src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_vdw->data;
 134     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 139     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 140     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 141     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149     j_coord_offsetE = 0;
 150     j_coord_offsetF = 0;
 151     j_coord_offsetG = 0;
 152     j_coord_offsetH = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183         fix1             = _mm256_setzero_ps();
 184         fiy1             = _mm256_setzero_ps();
 185         fiz1             = _mm256_setzero_ps();
 186         fix2             = _mm256_setzero_ps();
 187         fiy2             = _mm256_setzero_ps();
 188         fiz2             = _mm256_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_ps();
 192         vvdwsum          = _mm256_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             jnrE             = jjnr[jidx+4];
 204             jnrF             = jjnr[jidx+5];
 205             jnrG             = jjnr[jidx+6];
 206             jnrH             = jjnr[jidx+7];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211             j_coord_offsetE  = DIM*jnrE;
 212             j_coord_offsetF  = DIM*jnrF;
 213             j_coord_offsetG  = DIM*jnrG;
 214             j_coord_offsetH  = DIM*jnrH;
 215
 216             /* load j atom coordinates */
 217             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 219                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 220                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 221                                                  &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm256_sub_ps(ix0,jx0);
 225             dy00             = _mm256_sub_ps(iy0,jy0);
 226             dz00             = _mm256_sub_ps(iz0,jz0);
 227             dx10             = _mm256_sub_ps(ix1,jx0);
 228             dy10             = _mm256_sub_ps(iy1,jy0);
 229             dz10             = _mm256_sub_ps(iz1,jz0);
 230             dx20             = _mm256_sub_ps(ix2,jx0);
 231             dy20             = _mm256_sub_ps(iy2,jy0);
 232             dz20             = _mm256_sub_ps(iz2,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 236             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 237             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 238
 239             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 240             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 241             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 242
 243             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 244             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 245             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 249                                                                  charge+jnrC+0,charge+jnrD+0,
 250                                                                  charge+jnrE+0,charge+jnrF+0,
 251                                                                  charge+jnrG+0,charge+jnrH+0);
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253             vdwjidx0B        = 2*vdwtype[jnrB+0];
 254             vdwjidx0C        = 2*vdwtype[jnrC+0];
 255             vdwjidx0D        = 2*vdwtype[jnrD+0];
 256             vdwjidx0E        = 2*vdwtype[jnrE+0];
 257             vdwjidx0F        = 2*vdwtype[jnrF+0];
 258             vdwjidx0G        = 2*vdwtype[jnrG+0];
 259             vdwjidx0H        = 2*vdwtype[jnrH+0];
 260
 261             fjx0             = _mm256_setzero_ps();
 262             fjy0             = _mm256_setzero_ps();
 263             fjz0             = _mm256_setzero_ps();
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             r00              = _mm256_mul_ps(rsq00,rinv00);
 270
 271             /* Compute parameters for interactions between i and j atoms */
 272             qq00             = _mm256_mul_ps(iq0,jq0);
 273             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 274                                             vdwioffsetptr0+vdwjidx0B,
 275                                             vdwioffsetptr0+vdwjidx0C,
 276                                             vdwioffsetptr0+vdwjidx0D,
 277                                             vdwioffsetptr0+vdwjidx0E,
 278                                             vdwioffsetptr0+vdwjidx0F,
 279                                             vdwioffsetptr0+vdwjidx0G,
 280                                             vdwioffsetptr0+vdwjidx0H,
 281                                             &c6_00,&c12_00);
 282
 283             /* Calculate table index by multiplying r with table scale and truncate to integer */
 284             rt               = _mm256_mul_ps(r00,vftabscale);
 285             vfitab           = _mm256_cvttps_epi32(rt);
 286             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 287             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 288             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 289             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 290             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 291             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 292
 293             /* COULOMB ELECTROSTATICS */
 294             velec            = _mm256_mul_ps(qq00,rinv00);
 295             felec            = _mm256_mul_ps(velec,rinvsq00);
 296
 297             /* CUBIC SPLINE TABLE DISPERSION */
 298             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 299                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 300             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 301                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 302             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 303                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 304             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 305                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 306             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 307             Heps             = _mm256_mul_ps(vfeps,H);
