr[jidx+6];
 501             jnrlistH         = jjnr[jidx+7];
 502             /* Sign of each element will be negative for non-real atoms.
 503              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 504              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 505              */
 506             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 507                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 508
 509             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 510             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 511             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 512             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 513             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 514             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 515             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 516             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 517             j_coord_offsetA  = DIM*jnrA;
 518             j_coord_offsetB  = DIM*jnrB;
 519             j_coord_offsetC  = DIM*jnrC;
 520             j_coord_offsetD  = DIM*jnrD;
 521             j_coord_offsetE  = DIM*jnrE;
 522             j_coord_offsetF  = DIM*jnrF;
 523             j_coord_offsetG  = DIM*jnrG;
 524             j_coord_offsetH  = DIM*jnrH;
 525
 526             /* load j atom coordinates */
 527             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 528                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 529                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 530                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 531                                                  &jx0,&jy0,&jz0);
 532
 533             /* Calculate displacement vector */
 534             dx00             = _mm256_sub_ps(ix0,jx0);
 535             dy00             = _mm256_sub_ps(iy0,jy0);
 536             dz00             = _mm256_sub_ps(iz0,jz0);
 537             dx10             = _mm256_sub_ps(ix1,jx0);
 538             dy10             = _mm256_sub_ps(iy1,jy0);
 539             dz10             = _mm256_sub_ps(iz1,jz0);
 540             dx20             = _mm256_sub_ps(ix2,jx0);
 541             dy20             = _mm256_sub_ps(iy2,jy0);
 542             dz20             = _mm256_sub_ps(iz2,jz0);
 543
 544             /* Calculate squared distance and things based on it */
 545             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 546             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 547             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 548
 549             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 550             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 551             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 552
 553             /* Load parameters for j particles */
 554             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 555                                                                  charge+jnrC+0,charge+jnrD+0,
 556                                                                  charge+jnrE+0,charge+jnrF+0,
 557                                                                  charge+jnrG+0,charge+jnrH+0);
 558             vdwjidx0A        = 2*vdwtype[jnrA+0];
 559             vdwjidx0B        = 2*vdwtype[jnrB+0];
 560             vdwjidx0C        = 2*vdwtype[jnrC+0];
 561             vdwjidx0D        = 2*vdwtype[jnrD+0];
 562             vdwjidx0E        = 2*vdwtype[jnrE+0];
 563             vdwjidx0F        = 2*vdwtype[jnrF+0];
 564             vdwjidx0G        = 2*vdwtype[jnrG+0];
 565             vdwjidx0H        = 2*vdwtype[jnrH+0];
 566
 567             fjx0             = _mm256_setzero_ps();
 568             fjy0             = _mm256_setzero_ps();
 569             fjz0             = _mm256_setzero_ps();
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             r00              = _mm256_mul_ps(rsq00,rinv00);
 576             r00              = _mm256_andnot_ps(dummy_mask,r00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq00             = _mm256_mul_ps(iq0,jq0);
 580             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 581                                             vdwioffsetptr0+vdwjidx0B,
 582                                             vdwioffsetptr0+vdwjidx0C,
 583                                             vdwioffsetptr0+vdwjidx0D,
 584                                             vdwioffsetptr0+vdwjidx0E,
 585                                             vdwioffsetptr0+vdwjidx0F,
 586                                             vdwioffsetptr0+vdwjidx0G,
 587                                             vdwioffsetptr0+vdwjidx0H,
 588                                             &c6_00,&c12_00);
 589
 590             /* Calculate table index by multiplying r with table scale and truncate to integer */
 591             rt               = _mm256_mul_ps(r00,vftabscale);
 592             vfitab           = _mm256_cvttps_epi32(rt);
 593             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 594             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 595             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 596             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 597             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 598             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 599
 600             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 601             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 602                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 603             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 604                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 605             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 606                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 607             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 608                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 609             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 610             Heps             = _mm256_mul_ps(vfeps,H);
 611             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 612             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 613             velec            = _mm256_mul_ps(qq00,VV);
 614             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 615             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 616
 617             /* CUBIC SPLINE TABLE DISPERSION */
