src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_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     int              nvdwtype;
  96     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 100     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 101     __m256i          vfitab;
 102     __m128i          vfitab_lo,vfitab_hi;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m256           dummy_mask,cutoff_mask;
 107     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 108     __m256           one     = _mm256_set1_ps(1.0);
 109     __m256           two     = _mm256_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_elec->data;
 128     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136     j_coord_offsetE = 0;
 137     j_coord_offsetF = 0;
 138     j_coord_offsetG = 0;
 139     j_coord_offsetH = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm256_setzero_ps();
 167         fiy0             = _mm256_setzero_ps();
 168         fiz0             = _mm256_setzero_ps();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 172         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_ps();
 176         vvdwsum          = _mm256_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             jnrE             = jjnr[jidx+4];
 188             jnrF             = jjnr[jidx+5];
 189             jnrG             = jjnr[jidx+6];
 190             jnrH             = jjnr[jidx+7];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195             j_coord_offsetE  = DIM*jnrE;
 196             j_coord_offsetF  = DIM*jnrF;
 197             j_coord_offsetG  = DIM*jnrG;
 198             j_coord_offsetH  = DIM*jnrH;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 204                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_ps(ix0,jx0);
 209             dy00             = _mm256_sub_ps(iy0,jy0);
 210             dz00             = _mm256_sub_ps(iz0,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 214
 215             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 216
 217             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                                  charge+jnrC+0,charge+jnrD+0,
 222                                                                  charge+jnrE+0,charge+jnrF+0,
 223                                                                  charge+jnrG+0,charge+jnrH+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228             vdwjidx0E        = 2*vdwtype[jnrE+0];
 229             vdwjidx0F        = 2*vdwtype[jnrF+0];
 230             vdwjidx0G        = 2*vdwtype[jnrG+0];
 231             vdwjidx0H        = 2*vdwtype[jnrH+0];
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _mm256_mul_ps(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm256_mul_ps(iq0,jq0);
 241             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 242                                             vdwioffsetptr0+vdwjidx0B,
 243                                             vdwioffsetptr0+vdwjidx0C,
 244                                             vdwioffsetptr0+vdwjidx0D,
 245                                             vdwioffsetptr0+vdwjidx0E,
 246                                             vdwioffsetptr0+vdwjidx0F,
 247                                             vdwioffsetptr0+vdwjidx0G,
 248                                             vdwioffsetptr0+vdwjidx0H,
 249                                             &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm256_mul_ps(r00,vftabscale);
 253             vfitab           = _mm256_cvttps_epi32(rt);
 254             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 255             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 256             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 257             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 258             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 259             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 260
 261             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 262             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 263                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 264             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 265                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 266             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 267                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 268             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 269                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 270             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 271             Heps             = _mm256_mul_ps(vfeps,H);
 272             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 273             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 274             velec            = _mm256_mul_ps(qq00,VV);
 275             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 276             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 277
 278             /* LENNARD-JONES DISPERSION/REPULSION */
 279
 280             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 281             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 282             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 283             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 284             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm256_add_ps(velecsum,velec);
 288             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 289
 290             fscal            = _mm256_add_ps(felec,fvdw);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm256_mul_ps(fscal,dx00);
 294             ty               = _mm256_mul_ps(fscal,dy00);
 295             tz               = _mm256_mul_ps(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm256_add_ps(fix0,tx);
 299             fiy0             = _mm256_add_ps(fiy0,ty);
 300             fiz0             = _mm256_add_ps(fiz0,tz);
 301
 302             fjptrA             = f+j_coord_offsetA;
 303             fjptrB             = f+j_coord_offsetB;
 304             fjptrC             = f+j_coord_offsetC;
 305             fjptrD             = f+j_coord_offsetD;
 306             fjptrE             = f+j_coord_offsetE;
 307             fjptrF             = f+j_coord_offsetF;
 308             fjptrG             = f+j_coord_offsetG;
 309             fjptrH             = f+j_coord_offsetH;
 310             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 311
 312             /* Inner loop uses 56 flops */
 313         }
 314
 315         if(jidx<j_index_end)
 316         {
 317
 318             /* Get j neighbor index, and coordinate index */
 319             jnrlistA         = jjnr[jidx];
 320             jnrlistB         = jjnr[jidx+1];
 321             jnrlistC         = jjnr[jidx+2];
 322             jnrlistD         = jjnr[jidx+3];
 323             jnrlistE         = jjnr[jidx+4];
 324             jnrlistF         = jjnr[jidx+5];
 325             jnrlistG         = jjnr[jidx+6];
 326             jnrlistH         = jjnr[jidx+7];
 327             /* Sign of each element will be negative for non-real atoms.
