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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m256i          gbitab;
  96     __m128i          gbitab_lo,gbitab_hi;
  97     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  98     __m256           minushalf = _mm256_set1_ps(-0.5);
  99     real             *invsqrta,*dvda,*gbtab;
 100     int              nvdwtype;
 101     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 105     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 106     __m256i          vfitab;
 107     __m128i          vfitab_lo,vfitab_hi;
 108     __m128i          ifour       = _mm_set1_epi32(4);
 109     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 110     real             *vftab;
 111     __m256           dummy_mask,cutoff_mask;
 112     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 113     __m256           one     = _mm256_set1_ps(1.0);
 114     __m256           two     = _mm256_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 134
 135     invsqrta         = fr->invsqrta;
 136     dvda             = fr->dvda;
 137     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 138     gbtab            = fr->gbtab.data;
 139     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 176
 177         fix0             = _mm256_setzero_ps();
 178         fiy0             = _mm256_setzero_ps();
 179         fiz0             = _mm256_setzero_ps();
 180
 181         /* Load parameters for i particles */
 182         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 183         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 184         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm256_setzero_ps();
 188         vgbsum           = _mm256_setzero_ps();
 189         vvdwsum          = _mm256_setzero_ps();
 190         dvdasum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 228
 229             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                                  charge+jnrC+0,charge+jnrD+0,
 234                                                                  charge+jnrE+0,charge+jnrF+0,
 235                                                                  charge+jnrG+0,charge+jnrH+0);
 236             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 237                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 238                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 239                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244             vdwjidx0E        = 2*vdwtype[jnrE+0];
 245             vdwjidx0F        = 2*vdwtype[jnrF+0];
 246             vdwjidx0G        = 2*vdwtype[jnrG+0];
 247             vdwjidx0H        = 2*vdwtype[jnrH+0];
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             r00              = _mm256_mul_ps(rsq00,rinv00);
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq00             = _mm256_mul_ps(iq0,jq0);
 257             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 258                                             vdwioffsetptr0+vdwjidx0B,
 259                                             vdwioffsetptr0+vdwjidx0C,
 260                                             vdwioffsetptr0+vdwjidx0D,
 261                                             vdwioffsetptr0+vdwjidx0E,
 262                                             vdwioffsetptr0+vdwjidx0F,
 263                                             vdwioffsetptr0+vdwjidx0G,
 264                                             vdwioffsetptr0+vdwjidx0H,
 265                                             &c6_00,&c12_00);
 266
 267             /* Calculate table index by multiplying r with table scale and truncate to integer */
 268             rt               = _mm256_mul_ps(r00,vftabscale);
 269             vfitab           = _mm256_cvttps_epi32(rt);
 270             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 271             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 272             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 273             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 274             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 275             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 276
 277             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 278             isaprod          = _mm256_mul_ps(isai0,isaj0);
 279             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 280             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 281
 282             /* Calculate generalized born table index - this is a separate table from the normal one,
 283              * but we use the same procedure by multiplying r with scale and truncating to integer.
 284              */
 285             rt               = _mm256_mul_ps(r00,gbscale);
 286             gbitab           = _mm256_cvttps_epi32(rt);
 287             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 288             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 289             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 290             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 291             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 292             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 293             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 294                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 295             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 296                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 297             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 298                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 299             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 300                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 301             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 302             Heps             = _mm256_mul_ps(gbeps,H);
 303             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 304             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 305             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 306
 307             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 308             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 309             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 310             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 311             fjptrA           = dvda+jnrA;
 312             fjptrB           = dvda+jnrB;
 313             fjptrC           = dvda+jnrC;
 314             fjptrD           = dvda+jnrD;
 315             fjptrE           = dvda+jnrE;
 316             fjptrF           = dvda+jnrF;
 317             fjptrG           = dvda+jnrG;
 318             fjptrH           = dvda+jnrH;
 319             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 320                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 321             velec            = _mm256_mul_ps(qq00,rinv00);
 322             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 323
 324             /* CUBIC SPLINE TABLE DISPERSION */
 325             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 327             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 329             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 331             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 332                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 333             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 334             Heps             = _mm256_mul_ps(vfeps,H);
 335             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 336             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 337             vvdw6            = _mm256_mul_ps(c6_00,VV);
 338             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 339             fvdw6            = _mm256_mul_ps(c6_00,FF);
 340
 341             /* CUBIC SPLINE TABLE REPULSION */