 618             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 619             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 620             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 621                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 622             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 623                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 624             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 625                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 626             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 627                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 628             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 629             Heps             = _mm256_mul_ps(vfeps,H);
 630             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 631             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 632             vvdw6            = _mm256_mul_ps(c6_00,VV);
 633             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 634             fvdw6            = _mm256_mul_ps(c6_00,FF);
 635
 636             /* CUBIC SPLINE TABLE REPULSION */
 637             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 638             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 639             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 640                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 641             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 642                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 643             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 644                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 645             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 646                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 647             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 648             Heps             = _mm256_mul_ps(vfeps,H);
 649             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 650             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 651             vvdw12           = _mm256_mul_ps(c12_00,VV);
 652             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 653             fvdw12           = _mm256_mul_ps(c12_00,FF);
 654             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 655             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 656
 657             /* Update potential sum for this i atom from the interaction with this j atom. */
 658             velec            = _mm256_andnot_ps(dummy_mask,velec);
 659             velecsum         = _mm256_add_ps(velecsum,velec);
 660             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 661             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 662
 663             fscal            = _mm256_add_ps(felec,fvdw);
 664
 665             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 666
 667             /* Calculate temporary vectorial force */
 668             tx               = _mm256_mul_ps(fscal,dx00);
 669             ty               = _mm256_mul_ps(fscal,dy00);
 670             tz               = _mm256_mul_ps(fscal,dz00);
 671
 672             /* Update vectorial force */
 673             fix0             = _mm256_add_ps(fix0,tx);
 674             fiy0             = _mm256_add_ps(fiy0,ty);
 675             fiz0             = _mm256_add_ps(fiz0,tz);
 676
 677             fjx0             = _mm256_add_ps(fjx0,tx);
 678             fjy0             = _mm256_add_ps(fjy0,ty);
 679             fjz0             = _mm256_add_ps(fjz0,tz);
 680
 681             /**************************
 682              * CALCULATE INTERACTIONS *
 683              **************************/
 684
 685             r10              = _mm256_mul_ps(rsq10,rinv10);
 686             r10              = _mm256_andnot_ps(dummy_mask,r10);
 687
 688             /* Compute parameters for interactions between i and j atoms */
 689             qq10             = _mm256_mul_ps(iq1,jq0);
 690
 691             /* Calculate table index by multiplying r with table scale and truncate to integer */
 692             rt               = _mm256_mul_ps(r10,vftabscale);
 693             vfitab           = _mm256_cvttps_epi32(rt);
 694             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 695             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 696             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 697             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 698             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 699             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 700
 701             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 702             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 703                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 704             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 705                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 706             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 707                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 708             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 709                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 710             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 711             Heps             = _mm256_mul_ps(vfeps,H);
 712             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 713             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 714             velec            = _mm256_mul_ps(qq10,VV);
 715             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 716             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 717
 718             /* Update potential sum for this i atom from the interaction with this j atom. */
 719             velec            = _mm256_andnot_ps(dummy_mask,velec);
 720             velecsum         = _mm256_add_ps(velecsum,velec);
 721
 722             fscal            = felec;
 723
 724             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 725
 726             /* Calculate temporary vectorial force */
 727             tx               = _mm256_mul_ps(fscal,dx10);
 728             ty               = _mm256_mul_ps(fscal,dy10);
 729             tz               = _mm256_mul_ps(fscal,dz10);
 730
 731             /* Update vectorial force */
 732             fix1             = _mm256_add_ps(fix1,tx);
 733             fiy1             = _mm256_add_ps(fiy1,ty);
