src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecGB_VdwCSTab_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: GeneralizedBorn
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m256i          gbitab;
  94     __m128i          gbitab_lo,gbitab_hi;
  95     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  96     __m256           minushalf = _mm256_set1_ps(-0.5);
  97     real             *invsqrta,*dvda,*gbtab;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256i          vfitab;
 105     __m128i          vfitab_lo,vfitab_hi;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256           dummy_mask,cutoff_mask;
 110     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 111     __m256           one     = _mm256_set1_ps(1.0);
 112     __m256           two     = _mm256_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 132
 133     invsqrta         = fr->invsqrta;
 134     dvda             = fr->dvda;
 135     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 136     gbtab            = fr->gbtab.data;
 137     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145     j_coord_offsetE = 0;
 146     j_coord_offsetF = 0;
 147     j_coord_offsetG = 0;
 148     j_coord_offsetH = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 174
 175         fix0             = _mm256_setzero_ps();
 176         fiy0             = _mm256_setzero_ps();
 177         fiz0             = _mm256_setzero_ps();
 178
 179         /* Load parameters for i particles */
 180         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 181         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 182         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm256_setzero_ps();
 186         vgbsum           = _mm256_setzero_ps();
 187         vvdwsum          = _mm256_setzero_ps();
 188         dvdasum          = _mm256_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             jnrE             = jjnr[jidx+4];
 200             jnrF             = jjnr[jidx+5];
 201             jnrG             = jjnr[jidx+6];
 202             jnrH             = jjnr[jidx+7];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207             j_coord_offsetE  = DIM*jnrE;
 208             j_coord_offsetF  = DIM*jnrF;
 209             j_coord_offsetG  = DIM*jnrG;
 210             j_coord_offsetH  = DIM*jnrH;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 216                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 217                                                  &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm256_sub_ps(ix0,jx0);
 221             dy00             = _mm256_sub_ps(iy0,jy0);
 222             dz00             = _mm256_sub_ps(iz0,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 226
 227             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 228
 229             /* Load parameters for j particles */
 230             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 231                                                                  charge+jnrC+0,charge+jnrD+0,
 232                                                                  charge+jnrE+0,charge+jnrF+0,
 233                                                                  charge+jnrG+0,charge+jnrH+0);
 234             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 235                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 236                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 237                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242             vdwjidx0E        = 2*vdwtype[jnrE+0];
 243             vdwjidx0F        = 2*vdwtype[jnrF+0];
 244             vdwjidx0G        = 2*vdwtype[jnrG+0];
 245             vdwjidx0H        = 2*vdwtype[jnrH+0];
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             r00              = _mm256_mul_ps(rsq00,rinv00);
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             qq00             = _mm256_mul_ps(iq0,jq0);
 255             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 256                                             vdwioffsetptr0+vdwjidx0B,
 257                                             vdwioffsetptr0+vdwjidx0C,
 258                                             vdwioffsetptr0+vdwjidx0D,
 259                                             vdwioffsetptr0+vdwjidx0E,
 260                                             vdwioffsetptr0+vdwjidx0F,
 261                                             vdwioffsetptr0+vdwjidx0G,
 262                                             vdwioffsetptr0+vdwjidx0H,
 263                                             &c6_00,&c12_00);
 264
 265             /* Calculate table index by multiplying r with table scale and truncate to integer */
 266             rt               = _mm256_mul_ps(r00,vftabscale);
 267             vfitab           = _mm256_cvttps_epi32(rt);
 268             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 269             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 270             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 271             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 272             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 273             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 274
 275             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 276             isaprod          = _mm256_mul_ps(isai0,isaj0);
 277             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 278             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 279
 280             /* Calculate generalized born table index - this is a separate table from the normal one,
 281              * but we use the same procedure by multiplying r with scale and truncating to integer.