 342             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 343             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 344             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 345                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 346             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 347                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 348             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 349                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 350             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 351                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 352             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 353             Heps             = _mm256_mul_ps(vfeps,H);
 354             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 355             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 356             vvdw12           = _mm256_mul_ps(c12_00,VV);
 357             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 358             fvdw12           = _mm256_mul_ps(c12_00,FF);
 359             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 360             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velecsum         = _mm256_add_ps(velecsum,velec);
 364             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 365             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 366
 367             fscal            = _mm256_add_ps(felec,fvdw);
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm256_mul_ps(fscal,dx00);
 371             ty               = _mm256_mul_ps(fscal,dy00);
 372             tz               = _mm256_mul_ps(fscal,dz00);
 373
 374             /* Update vectorial force */
 375             fix0             = _mm256_add_ps(fix0,tx);
 376             fiy0             = _mm256_add_ps(fiy0,ty);
 377             fiz0             = _mm256_add_ps(fiz0,tz);
 378
 379             fjptrA             = f+j_coord_offsetA;
 380             fjptrB             = f+j_coord_offsetB;
 381             fjptrC             = f+j_coord_offsetC;
 382             fjptrD             = f+j_coord_offsetD;
 383             fjptrE             = f+j_coord_offsetE;
 384             fjptrF             = f+j_coord_offsetF;
 385             fjptrG             = f+j_coord_offsetG;
 386             fjptrH             = f+j_coord_offsetH;
 387             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 388
 389             /* Inner loop uses 91 flops */
 390         }
 391
 392         if(jidx<j_index_end)
 393         {
 394
 395             /* Get j neighbor index, and coordinate index */
 396             jnrlistA         = jjnr[jidx];
 397             jnrlistB         = jjnr[jidx+1];
 398             jnrlistC         = jjnr[jidx+2];
 399             jnrlistD         = jjnr[jidx+3];
 400             jnrlistE         = jjnr[jidx+4];
 401             jnrlistF         = jjnr[jidx+5];
 402             jnrlistG         = jjnr[jidx+6];
 403             jnrlistH         = jjnr[jidx+7];
 404             /* Sign of each element will be negative for non-real atoms.
 405              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 406              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 407              */
 408             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 409                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 410
 411             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 412             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 413             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 414             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 415             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 416             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 417             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 418             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 419             j_coord_offsetA  = DIM*jnrA;
 420             j_coord_offsetB  = DIM*jnrB;
 421             j_coord_offsetC  = DIM*jnrC;
 422             j_coord_offsetD  = DIM*jnrD;
 423             j_coord_offsetE  = DIM*jnrE;
 424             j_coord_offsetF  = DIM*jnrF;
 425             j_coord_offsetG  = DIM*jnrG;
 426             j_coord_offsetH  = DIM*jnrH;
 427
 428             /* load j atom coordinates */
 429             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 430                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 431                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 432                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 433                                                  &jx0,&jy0,&jz0);
 434
 435             /* Calculate displacement vector */
 436             dx00             = _mm256_sub_ps(ix0,jx0);
 437             dy00             = _mm256_sub_ps(iy0,jy0);
 438             dz00             = _mm256_sub_ps(iz0,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 442
 443             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 444
 445             /* Load parameters for j particles */
 446             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 447                                                                  charge+jnrC+0,charge+jnrD+0,
 448                                                                  charge+jnrE+0,charge+jnrF+0,
 449                                                                  charge+jnrG+0,charge+jnrH+0);
 450             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 451                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 452                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 453                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 454             vdwjidx0A        = 2*vdwtype[jnrA+0];
 455             vdwjidx0B        = 2*vdwtype[jnrB+0];
 456             vdwjidx0C        = 2*vdwtype[jnrC+0];
 457             vdwjidx0D        = 2*vdwtype[jnrD+0];
 458             vdwjidx0E        = 2*vdwtype[jnrE+0];
 459             vdwjidx0F        = 2*vdwtype[jnrF+0];
 460             vdwjidx0G        = 2*vdwtype[jnrG+0];
 461             vdwjidx0H        = 2*vdwtype[jnrH+0];
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             r00              = _mm256_mul_ps(rsq00,rinv00);
 468             r00              = _mm256_andnot_ps(dummy_mask,r00);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq00             = _mm256_mul_ps(iq0,jq0);
 472             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 473                                             vdwioffsetptr0+vdwjidx0B,
 474                                             vdwioffsetptr0+vdwjidx0C,
 475                                             vdwioffsetptr0+vdwjidx0D,
 476                                             vdwioffsetptr0+vdwjidx0E,
 477                                             vdwioffsetptr0+vdwjidx0F,
 478                                             vdwioffsetptr0+vdwjidx0G,
 479                                             vdwioffsetptr0+vdwjidx0H,
 480                                             &c6_00,&c12_00);
 481
 482             /* Calculate table index by multiplying r with table scale and truncate to integer */
 483             rt               = _mm256_mul_ps(r00,vftabscale);
 484             vfitab           = _mm256_cvttps_epi32(rt);
 485             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 486             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 487             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 488             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 489             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 490             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 491
 492             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 493             isaprod          = _mm256_mul_ps(isai0,isaj0);
 494             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 495             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 496
 497             /* Calculate generalized born table index - this is a separate table from the normal one,
 498              * but we use the same procedure by multiplying r with scale and truncating to integer.