 308             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 309             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 310             vvdw6            = _mm256_mul_ps(c6_00,VV);
 311             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 312             fvdw6            = _mm256_mul_ps(c6_00,FF);
 313
 314             /* CUBIC SPLINE TABLE REPULSION */
 315             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 316             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 317             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 318                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 319             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 320                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 321             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 323             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 325             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 326             Heps             = _mm256_mul_ps(vfeps,H);
 327             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 328             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 329             vvdw12           = _mm256_mul_ps(c12_00,VV);
 330             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 331             fvdw12           = _mm256_mul_ps(c12_00,FF);
 332             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 333             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm256_add_ps(velecsum,velec);
 337             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 338
 339             fscal            = _mm256_add_ps(felec,fvdw);
 340
 341             /* Calculate temporary vectorial force */
 342             tx               = _mm256_mul_ps(fscal,dx00);
 343             ty               = _mm256_mul_ps(fscal,dy00);
 344             tz               = _mm256_mul_ps(fscal,dz00);
 345
 346             /* Update vectorial force */
 347             fix0             = _mm256_add_ps(fix0,tx);
 348             fiy0             = _mm256_add_ps(fiy0,ty);
 349             fiz0             = _mm256_add_ps(fiz0,tz);
 350
 351             fjx0             = _mm256_add_ps(fjx0,tx);
 352             fjy0             = _mm256_add_ps(fjy0,ty);
 353             fjz0             = _mm256_add_ps(fjz0,tz);
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq10             = _mm256_mul_ps(iq1,jq0);
 361
 362             /* COULOMB ELECTROSTATICS */
 363             velec            = _mm256_mul_ps(qq10,rinv10);
 364             felec            = _mm256_mul_ps(velec,rinvsq10);
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velecsum         = _mm256_add_ps(velecsum,velec);
 368
 369             fscal            = felec;
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm256_mul_ps(fscal,dx10);
 373             ty               = _mm256_mul_ps(fscal,dy10);
 374             tz               = _mm256_mul_ps(fscal,dz10);
 375
 376             /* Update vectorial force */
 377             fix1             = _mm256_add_ps(fix1,tx);
 378             fiy1             = _mm256_add_ps(fiy1,ty);
 379             fiz1             = _mm256_add_ps(fiz1,tz);
 380
 381             fjx0             = _mm256_add_ps(fjx0,tx);
 382             fjy0             = _mm256_add_ps(fjy0,ty);
 383             fjz0             = _mm256_add_ps(fjz0,tz);
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq20             = _mm256_mul_ps(iq2,jq0);
 391
 392             /* COULOMB ELECTROSTATICS */
 393             velec            = _mm256_mul_ps(qq20,rinv20);
 394             felec            = _mm256_mul_ps(velec,rinvsq20);
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             velecsum         = _mm256_add_ps(velecsum,velec);
 398
 399             fscal            = felec;
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm256_mul_ps(fscal,dx20);
 403             ty               = _mm256_mul_ps(fscal,dy20);
 404             tz               = _mm256_mul_ps(fscal,dz20);
 405
 406             /* Update vectorial force */
 407             fix2             = _mm256_add_ps(fix2,tx);
 408             fiy2             = _mm256_add_ps(fiy2,ty);
 409             fiz2             = _mm256_add_ps(fiz2,tz);
 410
 411             fjx0             = _mm256_add_ps(fjx0,tx);
 412             fjy0             = _mm256_add_ps(fjy0,ty);
 413             fjz0             = _mm256_add_ps(fjz0,tz);
 414
 415             fjptrA             = f+j_coord_offsetA;
 416             fjptrB             = f+j_coord_offsetB;
 417             fjptrC             = f+j_coord_offsetC;
 418             fjptrD             = f+j_coord_offsetD;
 419             fjptrE             = f+j_coord_offsetE;
 420             fjptrF             = f+j_coord_offsetF;
 421             fjptrG             = f+j_coord_offsetG;
 422             fjptrH             = f+j_coord_offsetH;
 423
 424             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 425
 426             /* Inner loop uses 119 flops */
 427         }
 428
 429         if(jidx<j_index_end)
 430         {
 431
 432             /* Get j neighbor index, and coordinate index */
 433             jnrlistA         = jjnr[jidx];
 434             jnrlistB         = jjnr[jidx+1];
 435             jnrlistC         = jjnr[jidx+2];
 436             jnrlistD         = jjnr[jidx+3];
 437             jnrlistE         = jjnr[jidx+4];
 438             jnrlistF         = jjnr[jidx+5];
 439             jnrlistG         = jjnr[jidx+6];
 440             jnrlistH         = jjnr[jidx+7];
 441             /* Sign of each element will be negative for non-real atoms.