 282              */
 283             rt               = _mm256_mul_ps(r00,gbscale);
 284             gbitab           = _mm256_cvttps_epi32(rt);
 285             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 286             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 287             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 288             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 289             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 290             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 291             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 292                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 293             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 294                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 295             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 296                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 297             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 298                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 299             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 300             Heps             = _mm256_mul_ps(gbeps,H);
 301             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 302             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 303             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 304
 305             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 306             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 307             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 308             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 309             fjptrA           = dvda+jnrA;
 310             fjptrB           = dvda+jnrB;
 311             fjptrC           = dvda+jnrC;
 312             fjptrD           = dvda+jnrD;
 313             fjptrE           = dvda+jnrE;
 314             fjptrF           = dvda+jnrF;
 315             fjptrG           = dvda+jnrG;
 316             fjptrH           = dvda+jnrH;
 317             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 318                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 319             velec            = _mm256_mul_ps(qq00,rinv00);
 320             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 321
 322             /* CUBIC SPLINE TABLE DISPERSION */
 323             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 325             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 327             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 329             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 331             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 332             Heps             = _mm256_mul_ps(vfeps,H);
 333             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 334             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 335             vvdw6            = _mm256_mul_ps(c6_00,VV);
 336             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 337             fvdw6            = _mm256_mul_ps(c6_00,FF);
 338
 339             /* CUBIC SPLINE TABLE REPULSION */
 340             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 341             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 342             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 343                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 344             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 345                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 346             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 347                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 348             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 349                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 350             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 351             Heps             = _mm256_mul_ps(vfeps,H);
 352             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 353             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 354             vvdw12           = _mm256_mul_ps(c12_00,VV);
 355             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 356             fvdw12           = _mm256_mul_ps(c12_00,FF);
 357             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 358             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm256_add_ps(velecsum,velec);
 362             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 363             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 364
 365             fscal            = _mm256_add_ps(felec,fvdw);
 366
 367             /* Calculate temporary vectorial force */
 368             tx               = _mm256_mul_ps(fscal,dx00);
 369             ty               = _mm256_mul_ps(fscal,dy00);
 370             tz               = _mm256_mul_ps(fscal,dz00);
 371
 372             /* Update vectorial force */
 373             fix0             = _mm256_add_ps(fix0,tx);
 374             fiy0             = _mm256_add_ps(fiy0,ty);
 375             fiz0             = _mm256_add_ps(fiz0,tz);
 376
 377             fjptrA             = f+j_coord_offsetA;
 378             fjptrB             = f+j_coord_offsetB;
 379             fjptrC             = f+j_coord_offsetC;
 380             fjptrD             = f+j_coord_offsetD;
 381             fjptrE             = f+j_coord_offsetE;
 382             fjptrF             = f+j_coord_offsetF;
 383             fjptrG             = f+j_coord_offsetG;
 384             fjptrH             = f+j_coord_offsetH;
 385             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 386
 387             /* Inner loop uses 91 flops */
 388         }
 389
 390         if(jidx<j_index_end)
 391         {
 392
 393             /* Get j neighbor index, and coordinate index */
 394             jnrlistA         = jjnr[jidx];
 395             jnrlistB         = jjnr[jidx+1];
 396             jnrlistC         = jjnr[jidx+2];
 397             jnrlistD         = jjnr[jidx+3];
 398             jnrlistE         = jjnr[jidx+4];
 399             jnrlistF         = jjnr[jidx+5];
 400             jnrlistG         = jjnr[jidx+6];
 401             jnrlistH         = jjnr[jidx+7];
 402             /* Sign of each element will be negative for non-real atoms.