 328              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 329              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 330              */
 331             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 332                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 333
 334             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 335             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 336             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 337             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 338             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 339             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 340             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 341             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 342             j_coord_offsetA  = DIM*jnrA;
 343             j_coord_offsetB  = DIM*jnrB;
 344             j_coord_offsetC  = DIM*jnrC;
 345             j_coord_offsetD  = DIM*jnrD;
 346             j_coord_offsetE  = DIM*jnrE;
 347             j_coord_offsetF  = DIM*jnrF;
 348             j_coord_offsetG  = DIM*jnrG;
 349             j_coord_offsetH  = DIM*jnrH;
 350
 351             /* load j atom coordinates */
 352             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 353                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 354                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 355                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 356                                                  &jx0,&jy0,&jz0);
 357
 358             /* Calculate displacement vector */
 359             dx00             = _mm256_sub_ps(ix0,jx0);
 360             dy00             = _mm256_sub_ps(iy0,jy0);
 361             dz00             = _mm256_sub_ps(iz0,jz0);
 362
 363             /* Calculate squared distance and things based on it */
 364             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 365
 366             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 367
 368             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 369
 370             /* Load parameters for j particles */
 371             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 372                                                                  charge+jnrC+0,charge+jnrD+0,
 373                                                                  charge+jnrE+0,charge+jnrF+0,
 374                                                                  charge+jnrG+0,charge+jnrH+0);
 375             vdwjidx0A        = 2*vdwtype[jnrA+0];
 376             vdwjidx0B        = 2*vdwtype[jnrB+0];
 377             vdwjidx0C        = 2*vdwtype[jnrC+0];
 378             vdwjidx0D        = 2*vdwtype[jnrD+0];
 379             vdwjidx0E        = 2*vdwtype[jnrE+0];
 380             vdwjidx0F        = 2*vdwtype[jnrF+0];
 381             vdwjidx0G        = 2*vdwtype[jnrG+0];
 382             vdwjidx0H        = 2*vdwtype[jnrH+0];
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             r00              = _mm256_mul_ps(rsq00,rinv00);
 389             r00              = _mm256_andnot_ps(dummy_mask,r00);
 390
 391             /* Compute parameters for interactions between i and j atoms */
 392             qq00             = _mm256_mul_ps(iq0,jq0);
 393             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 394                                             vdwioffsetptr0+vdwjidx0B,
 395                                             vdwioffsetptr0+vdwjidx0C,
 396                                             vdwioffsetptr0+vdwjidx0D,
 397                                             vdwioffsetptr0+vdwjidx0E,
 398                                             vdwioffsetptr0+vdwjidx0F,
 399                                             vdwioffsetptr0+vdwjidx0G,
 400                                             vdwioffsetptr0+vdwjidx0H,
 401                                             &c6_00,&c12_00);
 402
 403             /* Calculate table index by multiplying r with table scale and truncate to integer */
 404             rt               = _mm256_mul_ps(r00,vftabscale);
 405             vfitab           = _mm256_cvttps_epi32(rt);
 406             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 407             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 408             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 409             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 410             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 411             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 412
 413             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 414             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 415                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 416             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 417                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 418             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 419                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 420             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 421                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 422             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 423             Heps             = _mm256_mul_ps(vfeps,H);