 442              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 443              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 444              */
 445             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 446                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 447
 448             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 449             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 450             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 451             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 452             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 453             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 454             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 455             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 456             j_coord_offsetA  = DIM*jnrA;
 457             j_coord_offsetB  = DIM*jnrB;
 458             j_coord_offsetC  = DIM*jnrC;
 459             j_coord_offsetD  = DIM*jnrD;
 460             j_coord_offsetE  = DIM*jnrE;
 461             j_coord_offsetF  = DIM*jnrF;
 462             j_coord_offsetG  = DIM*jnrG;
 463             j_coord_offsetH  = DIM*jnrH;
 464
 465             /* load j atom coordinates */
 466             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 467                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 468                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 469                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 470                                                  &jx0,&jy0,&jz0);
 471
 472             /* Calculate displacement vector */
 473             dx00             = _mm256_sub_ps(ix0,jx0);
 474             dy00             = _mm256_sub_ps(iy0,jy0);
 475             dz00             = _mm256_sub_ps(iz0,jz0);
 476             dx10             = _mm256_sub_ps(ix1,jx0);
 477             dy10             = _mm256_sub_ps(iy1,jy0);
 478             dz10             = _mm256_sub_ps(iz1,jz0);
 479             dx20             = _mm256_sub_ps(ix2,jx0);
 480             dy20             = _mm256_sub_ps(iy2,jy0);
 481             dz20             = _mm256_sub_ps(iz2,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 485             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 486             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 487
 488             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 489             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 490             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 491
 492             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 493             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 494             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 495
 496             /* Load parameters for j particles */
 497             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 498                                                                  charge+jnrC+0,charge+jnrD+0,
 499                                                                  charge+jnrE+0,charge+jnrF+0,
 500                                                                  charge+jnrG+0,charge+jnrH+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502             vdwjidx0B        = 2*vdwtype[jnrB+0];
 503             vdwjidx0C        = 2*vdwtype[jnrC+0];
 504             vdwjidx0D        = 2*vdwtype[jnrD+0];
 505             vdwjidx0E        = 2*vdwtype[jnrE+0];
 506             vdwjidx0F        = 2*vdwtype[jnrF+0];
 507             vdwjidx0G        = 2*vdwtype[jnrG+0];
 508             vdwjidx0H        = 2*vdwtype[jnrH+0];
 509
 510             fjx0             = _mm256_setzero_ps();
 511             fjy0             = _mm256_setzero_ps();
 512             fjz0             = _mm256_setzero_ps();
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             r00              = _mm256_mul_ps(rsq00,rinv00);
 519             r00              = _mm256_andnot_ps(dummy_mask,r00);
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq00             = _mm256_mul_ps(iq0,jq0);
 523             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 524                                             vdwioffsetptr0+vdwjidx0B,
 525                                             vdwioffsetptr0+vdwjidx0C,
 526                                             vdwioffsetptr0+vdwjidx0D,
 527                                             vdwioffsetptr0+vdwjidx0E,
 528                                             vdwioffsetptr0+vdwjidx0F,
 529                                             vdwioffsetptr0+vdwjidx0G,
 530                                             vdwioffsetptr0+vdwjidx0H,
 531                                             &c6_00,&c12_00);
 532
 533             /* Calculate table index by multiplying r with table scale and truncate to integer */
 534             rt               = _mm256_mul_ps(r00,vftabscale);
 535             vfitab           = _mm256_cvttps_epi32(rt);
 536             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 537             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 538             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 539             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 540             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 541             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 542
 543             /* COULOMB ELECTROSTATICS */
 544             velec            = _mm256_mul_ps(qq00,rinv00);
 545             felec            = _mm256_mul_ps(velec,rinvsq00);
 546
 547             /* CUBIC SPLINE TABLE DISPERSION */
 548             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 549                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 550             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 551                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 552             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 553                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 554             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 555                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 556             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 557             Heps             = _mm256_mul_ps(vfeps,H);
 558             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 559             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 560             vvdw6            = _mm256_mul_ps(c6_00,VV);
 561             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 562             fvdw6            = _mm256_mul_ps(c6_00,FF);
 563
 564             /* CUBIC SPLINE TABLE REPULSION */
