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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->ic->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_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     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 139     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 140     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 141     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149     j_coord_offsetE = 0;
 150     j_coord_offsetF = 0;
 151     j_coord_offsetG = 0;
 152     j_coord_offsetH = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183         fix1             = _mm256_setzero_ps();
 184         fiy1             = _mm256_setzero_ps();
 185         fiz1             = _mm256_setzero_ps();
 186         fix2             = _mm256_setzero_ps();
 187         fiy2             = _mm256_setzero_ps();
 188         fiz2             = _mm256_setzero_ps();
 189         fix3             = _mm256_setzero_ps();
 190         fiy3             = _mm256_setzero_ps();
 191         fiz3             = _mm256_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_ps();
 195         vvdwsum          = _mm256_setzero_ps();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             jnrE             = jjnr[jidx+4];
 207             jnrF             = jjnr[jidx+5];
 208             jnrG             = jjnr[jidx+6];
 209             jnrH             = jjnr[jidx+7];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214             j_coord_offsetE  = DIM*jnrE;
 215             j_coord_offsetF  = DIM*jnrF;
 216             j_coord_offsetG  = DIM*jnrG;
 217             j_coord_offsetH  = DIM*jnrH;
 218
 219             /* load j atom coordinates */
 220             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 221                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 222                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 223                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 224                                                  &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_ps(ix0,jx0);
 228             dy00             = _mm256_sub_ps(iy0,jy0);
 229             dz00             = _mm256_sub_ps(iz0,jz0);
 230             dx10             = _mm256_sub_ps(ix1,jx0);
 231             dy10             = _mm256_sub_ps(iy1,jy0);
 232             dz10             = _mm256_sub_ps(iz1,jz0);
 233             dx20             = _mm256_sub_ps(ix2,jx0);
 234             dy20             = _mm256_sub_ps(iy2,jy0);
 235             dz20             = _mm256_sub_ps(iz2,jz0);
 236             dx30             = _mm256_sub_ps(ix3,jx0);
 237             dy30             = _mm256_sub_ps(iy3,jy0);
 238             dz30             = _mm256_sub_ps(iz3,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 243             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 244             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 245
 246             rinv00           = avx256_invsqrt_f(rsq00);
 247             rinv10           = avx256_invsqrt_f(rsq10);
 248             rinv20           = avx256_invsqrt_f(rsq20);
 249             rinv30           = avx256_invsqrt_f(rsq30);
 250
 251             /* Load parameters for j particles */
 252             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 253                                                                  charge+jnrC+0,charge+jnrD+0,
 254                                                                  charge+jnrE+0,charge+jnrF+0,
 255                                                                  charge+jnrG+0,charge+jnrH+0);
 256             vdwjidx0A        = 2*vdwtype[jnrA+0];
 257             vdwjidx0B        = 2*vdwtype[jnrB+0];
 258             vdwjidx0C        = 2*vdwtype[jnrC+0];
 259             vdwjidx0D        = 2*vdwtype[jnrD+0];
 260             vdwjidx0E        = 2*vdwtype[jnrE+0];
 261             vdwjidx0F        = 2*vdwtype[jnrF+0];
 262             vdwjidx0G        = 2*vdwtype[jnrG+0];
 263             vdwjidx0H        = 2*vdwtype[jnrH+0];
 264
 265             fjx0             = _mm256_setzero_ps();
 266             fjy0             = _mm256_setzero_ps();
 267             fjz0             = _mm256_setzero_ps();
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             r00              = _mm256_mul_ps(rsq00,rinv00);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 277                                             vdwioffsetptr0+vdwjidx0B,
 278                                             vdwioffsetptr0+vdwjidx0C,
 279                                             vdwioffsetptr0+vdwjidx0D,
 280                                             vdwioffsetptr0+vdwjidx0E,
 281                                             vdwioffsetptr0+vdwjidx0F,
 282                                             vdwioffsetptr0+vdwjidx0G,
 283                                             vdwioffsetptr0+vdwjidx0H,
 284                                             &c6_00,&c12_00);
 285
 286             /* Calculate table index by multiplying r with table scale and truncate to integer */
 287             rt               = _mm256_mul_ps(r00,vftabscale);
 288             vfitab           = _mm256_cvttps_epi32(rt);
 289             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 290             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 291             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 292             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 293             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 294             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 295
 296             /* CUBIC SPLINE TABLE DISPERSION */
 297             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 298             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 299             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 300                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 301             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 302                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 303             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 304                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 305             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 307             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 308             Heps             = _mm256_mul_ps(vfeps,H);
 309             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 310             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 311             vvdw6            = _mm256_mul_ps(c6_00,VV);
 312             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 313             fvdw6            = _mm256_mul_ps(c6_00,FF);
 314
 315             /* CUBIC SPLINE TABLE REPULSION */
 316             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 317             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 318             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 320             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 322             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 324             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 326             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 327             Heps             = _mm256_mul_ps(vfeps,H);
 328             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 329             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 330             vvdw12           = _mm256_mul_ps(c12_00,VV);
 331             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 332             fvdw12           = _mm256_mul_ps(c12_00,FF);
 333             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 334             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 338
 339             fscal            = fvdw;
 340
 341             /* Calculate temporary vectorial force */
 342             tx               = _mm256_mul_ps(fscal,dx00);
 343             ty               = _mm256_mul_ps(fscal,dy00);
 344             tz               = _mm256_mul_ps(fscal,dz00);
 345
 346             /* Update vectorial force */
 347             fix0             = _mm256_add_ps(fix0,tx);
 348             fiy0             = _mm256_add_ps(fiy0,ty);
 349             fiz0             = _mm256_add_ps(fiz0,tz);
 350
 351             fjx0             = _mm256_add_ps(fjx0,tx);
 352             fjy0             = _mm256_add_ps(fjy0,ty);
 353             fjz0             = _mm256_add_ps(fjz0,tz);
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             r10              = _mm256_mul_ps(rsq10,rinv10);
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq10             = _mm256_mul_ps(iq1,jq0);
 363
 364             /* Calculate table index by multiplying r with table scale and truncate to integer */
 365             rt               = _mm256_mul_ps(r10,vftabscale);
 366             vfitab           = _mm256_cvttps_epi32(rt);