 403              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 404              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 405              */
 406             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 407                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 408
 409             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 410             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 411             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 412             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 413             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 414             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 415             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 416             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 417             j_coord_offsetA  = DIM*jnrA;
 418             j_coord_offsetB  = DIM*jnrB;
 419             j_coord_offsetC  = DIM*jnrC;
 420             j_coord_offsetD  = DIM*jnrD;
 421             j_coord_offsetE  = DIM*jnrE;
 422             j_coord_offsetF  = DIM*jnrF;
 423             j_coord_offsetG  = DIM*jnrG;
 424             j_coord_offsetH  = DIM*jnrH;
 425
 426             /* load j atom coordinates */
 427             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 428                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 429                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 430                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 431                                                  &jx0,&jy0,&jz0);
 432
 433             /* Calculate displacement vector */
 434             dx00             = _mm256_sub_ps(ix0,jx0);
 435             dy00             = _mm256_sub_ps(iy0,jy0);
 436             dz00             = _mm256_sub_ps(iz0,jz0);
 437
 438             /* Calculate squared distance and things based on it */
 439             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 440
 441             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 442
 443             /* Load parameters for j particles */
 444             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 445                                                                  charge+jnrC+0,charge+jnrD+0,
 446                                                                  charge+jnrE+0,charge+jnrF+0,
 447                                                                  charge+jnrG+0,charge+jnrH+0);
 448             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 449                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 450                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 451                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 452             vdwjidx0A        = 2*vdwtype[jnrA+0];
 453             vdwjidx0B        = 2*vdwtype[jnrB+0];
 454             vdwjidx0C        = 2*vdwtype[jnrC+0];
 455             vdwjidx0D        = 2*vdwtype[jnrD+0];
 456             vdwjidx0E        = 2*vdwtype[jnrE+0];
 457             vdwjidx0F        = 2*vdwtype[jnrF+0];
 458             vdwjidx0G        = 2*vdwtype[jnrG+0];
 459             vdwjidx0H        = 2*vdwtype[jnrH+0];
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             r00              = _mm256_mul_ps(rsq00,rinv00);
 466             r00              = _mm256_andnot_ps(dummy_mask,r00);
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             qq00             = _mm256_mul_ps(iq0,jq0);
 470             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 471                                             vdwioffsetptr0+vdwjidx0B,
 472                                             vdwioffsetptr0+vdwjidx0C,
 473                                             vdwioffsetptr0+vdwjidx0D,
 474                                             vdwioffsetptr0+vdwjidx0E,
 475                                             vdwioffsetptr0+vdwjidx0F,
 476                                             vdwioffsetptr0+vdwjidx0G,
 477                                             vdwioffsetptr0+vdwjidx0H,
 478                                             &c6_00,&c12_00);
 479
 480             /* Calculate table index by multiplying r with table scale and truncate to integer */
 481             rt               = _mm256_mul_ps(r00,vftabscale);
 482             vfitab           = _mm256_cvttps_epi32(rt);
 483             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 484             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 485             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 486             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 487             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 488             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 489
 490             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 491             isaprod          = _mm256_mul_ps(isai0,isaj0);
 492             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 493             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 494
 495             /* Calculate generalized born table index - this is a separate table from the normal one,
 496              * but we use the same procedure by multiplying r with scale and truncating to integer.
