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) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256i          vfitab;
 105     __m128i          vfitab_lo,vfitab_hi;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256           dummy_mask,cutoff_mask;
 110     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 111     __m256           one     = _mm256_set1_ps(1.0);
 112     __m256           two     = _mm256_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_ps(fr->ic->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 136     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 137     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 138     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146     j_coord_offsetE = 0;
 147     j_coord_offsetF = 0;
 148     j_coord_offsetG = 0;
 149     j_coord_offsetH = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm256_setzero_ps();
 178         fiy0             = _mm256_setzero_ps();
 179         fiz0             = _mm256_setzero_ps();
 180         fix1             = _mm256_setzero_ps();
 181         fiy1             = _mm256_setzero_ps();
 182         fiz1             = _mm256_setzero_ps();
 183         fix2             = _mm256_setzero_ps();
 184         fiy2             = _mm256_setzero_ps();
 185         fiz2             = _mm256_setzero_ps();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_ps();
 189         vvdwsum          = _mm256_setzero_ps();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             jnrE             = jjnr[jidx+4];
 201             jnrF             = jjnr[jidx+5];
 202             jnrG             = jjnr[jidx+6];
 203             jnrH             = jjnr[jidx+7];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208             j_coord_offsetE  = DIM*jnrE;
 209             j_coord_offsetF  = DIM*jnrF;
 210             j_coord_offsetG  = DIM*jnrG;
 211             j_coord_offsetH  = DIM*jnrH;
 212
 213             /* load j atom coordinates */
 214             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 216                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 217                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 218                                                  &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm256_sub_ps(ix0,jx0);
 222             dy00             = _mm256_sub_ps(iy0,jy0);
 223             dz00             = _mm256_sub_ps(iz0,jz0);
 224             dx10             = _mm256_sub_ps(ix1,jx0);
 225             dy10             = _mm256_sub_ps(iy1,jy0);
 226             dz10             = _mm256_sub_ps(iz1,jz0);
 227             dx20             = _mm256_sub_ps(ix2,jx0);
 228             dy20             = _mm256_sub_ps(iy2,jy0);
 229             dz20             = _mm256_sub_ps(iz2,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 233             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 234             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 235
 236             rinv00           = avx256_invsqrt_f(rsq00);
 237             rinv10           = avx256_invsqrt_f(rsq10);
 238             rinv20           = avx256_invsqrt_f(rsq20);
 239
 240             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 241             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 242             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 246                                                                  charge+jnrC+0,charge+jnrD+0,
 247                                                                  charge+jnrE+0,charge+jnrF+0,
 248                                                                  charge+jnrG+0,charge+jnrH+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251             vdwjidx0C        = 2*vdwtype[jnrC+0];
 252             vdwjidx0D        = 2*vdwtype[jnrD+0];
 253             vdwjidx0E        = 2*vdwtype[jnrE+0];
 254             vdwjidx0F        = 2*vdwtype[jnrF+0];
 255             vdwjidx0G        = 2*vdwtype[jnrG+0];
 256             vdwjidx0H        = 2*vdwtype[jnrH+0];
 257
 258             fjx0             = _mm256_setzero_ps();
 259             fjy0             = _mm256_setzero_ps();
 260             fjz0             = _mm256_setzero_ps();
 261
 262             /**************************
 263              * CALCULATE INTERACTIONS *
 264              **************************/
 265
 266             r00              = _mm256_mul_ps(rsq00,rinv00);
 267
 268             /* Compute parameters for interactions between i and j atoms */
 269             qq00             = _mm256_mul_ps(iq0,jq0);
 270             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 271                                             vdwioffsetptr0+vdwjidx0B,
 272                                             vdwioffsetptr0+vdwjidx0C,
 273                                             vdwioffsetptr0+vdwjidx0D,
 274                                             vdwioffsetptr0+vdwjidx0E,
 275                                             vdwioffsetptr0+vdwjidx0F,
 276                                             vdwioffsetptr0+vdwjidx0G,
 277                                             vdwioffsetptr0+vdwjidx0H,
 278                                             &c6_00,&c12_00);
 279
 280             /* Calculate table index by multiplying r with table scale and truncate to integer */
 281             rt               = _mm256_mul_ps(r00,vftabscale);
 282             vfitab           = _mm256_cvttps_epi32(rt);
 283             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 284             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 285             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 286             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 287             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 288             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 289
 290             /* COULOMB ELECTROSTATICS */
 291             velec            = _mm256_mul_ps(qq00,rinv00);
 292             felec            = _mm256_mul_ps(velec,rinvsq00);
 293
 294             /* CUBIC SPLINE TABLE DISPERSION */