 367             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 368             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 369             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 370             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 371             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 372             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 373
 374             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 375             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 376                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 377             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 378                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 379             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 380                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 381             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 382                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 383             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 384             Heps             = _mm256_mul_ps(vfeps,H);
 385             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 386             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 387             velec            = _mm256_mul_ps(qq10,VV);
 388             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 389             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm256_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm256_mul_ps(fscal,dx10);
 398             ty               = _mm256_mul_ps(fscal,dy10);
 399             tz               = _mm256_mul_ps(fscal,dz10);
 400
 401             /* Update vectorial force */
 402             fix1             = _mm256_add_ps(fix1,tx);
 403             fiy1             = _mm256_add_ps(fiy1,ty);
 404             fiz1             = _mm256_add_ps(fiz1,tz);
 405
 406             fjx0             = _mm256_add_ps(fjx0,tx);
 407             fjy0             = _mm256_add_ps(fjy0,ty);
 408             fjz0             = _mm256_add_ps(fjz0,tz);
 409
 410             /**************************
 411              * CALCULATE INTERACTIONS *
 412              **************************/
 413
 414             r20              = _mm256_mul_ps(rsq20,rinv20);
 415
 416             /* Compute parameters for interactions between i and j atoms */
 417             qq20             = _mm256_mul_ps(iq2,jq0);
 418
 419             /* Calculate table index by multiplying r with table scale and truncate to integer */
 420             rt               = _mm256_mul_ps(r20,vftabscale);
 421             vfitab           = _mm256_cvttps_epi32(rt);
 422             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 423             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 424             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 425             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 426             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 427             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 428
 429             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 430             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 431                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 432             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 433                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 434             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 435                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 436             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 437                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 438             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 439             Heps             = _mm256_mul_ps(vfeps,H);
 440             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 441             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 442             velec            = _mm256_mul_ps(qq20,VV);
 443             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 444             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 445
 446             /* Update potential sum for this i atom from the interaction with this j atom. */
 447             velecsum         = _mm256_add_ps(velecsum,velec);
 448
 449             fscal            = felec;
 450
 451             /* Calculate temporary vectorial force */
 452             tx               = _mm256_mul_ps(fscal,dx20);
 453             ty               = _mm256_mul_ps(fscal,dy20);
 454             tz               = _mm256_mul_ps(fscal,dz20);
 455
 456             /* Update vectorial force */
 457             fix2             = _mm256_add_ps(fix2,tx);
 458             fiy2             = _mm256_add_ps(fiy2,ty);
 459             fiz2             = _mm256_add_ps(fiz2,tz);
 460
 461             fjx0             = _mm256_add_ps(fjx0,tx);
 462             fjy0             = _mm256_add_ps(fjy0,ty);
 463             fjz0             = _mm256_add_ps(fjz0,tz);
 464
 465             /**************************
 466              * CALCULATE INTERACTIONS *
 467              **************************/
 468
 469             r30              = _mm256_mul_ps(rsq30,rinv30);
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq30             = _mm256_mul_ps(iq3,jq0);
 473
 474             /* Calculate table index by multiplying r with table scale and truncate to integer */
 475             rt               = _mm256_mul_ps(r30,vftabscale);
 476             vfitab           = _mm256_cvttps_epi32(rt);
 477             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 478             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 479             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 480             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 481             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 482             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 483
 484             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 485             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 486                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 487             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 488                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 489             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 490                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 491             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 492                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 493             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 494             Heps             = _mm256_mul_ps(vfeps,H);
 495             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 496             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 497             velec            = _mm256_mul_ps(qq30,VV);
 498             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 499             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velecsum         = _mm256_add_ps(velecsum,velec);
 503
 504             fscal            = felec;
 505
 506             /* Calculate temporary vectorial force */
 507             tx               = _mm256_mul_ps(fscal,dx30);
 508             ty               = _mm256_mul_ps(fscal,dy30);
 509             tz               = _mm256_mul_ps(fscal,dz30);
 510
 511             /* Update vectorial force */
 512             fix3             = _mm256_add_ps(fix3,tx);
 513             fiy3             = _mm256_add_ps(fiy3,ty);
 514             fiz3             = _mm256_add_ps(fiz3,tz);
 515
 516             fjx0             = _mm256_add_ps(fjx0,tx);
 517             fjy0             = _mm256_add_ps(fjy0,ty);
 518             fjz0             = _mm256_add_ps(fjz0,tz);
 519
 520             fjptrA             = f+j_coord_offsetA;
 521             fjptrB             = f+j_coord_offsetB;
 522             fjptrC             = f+j_coord_offsetC;
 523             fjptrD             = f+j_coord_offsetD;
 524             fjptrE             = f+j_coord_offsetE;
 525             fjptrF             = f+j_coord_offsetF;
 526             fjptrG             = f+j_coord_offsetG;
 527             fjptrH             = f+j_coord_offsetH;
 528
 529             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 530
 531             /* Inner loop uses 188 flops */
 532         }
 533
 534         if(jidx<j_index_end)
 535         {
 536
 537             /* Get j neighbor index, and coordinate index */
 538             jnrlistA         = jjnr[jidx];
 539             jnrlistB         = jjnr[jidx+1];
 540             jnrlistC         = jjnr[jidx+2];
 541             jnrlistD         = jjnr[jidx+3];
 542             jnrlistE         = jjnr[jidx+4];
 543             jnrlistF         = jjnr[jidx+5];
 544             jnrlistG         = jjnr[jidx+6];
 545             jnrlistH         = jjnr[jidx+7];
 546             /* Sign of each element will be negative for non-real atoms.