 497              */
 498             rt               = _mm256_mul_ps(r00,gbscale);
 499             gbitab           = _mm256_cvttps_epi32(rt);
 500             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 501             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 502             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 503             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 504             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 505             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 506             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 507                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 508             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 509                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 510             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 511                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 512             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 513                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 514             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 515             Heps             = _mm256_mul_ps(gbeps,H);
 516             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 517             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 518             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 519
 520             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 521             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 522             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 523             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 524             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 525             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 526             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 527             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 528             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 529             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 530             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 531             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 532             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 533             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 534             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 535                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 536             velec            = _mm256_mul_ps(qq00,rinv00);
 537             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 538
 539             /* CUBIC SPLINE TABLE DISPERSION */
 540             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 541                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 542             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 543                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 544             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 545                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 546             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 547                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 548             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 549             Heps             = _mm256_mul_ps(vfeps,H);
 550             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 551             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 552             vvdw6            = _mm256_mul_ps(c6_00,VV);
 553             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 554             fvdw6            = _mm256_mul_ps(c6_00,FF);
 555
 556             /* CUBIC SPLINE TABLE REPULSION */
 557             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 558             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 559             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 560                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 561             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 562                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 563             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 564                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 565             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 566                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 567             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 568             Heps             = _mm256_mul_ps(vfeps,H);
 569             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 570             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 571             vvdw12           = _mm256_mul_ps(c12_00,VV);
 572             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 573             fvdw12           = _mm256_mul_ps(c12_00,FF);
 574             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 575             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_andnot_ps(dummy_mask,velec);
 579             velecsum         = _mm256_add_ps(velecsum,velec);
 580             vgb              = _mm256_andnot_ps(dummy_mask,vgb);
 581             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 582             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 583             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 584
 585             fscal            = _mm256_add_ps(felec,fvdw);
 586
 587             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 588
 589             /* Calculate temporary vectorial force */
 590             tx               = _mm256_mul_ps(fscal,dx00);
 591             ty               = _mm256_mul_ps(fscal,dy00);
 592             tz               = _mm256_mul_ps(fscal,dz00);
 593
 594             /* Update vectorial force */
 595             fix0             = _mm256_add_ps(fix0,tx);
 596             fiy0             = _mm256_add_ps(fiy0,ty);
 597             fiz0             = _mm256_add_ps(fiz0,tz);
 598
 599             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 600             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 601             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 602             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 603             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 604             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 605             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 606             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 607             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 608
 609             /* Inner loop uses 92 flops */
 610         }
 611
 612         /* End of innermost loop */
 613
 614         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 615                                                  f+i_coord_offset,fshift+i_shift_offset);
 616
 617         ggid                        = gid[iidx];
 618         /* Update potential energies */
 619         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 620         gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 621         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 622         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 623         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 624
 625         /* Increment number of inner iterations */
 626         inneriter                  += j_index_end - j_index_start;
 627
 628         /* Outer loop uses 10 flops */
 629     }
 630
 631     /* Increment number of outer iterations */
 632     outeriter        += nri;
 633
 634     /* Update outer/inner flops */
 635
 636     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*92);
 637 }
 638 /*
 639  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_avx_256_single
 640  * Electrostatics interaction: GeneralizedBorn
 641  * VdW interaction:            CubicSplineTable
 642  * Geometry:                   Particle-Particle
 643  * Calculate force/pot:        Force
 644  */
 645 void
 646 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_avx_256_single
 647                     (t_nblist                    * gmx_restrict       nlist,
 648                      rvec                        * gmx_restrict          xx,
 649                      rvec                        * gmx_restrict          ff,
 650                      t_forcerec                  * gmx_restrict          fr,
 651                      t_mdatoms                   * gmx_restrict     mdatoms,
 652                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 653                      t_nrnb                      * gmx_restrict        nrnb)
 654 {
 655     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 656      * just 0 for non-waters.
 657      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 658      * jnr indices corresponding to data put in the four positions in the SIMD register.