 499              */
 500             rt               = _mm256_mul_ps(r00,gbscale);
 501             gbitab           = _mm256_cvttps_epi32(rt);
 502             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 503             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 504             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 505             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 506             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 507             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 508             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 509                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 510             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 511                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 512             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 513                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 514             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 515                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 516             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 517             Heps             = _mm256_mul_ps(gbeps,H);
 518             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 519             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 520             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 521
 522             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 523             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 524             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 525             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 526             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 527             /* 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. */
 528             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 529             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 530             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 531             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 532             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 533             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 534             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 535             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 536             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 537                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 538             velec            = _mm256_mul_ps(qq00,rinv00);
 539             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 540
 541             /* CUBIC SPLINE TABLE DISPERSION */
 542             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 543                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 544             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 545                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 546             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 547                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 548             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 549                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 550             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 551             Heps             = _mm256_mul_ps(vfeps,H);
 552             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 553             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 554             vvdw6            = _mm256_mul_ps(c6_00,VV);
 555             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 556             fvdw6            = _mm256_mul_ps(c6_00,FF);
 557
 558             /* CUBIC SPLINE TABLE REPULSION */
 559             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 560             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 561             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 562                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 563             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 564                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 565             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 566                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 567             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 568                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 569             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 570             Heps             = _mm256_mul_ps(vfeps,H);
 571             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 572             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 573             vvdw12           = _mm256_mul_ps(c12_00,VV);
 574             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 575             fvdw12           = _mm256_mul_ps(c12_00,FF);
 576             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 577             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velec            = _mm256_andnot_ps(dummy_mask,velec);
 581             velecsum         = _mm256_add_ps(velecsum,velec);
 582             vgb              = _mm256_andnot_ps(dummy_mask,vgb);
 583             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 584             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 585             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 586
 587             fscal            = _mm256_add_ps(felec,fvdw);
 588
 589             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm256_mul_ps(fscal,dx00);
 593             ty               = _mm256_mul_ps(fscal,dy00);
 594             tz               = _mm256_mul_ps(fscal,dz00);
 595
 596             /* Update vectorial force */
 597             fix0             = _mm256_add_ps(fix0,tx);
 598             fiy0             = _mm256_add_ps(fiy0,ty);
 599             fiz0             = _mm256_add_ps(fiz0,tz);
 600
 601             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 602             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 603             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 604             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 605             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 606             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 607             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 608             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 609             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 610
 611             /* Inner loop uses 92 flops */
 612         }
 613
 614         /* End of innermost loop */
 615
 616         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 617                                                  f+i_coord_offset,fshift+i_shift_offset);
 618
 619         ggid                        = gid[iidx];
 620         /* Update potential energies */
 621         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 622         gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 623         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 624         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 625         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 626
 627         /* Increment number of inner iterations */
 628         inneriter                  += j_index_end - j_index_start;
 629
 630         /* Outer loop uses 10 flops */
 631     }
 632
 633     /* Increment number of outer iterations */
 634     outeriter        += nri;
 635
 636     /* Update outer/inner flops */
 637
 638     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*92);
 639 }
 640 /*
 641  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_avx_256_single
 642  * Electrostatics interaction: GeneralizedBorn
 643  * VdW interaction:            CubicSplineTable
 644  * Geometry:                   Particle-Particle
 645  * Calculate force/pot:        Force
 646  */
 647 void
 648 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_avx_256_single
 649                     (t_nblist                    * gmx_restrict       nlist,
 650                      rvec                        * gmx_restrict          xx,
 651                      rvec                        * gmx_restrict          ff,
 652                      t_forcerec                  * gmx_restrict          fr,
 653                      t_mdatoms                   * gmx_restrict     mdatoms,
 654                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 655                      t_nrnb                      * gmx_restrict        nrnb)
 656 {
 657     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 658      * just 0 for non-waters.