 424             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 425             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 426             velec            = _mm256_mul_ps(qq00,VV);
 427             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 428             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 429
 430             /* LENNARD-JONES DISPERSION/REPULSION */
 431
 432             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 433             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 434             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 435             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 436             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velec            = _mm256_andnot_ps(dummy_mask,velec);
 440             velecsum         = _mm256_add_ps(velecsum,velec);
 441             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 442             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 443
 444             fscal            = _mm256_add_ps(felec,fvdw);
 445
 446             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 447
 448             /* Calculate temporary vectorial force */
 449             tx               = _mm256_mul_ps(fscal,dx00);
 450             ty               = _mm256_mul_ps(fscal,dy00);
 451             tz               = _mm256_mul_ps(fscal,dz00);
 452
 453             /* Update vectorial force */
 454             fix0             = _mm256_add_ps(fix0,tx);
 455             fiy0             = _mm256_add_ps(fiy0,ty);
 456             fiz0             = _mm256_add_ps(fiz0,tz);
 457
 458             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 459             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 460             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 461             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 462             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 463             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 464             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 465             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 466             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 467
 468             /* Inner loop uses 57 flops */
 469         }
 470
 471         /* End of innermost loop */
 472
 473         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 474                                                  f+i_coord_offset,fshift+i_shift_offset);
 475
 476         ggid                        = gid[iidx];
 477         /* Update potential energies */
 478         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 479         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 480
 481         /* Increment number of inner iterations */
 482         inneriter                  += j_index_end - j_index_start;
 483
 484         /* Outer loop uses 9 flops */
 485     }
 486
 487     /* Increment number of outer iterations */
 488     outeriter        += nri;
 489
 490     /* Update outer/inner flops */
 491
 492     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 493 }
 494 /*
 495  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_256_single
 496  * Electrostatics interaction: CubicSplineTable
 497  * VdW interaction:            LennardJones
 498  * Geometry:                   Particle-Particle
 499  * Calculate force/pot:        Force
 500  */
 501 void
 502 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_256_single
 503                     (t_nblist                    * gmx_restrict       nlist,
 504                      rvec                        * gmx_restrict          xx,
 505                      rvec                        * gmx_restrict          ff,
 506                      t_forcerec                  * gmx_restrict          fr,
 507                      t_mdatoms                   * gmx_restrict     mdatoms,
 508                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 509                      t_nrnb                      * gmx_restrict        nrnb)
 510 {
 511     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 512      * just 0 for non-waters.
 513      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 514      * jnr indices corresponding to data put in the four positions in the SIMD register.
 515      */
 516     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 517     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 518     int              jnrA,jnrB,jnrC,jnrD;
 519     int              jnrE,jnrF,jnrG,jnrH;
 520     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 521     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 522     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 523     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 524     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 525     real             rcutoff_scalar;
 526     real             *shiftvec,*fshift,*x,*f;
 527     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 528     real             scratch[4*DIM];
 529     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 530     real *           vdwioffsetptr0;