 659      */
 660     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 661     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 662     int              jnrA,jnrB,jnrC,jnrD;
 663     int              jnrE,jnrF,jnrG,jnrH;
 664     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 665     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 666     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 667     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 668     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 669     real             rcutoff_scalar;
 670     real             *shiftvec,*fshift,*x,*f;
 671     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 672     real             scratch[4*DIM];
 673     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 674     real *           vdwioffsetptr0;
 675     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 676     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 677     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 678     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 679     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 680     real             *charge;
 681     __m256i          gbitab;
 682     __m128i          gbitab_lo,gbitab_hi;
 683     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 684     __m256           minushalf = _mm256_set1_ps(-0.5);
 685     real             *invsqrta,*dvda,*gbtab;
 686     int              nvdwtype;
 687     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 688     int              *vdwtype;
 689     real             *vdwparam;
 690     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 691     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 692     __m256i          vfitab;
 693     __m128i          vfitab_lo,vfitab_hi;
 694     __m128i          ifour       = _mm_set1_epi32(4);
 695     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 696     real             *vftab;
 697     __m256           dummy_mask,cutoff_mask;
 698     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 699     __m256           one     = _mm256_set1_ps(1.0);
 700     __m256           two     = _mm256_set1_ps(2.0);
 701     x                = xx[0];
 702     f                = ff[0];
 703
 704     nri              = nlist->nri;
 705     iinr             = nlist->iinr;
 706     jindex           = nlist->jindex;
 707     jjnr             = nlist->jjnr;
 708     shiftidx         = nlist->shift;
 709     gid              = nlist->gid;
 710     shiftvec         = fr->shift_vec[0];
 711     fshift           = fr->fshift[0];
 712     facel            = _mm256_set1_ps(fr->epsfac);
 713     charge           = mdatoms->chargeA;
 714     nvdwtype         = fr->ntype;
 715     vdwparam         = fr->nbfp;
 716     vdwtype          = mdatoms->typeA;
 717
 718     vftab            = kernel_data->table_vdw->data;
 719     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 720
 721     invsqrta         = fr->invsqrta;
 722     dvda             = fr->dvda;
 723     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 724     gbtab            = fr->gbtab.data;
 725     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 726
 727     /* Avoid stupid compiler warnings */
 728     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 729     j_coord_offsetA = 0;
 730     j_coord_offsetB = 0;
 731     j_coord_offsetC = 0;
 732     j_coord_offsetD = 0;
 733     j_coord_offsetE = 0;
 734     j_coord_offsetF = 0;
 735     j_coord_offsetG = 0;
 736     j_coord_offsetH = 0;
 737
 738     outeriter        = 0;
 739     inneriter        = 0;
 740
 741     for(iidx=0;iidx<4*DIM;iidx++)
 742     {
 743         scratch[iidx] = 0.0;
 744     }
 745
 746     /* Start outer loop over neighborlists */
 747     for(iidx=0; iidx<nri; iidx++)
 748     {
 749         /* Load shift vector for this list */
 750         i_shift_offset   = DIM*shiftidx[iidx];
 751
 752         /* Load limits for loop over neighbors */
 753         j_index_start    = jindex[iidx];
 754         j_index_end      = jindex[iidx+1];
 755
 756         /* Get outer coordinate index */
 757         inr              = iinr[iidx];
 758         i_coord_offset   = DIM*inr;
 759
 760         /* Load i particle coords and add shift vector */
 761         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 762
 763         fix0             = _mm256_setzero_ps();
 764         fiy0             = _mm256_setzero_ps();
 765         fiz0             = _mm256_setzero_ps();
 766
 767         /* Load parameters for i particles */
 768         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 769         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 770         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 771
 772         dvdasum          = _mm256_setzero_ps();
 773
 774         /* Start inner kernel loop */
 775         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 776         {
 777
 778             /* Get j neighbor index, and coordinate index */
 779             jnrA             = jjnr[jidx];
 780             jnrB             = jjnr[jidx+1];
 781             jnrC             = jjnr[jidx+2];
 782             jnrD             = jjnr[jidx+3];
 783             jnrE             = jjnr[jidx+4];
 784             jnrF             = jjnr[jidx+5];
 785             jnrG             = jjnr[jidx+6];
 786             jnrH             = jjnr[jidx+7];
 787             j_coord_offsetA  = DIM*jnrA;
 788             j_coord_offsetB  = DIM*jnrB;
 789             j_coord_offsetC  = DIM*jnrC;
 790             j_coord_offsetD  = DIM*jnrD;
 791             j_coord_offsetE  = DIM*jnrE;
 792             j_coord_offsetF  = DIM*jnrF;
 793             j_coord_offsetG  = DIM*jnrG;
 794             j_coord_offsetH  = DIM*jnrH;
 795
 796             /* load j atom coordinates */
 797             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 798                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 799                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 800                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 801                                                  &jx0,&jy0,&jz0);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm256_sub_ps(ix0,jx0);
 805             dy00             = _mm256_sub_ps(iy0,jy0);
 806             dz00             = _mm256_sub_ps(iz0,jz0);
 807
 808             /* Calculate squared distance and things based on it */
 809             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 810
 811             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 812
 813             /* Load parameters for j particles */
 814             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 815                                                                  charge+jnrC+0,charge+jnrD+0,
 816                                                                  charge+jnrE+0,charge+jnrF+0,
 817                                                                  charge+jnrG+0,charge+jnrH+0);