 659      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 660      * jnr indices corresponding to data put in the four positions in the SIMD register.
 661      */
 662     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 663     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 664     int              jnrA,jnrB,jnrC,jnrD;
 665     int              jnrE,jnrF,jnrG,jnrH;
 666     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 667     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 668     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 669     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 670     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 671     real             rcutoff_scalar;
 672     real             *shiftvec,*fshift,*x,*f;
 673     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 674     real             scratch[4*DIM];
 675     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 676     real *           vdwioffsetptr0;
 677     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 678     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 679     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 680     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 681     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 682     real             *charge;
 683     __m256i          gbitab;
 684     __m128i          gbitab_lo,gbitab_hi;
 685     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 686     __m256           minushalf = _mm256_set1_ps(-0.5);
 687     real             *invsqrta,*dvda,*gbtab;
 688     int              nvdwtype;
 689     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 690     int              *vdwtype;
 691     real             *vdwparam;
 692     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 693     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 694     __m256i          vfitab;
 695     __m128i          vfitab_lo,vfitab_hi;
 696     __m128i          ifour       = _mm_set1_epi32(4);
 697     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 698     real             *vftab;
 699     __m256           dummy_mask,cutoff_mask;
 700     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 701     __m256           one     = _mm256_set1_ps(1.0);
 702     __m256           two     = _mm256_set1_ps(2.0);
 703     x                = xx[0];
 704     f                = ff[0];
 705
 706     nri              = nlist->nri;
 707     iinr             = nlist->iinr;
 708     jindex           = nlist->jindex;
 709     jjnr             = nlist->jjnr;
 710     shiftidx         = nlist->shift;
 711     gid              = nlist->gid;
 712     shiftvec         = fr->shift_vec[0];
 713     fshift           = fr->fshift[0];
 714     facel            = _mm256_set1_ps(fr->epsfac);
 715     charge           = mdatoms->chargeA;
 716     nvdwtype         = fr->ntype;
 717     vdwparam         = fr->nbfp;
 718     vdwtype          = mdatoms->typeA;
 719
 720     vftab            = kernel_data->table_vdw->data;
 721     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 722
 723     invsqrta         = fr->invsqrta;
 724     dvda             = fr->dvda;
 725     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 726     gbtab            = fr->gbtab.data;
 727     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 728
 729     /* Avoid stupid compiler warnings */
 730     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 731     j_coord_offsetA = 0;
 732     j_coord_offsetB = 0;
 733     j_coord_offsetC = 0;
 734     j_coord_offsetD = 0;
 735     j_coord_offsetE = 0;
 736     j_coord_offsetF = 0;
 737     j_coord_offsetG = 0;
 738     j_coord_offsetH = 0;
 739
 740     outeriter        = 0;
 741     inneriter        = 0;
 742
 743     for(iidx=0;iidx<4*DIM;iidx++)
 744     {
 745         scratch[iidx] = 0.0;
 746     }
 747
 748     /* Start outer loop over neighborlists */
 749     for(iidx=0; iidx<nri; iidx++)
 750     {
 751         /* Load shift vector for this list */
 752         i_shift_offset   = DIM*shiftidx[iidx];
 753
 754         /* Load limits for loop over neighbors */
 755         j_index_start    = jindex[iidx];
 756         j_index_end      = jindex[iidx+1];
 757
 758         /* Get outer coordinate index */
 759         inr              = iinr[iidx];
 760         i_coord_offset   = DIM*inr;
 761
 762         /* Load i particle coords and add shift vector */
 763         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 764
 765         fix0             = _mm256_setzero_ps();
 766         fiy0             = _mm256_setzero_ps();
 767         fiz0             = _mm256_setzero_ps();
 768
 769         /* Load parameters for i particles */
 770         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 771         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 772         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 773
 774         dvdasum          = _mm256_setzero_ps();
 775
 776         /* Start inner kernel loop */