 531     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 532     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 533     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 534     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 535     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 536     real             *charge;
 537     int              nvdwtype;
 538     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 539     int              *vdwtype;
 540     real             *vdwparam;
 541     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 542     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 543     __m256i          vfitab;
 544     __m128i          vfitab_lo,vfitab_hi;
 545     __m128i          ifour       = _mm_set1_epi32(4);
 546     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 547     real             *vftab;
 548     __m256           dummy_mask,cutoff_mask;
 549     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 550     __m256           one     = _mm256_set1_ps(1.0);
 551     __m256           two     = _mm256_set1_ps(2.0);
 552     x                = xx[0];
 553     f                = ff[0];
 554
 555     nri              = nlist->nri;
 556     iinr             = nlist->iinr;
 557     jindex           = nlist->jindex;
 558     jjnr             = nlist->jjnr;
 559     shiftidx         = nlist->shift;
 560     gid              = nlist->gid;
 561     shiftvec         = fr->shift_vec[0];
 562     fshift           = fr->fshift[0];
 563     facel            = _mm256_set1_ps(fr->epsfac);
 564     charge           = mdatoms->chargeA;
 565     nvdwtype         = fr->ntype;
 566     vdwparam         = fr->nbfp;
 567     vdwtype          = mdatoms->typeA;
 568
 569     vftab            = kernel_data->table_elec->data;
 570     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 571
 572     /* Avoid stupid compiler warnings */
 573     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 574     j_coord_offsetA = 0;
 575     j_coord_offsetB = 0;
 576     j_coord_offsetC = 0;
 577     j_coord_offsetD = 0;
 578     j_coord_offsetE = 0;
 579     j_coord_offsetF = 0;
 580     j_coord_offsetG = 0;
 581     j_coord_offsetH = 0;
 582
 583     outeriter        = 0;
 584     inneriter        = 0;
 585
 586     for(iidx=0;iidx<4*DIM;iidx++)
 587     {
 588         scratch[iidx] = 0.0;
 589     }
 590
 591     /* Start outer loop over neighborlists */
 592     for(iidx=0; iidx<nri; iidx++)
 593     {
 594         /* Load shift vector for this list */
 595         i_shift_offset   = DIM*shiftidx[iidx];
 596
 597         /* Load limits for loop over neighbors */
 598         j_index_start    = jindex[iidx];
 599         j_index_end      = jindex[iidx+1];
 600
 601         /* Get outer coordinate index */
 602         inr              = iinr[iidx];
 603         i_coord_offset   = DIM*inr;
 604
 605         /* Load i particle coords and add shift vector */
 606         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 607
 608         fix0             = _mm256_setzero_ps();
 609         fiy0             = _mm256_setzero_ps();
 610         fiz0             = _mm256_setzero_ps();
 611
 612         /* Load parameters for i particles */
 613         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 614         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 615
 616         /* Start inner kernel loop */
 617         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 618         {
 619
 620             /* Get j neighbor index, and coordinate index */
 621             jnrA             = jjnr[jidx];
 622             jnrB             = jjnr[jidx+1];
 623             jnrC             = jjnr[jidx+2];
 624             jnrD             = jjnr[jidx+3];
 625             jnrE             = jjnr[jidx+4];
 626             jnrF             = jjnr[jidx+5];
 627             jnrG             = jjnr[jidx+6];
 628             jnrH             = jjnr[jidx+7];
 629             j_coord_offsetA  = DIM*jnrA;
 630             j_coord_offsetB  = DIM*jnrB;
 631             j_coord_offsetC  = DIM*jnrC;
 632             j_coord_offsetD  = DIM*jnrD;
 633             j_coord_offsetE  = DIM*jnrE;
 634             j_coord_offsetF  = DIM*jnrF;
 635             j_coord_offsetG  = DIM*jnrG;
 636             j_coord_offsetH  = DIM*jnrH;
 637
 638             /* load j atom coordinates */
 639             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 640                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 641                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 642                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 643                                                  &jx0,&jy0,&jz0);
 644
 645             /* Calculate displacement vector */
 646             dx00             = _mm256_sub_ps(ix0,jx0);
 647             dy00             = _mm256_sub_ps(iy0,jy0);
 648             dz00             = _mm256_sub_ps(iz0,jz0);
 649
 650             /* Calculate squared distance and things based on it */
 651             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 652
 653             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 654
 655             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 656
 657             /* Load parameters for j particles */