 734             fiz1             = _mm256_add_ps(fiz1,tz);
 735
 736             fjx0             = _mm256_add_ps(fjx0,tx);
 737             fjy0             = _mm256_add_ps(fjy0,ty);
 738             fjz0             = _mm256_add_ps(fjz0,tz);
 739
 740             /**************************
 741              * CALCULATE INTERACTIONS *
 742              **************************/
 743
 744             r20              = _mm256_mul_ps(rsq20,rinv20);
 745             r20              = _mm256_andnot_ps(dummy_mask,r20);
 746
 747             /* Compute parameters for interactions between i and j atoms */
 748             qq20             = _mm256_mul_ps(iq2,jq0);
 749
 750             /* Calculate table index by multiplying r with table scale and truncate to integer */
 751             rt               = _mm256_mul_ps(r20,vftabscale);
 752             vfitab           = _mm256_cvttps_epi32(rt);
 753             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 754             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 755             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 756             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 757             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 758             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 759
 760             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 761             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 762                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 763             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 764                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 765             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 766                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 767             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 768                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 769             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 770             Heps             = _mm256_mul_ps(vfeps,H);
 771             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 772             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 773             velec            = _mm256_mul_ps(qq20,VV);
 774             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 775             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 776
 777             /* Update potential sum for this i atom from the interaction with this j atom. */
 778             velec            = _mm256_andnot_ps(dummy_mask,velec);
 779             velecsum         = _mm256_add_ps(velecsum,velec);
 780
 781             fscal            = felec;
 782
 783             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 784
 785             /* Calculate temporary vectorial force */
 786             tx               = _mm256_mul_ps(fscal,dx20);
 787             ty               = _mm256_mul_ps(fscal,dy20);
 788             tz               = _mm256_mul_ps(fscal,dz20);
 789
 790             /* Update vectorial force */
 791             fix2             = _mm256_add_ps(fix2,tx);
 792             fiy2             = _mm256_add_ps(fiy2,ty);
 793             fiz2             = _mm256_add_ps(fiz2,tz);
 794
 795             fjx0             = _mm256_add_ps(fjx0,tx);
 796             fjy0             = _mm256_add_ps(fjy0,ty);
 797             fjz0             = _mm256_add_ps(fjz0,tz);
 798
 799             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 800             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 801             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 802             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 803             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 804             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 805             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 806             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 807
 808             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 809
 810             /* Inner loop uses 165 flops */
 811         }
 812
 813         /* End of innermost loop */
 814
 815         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 816                                                  f+i_coord_offset,fshift+i_shift_offset);
 817
 818         ggid                        = gid[iidx];
 819         /* Update potential energies */
 820         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 821         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 822
 823         /* Increment number of inner iterations */
 824         inneriter                  += j_index_end - j_index_start;
 825
 826         /* Outer loop uses 20 flops */
 827     }
 828
 829     /* Increment number of outer iterations */
 830     outeriter        += nri;
 831
 832     /* Update outer/inner flops */
 833
 834     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*165);
 835 }
 836 /*
 837  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_single
 838  * Electrostatics interaction: CubicSplineTable
 839  * VdW interaction:            CubicSplineTable
 840  * Geometry:                   Water3-Particle
 841  * Calculate force/pot:        Force
 842  */
 843 void
 844 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_single
 845                     (t_nblist                    * gmx_restrict       nlist,
 846                      rvec                        * gmx_restrict          xx,
 847                      rvec                        * gmx_restrict          ff,
 848                      t_forcerec                  * gmx_restrict          fr,
 849                      t_mdatoms                   * gmx_restrict     mdatoms,
 850                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 851                      t_nrnb                      * gmx_restrict        nrnb)
 852 {
 853     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 854      * just 0 for non-waters.
 855      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 856      * jnr indices corresponding to data put in the four positions in the SIMD register.
 857      */
 858     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 859     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 860     int              jnrA,jnrB,jnrC,jnrD;
 861     int              jnrE,jnrF,jnrG,jnrH;
 862     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 863     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 864     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 865     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 866     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 867     real             rcutoff_scalar;