 547              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 548              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 549              */
 550             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 551                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 552
 553             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 554             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 555             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 556             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 557             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 558             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 559             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 560             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 561             j_coord_offsetA  = DIM*jnrA;
 562             j_coord_offsetB  = DIM*jnrB;
 563             j_coord_offsetC  = DIM*jnrC;
 564             j_coord_offsetD  = DIM*jnrD;
 565             j_coord_offsetE  = DIM*jnrE;
 566             j_coord_offsetF  = DIM*jnrF;
 567             j_coord_offsetG  = DIM*jnrG;
 568             j_coord_offsetH  = DIM*jnrH;
 569
 570             /* load j atom coordinates */
 571             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 572                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 573                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 574                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 575                                                  &jx0,&jy0,&jz0);
 576
 577             /* Calculate displacement vector */
 578             dx00             = _mm256_sub_ps(ix0,jx0);
 579             dy00             = _mm256_sub_ps(iy0,jy0);
 580             dz00             = _mm256_sub_ps(iz0,jz0);
 581             dx10             = _mm256_sub_ps(ix1,jx0);
 582             dy10             = _mm256_sub_ps(iy1,jy0);
 583             dz10             = _mm256_sub_ps(iz1,jz0);
 584             dx20             = _mm256_sub_ps(ix2,jx0);
 585             dy20             = _mm256_sub_ps(iy2,jy0);
 586             dz20             = _mm256_sub_ps(iz2,jz0);
 587             dx30             = _mm256_sub_ps(ix3,jx0);
 588             dy30             = _mm256_sub_ps(iy3,jy0);
 589             dz30             = _mm256_sub_ps(iz3,jz0);
 590
 591             /* Calculate squared distance and things based on it */
 592             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 593             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 594             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 595             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 596
 597             rinv00           = avx256_invsqrt_f(rsq00);
 598             rinv10           = avx256_invsqrt_f(rsq10);
 599             rinv20           = avx256_invsqrt_f(rsq20);
 600             rinv30           = avx256_invsqrt_f(rsq30);
 601
 602             /* Load parameters for j particles */
 603             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 604                                                                  charge+jnrC+0,charge+jnrD+0,
 605                                                                  charge+jnrE+0,charge+jnrF+0,
 606                                                                  charge+jnrG+0,charge+jnrH+0);
 607             vdwjidx0A        = 2*vdwtype[jnrA+0];
 608             vdwjidx0B        = 2*vdwtype[jnrB+0];
 609             vdwjidx0C        = 2*vdwtype[jnrC+0];
 610             vdwjidx0D        = 2*vdwtype[jnrD+0];
 611             vdwjidx0E        = 2*vdwtype[jnrE+0];
 612             vdwjidx0F        = 2*vdwtype[jnrF+0];
 613             vdwjidx0G        = 2*vdwtype[jnrG+0];
 614             vdwjidx0H        = 2*vdwtype[jnrH+0];
 615
 616             fjx0             = _mm256_setzero_ps();
 617             fjy0             = _mm256_setzero_ps();
 618             fjz0             = _mm256_setzero_ps();
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             r00              = _mm256_mul_ps(rsq00,rinv00);
 625             r00              = _mm256_andnot_ps(dummy_mask,r00);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 629                                             vdwioffsetptr0+vdwjidx0B,
 630                                             vdwioffsetptr0+vdwjidx0C,
 631                                             vdwioffsetptr0+vdwjidx0D,
 632                                             vdwioffsetptr0+vdwjidx0E,
 633                                             vdwioffsetptr0+vdwjidx0F,
 634                                             vdwioffsetptr0+vdwjidx0G,
 635                                             vdwioffsetptr0+vdwjidx0H,
 636                                             &c6_00,&c12_00);
 637
 638             /* Calculate table index by multiplying r with table scale and truncate to integer */
 639             rt               = _mm256_mul_ps(r00,vftabscale);
 640             vfitab           = _mm256_cvttps_epi32(rt);
 641             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 642             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 643             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 644             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 645             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 646             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 647
 648             /* CUBIC SPLINE TABLE DISPERSION */
 649             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 650             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 651             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 652                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 653             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 654                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 655             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 656                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 657             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 658                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 659             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 660             Heps             = _mm256_mul_ps(vfeps,H);
 661             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 662             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 663             vvdw6            = _mm256_mul_ps(c6_00,VV);
 664             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 665             fvdw6            = _mm256_mul_ps(c6_00,FF);
 666
 667             /* CUBIC SPLINE TABLE REPULSION */
 668             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 669             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 670             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 671                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 672             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 673                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 674             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 675                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 676             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 677                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 678             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 679             Heps             = _mm256_mul_ps(vfeps,H);
 680             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 681             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 682             vvdw12           = _mm256_mul_ps(c12_00,VV);
 683             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 684             fvdw12           = _mm256_mul_ps(c12_00,FF);
 685             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 686             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 687
 688             /* Update potential sum for this i atom from the interaction with this j atom. */
 689             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 690             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 691
 692             fscal            = fvdw;
 693
 694             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 695
 696             /* Calculate temporary vectorial force */
 697             tx               = _mm256_mul_ps(fscal,dx00);
 698             ty               = _mm256_mul_ps(fscal,dy00);
 699             tz               = _mm256_mul_ps(fscal,dz00);
 700
 701             /* Update vectorial force */
 702             fix0             = _mm256_add_ps(fix0,tx);
 703             fiy0             = _mm256_add_ps(fiy0,ty);
 704             fiz0             = _mm256_add_ps(fiz0,tz);
 705
 706             fjx0             = _mm256_add_ps(fjx0,tx);
 707             fjy0             = _mm256_add_ps(fjy0,ty);
 708             fjz0             = _mm256_add_ps(fjz0,tz);
 709
 710             /**************************
 711              * CALCULATE INTERACTIONS *
 712              **************************/
 713
 714             r10              = _mm256_mul_ps(rsq10,rinv10);
 715             r10              = _mm256_andnot_ps(dummy_mask,r10);
 716
 717             /* Compute parameters for interactions between i and j atoms */
 718             qq10             = _mm256_mul_ps(iq1,jq0);
 719
 720             /* Calculate table index by multiplying r with table scale and truncate to integer */
 721             rt               = _mm256_mul_ps(r10,vftabscale);
 722             vfitab           = _mm256_cvttps_epi32(rt);
 723             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 724             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 725             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 726             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 727             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 728             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 729
 730             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 731             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 732                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 733             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 734                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 735             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 736                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 737             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 738                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 739             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 740             Heps             = _mm256_mul_ps(vfeps,H);