 295             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 297             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 298                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 299             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 300                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 301             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 302                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 303             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 304             Heps             = _mm256_mul_ps(vfeps,H);
 305             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 306             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 307             vvdw6            = _mm256_mul_ps(c6_00,VV);
 308             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 309             fvdw6            = _mm256_mul_ps(c6_00,FF);
 310
 311             /* CUBIC SPLINE TABLE REPULSION */
 312             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 313             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 314             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 316             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 317                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 318             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 320             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 322             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 323             Heps             = _mm256_mul_ps(vfeps,H);
 324             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 325             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 326             vvdw12           = _mm256_mul_ps(c12_00,VV);
 327             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 328             fvdw12           = _mm256_mul_ps(c12_00,FF);
 329             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 330             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm256_add_ps(velecsum,velec);
 334             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 335
 336             fscal            = _mm256_add_ps(felec,fvdw);
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm256_mul_ps(fscal,dx00);
 340             ty               = _mm256_mul_ps(fscal,dy00);
 341             tz               = _mm256_mul_ps(fscal,dz00);
 342
 343             /* Update vectorial force */
 344             fix0             = _mm256_add_ps(fix0,tx);
 345             fiy0             = _mm256_add_ps(fiy0,ty);
 346             fiz0             = _mm256_add_ps(fiz0,tz);
 347
 348             fjx0             = _mm256_add_ps(fjx0,tx);
 349             fjy0             = _mm256_add_ps(fjy0,ty);
 350             fjz0             = _mm256_add_ps(fjz0,tz);
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq10             = _mm256_mul_ps(iq1,jq0);
 358
 359             /* COULOMB ELECTROSTATICS */
 360             velec            = _mm256_mul_ps(qq10,rinv10);
 361             felec            = _mm256_mul_ps(velec,rinvsq10);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velecsum         = _mm256_add_ps(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm256_mul_ps(fscal,dx10);
 370             ty               = _mm256_mul_ps(fscal,dy10);
 371             tz               = _mm256_mul_ps(fscal,dz10);
 372
 373             /* Update vectorial force */
 374             fix1             = _mm256_add_ps(fix1,tx);
 375             fiy1             = _mm256_add_ps(fiy1,ty);
 376             fiz1             = _mm256_add_ps(fiz1,tz);
 377
 378             fjx0             = _mm256_add_ps(fjx0,tx);
 379             fjy0             = _mm256_add_ps(fjy0,ty);
 380             fjz0             = _mm256_add_ps(fjz0,tz);
 381
 382             /**************************
 383              * CALCULATE INTERACTIONS *
 384              **************************/
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq20             = _mm256_mul_ps(iq2,jq0);
 388
 389             /* COULOMB ELECTROSTATICS */
 390             velec            = _mm256_mul_ps(qq20,rinv20);
 391             felec            = _mm256_mul_ps(velec,rinvsq20);
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm256_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm256_mul_ps(fscal,dx20);
 400             ty               = _mm256_mul_ps(fscal,dy20);
 401             tz               = _mm256_mul_ps(fscal,dz20);
 402
 403             /* Update vectorial force */
 404             fix2             = _mm256_add_ps(fix2,tx);
 405             fiy2             = _mm256_add_ps(fiy2,ty);
 406             fiz2             = _mm256_add_ps(fiz2,tz);
 407
 408             fjx0             = _mm256_add_ps(fjx0,tx);
 409             fjy0             = _mm256_add_ps(fjy0,ty);
 410             fjz0             = _mm256_add_ps(fjz0,tz);
 411
 412             fjptrA             = f+j_coord_offsetA;
 413             fjptrB             = f+j_coord_offsetB;
 414             fjptrC             = f+j_coord_offsetC;
 415             fjptrD             = f+j_coord_offsetD;
 416             fjptrE             = f+j_coord_offsetE;
 417             fjptrF             = f+j_coord_offsetF;
 418             fjptrG             = f+j_coord_offsetG;
 419             fjptrH             = f+j_coord_offsetH;
 420
 421             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 422
 423             /* Inner loop uses 119 flops */
 424         }
 425
 426         if(jidx<j_index_end)
 427         {
 428
 429             /* Get j neighbor index, and coordinate index */
 430             jnrlistA         = jjnr[jidx];
 431             jnrlistB         = jjnr[jidx+1];
 432             jnrlistC         = jjnr[jidx+2];
 433             jnrlistD         = jjnr[jidx+3];
 434             jnrlistE         = jjnr[jidx+4];
 435             jnrlistF         = jjnr[jidx+5];
 436             jnrlistG         = jjnr[jidx+6];
 437             jnrlistH         = jjnr[jidx+7];
 438             /* Sign of each element will be negative for non-real atoms.