 868     real             *shiftvec,*fshift,*x,*f;
 869     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 870     real             scratch[4*DIM];
 871     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 872     real *           vdwioffsetptr0;
 873     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 874     real *           vdwioffsetptr1;
 875     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 876     real *           vdwioffsetptr2;
 877     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 878     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 879     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 880     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 881     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 882     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 883     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 884     real             *charge;
 885     int              nvdwtype;
 886     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 887     int              *vdwtype;
 888     real             *vdwparam;
 889     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 890     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 891     __m256i          vfitab;
 892     __m128i          vfitab_lo,vfitab_hi;
 893     __m128i          ifour       = _mm_set1_epi32(4);
 894     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 895     real             *vftab;
 896     __m256           dummy_mask,cutoff_mask;
 897     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 898     __m256           one     = _mm256_set1_ps(1.0);
 899     __m256           two     = _mm256_set1_ps(2.0);
 900     x                = xx[0];
 901     f                = ff[0];
 902
 903     nri              = nlist->nri;
 904     iinr             = nlist->iinr;
 905     jindex           = nlist->jindex;
 906     jjnr             = nlist->jjnr;
 907     shiftidx         = nlist->shift;
 908     gid              = nlist->gid;
 909     shiftvec         = fr->shift_vec[0];
 910     fshift           = fr->fshift[0];
 911     facel            = _mm256_set1_ps(fr->epsfac);
 912     charge           = mdatoms->chargeA;
 913     nvdwtype         = fr->ntype;
 914     vdwparam         = fr->nbfp;
 915     vdwtype          = mdatoms->typeA;
 916
 917     vftab            = kernel_data->table_elec_vdw->data;
 918     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 919
 920     /* Setup water-specific parameters */
 921     inr              = nlist->iinr[0];
 922     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 923     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 924     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 925     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 926
 927     /* Avoid stupid compiler warnings */
 928     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 929     j_coord_offsetA = 0;
 930     j_coord_offsetB = 0;
 931     j_coord_offsetC = 0;
 932     j_coord_offsetD = 0;
 933     j_coord_offsetE = 0;
 934     j_coord_offsetF = 0;
 935     j_coord_offsetG = 0;
 936     j_coord_offsetH = 0;
 937
 938     outeriter        = 0;
 939     inneriter        = 0;
 940
 941     for(iidx=0;iidx<4*DIM;iidx++)
 942     {
 943         scratch[iidx] = 0.0;
 944     }
 945
 946     /* Start outer loop over neighborlists */
 947     for(iidx=0; iidx<nri; iidx++)
 948     {
 949         /* Load shift vector for this list */
 950         i_shift_offset   = DIM*shiftidx[iidx];
 951
 952         /* Load limits for loop over neighbors */
 953         j_index_start    = jindex[iidx];
 954         j_index_end      = jindex[iidx+1];
 955
 956         /* Get outer coordinate index */
 957         inr              = iinr[iidx];
 958         i_coord_offset   = DIM*inr;
 959
 960         /* Load i particle coords and add shift vector */
 961         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 962                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 963
 964         fix0             = _mm256_setzero_ps();
 965         fiy0             = _mm256_setzero_ps();
 966         fiz0             = _mm256_setzero_ps();
 967         fix1             = _mm256_setzero_ps();
 968         fiy1             = _mm256_setzero_ps();
 969         fiz1             = _mm256_setzero_ps();
 970         fix2             = _mm256_setzero_ps();
 971         fiy2             = _mm256_setzero_ps();
 972         fiz2             = _mm256_setzero_ps();
 973
 974         /* Start inner kernel loop */
 975         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 976         {
 977
 978             /* Get j neighbor index, and coordinate index */
 979             jnrA             = jjnr[jidx];
 980             jnrB             = jjnr[jidx+1];
 981             jnrC             = jjnr[jidx+2];
 982             jnrD             = jjnr[jidx+3];
 983             jnrE             = jjnr[jidx+4];
 984             jnrF             = jjnr[jidx+5];
 985             jnrG             = jjnr[jidx+6];
 986             jnrH             = jjnr[jidx+7];
 987             j_coord_offsetA  = DIM*jnrA;
 988             j_coord_offsetB  = DIM*jnrB;
 989             j_coord_offsetC  = DIM*jnrC;
 990             j_coord_offsetD  = DIM*jnrD;
 991             j_coord_offsetE  = DIM*jnrE;
 992             j_coord_offsetF  = DIM*jnrF;
 993             j_coord_offsetG  = DIM*jnrG;
 994             j_coord_offsetH  = DIM*jnrH;
 995
 996             /* load j atom coordinates */
 997             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 998                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 999                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1000                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1001                                                  &jx0,&jy0,&jz0);
1002
1003             /* Calculate displacement vector */
1004             dx00             = _mm256_sub_ps(ix0,jx0);
1005             dy00             = _mm256_sub_ps(iy0,jy0);
1006             dz00             = _mm256_sub_ps(iz0,jz0);
1007             dx10             = _mm256_sub_ps(ix1,jx0);
1008             dy10             = _mm256_sub_ps(iy1,jy0);
1009             dz10             = _mm256_sub_ps(iz1,jz0);
1010             dx20             = _mm256_sub_ps(ix2,jx0);
1011             dy20             = _mm256_sub_ps(iy2,jy0);
1012             dz20             = _mm256_sub_ps(iz2,jz0);
1013
1014             /* Calculate squared distance and things based on it */