 741             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 742             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 743             velec            = _mm256_mul_ps(qq10,VV);
 744             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 745             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 746
 747             /* Update potential sum for this i atom from the interaction with this j atom. */
 748             velec            = _mm256_andnot_ps(dummy_mask,velec);
 749             velecsum         = _mm256_add_ps(velecsum,velec);
 750
 751             fscal            = felec;
 752
 753             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 754
 755             /* Calculate temporary vectorial force */
 756             tx               = _mm256_mul_ps(fscal,dx10);
 757             ty               = _mm256_mul_ps(fscal,dy10);
 758             tz               = _mm256_mul_ps(fscal,dz10);
 759
 760             /* Update vectorial force */
 761             fix1             = _mm256_add_ps(fix1,tx);
 762             fiy1             = _mm256_add_ps(fiy1,ty);
 763             fiz1             = _mm256_add_ps(fiz1,tz);
 764
 765             fjx0             = _mm256_add_ps(fjx0,tx);
 766             fjy0             = _mm256_add_ps(fjy0,ty);
 767             fjz0             = _mm256_add_ps(fjz0,tz);
 768
 769             /**************************
 770              * CALCULATE INTERACTIONS *
 771              **************************/
 772
 773             r20              = _mm256_mul_ps(rsq20,rinv20);
 774             r20              = _mm256_andnot_ps(dummy_mask,r20);
 775
 776             /* Compute parameters for interactions between i and j atoms */
 777             qq20             = _mm256_mul_ps(iq2,jq0);
 778
 779             /* Calculate table index by multiplying r with table scale and truncate to integer */
 780             rt               = _mm256_mul_ps(r20,vftabscale);
 781             vfitab           = _mm256_cvttps_epi32(rt);
 782             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 783             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 784             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 785             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 786             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 787             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 788
 789             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 790             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 791                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 792             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 793                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 794             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 795                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 796             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 797                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 798             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 799             Heps             = _mm256_mul_ps(vfeps,H);
 800             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 801             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 802             velec            = _mm256_mul_ps(qq20,VV);
 803             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 804             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 805
 806             /* Update potential sum for this i atom from the interaction with this j atom. */
 807             velec            = _mm256_andnot_ps(dummy_mask,velec);
 808             velecsum         = _mm256_add_ps(velecsum,velec);
 809
 810             fscal            = felec;
 811
 812             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 813
 814             /* Calculate temporary vectorial force */
 815             tx               = _mm256_mul_ps(fscal,dx20);
 816             ty               = _mm256_mul_ps(fscal,dy20);
 817             tz               = _mm256_mul_ps(fscal,dz20);
 818
 819             /* Update vectorial force */
 820             fix2             = _mm256_add_ps(fix2,tx);
 821             fiy2             = _mm256_add_ps(fiy2,ty);
 822             fiz2             = _mm256_add_ps(fiz2,tz);
 823
 824             fjx0             = _mm256_add_ps(fjx0,tx);
 825             fjy0             = _mm256_add_ps(fjy0,ty);
 826             fjz0             = _mm256_add_ps(fjz0,tz);
 827
 828             /**************************
 829              * CALCULATE INTERACTIONS *
 830              **************************/
 831
 832             r30              = _mm256_mul_ps(rsq30,rinv30);
 833             r30              = _mm256_andnot_ps(dummy_mask,r30);
 834
 835             /* Compute parameters for interactions between i and j atoms */
 836             qq30             = _mm256_mul_ps(iq3,jq0);
 837
 838             /* Calculate table index by multiplying r with table scale and truncate to integer */
 839             rt               = _mm256_mul_ps(r30,vftabscale);
 840             vfitab           = _mm256_cvttps_epi32(rt);
 841             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 842             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 843             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 844             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 845             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 846             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 847
 848             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 849             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 850                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 851             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 852                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 853             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 854                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 855             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 856                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 857             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 858             Heps             = _mm256_mul_ps(vfeps,H);
 859             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 860             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 861             velec            = _mm256_mul_ps(qq30,VV);
 862             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 863             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 864
 865             /* Update potential sum for this i atom from the interaction with this j atom. */
 866             velec            = _mm256_andnot_ps(dummy_mask,velec);
 867             velecsum         = _mm256_add_ps(velecsum,velec);
 868
 869             fscal            = felec;
 870
 871             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 872
 873             /* Calculate temporary vectorial force */
 874             tx               = _mm256_mul_ps(fscal,dx30);
 875             ty               = _mm256_mul_ps(fscal,dy30);
 876             tz               = _mm256_mul_ps(fscal,dz30);
 877
 878             /* Update vectorial force */
 879             fix3             = _mm256_add_ps(fix3,tx);
 880             fiy3             = _mm256_add_ps(fiy3,ty);
 881             fiz3             = _mm256_add_ps(fiz3,tz);
 882
 883             fjx0             = _mm256_add_ps(fjx0,tx);
 884             fjy0             = _mm256_add_ps(fjy0,ty);
 885             fjz0             = _mm256_add_ps(fjz0,tz);
 886
 887             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 888             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 889             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 890             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 891             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 892             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 893             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 894             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 895
 896             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 897
 898             /* Inner loop uses 192 flops */
 899         }
 900
 901         /* End of innermost loop */
 902
 903         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 904                                                  f+i_coord_offset,fshift+i_shift_offset);
 905
 906         ggid                        = gid[iidx];
 907         /* Update potential energies */
 908         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 909         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 910
 911         /* Increment number of inner iterations */
 912         inneriter                  += j_index_end - j_index_start;
 913
 914         /* Outer loop uses 26 flops */
 915     }
 916
 917     /* Increment number of outer iterations */
 918     outeriter        += nri;
 919
 920     /* Update outer/inner flops */
 921
 922     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*192);
 923 }
 924 /*
 925  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_single
 926  * Electrostatics interaction: CubicSplineTable
 927  * VdW interaction:            CubicSplineTable
 928  * Geometry:                   Water4-Particle
 929  * Calculate force/pot:        Force
 930  */
 931 void
 932 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_single
 933                     (t_nblist                    * gmx_restrict       nlist,
 934                      rvec                        * gmx_restrict          xx,
 935                      rvec                        * gmx_restrict          ff,
 936                      struct t_forcerec           * gmx_restrict          fr,
 937                      t_mdatoms                   * gmx_restrict     mdatoms,
 938                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 939                      t_nrnb                      * gmx_restrict        nrnb)
 940 {
 941     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 942      * just 0 for non-waters.
 943      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 944      * jnr indices corresponding to data put in the four positions in the SIMD register.