 565             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 566             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 567             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 568                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 569             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 570                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 571             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 572                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 573             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 574                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 575             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 576             Heps             = _mm256_mul_ps(vfeps,H);
 577             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 578             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 579             vvdw12           = _mm256_mul_ps(c12_00,VV);
 580             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 581             fvdw12           = _mm256_mul_ps(c12_00,FF);
 582             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 583             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velec            = _mm256_andnot_ps(dummy_mask,velec);
 587             velecsum         = _mm256_add_ps(velecsum,velec);
 588             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 589             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 590
 591             fscal            = _mm256_add_ps(felec,fvdw);
 592
 593             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm256_mul_ps(fscal,dx00);
 597             ty               = _mm256_mul_ps(fscal,dy00);
 598             tz               = _mm256_mul_ps(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm256_add_ps(fix0,tx);
 602             fiy0             = _mm256_add_ps(fiy0,ty);
 603             fiz0             = _mm256_add_ps(fiz0,tz);
 604
 605             fjx0             = _mm256_add_ps(fjx0,tx);
 606             fjy0             = _mm256_add_ps(fjy0,ty);
 607             fjz0             = _mm256_add_ps(fjz0,tz);
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq10             = _mm256_mul_ps(iq1,jq0);
 615
 616             /* COULOMB ELECTROSTATICS */
 617             velec            = _mm256_mul_ps(qq10,rinv10);
 618             felec            = _mm256_mul_ps(velec,rinvsq10);
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             velec            = _mm256_andnot_ps(dummy_mask,velec);
 622             velecsum         = _mm256_add_ps(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 627
 628             /* Calculate temporary vectorial force */
 629             tx               = _mm256_mul_ps(fscal,dx10);
 630             ty               = _mm256_mul_ps(fscal,dy10);
 631             tz               = _mm256_mul_ps(fscal,dz10);
 632
 633             /* Update vectorial force */
 634             fix1             = _mm256_add_ps(fix1,tx);
 635             fiy1             = _mm256_add_ps(fiy1,ty);
 636             fiz1             = _mm256_add_ps(fiz1,tz);
 637
 638             fjx0             = _mm256_add_ps(fjx0,tx);
 639             fjy0             = _mm256_add_ps(fjy0,ty);
 640             fjz0             = _mm256_add_ps(fjz0,tz);
 641
 642             /**************************
 643              * CALCULATE INTERACTIONS *
 644              **************************/
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq20             = _mm256_mul_ps(iq2,jq0);
 648
 649             /* COULOMB ELECTROSTATICS */
 650             velec            = _mm256_mul_ps(qq20,rinv20);
 651             felec            = _mm256_mul_ps(velec,rinvsq20);
 652
 653             /* Update potential sum for this i atom from the interaction with this j atom. */
 654             velec            = _mm256_andnot_ps(dummy_mask,velec);
 655             velecsum         = _mm256_add_ps(velecsum,velec);
 656
 657             fscal            = felec;
 658
 659             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm256_mul_ps(fscal,dx20);
 663             ty               = _mm256_mul_ps(fscal,dy20);
 664             tz               = _mm256_mul_ps(fscal,dz20);
 665
 666             /* Update vectorial force */
 667             fix2             = _mm256_add_ps(fix2,tx);
 668             fiy2             = _mm256_add_ps(fiy2,ty);
 669             fiz2             = _mm256_add_ps(fiz2,tz);
 670
 671             fjx0             = _mm256_add_ps(fjx0,tx);
 672             fjy0             = _mm256_add_ps(fjy0,ty);
 673             fjz0             = _mm256_add_ps(fjz0,tz);
 674
 675             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 676             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 677             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 678             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 679             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 680             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 681             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 682             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 683
 684             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 685
 686             /* Inner loop uses 120 flops */
 687         }
 688
 689         /* End of innermost loop */
 690
 691         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 692                                                  f+i_coord_offset,fshift+i_shift_offset);
 693
 694         ggid                        = gid[iidx];
 695         /* Update potential energies */
 696         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 697         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 698
 699         /* Increment number of inner iterations */
 700         inneriter                  += j_index_end - j_index_start;
 701
 702         /* Outer loop uses 20 flops */
 703     }
 704
 705     /* Increment number of outer iterations */
 706     outeriter        += nri;
 707
 708     /* Update outer/inner flops */
 709
 710     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 711 }
 712 /*
 713  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_single
 714  * Electrostatics interaction: Coulomb
 715  * VdW interaction:            CubicSplineTable
 716  * Geometry:                   Water3-Particle
 717  * Calculate force/pot:        Force
 718  */
 719 void
 720 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_single
 721                     (t_nblist                    * gmx_restrict       nlist,
 722                      rvec                        * gmx_restrict          xx,
 723                      rvec                        * gmx_restrict          ff,
 724                      t_forcerec                  * gmx_restrict          fr,
 725                      t_mdatoms                   * gmx_restrict     mdatoms,
 726                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 727                      t_nrnb                      * gmx_restrict        nrnb)
 728 {
 729     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 730      * just 0 for non-waters.
 731      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 732      * jnr indices corresponding to data put in the four positions in the SIMD register.