1015             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1016             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1017             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1018
1019             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1020             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1021             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1022
1023             /* Load parameters for j particles */
1024             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1025                                                                  charge+jnrC+0,charge+jnrD+0,
1026                                                                  charge+jnrE+0,charge+jnrF+0,
1027                                                                  charge+jnrG+0,charge+jnrH+0);
1028             vdwjidx0A        = 2*vdwtype[jnrA+0];
1029             vdwjidx0B        = 2*vdwtype[jnrB+0];
1030             vdwjidx0C        = 2*vdwtype[jnrC+0];
1031             vdwjidx0D        = 2*vdwtype[jnrD+0];
1032             vdwjidx0E        = 2*vdwtype[jnrE+0];
1033             vdwjidx0F        = 2*vdwtype[jnrF+0];
1034             vdwjidx0G        = 2*vdwtype[jnrG+0];
1035             vdwjidx0H        = 2*vdwtype[jnrH+0];
1036
1037             fjx0             = _mm256_setzero_ps();
1038             fjy0             = _mm256_setzero_ps();
1039             fjz0             = _mm256_setzero_ps();
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             r00              = _mm256_mul_ps(rsq00,rinv00);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq00             = _mm256_mul_ps(iq0,jq0);
1049             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1050                                             vdwioffsetptr0+vdwjidx0B,
1051                                             vdwioffsetptr0+vdwjidx0C,
1052                                             vdwioffsetptr0+vdwjidx0D,
1053                                             vdwioffsetptr0+vdwjidx0E,
1054                                             vdwioffsetptr0+vdwjidx0F,
1055                                             vdwioffsetptr0+vdwjidx0G,
1056                                             vdwioffsetptr0+vdwjidx0H,
1057                                             &c6_00,&c12_00);
1058
1059             /* Calculate table index by multiplying r with table scale and truncate to integer */
1060             rt               = _mm256_mul_ps(r00,vftabscale);
1061             vfitab           = _mm256_cvttps_epi32(rt);
1062             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1063             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1064             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1065             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1066             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1067             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1068
1069             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1070             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1071                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1072             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1073                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1074             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1075                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1076             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1077                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1078             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1079             Heps             = _mm256_mul_ps(vfeps,H);
1080             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1081             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1082             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1083
1084             /* CUBIC SPLINE TABLE DISPERSION */
1085             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1086             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1087             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1088                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1089             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1090                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1091             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1092                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1093             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1094                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1095             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1096             Heps             = _mm256_mul_ps(vfeps,H);
1097             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1098             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1099             fvdw6            = _mm256_mul_ps(c6_00,FF);
1100
1101             /* CUBIC SPLINE TABLE REPULSION */
1102             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1103             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1104             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1105                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1106             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1107                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1108             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1109                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1110             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1111                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1112             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1113             Heps             = _mm256_mul_ps(vfeps,H);
1114             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1115             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1116             fvdw12           = _mm256_mul_ps(c12_00,FF);
1117             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1118
1119             fscal            = _mm256_add_ps(felec,fvdw);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm256_mul_ps(fscal,dx00);
1123             ty               = _mm256_mul_ps(fscal,dy00);
1124             tz               = _mm256_mul_ps(fscal,dz00);
1125
1126             /* Update vectorial force */
1127             fix0             = _mm256_add_ps(fix0,tx);
1128             fiy0             = _mm256_add_ps(fiy0,ty);
1129             fiz0             = _mm256_add_ps(fiz0,tz);
1130