 777         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 778         {
 779
 780             /* Get j neighbor index, and coordinate index */
 781             jnrA             = jjnr[jidx];
 782             jnrB             = jjnr[jidx+1];
 783             jnrC             = jjnr[jidx+2];
 784             jnrD             = jjnr[jidx+3];
 785             jnrE             = jjnr[jidx+4];
 786             jnrF             = jjnr[jidx+5];
 787             jnrG             = jjnr[jidx+6];
 788             jnrH             = jjnr[jidx+7];
 789             j_coord_offsetA  = DIM*jnrA;
 790             j_coord_offsetB  = DIM*jnrB;
 791             j_coord_offsetC  = DIM*jnrC;
 792             j_coord_offsetD  = DIM*jnrD;
 793             j_coord_offsetE  = DIM*jnrE;
 794             j_coord_offsetF  = DIM*jnrF;
 795             j_coord_offsetG  = DIM*jnrG;
 796             j_coord_offsetH  = DIM*jnrH;
 797
 798             /* load j atom coordinates */
 799             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 800                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 801                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 802                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 803                                                  &jx0,&jy0,&jz0);
 804
 805             /* Calculate displacement vector */
 806             dx00             = _mm256_sub_ps(ix0,jx0);
 807             dy00             = _mm256_sub_ps(iy0,jy0);
 808             dz00             = _mm256_sub_ps(iz0,jz0);
 809
 810             /* Calculate squared distance and things based on it */
 811             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 812
 813             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 814
 815             /* Load parameters for j particles */
 816             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 817                                                                  charge+jnrC+0,charge+jnrD+0,
 818                                                                  charge+jnrE+0,charge+jnrF+0,
 819                                                                  charge+jnrG+0,charge+jnrH+0);
 820             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 821                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 822                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 823                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 824             vdwjidx0A        = 2*vdwtype[jnrA+0];
 825             vdwjidx0B        = 2*vdwtype[jnrB+0];
 826             vdwjidx0C        = 2*vdwtype[jnrC+0];
 827             vdwjidx0D        = 2*vdwtype[jnrD+0];
 828             vdwjidx0E        = 2*vdwtype[jnrE+0];
 829             vdwjidx0F        = 2*vdwtype[jnrF+0];
 830             vdwjidx0G        = 2*vdwtype[jnrG+0];
 831             vdwjidx0H        = 2*vdwtype[jnrH+0];
 832
 833             /**************************
 834              * CALCULATE INTERACTIONS *
 835              **************************/
 836
 837             r00              = _mm256_mul_ps(rsq00,rinv00);
 838
 839             /* Compute parameters for interactions between i and j atoms */
 840             qq00             = _mm256_mul_ps(iq0,jq0);
 841             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 842                                             vdwioffsetptr0+vdwjidx0B,
 843                                             vdwioffsetptr0+vdwjidx0C,
 844                                             vdwioffsetptr0+vdwjidx0D,
 845                                             vdwioffsetptr0+vdwjidx0E,
 846                                             vdwioffsetptr0+vdwjidx0F,
 847                                             vdwioffsetptr0+vdwjidx0G,
 848                                             vdwioffsetptr0+vdwjidx0H,
 849                                             &c6_00,&c12_00);
 850
 851             /* Calculate table index by multiplying r with table scale and truncate to integer */
 852             rt               = _mm256_mul_ps(r00,vftabscale);
 853             vfitab           = _mm256_cvttps_epi32(rt);
 854             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 855             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 856             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 857             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 858             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 859             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 860
 861             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 862             isaprod          = _mm256_mul_ps(isai0,isaj0);
 863             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 864             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 865
 866             /* Calculate generalized born table index - this is a separate table from the normal one,
 867              * but we use the same procedure by multiplying r with scale and truncating to integer.