 733      */
 734     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 735     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 736     int              jnrA,jnrB,jnrC,jnrD;
 737     int              jnrE,jnrF,jnrG,jnrH;
 738     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 739     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 740     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 741     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 742     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 743     real             rcutoff_scalar;
 744     real             *shiftvec,*fshift,*x,*f;
 745     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 746     real             scratch[4*DIM];
 747     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 748     real *           vdwioffsetptr0;
 749     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 750     real *           vdwioffsetptr1;
 751     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 752     real *           vdwioffsetptr2;
 753     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 754     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 755     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 756     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 757     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 758     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 759     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 760     real             *charge;
 761     int              nvdwtype;
 762     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 763     int              *vdwtype;
 764     real             *vdwparam;
 765     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 766     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 767     __m256i          vfitab;
 768     __m128i          vfitab_lo,vfitab_hi;
 769     __m128i          ifour       = _mm_set1_epi32(4);
 770     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 771     real             *vftab;
 772     __m256           dummy_mask,cutoff_mask;
 773     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 774     __m256           one     = _mm256_set1_ps(1.0);
 775     __m256           two     = _mm256_set1_ps(2.0);
 776     x                = xx[0];
 777     f                = ff[0];
 778
 779     nri              = nlist->nri;
 780     iinr             = nlist->iinr;
 781     jindex           = nlist->jindex;
 782     jjnr             = nlist->jjnr;
 783     shiftidx         = nlist->shift;
 784     gid              = nlist->gid;
 785     shiftvec         = fr->shift_vec[0];
 786     fshift           = fr->fshift[0];
 787     facel            = _mm256_set1_ps(fr->epsfac);
 788     charge           = mdatoms->chargeA;
 789     nvdwtype         = fr->ntype;
 790     vdwparam         = fr->nbfp;
 791     vdwtype          = mdatoms->typeA;
 792
 793     vftab            = kernel_data->table_vdw->data;
 794     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 795
 796     /* Setup water-specific parameters */
 797     inr              = nlist->iinr[0];
 798     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 799     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 800     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 801     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 802
 803     /* Avoid stupid compiler warnings */
 804     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 805     j_coord_offsetA = 0;
 806     j_coord_offsetB = 0;
 807     j_coord_offsetC = 0;
 808     j_coord_offsetD = 0;
 809     j_coord_offsetE = 0;
 810     j_coord_offsetF = 0;
 811     j_coord_offsetG = 0;
 812     j_coord_offsetH = 0;
 813
 814     outeriter        = 0;
 815     inneriter        = 0;
 816
 817     for(iidx=0;iidx<4*DIM;iidx++)
 818     {
 819         scratch[iidx] = 0.0;
 820     }
 821
 822     /* Start outer loop over neighborlists */
 823     for(iidx=0; iidx<nri; iidx++)
 824     {
 825         /* Load shift vector for this list */
 826         i_shift_offset   = DIM*shiftidx[iidx];
 827
 828         /* Load limits for loop over neighbors */
 829         j_index_start    = jindex[iidx];
 830         j_index_end      = jindex[iidx+1];
 831
 832         /* Get outer coordinate index */
 833         inr              = iinr[iidx];
 834         i_coord_offset   = DIM*inr;
 835
 836         /* Load i particle coords and add shift vector */
 837         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 838                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 839
 840         fix0             = _mm256_setzero_ps();
 841         fiy0             = _mm256_setzero_ps();
 842         fiz0             = _mm256_setzero_ps();
 843         fix1             = _mm256_setzero_ps();
 844         fiy1             = _mm256_setzero_ps();
 845         fiz1             = _mm256_setzero_ps();
 846         fix2             = _mm256_setzero_ps();
 847         fiy2             = _mm256_setzero_ps();
 848         fiz2             = _mm256_setzero_ps();
 849
 850         /* Start inner kernel loop */
 851         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 852         {
 853
 854             /* Get j neighbor index, and coordinate index */
 855             jnrA             = jjnr[jidx];
 856             jnrB             = jjnr[jidx+1];
 857             jnrC             = jjnr[jidx+2];
 858             jnrD             = jjnr[jidx+3];
 859             jnrE             = jjnr[jidx+4];
 860             jnrF             = jjnr[jidx+5];
 861             jnrG             = jjnr[jidx+6];
 862             jnrH             = jjnr[jidx+7];
 863             j_coord_offsetA  = DIM*jnrA;
 864             j_coord_offsetB  = DIM*jnrB;
 865             j_coord_offsetC  = DIM*jnrC;
 866             j_coord_offsetD  = DIM*jnrD;
 867             j_coord_offsetE  = DIM*jnrE;
 868             j_coord_offsetF  = DIM*jnrF;
 869             j_coord_offsetG  = DIM*jnrG;
 870             j_coord_offsetH  = DIM*jnrH;
 871
 872             /* load j atom coordinates */
 873             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 874                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 875                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 876                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 877                                                  &jx0,&jy0,&jz0);
 878
 879             /* Calculate displacement vector */
 880             dx00             = _mm256_sub_ps(ix0,jx0);
 881             dy00             = _mm256_sub_ps(iy0,jy0);
 882             dz00             = _mm256_sub_ps(iz0,jz0);
 883             dx10             = _mm256_sub_ps(ix1,jx0);
 884             dy10             = _mm256_sub_ps(iy1,jy0);
 885             dz10             = _mm256_sub_ps(iz1,jz0);
 886             dx20             = _mm256_sub_ps(ix2,jx0);
 887             dy20             = _mm256_sub_ps(iy2,jy0);
 888             dz20             = _mm256_sub_ps(iz2,jz0);
 889
 890             /* Calculate squared distance and things based on it */