1131             fjx0             = _mm256_add_ps(fjx0,tx);
1132             fjy0             = _mm256_add_ps(fjy0,ty);
1133             fjz0             = _mm256_add_ps(fjz0,tz);
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             r10              = _mm256_mul_ps(rsq10,rinv10);
1140
1141             /* Compute parameters for interactions between i and j atoms */
1142             qq10             = _mm256_mul_ps(iq1,jq0);
1143
1144             /* Calculate table index by multiplying r with table scale and truncate to integer */
1145             rt               = _mm256_mul_ps(r10,vftabscale);
1146             vfitab           = _mm256_cvttps_epi32(rt);
1147             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1148             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1149             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1150             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1151             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1152             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1153
1154             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1155             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1156                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1157             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1158                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1159             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1160                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1161             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1162                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1163             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1164             Heps             = _mm256_mul_ps(vfeps,H);
1165             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1166             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1167             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1168
1169             fscal            = felec;
1170
1171             /* Calculate temporary vectorial force */
1172             tx               = _mm256_mul_ps(fscal,dx10);
1173             ty               = _mm256_mul_ps(fscal,dy10);
1174             tz               = _mm256_mul_ps(fscal,dz10);
1175
1176             /* Update vectorial force */
1177             fix1             = _mm256_add_ps(fix1,tx);
1178             fiy1             = _mm256_add_ps(fiy1,ty);
1179             fiz1             = _mm256_add_ps(fiz1,tz);
1180
1181             fjx0             = _mm256_add_ps(fjx0,tx);
1182             fjy0             = _mm256_add_ps(fjy0,ty);
1183             fjz0             = _mm256_add_ps(fjz0,tz);
1184
1185             /**************************
1186              * CALCULATE INTERACTIONS *
1187              **************************/
1188
1189             r20              = _mm256_mul_ps(rsq20,rinv20);
1190
1191             /* Compute parameters for interactions between i and j atoms */
1192             qq20             = _mm256_mul_ps(iq2,jq0);
1193
1194             /* Calculate table index by multiplying r with table scale and truncate to integer */
1195             rt               = _mm256_mul_ps(r20,vftabscale);
1196             vfitab           = _mm256_cvttps_epi32(rt);
1197             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1198             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1199             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1200             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1201             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1202             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1203
1204             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1205             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1206                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1207             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1208                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1209             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1210                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1211             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1212                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1213             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1214             Heps             = _mm256_mul_ps(vfeps,H);
1215             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1216             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1217             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1218
1219             fscal            = felec;
1220
1221             /* Calculate temporary vectorial force */
1222             tx               = _mm256_mul_ps(fscal,dx20);
1223             ty               = _mm256_mul_ps(fscal,dy20);
1224             tz               = _mm256_mul_ps(fscal,dz20);
1225
1226             /* Update vectorial force */
1227             fix2             = _mm256_add_ps(fix2,tx);
1228             fiy2             = _mm256_add_ps(fiy2,ty);
1229             fiz2             = _mm256_add_ps(fiz2,tz);
1230
1231             fjx0             = _mm256_add_ps(fjx0,tx);
1232             fjy0             = _mm256_add_ps(fjy0,ty);
1233             fjz0             = _mm256_add_ps(fjz0,tz);
1234
1235             fjptrA             = f+j_coord_offsetA;
1236             fjptrB             = f+j_coord_offsetB;
1237             fjptrC             = f+j_coord_offsetC;
1238             fjptrD             = f+j_coord_offsetD;
1239             fjptrE             = f+j_coord_offsetE;
1240             fjptrF             = f+j_coord_offsetF;
1241             fjptrG             = f+j_coord_offsetG;
1242             fjptrH             = f+j_coord_offsetH;
1243
1244             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1245
1246             /* Inner loop uses 142 flops */
1247         }
1248
1249         if(jidx<j_index_end)
1250         {
1251
1252             /* Get j neighbor index, and coordinate index */
1253             jnrlistA         = jjnr[jidx];
1254             jnrlistB         = jjnr[jidx+1];
1255             jnrlistC         = jjnr[jidx+2];
1256             jnrlistD         = jjnr[jidx+3];
1257             jnrlistE         = jjnr[jidx+4];
1258             jnrlistF         = jjnr[jidx+5];
1259             jnrlistG         = jjnr[jidx+6];
1260             jnrlistH         = jjnr[jidx+7];
1261             /* Sign of each element will be negative for non-real atoms.