 945      */
 946     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 947     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 948     int              jnrA,jnrB,jnrC,jnrD;
 949     int              jnrE,jnrF,jnrG,jnrH;
 950     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 951     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 952     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 953     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 954     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 955     real             rcutoff_scalar;
 956     real             *shiftvec,*fshift,*x,*f;
 957     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 958     real             scratch[4*DIM];
 959     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 960     real *           vdwioffsetptr0;
 961     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 962     real *           vdwioffsetptr1;
 963     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 964     real *           vdwioffsetptr2;
 965     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 966     real *           vdwioffsetptr3;
 967     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 968     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 969     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 970     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 971     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 972     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 973     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 974     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 975     real             *charge;
 976     int              nvdwtype;
 977     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 978     int              *vdwtype;
 979     real             *vdwparam;
 980     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 981     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 982     __m256i          vfitab;
 983     __m128i          vfitab_lo,vfitab_hi;
 984     __m128i          ifour       = _mm_set1_epi32(4);
 985     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 986     real             *vftab;
 987     __m256           dummy_mask,cutoff_mask;
 988     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 989     __m256           one     = _mm256_set1_ps(1.0);
 990     __m256           two     = _mm256_set1_ps(2.0);
 991     x                = xx[0];
 992     f                = ff[0];
 993
 994     nri              = nlist->nri;
 995     iinr             = nlist->iinr;
 996     jindex           = nlist->jindex;
 997     jjnr             = nlist->jjnr;
 998     shiftidx         = nlist->shift;
 999     gid              = nlist->gid;
1000     shiftvec         = fr->shift_vec[0];
1001     fshift           = fr->fshift[0];
1002     facel            = _mm256_set1_ps(fr->ic->epsfac);
1003     charge           = mdatoms->chargeA;
1004     nvdwtype         = fr->ntype;
1005     vdwparam         = fr->nbfp;
1006     vdwtype          = mdatoms->typeA;
1007
1008     vftab            = kernel_data->table_elec_vdw->data;
1009     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
1010
1011     /* Setup water-specific parameters */
1012     inr              = nlist->iinr[0];
1013     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1014     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1015     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1016     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1017
1018     /* Avoid stupid compiler warnings */
1019     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1020     j_coord_offsetA = 0;
1021     j_coord_offsetB = 0;
1022     j_coord_offsetC = 0;
1023     j_coord_offsetD = 0;
1024     j_coord_offsetE = 0;
1025     j_coord_offsetF = 0;
1026     j_coord_offsetG = 0;
1027     j_coord_offsetH = 0;
1028
1029     outeriter        = 0;
1030     inneriter        = 0;
1031
1032     for(iidx=0;iidx<4*DIM;iidx++)
1033     {
1034         scratch[iidx] = 0.0;
1035     }
1036
1037     /* Start outer loop over neighborlists */
1038     for(iidx=0; iidx<nri; iidx++)
1039     {
1040         /* Load shift vector for this list */
1041         i_shift_offset   = DIM*shiftidx[iidx];
1042
1043         /* Load limits for loop over neighbors */
1044         j_index_start    = jindex[iidx];
1045         j_index_end      = jindex[iidx+1];
1046
1047         /* Get outer coordinate index */
1048         inr              = iinr[iidx];
1049         i_coord_offset   = DIM*inr;
1050
1051         /* Load i particle coords and add shift vector */
1052         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1053                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1054
1055         fix0             = _mm256_setzero_ps();
1056         fiy0             = _mm256_setzero_ps();
1057         fiz0             = _mm256_setzero_ps();
1058         fix1             = _mm256_setzero_ps();
1059         fiy1             = _mm256_setzero_ps();
1060         fiz1             = _mm256_setzero_ps();
1061         fix2             = _mm256_setzero_ps();
1062         fiy2             = _mm256_setzero_ps();
1063         fiz2             = _mm256_setzero_ps();
1064         fix3             = _mm256_setzero_ps();
1065         fiy3             = _mm256_setzero_ps();
1066         fiz3             = _mm256_setzero_ps();
1067
1068         /* Start inner kernel loop */
1069         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1070         {
1071
1072             /* Get j neighbor index, and coordinate index */
1073             jnrA             = jjnr[jidx];
1074             jnrB             = jjnr[jidx+1];
1075             jnrC             = jjnr[jidx+2];
1076             jnrD             = jjnr[jidx+3];
1077             jnrE             = jjnr[jidx+4];
1078             jnrF             = jjnr[jidx+5];
1079             jnrG             = jjnr[jidx+6];
1080             jnrH             = jjnr[jidx+7];
1081             j_coord_offsetA  = DIM*jnrA;
1082             j_coord_offsetB  = DIM*jnrB;
1083             j_coord_offsetC  = DIM*jnrC;
1084             j_coord_offsetD  = DIM*jnrD;
1085             j_coord_offsetE  = DIM*jnrE;
1086             j_coord_offsetF  = DIM*jnrF;
1087             j_coord_offsetG  = DIM*jnrG;
1088             j_coord_offsetH  = DIM*jnrH;
1089
1090             /* load j atom coordinates */
1091             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1092                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1093                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1094                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1095                                                  &jx0,&jy0,&jz0);
1096
1097             /* Calculate displacement vector */
1098             dx00             = _mm256_sub_ps(ix0,jx0);
1099             dy00             = _mm256_sub_ps(iy0,jy0);
1100             dz00             = _mm256_sub_ps(iz0,jz0);
1101             dx10             = _mm256_sub_ps(ix1,jx0);
1102             dy10             = _mm256_sub_ps(iy1,jy0);
1103             dz10             = _mm256_sub_ps(iz1,jz0);
1104             dx20             = _mm256_sub_ps(ix2,jx0);
1105             dy20             = _mm256_sub_ps(iy2,jy0);
1106             dz20             = _mm256_sub_ps(iz2,jz0);
1107             dx30             = _mm256_sub_ps(ix3,jx0);
1108             dy30             = _mm256_sub_ps(iy3,jy0);
1109             dz30             = _mm256_sub_ps(iz3,jz0);
1110
1111             /* Calculate squared distance and things based on it */
1112             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1113             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1114             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1115             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1116
1117             rinv00           = avx256_invsqrt_f(rsq00);
1118             rinv10           = avx256_invsqrt_f(rsq10);
1119             rinv20           = avx256_invsqrt_f(rsq20);
1120             rinv30           = avx256_invsqrt_f(rsq30);
1121
1122             /* Load parameters for j particles */
1123             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1124                                                                  charge+jnrC+0,charge+jnrD+0,
1125                                                                  charge+jnrE+0,charge+jnrF+0,
1126                                                                  charge+jnrG+0,charge+jnrH+0);
1127             vdwjidx0A        = 2*vdwtype[jnrA+0];
1128             vdwjidx0B        = 2*vdwtype[jnrB+0];
1129             vdwjidx0C        = 2*vdwtype[jnrC+0];
1130             vdwjidx0D        = 2*vdwtype[jnrD+0];
1131             vdwjidx0E        = 2*vdwtype[jnrE+0];
1132             vdwjidx0F        = 2*vdwtype[jnrF+0];
1133             vdwjidx0G        = 2*vdwtype[jnrG+0];
1134             vdwjidx0H        = 2*vdwtype[jnrH+0];
1135
1136             fjx0             = _mm256_setzero_ps();
1137             fjy0             = _mm256_setzero_ps();
1138             fjz0             = _mm256_setzero_ps();
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             r00              = _mm256_mul_ps(rsq00,rinv00);
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1148                                             vdwioffsetptr0+vdwjidx0B,
1149                                             vdwioffsetptr0+vdwjidx0C,
1150                                             vdwioffsetptr0+vdwjidx0D,
1151                                             vdwioffsetptr0+vdwjidx0E,
1152                                             vdwioffsetptr0+vdwjidx0F,
1153                                             vdwioffsetptr0+vdwjidx0G,
1154                                             vdwioffsetptr0+vdwjidx0H,
1155                                             &c6_00,&c12_00);
1156
1157             /* Calculate table index by multiplying r with table scale and truncate to integer */
1158             rt               = _mm256_mul_ps(r00,vftabscale);
1159             vfitab           = _mm256_cvttps_epi32(rt);
1160             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1161             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1162             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1163             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1164             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1165             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1166
1167             /* CUBIC SPLINE TABLE DISPERSION */
1168             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1169             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1170             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1171                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1172             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1173                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1174             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1175                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1176             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1177                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1178             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1179             Heps             = _mm256_mul_ps(vfeps,H);