 658             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 659                                                                  charge+jnrC+0,charge+jnrD+0,
 660                                                                  charge+jnrE+0,charge+jnrF+0,
 661                                                                  charge+jnrG+0,charge+jnrH+0);
 662             vdwjidx0A        = 2*vdwtype[jnrA+0];
 663             vdwjidx0B        = 2*vdwtype[jnrB+0];
 664             vdwjidx0C        = 2*vdwtype[jnrC+0];
 665             vdwjidx0D        = 2*vdwtype[jnrD+0];
 666             vdwjidx0E        = 2*vdwtype[jnrE+0];
 667             vdwjidx0F        = 2*vdwtype[jnrF+0];
 668             vdwjidx0G        = 2*vdwtype[jnrG+0];
 669             vdwjidx0H        = 2*vdwtype[jnrH+0];
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             r00              = _mm256_mul_ps(rsq00,rinv00);
 676
 677             /* Compute parameters for interactions between i and j atoms */
 678             qq00             = _mm256_mul_ps(iq0,jq0);
 679             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 680                                             vdwioffsetptr0+vdwjidx0B,
 681                                             vdwioffsetptr0+vdwjidx0C,
 682                                             vdwioffsetptr0+vdwjidx0D,
 683                                             vdwioffsetptr0+vdwjidx0E,
 684                                             vdwioffsetptr0+vdwjidx0F,
 685                                             vdwioffsetptr0+vdwjidx0G,
 686                                             vdwioffsetptr0+vdwjidx0H,
 687                                             &c6_00,&c12_00);
 688
 689             /* Calculate table index by multiplying r with table scale and truncate to integer */
 690             rt               = _mm256_mul_ps(r00,vftabscale);
 691             vfitab           = _mm256_cvttps_epi32(rt);
 692             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 693             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 694             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 695             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 696             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 697             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 698
 699             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 700             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 701                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 702             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 703                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 704             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 705                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 706             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 707                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 708             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 709             Heps             = _mm256_mul_ps(vfeps,H);
 710             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 711             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 712             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 713
 714             /* LENNARD-JONES DISPERSION/REPULSION */
 715
 716             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 717             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 718
 719             fscal            = _mm256_add_ps(felec,fvdw);
 720
 721             /* Calculate temporary vectorial force */
 722             tx               = _mm256_mul_ps(fscal,dx00);
 723             ty               = _mm256_mul_ps(fscal,dy00);
 724             tz               = _mm256_mul_ps(fscal,dz00);
 725
 726             /* Update vectorial force */
 727             fix0             = _mm256_add_ps(fix0,tx);
 728             fiy0             = _mm256_add_ps(fiy0,ty);
 729             fiz0             = _mm256_add_ps(fiz0,tz);
 730
 731             fjptrA             = f+j_coord_offsetA;
 732             fjptrB             = f+j_coord_offsetB;
 733             fjptrC             = f+j_coord_offsetC;
 734             fjptrD             = f+j_coord_offsetD;
 735             fjptrE             = f+j_coord_offsetE;
 736             fjptrF             = f+j_coord_offsetF;
 737             fjptrG             = f+j_coord_offsetG;
 738             fjptrH             = f+j_coord_offsetH;
 739             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 740
 741             /* Inner loop uses 47 flops */
 742         }
 743
 744         if(jidx<j_index_end)
 745         {
 746
 747             /* Get j neighbor index, and coordinate index */
 748             jnrlistA         = jjnr[jidx];
 749             jnrlistB         = jjnr[jidx+1];
 750             jnrlistC         = jjnr[jidx+2];
 751             jnrlistD         = jjnr[jidx+3];
 752             jnrlistE         = jjnr[jidx+4];
 753             jnrlistF         = jjnr[jidx+5];
 754             jnrlistG         = jjnr[jidx+6];
 755             jnrlistH         = jjnr[jidx+7];
 756             /* Sign of each element will be negative for non-real atoms.