1262              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1263              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1264              */
1265             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1266                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1267
1268             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1269             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1270             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1271             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1272             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1273             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1274             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1275             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1276             j_coord_offsetA  = DIM*jnrA;
1277             j_coord_offsetB  = DIM*jnrB;
1278             j_coord_offsetC  = DIM*jnrC;
1279             j_coord_offsetD  = DIM*jnrD;
1280             j_coord_offsetE  = DIM*jnrE;
1281             j_coord_offsetF  = DIM*jnrF;
1282             j_coord_offsetG  = DIM*jnrG;
1283             j_coord_offsetH  = DIM*jnrH;
1284
1285             /* load j atom coordinates */
1286             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1287                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1288                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1289                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1290                                                  &jx0,&jy0,&jz0);
1291
1292             /* Calculate displacement vector */
1293             dx00             = _mm256_sub_ps(ix0,jx0);
1294             dy00             = _mm256_sub_ps(iy0,jy0);
1295             dz00             = _mm256_sub_ps(iz0,jz0);
1296             dx10             = _mm256_sub_ps(ix1,jx0);
1297             dy10             = _mm256_sub_ps(iy1,jy0);
1298             dz10             = _mm256_sub_ps(iz1,jz0);
1299             dx20             = _mm256_sub_ps(ix2,jx0);
1300             dy20             = _mm256_sub_ps(iy2,jy0);
1301             dz20             = _mm256_sub_ps(iz2,jz0);
1302
1303             /* Calculate squared distance and things based on it */
1304             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1305             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1306             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1307
1308             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1309             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1310             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1311
1312             /* Load parameters for j particles */
1313             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1314                                                                  charge+jnrC+0,charge+jnrD+0,
1315                                                                  charge+jnrE+0,charge+jnrF+0,
1316                                                                  charge+jnrG+0,charge+jnrH+0);
1317             vdwjidx0A        = 2*vdwtype[jnrA+0];
1318             vdwjidx0B        = 2*vdwtype[jnrB+0];
1319             vdwjidx0C        = 2*vdwtype[jnrC+0];
1320             vdwjidx0D        = 2*vdwtype[jnrD+0];
1321             vdwjidx0E        = 2*vdwtype[jnrE+0];
1322             vdwjidx0F        = 2*vdwtype[jnrF+0];
1323             vdwjidx0G        = 2*vdwtype[jnrG+0];
1324             vdwjidx0H        = 2*vdwtype[jnrH+0];
1325
1326             fjx0             = _mm256_setzero_ps();
1327             fjy0             = _mm256_setzero_ps();
1328             fjz0             = _mm256_setzero_ps();
1329
1330             /**************************
1331              * CALCULATE INTERACTIONS *
1332              **************************/
1333
1334             r00              = _mm256_mul_ps(rsq00,rinv00);
1335             r00              = _mm256_andnot_ps(dummy_mask,r00);
1336
1337             /* Compute parameters for interactions between i and j atoms */
1338             qq00             = _mm256_mul_ps(iq0,jq0);
1339             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1340                                             vdwioffsetptr0+vdwjidx0B,
1341                                             vdwioffsetptr0+vdwjidx0C,
1342                                             vdwioffsetptr0+vdwjidx0D,
1343                                             vdwioffsetptr0+vdwjidx0E,
1344                                             vdwioffsetptr0+vdwjidx0F,
1345                                             vdwioffsetptr0+vdwjidx0G,
1346                                             vdwioffsetptr0+vdwjidx0H,
1347                                             &c6_00,&c12_00);
1348
1349             /* Calculate table index by multiplying r with table scale and truncate to integer */
1350             rt               = _mm256_mul_ps(r00,vftabscale);
1351             vfitab           = _mm256_cvttps_epi32(rt);
1352             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1353             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1354             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1355             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1356             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1357             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1358
1359             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1360             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1361                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1362             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1364             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1366             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1368             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1369             Heps             = _mm256_mul_ps(vfeps,H);
1370             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1371             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1372             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1373
1374             /* CUBIC SPLINE TABLE DISPERSION */
1375             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1376             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1377             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1378                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1379             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1380                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1381             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1382                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1383             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1384                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1385             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1386             Heps             = _mm256_mul_ps(vfeps,H);
1387             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1388             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1389             fvdw6            = _mm256_mul_ps(c6_00,FF);
1390
1391             /* CUBIC SPLINE TABLE REPULSION */
1392             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1393             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1394             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1395                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1396             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1397                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1398             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1399                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1400             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1401                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1402             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1403             Heps             = _mm256_mul_ps(vfeps,H);