 891             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 892             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 893             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 894
 895             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 896             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 897             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 898
 899             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 900             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 901             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 902
 903             /* Load parameters for j particles */
 904             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 905                                                                  charge+jnrC+0,charge+jnrD+0,
 906                                                                  charge+jnrE+0,charge+jnrF+0,
 907                                                                  charge+jnrG+0,charge+jnrH+0);
 908             vdwjidx0A        = 2*vdwtype[jnrA+0];
 909             vdwjidx0B        = 2*vdwtype[jnrB+0];
 910             vdwjidx0C        = 2*vdwtype[jnrC+0];
 911             vdwjidx0D        = 2*vdwtype[jnrD+0];
 912             vdwjidx0E        = 2*vdwtype[jnrE+0];
 913             vdwjidx0F        = 2*vdwtype[jnrF+0];
 914             vdwjidx0G        = 2*vdwtype[jnrG+0];
 915             vdwjidx0H        = 2*vdwtype[jnrH+0];
 916
 917             fjx0             = _mm256_setzero_ps();
 918             fjy0             = _mm256_setzero_ps();
 919             fjz0             = _mm256_setzero_ps();
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             r00              = _mm256_mul_ps(rsq00,rinv00);
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             qq00             = _mm256_mul_ps(iq0,jq0);
 929             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 930                                             vdwioffsetptr0+vdwjidx0B,
 931                                             vdwioffsetptr0+vdwjidx0C,
 932                                             vdwioffsetptr0+vdwjidx0D,
 933                                             vdwioffsetptr0+vdwjidx0E,
 934                                             vdwioffsetptr0+vdwjidx0F,
 935                                             vdwioffsetptr0+vdwjidx0G,
 936                                             vdwioffsetptr0+vdwjidx0H,
 937                                             &c6_00,&c12_00);
 938
 939             /* Calculate table index by multiplying r with table scale and truncate to integer */
 940             rt               = _mm256_mul_ps(r00,vftabscale);
 941             vfitab           = _mm256_cvttps_epi32(rt);
 942             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 943             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 944             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 945             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 946             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 947             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 948
 949             /* COULOMB ELECTROSTATICS */
 950             velec            = _mm256_mul_ps(qq00,rinv00);
 951             felec            = _mm256_mul_ps(velec,rinvsq00);
 952
 953             /* CUBIC SPLINE TABLE DISPERSION */
 954             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 955                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 956             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 957                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 958             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 959                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 960             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 961                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 962             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 963             Heps             = _mm256_mul_ps(vfeps,H);
 964             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 965             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 966             fvdw6            = _mm256_mul_ps(c6_00,FF);
 967
 968             /* CUBIC SPLINE TABLE REPULSION */
 969             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 970             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 971             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 972                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 973             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 974                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 975             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 976                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 977             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 978                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 979             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 980             Heps             = _mm256_mul_ps(vfeps,H);
 981             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 982             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 983             fvdw12           = _mm256_mul_ps(c12_00,FF);
 984             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 985
 986             fscal            = _mm256_add_ps(felec,fvdw);
 987
 988             /* Calculate temporary vectorial force */
 989             tx               = _mm256_mul_ps(fscal,dx00);
 990             ty               = _mm256_mul_ps(fscal,dy00);
 991             tz               = _mm256_mul_ps(fscal,dz00);
 992
 993             /* Update vectorial force */
 994             fix0             = _mm256_add_ps(fix0,tx);
 995             fiy0             = _mm256_add_ps(fiy0,ty);
 996             fiz0             = _mm256_add_ps(fiz0,tz);
 997
 998             fjx0             = _mm256_add_ps(fjx0,tx);
 999             fjy0             = _mm256_add_ps(fjy0,ty);
1000             fjz0             = _mm256_add_ps(fjz0,tz);
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq10             = _mm256_mul_ps(iq1,jq0);
1008
1009             /* COULOMB ELECTROSTATICS */
1010             velec            = _mm256_mul_ps(qq10,rinv10);
1011             felec            = _mm256_mul_ps(velec,rinvsq10);
1012
1013             fscal            = felec;
1014
1015             /* Calculate temporary vectorial force */
1016             tx               = _mm256_mul_ps(fscal,dx10);
1017             ty               = _mm256_mul_ps(fscal,dy10);
1018             tz               = _mm256_mul_ps(fscal,dz10);
1019
1020             /* Update vectorial force */
1021             fix1             = _mm256_add_ps(fix1,tx);
1022             fiy1             = _mm256_add_ps(fiy1,ty);
1023             fiz1             = _mm256_add_ps(fiz1,tz);
1024
1025             fjx0             = _mm256_add_ps(fjx0,tx);
1026             fjy0             = _mm256_add_ps(fjy0,ty);
1027             fjz0             = _mm256_add_ps(fjz0,tz);