1180             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1181             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1182             fvdw6            = _mm256_mul_ps(c6_00,FF);
1183
1184             /* CUBIC SPLINE TABLE REPULSION */
1185             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1186             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1187             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1188                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1189             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1190                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1191             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1193             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1195             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1196             Heps             = _mm256_mul_ps(vfeps,H);
1197             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1198             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1199             fvdw12           = _mm256_mul_ps(c12_00,FF);
1200             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1201
1202             fscal            = fvdw;
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm256_mul_ps(fscal,dx00);
1206             ty               = _mm256_mul_ps(fscal,dy00);
1207             tz               = _mm256_mul_ps(fscal,dz00);
1208
1209             /* Update vectorial force */
1210             fix0             = _mm256_add_ps(fix0,tx);
1211             fiy0             = _mm256_add_ps(fiy0,ty);
1212             fiz0             = _mm256_add_ps(fiz0,tz);
1213
1214             fjx0             = _mm256_add_ps(fjx0,tx);
1215             fjy0             = _mm256_add_ps(fjy0,ty);
1216             fjz0             = _mm256_add_ps(fjz0,tz);
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             r10              = _mm256_mul_ps(rsq10,rinv10);
1223
1224             /* Compute parameters for interactions between i and j atoms */
1225             qq10             = _mm256_mul_ps(iq1,jq0);
1226
1227             /* Calculate table index by multiplying r with table scale and truncate to integer */
1228             rt               = _mm256_mul_ps(r10,vftabscale);
1229             vfitab           = _mm256_cvttps_epi32(rt);
1230             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1231             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1232             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1233             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1234             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1235             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1236
1237             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1238             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1239                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1240             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1241                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1242             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1243                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1244             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1245                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1246             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1247             Heps             = _mm256_mul_ps(vfeps,H);
1248             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1249             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1250             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1251
1252             fscal            = felec;
1253
1254             /* Calculate temporary vectorial force */
1255             tx               = _mm256_mul_ps(fscal,dx10);
1256             ty               = _mm256_mul_ps(fscal,dy10);
1257             tz               = _mm256_mul_ps(fscal,dz10);
1258
1259             /* Update vectorial force */
1260             fix1             = _mm256_add_ps(fix1,tx);
1261             fiy1             = _mm256_add_ps(fiy1,ty);
1262             fiz1             = _mm256_add_ps(fiz1,tz);
1263
1264             fjx0             = _mm256_add_ps(fjx0,tx);
1265             fjy0             = _mm256_add_ps(fjy0,ty);
1266             fjz0             = _mm256_add_ps(fjz0,tz);
1267
1268             /**************************
1269              * CALCULATE INTERACTIONS *
1270              **************************/
1271
1272             r20              = _mm256_mul_ps(rsq20,rinv20);
1273
1274             /* Compute parameters for interactions between i and j atoms */
1275             qq20             = _mm256_mul_ps(iq2,jq0);
1276
1277             /* Calculate table index by multiplying r with table scale and truncate to integer */
1278             rt               = _mm256_mul_ps(r20,vftabscale);
1279             vfitab           = _mm256_cvttps_epi32(rt);
1280             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1281             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1282             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1283             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1284             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1285             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1286
1287             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1288             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1289                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1290             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1292             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1294             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1296             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1297             Heps             = _mm256_mul_ps(vfeps,H);
1298             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1299             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1300             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1301
1302             fscal            = felec;
1303
1304             /* Calculate temporary vectorial force */
1305             tx               = _mm256_mul_ps(fscal,dx20);
1306             ty               = _mm256_mul_ps(fscal,dy20);
1307             tz               = _mm256_mul_ps(fscal,dz20);
1308
1309             /* Update vectorial force */
1310             fix2             = _mm256_add_ps(fix2,tx);
1311             fiy2             = _mm256_add_ps(fiy2,ty);
1312             fiz2             = _mm256_add_ps(fiz2,tz);
1313
1314             fjx0             = _mm256_add_ps(fjx0,tx);
1315             fjy0             = _mm256_add_ps(fjy0,ty);
1316             fjz0             = _mm256_add_ps(fjz0,tz);
1317
1318             /**************************
1319              * CALCULATE INTERACTIONS *
1320              **************************/
1321
1322             r30              = _mm256_mul_ps(rsq30,rinv30);
1323
1324             /* Compute parameters for interactions between i and j atoms */
1325             qq30             = _mm256_mul_ps(iq3,jq0);
1326
1327             /* Calculate table index by multiplying r with table scale and truncate to integer */
1328             rt               = _mm256_mul_ps(r30,vftabscale);
1329             vfitab           = _mm256_cvttps_epi32(rt);
1330             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1331             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1332             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1333             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1334             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1335             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1336
1337             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1338             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1339                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1340             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1341                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1342             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1343                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1344             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1345                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1346             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1347             Heps             = _mm256_mul_ps(vfeps,H);
1348             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1349             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1350             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1351
1352             fscal            = felec;
1353
1354             /* Calculate temporary vectorial force */
1355             tx               = _mm256_mul_ps(fscal,dx30);
1356             ty               = _mm256_mul_ps(fscal,dy30);
1357             tz               = _mm256_mul_ps(fscal,dz30);
1358
1359             /* Update vectorial force */
1360             fix3             = _mm256_add_ps(fix3,tx);
1361             fiy3             = _mm256_add_ps(fiy3,ty);
1362             fiz3             = _mm256_add_ps(fiz3,tz);
1363
1364             fjx0             = _mm256_add_ps(fjx0,tx);
1365             fjy0             = _mm256_add_ps(fjy0,ty);
1366             fjz0             = _mm256_add_ps(fjz0,tz);
1367
1368             fjptrA             = f+j_coord_offsetA;
1369             fjptrB             = f+j_coord_offsetB;
1370             fjptrC             = f+j_coord_offsetC;
1371             fjptrD             = f+j_coord_offsetD;
1372             fjptrE             = f+j_coord_offsetE;
1373             fjptrF             = f+j_coord_offsetF;
1374             fjptrG             = f+j_coord_offsetG;
1375             fjptrH             = f+j_coord_offsetH;
1376
1377             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1378
1379             /* Inner loop uses 168 flops */
1380         }
1381
1382         if(jidx<j_index_end)
1383         {
1384
1385             /* Get j neighbor index, and coordinate index */
1386             jnrlistA         = jjnr[jidx];
1387             jnrlistB         = jjnr[jidx+1];
1388             jnrlistC         = jjnr[jidx+2];
1389             jnrlistD         = jjnr[jidx+3];
1390             jnrlistE         = jjnr[jidx+4];
1391             jnrlistF         = jjnr[jidx+5];
1392             jnrlistG         = jjnr[jidx+6];
1393             jnrlistH         = jjnr[jidx+7];
1394             /* Sign of each element will be negative for non-real atoms.