 818             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 819                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 820                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 821                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 822             vdwjidx0A        = 2*vdwtype[jnrA+0];
 823             vdwjidx0B        = 2*vdwtype[jnrB+0];
 824             vdwjidx0C        = 2*vdwtype[jnrC+0];
 825             vdwjidx0D        = 2*vdwtype[jnrD+0];
 826             vdwjidx0E        = 2*vdwtype[jnrE+0];
 827             vdwjidx0F        = 2*vdwtype[jnrF+0];
 828             vdwjidx0G        = 2*vdwtype[jnrG+0];
 829             vdwjidx0H        = 2*vdwtype[jnrH+0];
 830
 831             /**************************
 832              * CALCULATE INTERACTIONS *
 833              **************************/
 834
 835             r00              = _mm256_mul_ps(rsq00,rinv00);
 836
 837             /* Compute parameters for interactions between i and j atoms */
 838             qq00             = _mm256_mul_ps(iq0,jq0);
 839             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 840                                             vdwioffsetptr0+vdwjidx0B,
 841                                             vdwioffsetptr0+vdwjidx0C,
 842                                             vdwioffsetptr0+vdwjidx0D,
 843                                             vdwioffsetptr0+vdwjidx0E,
 844                                             vdwioffsetptr0+vdwjidx0F,
 845                                             vdwioffsetptr0+vdwjidx0G,
 846                                             vdwioffsetptr0+vdwjidx0H,
 847                                             &c6_00,&c12_00);
 848
 849             /* Calculate table index by multiplying r with table scale and truncate to integer */
 850             rt               = _mm256_mul_ps(r00,vftabscale);
 851             vfitab           = _mm256_cvttps_epi32(rt);
 852             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 853             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 854             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 855             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 856             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 857             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 858
 859             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 860             isaprod          = _mm256_mul_ps(isai0,isaj0);
 861             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 862             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 863
 864             /* Calculate generalized born table index - this is a separate table from the normal one,
 865              * but we use the same procedure by multiplying r with scale and truncating to integer.
 866              */
 867             rt               = _mm256_mul_ps(r00,gbscale);
 868             gbitab           = _mm256_cvttps_epi32(rt);
 869             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 870             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 871             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 872             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 873             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 874             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 875             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 876                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 877             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 878                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 879             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 880                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 881             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 882                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 883             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 884             Heps             = _mm256_mul_ps(gbeps,H);
 885             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 886             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 887             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 888
 889             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 890             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 891             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 892             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 893             fjptrA           = dvda+jnrA;
 894             fjptrB           = dvda+jnrB;
 895             fjptrC           = dvda+jnrC;
 896             fjptrD           = dvda+jnrD;
 897             fjptrE           = dvda+jnrE;
 898             fjptrF           = dvda+jnrF;
 899             fjptrG           = dvda+jnrG;
 900             fjptrH           = dvda+jnrH;
 901             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 902                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 903             velec            = _mm256_mul_ps(qq00,rinv00);
 904             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 905
 906             /* CUBIC SPLINE TABLE DISPERSION */
 907             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 908                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 909             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 910                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 911             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 912                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 913             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 914                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 915             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 916             Heps             = _mm256_mul_ps(vfeps,H);
 917             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 918             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 919             fvdw6            = _mm256_mul_ps(c6_00,FF);
 920
 921             /* CUBIC SPLINE TABLE REPULSION */
 922             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 923             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 924             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 925                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 926             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 927                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 928             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 929                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 930             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 931                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 932             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 933             Heps             = _mm256_mul_ps(vfeps,H);