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             /* Compute parameters for interactions between i and j atoms */
1034             qq20             = _mm256_mul_ps(iq2,jq0);
1035
1036             /* COULOMB ELECTROSTATICS */
1037             velec            = _mm256_mul_ps(qq20,rinv20);
1038             felec            = _mm256_mul_ps(velec,rinvsq20);
1039
1040             fscal            = felec;
1041
1042             /* Calculate temporary vectorial force */
1043             tx               = _mm256_mul_ps(fscal,dx20);
1044             ty               = _mm256_mul_ps(fscal,dy20);
1045             tz               = _mm256_mul_ps(fscal,dz20);
1046
1047             /* Update vectorial force */
1048             fix2             = _mm256_add_ps(fix2,tx);
1049             fiy2             = _mm256_add_ps(fiy2,ty);
1050             fiz2             = _mm256_add_ps(fiz2,tz);
1051
1052             fjx0             = _mm256_add_ps(fjx0,tx);
1053             fjy0             = _mm256_add_ps(fjy0,ty);
1054             fjz0             = _mm256_add_ps(fjz0,tz);
1055
1056             fjptrA             = f+j_coord_offsetA;
1057             fjptrB             = f+j_coord_offsetB;
1058             fjptrC             = f+j_coord_offsetC;
1059             fjptrD             = f+j_coord_offsetD;
1060             fjptrE             = f+j_coord_offsetE;
1061             fjptrF             = f+j_coord_offsetF;
1062             fjptrG             = f+j_coord_offsetG;
1063             fjptrH             = f+j_coord_offsetH;
1064
1065             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1066
1067             /* Inner loop uses 108 flops */
1068         }
1069
1070         if(jidx<j_index_end)
1071         {
1072
1073             /* Get j neighbor index, and coordinate index */
1074             jnrlistA         = jjnr[jidx];
1075             jnrlistB         = jjnr[jidx+1];
1076             jnrlistC         = jjnr[jidx+2];
1077             jnrlistD         = jjnr[jidx+3];
1078             jnrlistE         = jjnr[jidx+4];
1079             jnrlistF         = jjnr[jidx+5];
1080             jnrlistG         = jjnr[jidx+6];
1081             jnrlistH         = jjnr[jidx+7];
1082             /* Sign of each element will be negative for non-real atoms.
1083              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1084              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1085              */
1086             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1087                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1088
1089             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1090             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1091             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1092             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1093             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1094             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1095             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1096             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1097             j_coord_offsetA  = DIM*jnrA;
1098             j_coord_offsetB  = DIM*jnrB;
1099             j_coord_offsetC  = DIM*jnrC;
1100             j_coord_offsetD  = DIM*jnrD;
1101             j_coord_offsetE  = DIM*jnrE;
1102             j_coord_offsetF  = DIM*jnrF;
1103             j_coord_offsetG  = DIM*jnrG;
1104             j_coord_offsetH  = DIM*jnrH;
1105
1106             /* load j atom coordinates */
1107             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1108                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1109                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1110                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1111                                                  &jx0,&jy0,&jz0);
1112
1113             /* Calculate displacement vector */
1114             dx00             = _mm256_sub_ps(ix0,jx0);
1115             dy00             = _mm256_sub_ps(iy0,jy0);
1116             dz00             = _mm256_sub_ps(iz0,jz0);
1117             dx10             = _mm256_sub_ps(ix1,jx0);
1118             dy10             = _mm256_sub_ps(iy1,jy0);
1119             dz10             = _mm256_sub_ps(iz1,jz0);
1120             dx20             = _mm256_sub_ps(ix2,jx0);
1121             dy20             = _mm256_sub_ps(iy2,jy0);
1122             dz20             = _mm256_sub_ps(iz2,jz0);
1123
1124             /* Calculate squared distance and things based on it */
1125             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1126             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1127             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1128
1129             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1130             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1131             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1132
1133             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1134             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1135             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1136
1137             /* Load parameters for j particles */
1138             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1139                                                                  charge+jnrC+0,charge+jnrD+0,
1140                                                                  charge+jnrE+0,charge+jnrF+0,
1141                                                                  charge+jnrG+0,charge+jnrH+0);
1142             vdwjidx0A        = 2*vdwtype[jnrA+0];
1143             vdwjidx0B        = 2*vdwtype[jnrB+0];
1144             vdwjidx0C        = 2*vdwtype[jnrC+0];
1145             vdwjidx0D        = 2*vdwtype[jnrD+0];
1146             vdwjidx0E        = 2*vdwtype[jnrE+0];
1147             vdwjidx0F        = 2*vdwtype[jnrF+0];
1148             vdwjidx0G        = 2*vdwtype[jnrG+0];
1149             vdwjidx0H        = 2*vdwtype[jnrH+0];
1150
1151             fjx0             = _mm256_setzero_ps();
1152             fjy0             = _mm256_setzero_ps();
1153             fjz0             = _mm256_setzero_ps();
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             r00              = _mm256_mul_ps(rsq00,rinv00);
1160             r00              = _mm256_andnot_ps(dummy_mask,r00);
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             qq00             = _mm256_mul_ps(iq0,jq0);
1164             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1165                                             vdwioffsetptr0+vdwjidx0B,
1166                                             vdwioffsetptr0+vdwjidx0C,
1167                                             vdwioffsetptr0+vdwjidx0D,
1168                                             vdwioffsetptr0+vdwjidx0E,
1169                                             vdwioffsetptr0+vdwjidx0F,
1170                                             vdwioffsetptr0+vdwjidx0G,
1171                                             vdwioffsetptr0+vdwjidx0H,