1404             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1405             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1406             fvdw12           = _mm256_mul_ps(c12_00,FF);
1407             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1408
1409             fscal            = _mm256_add_ps(felec,fvdw);
1410
1411             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1412
1413             /* Calculate temporary vectorial force */
1414             tx               = _mm256_mul_ps(fscal,dx00);
1415             ty               = _mm256_mul_ps(fscal,dy00);
1416             tz               = _mm256_mul_ps(fscal,dz00);
1417
1418             /* Update vectorial force */
1419             fix0             = _mm256_add_ps(fix0,tx);
1420             fiy0             = _mm256_add_ps(fiy0,ty);
1421             fiz0             = _mm256_add_ps(fiz0,tz);
1422
1423             fjx0             = _mm256_add_ps(fjx0,tx);
1424             fjy0             = _mm256_add_ps(fjy0,ty);
1425             fjz0             = _mm256_add_ps(fjz0,tz);
1426
1427             /**************************
1428              * CALCULATE INTERACTIONS *
1429              **************************/
1430
1431             r10              = _mm256_mul_ps(rsq10,rinv10);
1432             r10              = _mm256_andnot_ps(dummy_mask,r10);
1433
1434             /* Compute parameters for interactions between i and j atoms */
1435             qq10             = _mm256_mul_ps(iq1,jq0);
1436
1437             /* Calculate table index by multiplying r with table scale and truncate to integer */
1438             rt               = _mm256_mul_ps(r10,vftabscale);
1439             vfitab           = _mm256_cvttps_epi32(rt);
1440             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1441             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1442             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1443             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1444             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1445             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1446
1447             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1448             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1449                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1450             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1451                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1452             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1453                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1454             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1455                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1456             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1457             Heps             = _mm256_mul_ps(vfeps,H);
1458             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1459             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1460             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1461
1462             fscal            = felec;
1463
1464             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1465
1466             /* Calculate temporary vectorial force */
1467             tx               = _mm256_mul_ps(fscal,dx10);
1468             ty               = _mm256_mul_ps(fscal,dy10);
1469             tz               = _mm256_mul_ps(fscal,dz10);
1470
1471             /* Update vectorial force */
1472             fix1             = _mm256_add_ps(fix1,tx);
1473             fiy1             = _mm256_add_ps(fiy1,ty);
1474             fiz1             = _mm256_add_ps(fiz1,tz);
1475
1476             fjx0             = _mm256_add_ps(fjx0,tx);
1477             fjy0             = _mm256_add_ps(fjy0,ty);
1478             fjz0             = _mm256_add_ps(fjz0,tz);
1479
1480             /**************************
1481              * CALCULATE INTERACTIONS *
1482              **************************/
1483
1484             r20              = _mm256_mul_ps(rsq20,rinv20);
1485             r20              = _mm256_andnot_ps(dummy_mask,r20);
1486
1487             /* Compute parameters for interactions between i and j atoms */
1488             qq20             = _mm256_mul_ps(iq2,jq0);
1489
1490             /* Calculate table index by multiplying r with table scale and truncate to integer */
1491             rt               = _mm256_mul_ps(r20,vftabscale);
1492             vfitab           = _mm256_cvttps_epi32(rt);
1493             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1494             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1495             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1496             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1497             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1498             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1499
1500             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1501             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1502                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1503             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1504                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1505             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1506                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1507             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1508                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1509             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1510             Heps             = _mm256_mul_ps(vfeps,H);
1511             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1512             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1513             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1514
1515             fscal            = felec;
1516
1517             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1518
1519             /* Calculate temporary vectorial force */
1520             tx               = _mm256_mul_ps(fscal,dx20);
1521             ty               = _mm256_mul_ps(fscal,dy20);
1522             tz               = _mm256_mul_ps(fscal,dz20);
1523
1524             /* Update vectorial force */
1525             fix2             = _mm256_add_ps(fix2,tx);
1526             fiy2             = _mm256_add_ps(fiy2,ty);
1527             fiz2             = _mm256_add_ps(fiz2,tz);
1528
1529             fjx0             = _mm256_add_ps(fjx0,tx);
1530             fjy0             = _mm256_add_ps(fjy0,ty);
1531             fjz0             = _mm256_add_ps(fjz0,tz);
1532
1533             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1534             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1535             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1536             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1537             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1538             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1539             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1540             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1541
1542             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1543
1544             /* Inner loop uses 145 flops */
1545         }
1546
1547         /* End of innermost loop */
1548
1549         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1550                                                  f+i_coord_offset,fshift+i_shift_offset);
1551
1552         /* Increment number of inner iterations */
1553         inneriter                  += j_index_end - j_index_start;
1554
1555         /* Outer loop uses 18 flops */
1556     }
1557
1558     /* Increment number of outer iterations */
1559     outeriter        += nri;
1560
1561     /* Update outer/inner flops */
1562
1563     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*145);
1564 }