 439              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 440              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 441              */
 442             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 443                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 444
 445             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 446             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 447             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 448             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 449             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 450             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 451             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 452             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 453             j_coord_offsetA  = DIM*jnrA;
 454             j_coord_offsetB  = DIM*jnrB;
 455             j_coord_offsetC  = DIM*jnrC;
 456             j_coord_offsetD  = DIM*jnrD;
 457             j_coord_offsetE  = DIM*jnrE;
 458             j_coord_offsetF  = DIM*jnrF;
 459             j_coord_offsetG  = DIM*jnrG;
 460             j_coord_offsetH  = DIM*jnrH;
 461
 462             /* load j atom coordinates */
 463             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 464                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 465                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 466                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 467                                                  &jx0,&jy0,&jz0);
 468
 469             /* Calculate displacement vector */
 470             dx00             = _mm256_sub_ps(ix0,jx0);
 471             dy00             = _mm256_sub_ps(iy0,jy0);
 472             dz00             = _mm256_sub_ps(iz0,jz0);
 473             dx10             = _mm256_sub_ps(ix1,jx0);
 474             dy10             = _mm256_sub_ps(iy1,jy0);
 475             dz10             = _mm256_sub_ps(iz1,jz0);
 476             dx20             = _mm256_sub_ps(ix2,jx0);
 477             dy20             = _mm256_sub_ps(iy2,jy0);
 478             dz20             = _mm256_sub_ps(iz2,jz0);
 479
 480             /* Calculate squared distance and things based on it */
 481             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 482             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 483             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 484
 485             rinv00           = avx256_invsqrt_f(rsq00);
 486             rinv10           = avx256_invsqrt_f(rsq10);
 487             rinv20           = avx256_invsqrt_f(rsq20);
 488
 489             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 490             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 491             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 492
 493             /* Load parameters for j particles */
 494             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 495                                                                  charge+jnrC+0,charge+jnrD+0,
 496                                                                  charge+jnrE+0,charge+jnrF+0,
 497                                                                  charge+jnrG+0,charge+jnrH+0);
 498             vdwjidx0A        = 2*vdwtype[jnrA+0];
 499             vdwjidx0B        = 2*vdwtype[jnrB+0];
 500             vdwjidx0C        = 2*vdwtype[jnrC+0];
 501             vdwjidx0D        = 2*vdwtype[jnrD+0];
 502             vdwjidx0E        = 2*vdwtype[jnrE+0];
 503             vdwjidx0F        = 2*vdwtype[jnrF+0];
 504             vdwjidx0G        = 2*vdwtype[jnrG+0];
 505             vdwjidx0H        = 2*vdwtype[jnrH+0];
 506
 507             fjx0             = _mm256_setzero_ps();
 508             fjy0             = _mm256_setzero_ps();
 509             fjz0             = _mm256_setzero_ps();
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             r00              = _mm256_mul_ps(rsq00,rinv00);
 516             r00              = _mm256_andnot_ps(dummy_mask,r00);
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             qq00             = _mm256_mul_ps(iq0,jq0);
 520             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 521                                             vdwioffsetptr0+vdwjidx0B,
 522                                             vdwioffsetptr0+vdwjidx0C,
 523                                             vdwioffsetptr0+vdwjidx0D,
 524                                             vdwioffsetptr0+vdwjidx0E,
 525                                             vdwioffsetptr0+vdwjidx0F,
 526                                             vdwioffsetptr0+vdwjidx0G,
 527                                             vdwioffsetptr0+vdwjidx0H,
 528                                             &c6_00,&c12_00);
 529
 530             /* Calculate table index by multiplying r with table scale and truncate to integer */
 531             rt               = _mm256_mul_ps(r00,vftabscale);
 532             vfitab           = _mm256_cvttps_epi32(rt);
 533             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 534             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 535             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 536             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 537             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 538             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 539
 540             /* COULOMB ELECTROSTATICS */
 541             velec            = _mm256_mul_ps(qq00,rinv00);
 542             felec            = _mm256_mul_ps(velec,rinvsq00);