1395              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1396              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1397              */
1398             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1399                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1400
1401             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1402             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1403             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1404             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1405             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1406             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1407             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1408             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1409             j_coord_offsetA  = DIM*jnrA;
1410             j_coord_offsetB  = DIM*jnrB;
1411             j_coord_offsetC  = DIM*jnrC;
1412             j_coord_offsetD  = DIM*jnrD;
1413             j_coord_offsetE  = DIM*jnrE;
1414             j_coord_offsetF  = DIM*jnrF;
1415             j_coord_offsetG  = DIM*jnrG;
1416             j_coord_offsetH  = DIM*jnrH;
1417
1418             /* load j atom coordinates */
1419             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1420                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1421                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1422                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1423                                                  &jx0,&jy0,&jz0);
1424
1425             /* Calculate displacement vector */
1426             dx00             = _mm256_sub_ps(ix0,jx0);
1427             dy00             = _mm256_sub_ps(iy0,jy0);
1428             dz00             = _mm256_sub_ps(iz0,jz0);
1429             dx10             = _mm256_sub_ps(ix1,jx0);
1430             dy10             = _mm256_sub_ps(iy1,jy0);
1431             dz10             = _mm256_sub_ps(iz1,jz0);
1432             dx20             = _mm256_sub_ps(ix2,jx0);
1433             dy20             = _mm256_sub_ps(iy2,jy0);
1434             dz20             = _mm256_sub_ps(iz2,jz0);
1435             dx30             = _mm256_sub_ps(ix3,jx0);
1436             dy30             = _mm256_sub_ps(iy3,jy0);
1437             dz30             = _mm256_sub_ps(iz3,jz0);
1438
1439             /* Calculate squared distance and things based on it */
1440             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1441             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1442             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1443             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1444
1445             rinv00           = avx256_invsqrt_f(rsq00);
1446             rinv10           = avx256_invsqrt_f(rsq10);
1447             rinv20           = avx256_invsqrt_f(rsq20);
1448             rinv30           = avx256_invsqrt_f(rsq30);
1449
1450             /* Load parameters for j particles */
1451             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1452                                                                  charge+jnrC+0,charge+jnrD+0,
1453                                                                  charge+jnrE+0,charge+jnrF+0,
1454                                                                  charge+jnrG+0,charge+jnrH+0);
1455             vdwjidx0A        = 2*vdwtype[jnrA+0];
1456             vdwjidx0B        = 2*vdwtype[jnrB+0];
1457             vdwjidx0C        = 2*vdwtype[jnrC+0];
1458             vdwjidx0D        = 2*vdwtype[jnrD+0];
1459             vdwjidx0E        = 2*vdwtype[jnrE+0];
1460             vdwjidx0F        = 2*vdwtype[jnrF+0];
1461             vdwjidx0G        = 2*vdwtype[jnrG+0];
1462             vdwjidx0H        = 2*vdwtype[jnrH+0];
1463
1464             fjx0             = _mm256_setzero_ps();
1465             fjy0             = _mm256_setzero_ps();
1466             fjz0             = _mm256_setzero_ps();
1467
1468             /**************************
1469              * CALCULATE INTERACTIONS *
1470              **************************/
1471
1472             r00              = _mm256_mul_ps(rsq00,rinv00);
1473             r00              = _mm256_andnot_ps(dummy_mask,r00);
1474
1475             /* Compute parameters for interactions between i and j atoms */
1476             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1477                                             vdwioffsetptr0+vdwjidx0B,
1478                                             vdwioffsetptr0+vdwjidx0C,
1479                                             vdwioffsetptr0+vdwjidx0D,
1480                                             vdwioffsetptr0+vdwjidx0E,
1481                                             vdwioffsetptr0+vdwjidx0F,
1482                                             vdwioffsetptr0+vdwjidx0G,
1483                                             vdwioffsetptr0+vdwjidx0H,
1484                                             &c6_00,&c12_00);
1485
1486             /* Calculate table index by multiplying r with table scale and truncate to integer */
1487             rt               = _mm256_mul_ps(r00,vftabscale);
1488             vfitab           = _mm256_cvttps_epi32(rt);
1489             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1490             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1491             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1492             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1493             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1494             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1495
1496             /* CUBIC SPLINE TABLE DISPERSION */
1497             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1498             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1499             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1500                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1501             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1502                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1503             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1504                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1505             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1506                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1507             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1508             Heps             = _mm256_mul_ps(vfeps,H);
1509             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1510             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1511             fvdw6            = _mm256_mul_ps(c6_00,FF);
1512
1513             /* CUBIC SPLINE TABLE REPULSION */
1514             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1515             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1516             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1517                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1518             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1519                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1520             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1521                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1522             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1523                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1524             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1525             Heps             = _mm256_mul_ps(vfeps,H);
1526             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1527             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1528             fvdw12           = _mm256_mul_ps(c12_00,FF);
1529             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1530
1531             fscal            = fvdw;
1532
1533             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1534
1535             /* Calculate temporary vectorial force */
1536             tx               = _mm256_mul_ps(fscal,dx00);
1537             ty               = _mm256_mul_ps(fscal,dy00);
1538             tz               = _mm256_mul_ps(fscal,dz00);
1539
1540             /* Update vectorial force */
1541             fix0             = _mm256_add_ps(fix0,tx);
1542             fiy0             = _mm256_add_ps(fiy0,ty);
1543             fiz0             = _mm256_add_ps(fiz0,tz);
1544
1545             fjx0             = _mm256_add_ps(fjx0,tx);
1546             fjy0             = _mm256_add_ps(fjy0,ty);
1547             fjz0             = _mm256_add_ps(fjz0,tz);
1548
1549             /**************************
1550              * CALCULATE INTERACTIONS *
1551              **************************/
1552
1553             r10              = _mm256_mul_ps(rsq10,rinv10);
1554             r10              = _mm256_andnot_ps(dummy_mask,r10);
1555
1556             /* Compute parameters for interactions between i and j atoms */
1557             qq10             = _mm256_mul_ps(iq1,jq0);