 934             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 935             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 936             fvdw12           = _mm256_mul_ps(c12_00,FF);
 937             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 938
 939             fscal            = _mm256_add_ps(felec,fvdw);
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm256_mul_ps(fscal,dx00);
 943             ty               = _mm256_mul_ps(fscal,dy00);
 944             tz               = _mm256_mul_ps(fscal,dz00);
 945
 946             /* Update vectorial force */
 947             fix0             = _mm256_add_ps(fix0,tx);
 948             fiy0             = _mm256_add_ps(fiy0,ty);
 949             fiz0             = _mm256_add_ps(fiz0,tz);
 950
 951             fjptrA             = f+j_coord_offsetA;
 952             fjptrB             = f+j_coord_offsetB;
 953             fjptrC             = f+j_coord_offsetC;
 954             fjptrD             = f+j_coord_offsetD;
 955             fjptrE             = f+j_coord_offsetE;
 956             fjptrF             = f+j_coord_offsetF;
 957             fjptrG             = f+j_coord_offsetG;
 958             fjptrH             = f+j_coord_offsetH;
 959             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 960
 961             /* Inner loop uses 81 flops */
 962         }
 963
 964         if(jidx<j_index_end)
 965         {
 966
 967             /* Get j neighbor index, and coordinate index */
 968             jnrlistA         = jjnr[jidx];
 969             jnrlistB         = jjnr[jidx+1];
 970             jnrlistC         = jjnr[jidx+2];
 971             jnrlistD         = jjnr[jidx+3];
 972             jnrlistE         = jjnr[jidx+4];
 973             jnrlistF         = jjnr[jidx+5];
 974             jnrlistG         = jjnr[jidx+6];
 975             jnrlistH         = jjnr[jidx+7];
 976             /* Sign of each element will be negative for non-real atoms.
 977              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 978              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 979              */
 980             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 981                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 982
 983             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 984             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 985             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 986             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 987             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 988             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 989             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 990             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 991             j_coord_offsetA  = DIM*jnrA;
 992             j_coord_offsetB  = DIM*jnrB;
 993             j_coord_offsetC  = DIM*jnrC;
 994             j_coord_offsetD  = DIM*jnrD;
 995             j_coord_offsetE  = DIM*jnrE;
 996             j_coord_offsetF  = DIM*jnrF;
 997             j_coord_offsetG  = DIM*jnrG;
 998             j_coord_offsetH  = DIM*jnrH;
 999
1000             /* load j atom coordinates */
1001             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1002                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1003                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1004                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1005                                                  &jx0,&jy0,&jz0);
1006
1007             /* Calculate displacement vector */
1008             dx00             = _mm256_sub_ps(ix0,jx0);
1009             dy00             = _mm256_sub_ps(iy0,jy0);
1010             dz00             = _mm256_sub_ps(iz0,jz0);
1011
1012             /* Calculate squared distance and things based on it */
1013             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1014
1015             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1016
1017             /* Load parameters for j particles */
1018             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1019                                                                  charge+jnrC+0,charge+jnrD+0,
1020                                                                  charge+jnrE+0,charge+jnrF+0,
1021                                                                  charge+jnrG+0,charge+jnrH+0);
1022             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
1023                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
1024                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
1025                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
1026             vdwjidx0A        = 2*vdwtype[jnrA+0];
1027             vdwjidx0B        = 2*vdwtype[jnrB+0];
1028             vdwjidx0C        = 2*vdwtype[jnrC+0];
1029             vdwjidx0D        = 2*vdwtype[jnrD+0];
1030             vdwjidx0E        = 2*vdwtype[jnrE+0];
1031             vdwjidx0F        = 2*vdwtype[jnrF+0];
1032             vdwjidx0G        = 2*vdwtype[jnrG+0];
1033             vdwjidx0H        = 2*vdwtype[jnrH+0];
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             r00              = _mm256_mul_ps(rsq00,rinv00);
1040             r00              = _mm256_andnot_ps(dummy_mask,r00);
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             qq00             = _mm256_mul_ps(iq0,jq0);
1044             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1045                                             vdwioffsetptr0+vdwjidx0B,
1046                                             vdwioffsetptr0+vdwjidx0C,
1047                                             vdwioffsetptr0+vdwjidx0D,
1048                                             vdwioffsetptr0+vdwjidx0E,
1049                                             vdwioffsetptr0+vdwjidx0F,
1050                                             vdwioffsetptr0+vdwjidx0G,
1051                                             vdwioffsetptr0+vdwjidx0H,
1052                                             &c6_00,&c12_00);
1053
1054             /* Calculate table index by multiplying r with table scale and truncate to integer */
1055             rt               = _mm256_mul_ps(r00,vftabscale);
1056             vfitab           = _mm256_cvttps_epi32(rt);
1057             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1058             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1059             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1060             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1061             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1062             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1063
1064             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
1065             isaprod          = _mm256_mul_ps(isai0,isaj0);
1066             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
1067             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