 543
 544             /* CUBIC SPLINE TABLE DISPERSION */
 545             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 546                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 547             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 548                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 549             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 550                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 551             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 552                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 553             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 554             Heps             = _mm256_mul_ps(vfeps,H);
 555             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 556             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 557             vvdw6            = _mm256_mul_ps(c6_00,VV);
 558             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 559             fvdw6            = _mm256_mul_ps(c6_00,FF);
 560
 561             /* CUBIC SPLINE TABLE REPULSION */
 562             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 563             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 564             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 565                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 566             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 567                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 568             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 569                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 570             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 571                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 572             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 573             Heps             = _mm256_mul_ps(vfeps,H);
 574             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 575             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 576             vvdw12           = _mm256_mul_ps(c12_00,VV);
 577             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 578             fvdw12           = _mm256_mul_ps(c12_00,FF);
 579             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 580             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm256_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm256_add_ps(velecsum,velec);
 585             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 586             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 587
 588             fscal            = _mm256_add_ps(felec,fvdw);
 589
 590             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm256_mul_ps(fscal,dx00);
 594             ty               = _mm256_mul_ps(fscal,dy00);
 595             tz               = _mm256_mul_ps(fscal,dz00);
 596
 597             /* Update vectorial force */
 598             fix0             = _mm256_add_ps(fix0,tx);
 599             fiy0             = _mm256_add_ps(fiy0,ty);
 600             fiz0             = _mm256_add_ps(fiz0,tz);
 601
 602             fjx0             = _mm256_add_ps(fjx0,tx);
 603             fjy0             = _mm256_add_ps(fjy0,ty);
 604             fjz0             = _mm256_add_ps(fjz0,tz);
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq10             = _mm256_mul_ps(iq1,jq0);
 612
 613             /* COULOMB ELECTROSTATICS */
 614             velec            = _mm256_mul_ps(qq10,rinv10);
 615             felec            = _mm256_mul_ps(velec,rinvsq10);
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velec            = _mm256_andnot_ps(dummy_mask,velec);
 619             velecsum         = _mm256_add_ps(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm256_mul_ps(fscal,dx10);
 627             ty               = _mm256_mul_ps(fscal,dy10);
 628             tz               = _mm256_mul_ps(fscal,dz10);
 629
 630             /* Update vectorial force */
 631             fix1             = _mm256_add_ps(fix1,tx);
 632             fiy1             = _mm256_add_ps(fiy1,ty);
 633             fiz1             = _mm256_add_ps(fiz1,tz);
 634
 635             fjx0             = _mm256_add_ps(fjx0,tx);
 636             fjy0             = _mm256_add_ps(fjy0,ty);
 637             fjz0             = _mm256_add_ps(fjz0,tz);
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             /* Compute parameters for interactions between i and j atoms */
 644             qq20             = _mm256_mul_ps(iq2,jq0);
 645
 646             /* COULOMB ELECTROSTATICS */
 647             velec            = _mm256_mul_ps(qq20,rinv20);
 648             felec            = _mm256_mul_ps(velec,rinvsq20);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm256_andnot_ps(dummy_mask,velec);
 652             velecsum         = _mm256_add_ps(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm256_mul_ps(fscal,dx20);
 660             ty               = _mm256_mul_ps(fscal,dy20);
 661             tz               = _mm256_mul_ps(fscal,dz20);
 662
 663             /* Update vectorial force */
 664             fix2             = _mm256_add_ps(fix2,tx);
 665             fiy2             = _mm256_add_ps(fiy2,ty);
 666             fiz2             = _mm256_add_ps(fiz2,tz);
 667
 668             fjx0             = _mm256_add_ps(fjx0,tx);
 669             fjy0             = _mm256_add_ps(fjy0,ty);
 670             fjz0             = _mm256_add_ps(fjz0,tz);
 671