1558
1559             /* Calculate table index by multiplying r with table scale and truncate to integer */
1560             rt               = _mm256_mul_ps(r10,vftabscale);
1561             vfitab           = _mm256_cvttps_epi32(rt);
1562             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1563             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1564             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1565             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1566             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1567             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1568
1569             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1570             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1571                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1572             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1573                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1574             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1576             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1578             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1579             Heps             = _mm256_mul_ps(vfeps,H);
1580             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1581             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1582             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1583
1584             fscal            = felec;
1585
1586             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1587
1588             /* Calculate temporary vectorial force */
1589             tx               = _mm256_mul_ps(fscal,dx10);
1590             ty               = _mm256_mul_ps(fscal,dy10);
1591             tz               = _mm256_mul_ps(fscal,dz10);
1592
1593             /* Update vectorial force */
1594             fix1             = _mm256_add_ps(fix1,tx);
1595             fiy1             = _mm256_add_ps(fiy1,ty);
1596             fiz1             = _mm256_add_ps(fiz1,tz);
1597
1598             fjx0             = _mm256_add_ps(fjx0,tx);
1599             fjy0             = _mm256_add_ps(fjy0,ty);
1600             fjz0             = _mm256_add_ps(fjz0,tz);
1601
1602             /**************************
1603              * CALCULATE INTERACTIONS *
1604              **************************/
1605
1606             r20              = _mm256_mul_ps(rsq20,rinv20);
1607             r20              = _mm256_andnot_ps(dummy_mask,r20);
1608
1609             /* Compute parameters for interactions between i and j atoms */
1610             qq20             = _mm256_mul_ps(iq2,jq0);
1611
1612             /* Calculate table index by multiplying r with table scale and truncate to integer */
1613             rt               = _mm256_mul_ps(r20,vftabscale);
1614             vfitab           = _mm256_cvttps_epi32(rt);
1615             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1616             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1617             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1618             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1619             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1620             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1621
1622             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1623             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1625             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1627             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1628                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1629             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1630                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1631             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1632             Heps             = _mm256_mul_ps(vfeps,H);
1633             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1634             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1635             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1636
1637             fscal            = felec;
1638
1639             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1640
1641             /* Calculate temporary vectorial force */
1642             tx               = _mm256_mul_ps(fscal,dx20);
1643             ty               = _mm256_mul_ps(fscal,dy20);
1644             tz               = _mm256_mul_ps(fscal,dz20);
1645
1646             /* Update vectorial force */
1647             fix2             = _mm256_add_ps(fix2,tx);
1648             fiy2             = _mm256_add_ps(fiy2,ty);
1649             fiz2             = _mm256_add_ps(fiz2,tz);
1650
1651             fjx0             = _mm256_add_ps(fjx0,tx);
1652             fjy0             = _mm256_add_ps(fjy0,ty);
1653             fjz0             = _mm256_add_ps(fjz0,tz);
1654
1655             /**************************
1656              * CALCULATE INTERACTIONS *
1657              **************************/
1658
1659             r30              = _mm256_mul_ps(rsq30,rinv30);
1660             r30              = _mm256_andnot_ps(dummy_mask,r30);
1661
1662             /* Compute parameters for interactions between i and j atoms */
1663             qq30             = _mm256_mul_ps(iq3,jq0);
1664
1665             /* Calculate table index by multiplying r with table scale and truncate to integer */
1666             rt               = _mm256_mul_ps(r30,vftabscale);
1667             vfitab           = _mm256_cvttps_epi32(rt);
1668             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1669             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1670             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1671             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1672             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1673             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1674
1675             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1676             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1677                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1678             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1679                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1680             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1681                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1682             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1683                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1684             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1685             Heps             = _mm256_mul_ps(vfeps,H);
1686             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1687             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1688             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1689
1690             fscal            = felec;
1691
1692             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1693
1694             /* Calculate temporary vectorial force */
1695             tx               = _mm256_mul_ps(fscal,dx30);
1696             ty               = _mm256_mul_ps(fscal,dy30);
1697             tz               = _mm256_mul_ps(fscal,dz30);
1698
1699             /* Update vectorial force */
1700             fix3             = _mm256_add_ps(fix3,tx);
1701             fiy3             = _mm256_add_ps(fiy3,ty);
1702             fiz3             = _mm256_add_ps(fiz3,tz);
1703
1704             fjx0             = _mm256_add_ps(fjx0,tx);
1705             fjy0             = _mm256_add_ps(fjy0,ty);
1706             fjz0             = _mm256_add_ps(fjz0,tz);
1707
1708             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1709             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1710             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1711             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1712             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1713             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1714             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1715             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1716
1717             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1718
1719             /* Inner loop uses 172 flops */
1720         }
1721
1722         /* End of innermost loop */
1723
1724         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1725                                                  f+i_coord_offset,fshift+i_shift_offset);
1726
1727         /* Increment number of inner iterations */
1728         inneriter                  += j_index_end - j_index_start;
1729
1730         /* Outer loop uses 24 flops */
1731     }
1732
1733     /* Increment number of outer iterations */
1734     outeriter        += nri;
1735
1736     /* Update outer/inner flops */
1737
1738     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*172);
1739 }