1172                                             &c6_00,&c12_00);
1173
1174             /* Calculate table index by multiplying r with table scale and truncate to integer */
1175             rt               = _mm256_mul_ps(r00,vftabscale);
1176             vfitab           = _mm256_cvttps_epi32(rt);
1177             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1178             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1179             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1180             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1181             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1182             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1183
1184             /* COULOMB ELECTROSTATICS */
1185             velec            = _mm256_mul_ps(qq00,rinv00);
1186             felec            = _mm256_mul_ps(velec,rinvsq00);
1187
1188             /* CUBIC SPLINE TABLE DISPERSION */
1189             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1190                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1191             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1193             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1195             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1196                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1197             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1198             Heps             = _mm256_mul_ps(vfeps,H);
1199             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1200             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1201             fvdw6            = _mm256_mul_ps(c6_00,FF);
1202
1203             /* CUBIC SPLINE TABLE REPULSION */
1204             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1205             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1206             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1207                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1208             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1210             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1212             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1214             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1215             Heps             = _mm256_mul_ps(vfeps,H);
1216             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1217             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1218             fvdw12           = _mm256_mul_ps(c12_00,FF);
1219             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1220
1221             fscal            = _mm256_add_ps(felec,fvdw);
1222
1223             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1224
1225             /* Calculate temporary vectorial force */
1226             tx               = _mm256_mul_ps(fscal,dx00);
1227             ty               = _mm256_mul_ps(fscal,dy00);
1228             tz               = _mm256_mul_ps(fscal,dz00);
1229
1230             /* Update vectorial force */
1231             fix0             = _mm256_add_ps(fix0,tx);
1232             fiy0             = _mm256_add_ps(fiy0,ty);
1233             fiz0             = _mm256_add_ps(fiz0,tz);
1234
1235             fjx0             = _mm256_add_ps(fjx0,tx);
1236             fjy0             = _mm256_add_ps(fjy0,ty);
1237             fjz0             = _mm256_add_ps(fjz0,tz);
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             /* Compute parameters for interactions between i and j atoms */
1244             qq10             = _mm256_mul_ps(iq1,jq0);
1245
1246             /* COULOMB ELECTROSTATICS */
1247             velec            = _mm256_mul_ps(qq10,rinv10);
1248             felec            = _mm256_mul_ps(velec,rinvsq10);
1249
1250             fscal            = felec;
1251
1252             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1253
1254             /* Calculate temporary vectorial force */
1255             tx               = _mm256_mul_ps(fscal,dx10);
1256             ty               = _mm256_mul_ps(fscal,dy10);
1257             tz               = _mm256_mul_ps(fscal,dz10);
1258
1259             /* Update vectorial force */
1260             fix1             = _mm256_add_ps(fix1,tx);
1261             fiy1             = _mm256_add_ps(fiy1,ty);
1262             fiz1             = _mm256_add_ps(fiz1,tz);
1263
1264             fjx0             = _mm256_add_ps(fjx0,tx);
1265             fjy0             = _mm256_add_ps(fjy0,ty);
1266             fjz0             = _mm256_add_ps(fjz0,tz);
1267
1268             /**************************
1269              * CALCULATE INTERACTIONS *
1270              **************************/
1271
1272             /* Compute parameters for interactions between i and j atoms */
1273             qq20             = _mm256_mul_ps(iq2,jq0);
1274
1275             /* COULOMB ELECTROSTATICS */
1276             velec            = _mm256_mul_ps(qq20,rinv20);
1277             felec            = _mm256_mul_ps(velec,rinvsq20);
1278
1279             fscal            = felec;
1280
1281             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1282
1283             /* Calculate temporary vectorial force */
1284             tx               = _mm256_mul_ps(fscal,dx20);
1285             ty               = _mm256_mul_ps(fscal,dy20);
1286             tz               = _mm256_mul_ps(fscal,dz20);
1287
1288             /* Update vectorial force */
1289             fix2             = _mm256_add_ps(fix2,tx);
1290             fiy2             = _mm256_add_ps(fiy2,ty);
1291             fiz2             = _mm256_add_ps(fiz2,tz);
1292
1293             fjx0             = _mm256_add_ps(fjx0,tx);
1294             fjy0             = _mm256_add_ps(fjy0,ty);
1295             fjz0             = _mm256_add_ps(fjz0,tz);
1296
1297             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1298             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1299             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1300             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1301             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1302             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1303             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1304             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1305
1306             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1307
1308             /* Inner loop uses 109 flops */
1309         }
1310
1311         /* End of innermost loop */
1312
1313         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1314                                                  f+i_coord_offset,fshift+i_shift_offset);
1315
1316         /* Increment number of inner iterations */
1317         inneriter                  += j_index_end - j_index_start;
1318
1319         /* Outer loop uses 18 flops */
1320     }
1321
1322     /* Increment number of outer iterations */
1323     outeriter        += nri;
1324
1325     /* Update outer/inner flops */
1326
1327     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1328 }