1068
1069             /* Calculate generalized born table index - this is a separate table from the normal one,
1070              * but we use the same procedure by multiplying r with scale and truncating to integer.
1071              */
1072             rt               = _mm256_mul_ps(r00,gbscale);
1073             gbitab           = _mm256_cvttps_epi32(rt);
1074             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1075             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1076             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
1077             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
1078             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
1079             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
1080             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
1081                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
1082             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
1083                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
1084             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
1085                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
1086             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
1087                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
1088             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1089             Heps             = _mm256_mul_ps(gbeps,H);
1090             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
1091             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
1092             vgb              = _mm256_mul_ps(gbqqfactor,VV);
1093
1094             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1095             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
1096             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
1097             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
1098             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
1099             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
1100             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
1101             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
1102             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
1103             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
1104             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
1105             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
1106             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
1107             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
1108             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1109                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
1110             velec            = _mm256_mul_ps(qq00,rinv00);
1111             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
1112
1113             /* CUBIC SPLINE TABLE DISPERSION */
1114             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1115                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1116             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1117                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1118             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1119                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1120             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1121                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1122             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1123             Heps             = _mm256_mul_ps(vfeps,H);
1124             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1125             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1126             fvdw6            = _mm256_mul_ps(c6_00,FF);
1127
1128             /* CUBIC SPLINE TABLE REPULSION */
1129             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1130             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1131             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1132                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1133             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1134                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1135             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1136                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1137             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1138                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1139             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1140             Heps             = _mm256_mul_ps(vfeps,H);
1141             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1142             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1143             fvdw12           = _mm256_mul_ps(c12_00,FF);
1144             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1145
1146             fscal            = _mm256_add_ps(felec,fvdw);
1147
1148             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm256_mul_ps(fscal,dx00);
1152             ty               = _mm256_mul_ps(fscal,dy00);
1153             tz               = _mm256_mul_ps(fscal,dz00);
1154
1155             /* Update vectorial force */
1156             fix0             = _mm256_add_ps(fix0,tx);
1157             fiy0             = _mm256_add_ps(fiy0,ty);
1158             fiz0             = _mm256_add_ps(fiz0,tz);
1159
1160             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1161             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1162             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1163             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1164             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1165             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1166             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1167             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1168             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1169
1170             /* Inner loop uses 82 flops */
1171         }
1172
1173         /* End of innermost loop */
1174
1175         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1176                                                  f+i_coord_offset,fshift+i_shift_offset);
1177
1178         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
1179         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
1180
1181         /* Increment number of inner iterations */
1182         inneriter                  += j_index_end - j_index_start;
1183
1184         /* Outer loop uses 7 flops */
1185     }
1186
1187     /* Increment number of outer iterations */
1188     outeriter        += nri;
1189
1190     /* Update outer/inner flops */
1191
1192     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*82);
1193 }