 868              */
 869             rt               = _mm256_mul_ps(r00,gbscale);
 870             gbitab           = _mm256_cvttps_epi32(rt);
 871             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 872             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 873             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 874             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 875             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 876             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 877             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 878                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 879             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 880                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 881             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 882                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 883             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 884                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 885             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 886             Heps             = _mm256_mul_ps(gbeps,H);
 887             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 888             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 889             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 890
 891             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 892             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 893             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 894             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 895             fjptrA           = dvda+jnrA;
 896             fjptrB           = dvda+jnrB;
 897             fjptrC           = dvda+jnrC;
 898             fjptrD           = dvda+jnrD;
 899             fjptrE           = dvda+jnrE;
 900             fjptrF           = dvda+jnrF;
 901             fjptrG           = dvda+jnrG;
 902             fjptrH           = dvda+jnrH;
 903             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 904                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 905             velec            = _mm256_mul_ps(qq00,rinv00);
 906             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 907
 908             /* CUBIC SPLINE TABLE DISPERSION */
 909             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 910                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 911             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 912                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 913             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 914                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 915             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 916                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 917             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 918             Heps             = _mm256_mul_ps(vfeps,H);
 919             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 920             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 921             fvdw6            = _mm256_mul_ps(c6_00,FF);
 922
 923             /* CUBIC SPLINE TABLE REPULSION */
 924             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 925             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 926             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 927                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 928             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 929                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 930             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 931                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 932             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 933                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 934             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 935             Heps             = _mm256_mul_ps(vfeps,H);
 936             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 937             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 938             fvdw12           = _mm256_mul_ps(c12_00,FF);
 939             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 940
 941             fscal            = _mm256_add_ps(felec,fvdw);
 942
 943             /* Calculate temporary vectorial force */
 944             tx               = _mm256_mul_ps(fscal,dx00);
 945             ty               = _mm256_mul_ps(fscal,dy00);
 946             tz               = _mm256_mul_ps(fscal,dz00);
 947
 948             /* Update vectorial force */
 949             fix0             = _mm256_add_ps(fix0,tx);
 950             fiy0             = _mm256_add_ps(fiy0,ty);
 951             fiz0             = _mm256_add_ps(fiz0,tz);
 952
 953             fjptrA             = f+j_coord_offsetA;
 954             fjptrB             = f+j_coord_offsetB;
 955             fjptrC             = f+j_coord_offsetC;
 956             fjptrD             = f+j_coord_offsetD;
 957             fjptrE             = f+j_coord_offsetE;
 958             fjptrF             = f+j_coord_offsetF;
 959             fjptrG             = f+j_coord_offsetG;
 960             fjptrH             = f+j_coord_offsetH;
 961             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 962
 963             /* Inner loop uses 81 flops */
 964         }
 965
 966         if(jidx<j_index_end)
 967         {
 968
 969             /* Get j neighbor index, and coordinate index */
 970             jnrlistA         = jjnr[jidx];
 971             jnrlistB         = jjnr[jidx+1];
 972             jnrlistC         = jjnr[jidx+2];
 973             jnrlistD         = jjnr[jidx+3];
 974             jnrlistE         = jjnr[jidx+4];
 975             jnrlistF         = jjnr[jidx+5];
 976             jnrlistG         = jjnr[jidx+6];
 977             jnrlistH         = jjnr[jidx+7];
 978             /* Sign of each element will be negative for non-real atoms.
 979              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 980              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 981              */
 982             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 983                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 984
 985             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 986             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 987             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 988             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 989             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 990             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 991             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 992             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 993             j_coord_offsetA  = DIM*jnrA;
 994             j_coord_offsetB  = DIM*jnrB;
 995             j_coord_offsetC  = DIM*jnrC;
 996             j_coord_offsetD  = DIM*jnrD;
 997             j_coord_offsetE  = DIM*jnrE;
 998             j_coord_offsetF  = DIM*jnrF;
 999             j_coord_offsetG  = DIM*jnrG;
1000             j_coord_offsetH  = DIM*jnrH;
1001
1002             /* load j atom coordinates */
1003             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1004                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1005                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1006                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1007                                                  &jx0,&jy0,&jz0);
1008
1009             /* Calculate displacement vector */
1010             dx00             = _mm256_sub_ps(ix0,jx0);
1011             dy00             = _mm256_sub_ps(iy0,jy0);
1012             dz00             = _mm256_sub_ps(iz0,jz0);
1013
1014             /* Calculate squared distance and things based on it */
1015             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1016
1017             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1018
1019             /* Load parameters for j particles */
1020             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1021                                                                  charge+jnrC+0,charge+jnrD+0,
1022                                                                  charge+jnrE+0,charge+jnrF+0,
1023                                                                  charge+jnrG+0,charge+jnrH+0);
1024             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
1025                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
1026                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
1027                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
1028             vdwjidx0A        = 2*vdwtype[jnrA+0];
1029             vdwjidx0B        = 2*vdwtype[jnrB+0];
1030             vdwjidx0C        = 2*vdwtype[jnrC+0];
1031             vdwjidx0D        = 2*vdwtype[jnrD+0];