 757              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 758              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 759              */
 760             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 761                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 762
 763             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 764             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 765             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 766             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 767             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 768             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 769             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 770             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 771             j_coord_offsetA  = DIM*jnrA;
 772             j_coord_offsetB  = DIM*jnrB;
 773             j_coord_offsetC  = DIM*jnrC;
 774             j_coord_offsetD  = DIM*jnrD;
 775             j_coord_offsetE  = DIM*jnrE;
 776             j_coord_offsetF  = DIM*jnrF;
 777             j_coord_offsetG  = DIM*jnrG;
 778             j_coord_offsetH  = DIM*jnrH;
 779
 780             /* load j atom coordinates */
 781             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 782                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 783                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 784                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 785                                                  &jx0,&jy0,&jz0);
 786
 787             /* Calculate displacement vector */
 788             dx00             = _mm256_sub_ps(ix0,jx0);
 789             dy00             = _mm256_sub_ps(iy0,jy0);
 790             dz00             = _mm256_sub_ps(iz0,jz0);
 791
 792             /* Calculate squared distance and things based on it */
 793             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 794
 795             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 796
 797             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 798
 799             /* Load parameters for j particles */
 800             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 801                                                                  charge+jnrC+0,charge+jnrD+0,
 802                                                                  charge+jnrE+0,charge+jnrF+0,
 803                                                                  charge+jnrG+0,charge+jnrH+0);
 804             vdwjidx0A        = 2*vdwtype[jnrA+0];
 805             vdwjidx0B        = 2*vdwtype[jnrB+0];
 806             vdwjidx0C        = 2*vdwtype[jnrC+0];
 807             vdwjidx0D        = 2*vdwtype[jnrD+0];
 808             vdwjidx0E        = 2*vdwtype[jnrE+0];
 809             vdwjidx0F        = 2*vdwtype[jnrF+0];
 810             vdwjidx0G        = 2*vdwtype[jnrG+0];
 811             vdwjidx0H        = 2*vdwtype[jnrH+0];
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             r00              = _mm256_mul_ps(rsq00,rinv00);
 818             r00              = _mm256_andnot_ps(dummy_mask,r00);
 819
 820             /* Compute parameters for interactions between i and j atoms */
 821             qq00             = _mm256_mul_ps(iq0,jq0);
 822             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 823                                             vdwioffsetptr0+vdwjidx0B,
 824                                             vdwioffsetptr0+vdwjidx0C,
 825                                             vdwioffsetptr0+vdwjidx0D,
 826                                             vdwioffsetptr0+vdwjidx0E,
 827                                             vdwioffsetptr0+vdwjidx0F,
 828                                             vdwioffsetptr0+vdwjidx0G,
 829                                             vdwioffsetptr0+vdwjidx0H,
 830                                             &c6_00,&c12_00);
 831
 832             /* Calculate table index by multiplying r with table scale and truncate to integer */
 833             rt               = _mm256_mul_ps(r00,vftabscale);
 834             vfitab           = _mm256_cvttps_epi32(rt);
 835             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 836             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 837             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 838             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 839             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 840             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 841
 842             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 843             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 844                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 845             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 846                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 847             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 848                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 849             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 850                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 851             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 852             Heps             = _mm256_mul_ps(vfeps,H);
 853             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 854             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 855             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 856
 857             /* LENNARD-JONES DISPERSION/REPULSION */
 858
 859             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 860             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 861
 862             fscal            = _mm256_add_ps(felec,fvdw);
 863
 864             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 865
 866             /* Calculate temporary vectorial force */
 867             tx               = _mm256_mul_ps(fscal,dx00);
 868             ty               = _mm256_mul_ps(fscal,dy00);
 869             tz               = _mm256_mul_ps(fscal,dz00);
 870
 871             /* Update vectorial force */
 872             fix0             = _mm256_add_ps(fix0,tx);
 873             fiy0             = _mm256_add_ps(fiy0,ty);
 874             fiz0             = _mm256_add_ps(fiz0,tz);
 875
 876             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 877             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 878             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 879             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 880             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 881             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 882             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 883             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 884             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 885
 886             /* Inner loop uses 48 flops */
 887         }
 888
 889         /* End of innermost loop */
 890
 891         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 892                                                  f+i_coord_offset,fshift+i_shift_offset);
 893
 894         /* Increment number of inner iterations */
 895         inneriter                  += j_index_end - j_index_start;
 896
 897         /* Outer loop uses 7 flops */
 898     }
 899
 900     /* Increment number of outer iterations */
 901     outeriter        += nri;
 902
 903     /* Update outer/inner flops */
 904
 905     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*48);
 906 }