1032             vdwjidx0E        = 2*vdwtype[jnrE+0];
1033             vdwjidx0F        = 2*vdwtype[jnrF+0];
1034             vdwjidx0G        = 2*vdwtype[jnrG+0];
1035             vdwjidx0H        = 2*vdwtype[jnrH+0];
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             r00              = _mm256_mul_ps(rsq00,rinv00);
1042             r00              = _mm256_andnot_ps(dummy_mask,r00);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq00             = _mm256_mul_ps(iq0,jq0);
1046             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1047                                             vdwioffsetptr0+vdwjidx0B,
1048                                             vdwioffsetptr0+vdwjidx0C,
1049                                             vdwioffsetptr0+vdwjidx0D,
1050                                             vdwioffsetptr0+vdwjidx0E,
1051                                             vdwioffsetptr0+vdwjidx0F,
1052                                             vdwioffsetptr0+vdwjidx0G,
1053                                             vdwioffsetptr0+vdwjidx0H,
1054                                             &c6_00,&c12_00);
1055
1056             /* Calculate table index by multiplying r with table scale and truncate to integer */
1057             rt               = _mm256_mul_ps(r00,vftabscale);
1058             vfitab           = _mm256_cvttps_epi32(rt);
1059             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1060             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1061             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1062             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1063             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1064             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1065
1066             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
1067             isaprod          = _mm256_mul_ps(isai0,isaj0);
1068             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
1069             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
1070
1071             /* Calculate generalized born table index - this is a separate table from the normal one,
1072              * but we use the same procedure by multiplying r with scale and truncating to integer.
1073              */
1074             rt               = _mm256_mul_ps(r00,gbscale);
1075             gbitab           = _mm256_cvttps_epi32(rt);
1076             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1077             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1078             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
1079             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
1080             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
1081             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
1082             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
1083                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
1084             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
1085                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
1086             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
1087                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
1088             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
1089                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
1090             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1091             Heps             = _mm256_mul_ps(gbeps,H);
1092             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
1093             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
1094             vgb              = _mm256_mul_ps(gbqqfactor,VV);
1095
1096             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1097             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
1098             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
1099             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
1100             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
1101             /* 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. */
1102             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
1103             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
1104             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
1105             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
1106             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
1107             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
1108             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
1109             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
1110             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1111                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
1112             velec            = _mm256_mul_ps(qq00,rinv00);
1113             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
1114
1115             /* CUBIC SPLINE TABLE DISPERSION */
1116             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1117                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1118             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1119                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1120             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1121                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1122             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1123                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1124             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1125             Heps             = _mm256_mul_ps(vfeps,H);
1126             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1127             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1128             fvdw6            = _mm256_mul_ps(c6_00,FF);
1129
1130             /* CUBIC SPLINE TABLE REPULSION */
1131             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1132             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1133             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1134                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1135             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1136                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1137             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1138                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1139             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1140                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1141             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1142             Heps             = _mm256_mul_ps(vfeps,H);
1143             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1144             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1145             fvdw12           = _mm256_mul_ps(c12_00,FF);
1146             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1147
1148             fscal            = _mm256_add_ps(felec,fvdw);
1149
1150             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1151
1152             /* Calculate temporary vectorial force */
1153             tx               = _mm256_mul_ps(fscal,dx00);
1154             ty               = _mm256_mul_ps(fscal,dy00);
1155             tz               = _mm256_mul_ps(fscal,dz00);
1156
1157             /* Update vectorial force */
1158             fix0             = _mm256_add_ps(fix0,tx);
1159             fiy0             = _mm256_add_ps(fiy0,ty);
1160             fiz0             = _mm256_add_ps(fiz0,tz);
1161
1162             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1163             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1164             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1165             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1166             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1167             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1168             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1169             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1170             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1171
1172             /* Inner loop uses 82 flops */
1173         }
1174
1175         /* End of innermost loop */
1176
1177         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1178                                                  f+i_coord_offset,fshift+i_shift_offset);
1179
1180         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
1181         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
1182
1183         /* Increment number of inner iterations */
1184         inneriter                  += j_index_end - j_index_start;
1185
1186         /* Outer loop uses 7 flops */
1187     }
1188
1189     /* Increment number of outer iterations */
1190     outeriter        += nri;
1191
1192     /* Update outer/inner flops */
1193
1194     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*82);
1195 }