 672             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 673             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 674             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 675             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 676             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 677             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 678             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 679             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 680
 681             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 682
 683             /* Inner loop uses 120 flops */
 684         }
 685
 686         /* End of innermost loop */
 687
 688         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 689                                                  f+i_coord_offset,fshift+i_shift_offset);
 690
 691         ggid                        = gid[iidx];
 692         /* Update potential energies */
 693         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 694         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 695
 696         /* Increment number of inner iterations */
 697         inneriter                  += j_index_end - j_index_start;
 698
 699         /* Outer loop uses 20 flops */
 700     }
 701
 702     /* Increment number of outer iterations */
 703     outeriter        += nri;
 704
 705     /* Update outer/inner flops */
 706
 707     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 708 }
 709 /*
 710  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_single
 711  * Electrostatics interaction: Coulomb
 712  * VdW interaction:            CubicSplineTable
 713  * Geometry:                   Water3-Particle
 714  * Calculate force/pot:        Force
 715  */
 716 void
 717 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_single
 718                     (t_nblist                    * gmx_restrict       nlist,
 719                      rvec                        * gmx_restrict          xx,
 720                      rvec                        * gmx_restrict          ff,
 721                      struct t_forcerec           * gmx_restrict          fr,
 722                      t_mdatoms                   * gmx_restrict     mdatoms,
 723                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 724                      t_nrnb                      * gmx_restrict        nrnb)
 725 {
 726     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 727      * just 0 for non-waters.
 728      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 729      * jnr indices corresponding to data put in the four positions in the SIMD register.
 730      */
 731     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 732     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 733     int              jnrA,jnrB,jnrC,jnrD;
 734     int              jnrE,jnrF,jnrG,jnrH;
 735     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 736     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 737     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 738     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 739     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 740     real             rcutoff_scalar;
 741     real             *shiftvec,*fshift,*x,*f;
 742     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 743     real             scratch[4*DIM];
 744     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 745     real *           vdwioffsetptr0;
 746     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 747     real *           vdwioffsetptr1;
 748     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 749     real *           vdwioffsetptr2;
 750     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 751     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 752     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 753     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 754     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 755     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 756     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 757     real             *charge;
 758     int              nvdwtype;
 759     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 760     int              *vdwtype;
 761     real             *vdwparam;
 762     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 763     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 764     __m256i          vfitab;
 765     __m128i          vfitab_lo,vfitab_hi;
 766     __m128i          ifour       = _mm_set1_epi32(4);
 767     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 768     real             *vftab;
 769     __m256           dummy_mask,cutoff_mask;
 770     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 771     __m256           one     = _mm256_set1_ps(1.0);
 772     __m256           two     = _mm256_set1_ps(2.0);
 773     x                = xx[0];
 774     f                = ff[0];
 775
 776     nri              = nlist->nri;
 777     iinr             = nlist->iinr;
 778     jindex           = nlist->jindex;
 779     jjnr             = nlist->jjnr;
 780     shiftidx         = nlist->shift;
 781     gid              = nlist->gid;
 782     shiftvec         = fr->shift_vec[0];
 783     fshift           = fr->fshift[0];
 784     facel            = _mm256_set1_ps(fr->ic->epsfac);
 785     charge           = mdatoms->chargeA;
 786     nvdwtype         = fr->ntype;
 787     vdwparam         = fr->nbfp;
 788     vdwtype          = mdatoms->typeA;
 789
 790     vftab            = kernel_data->table_vdw->data;
 791     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 792
 793     /* Setup water-specific parameters */
 794     inr              = nlist->iinr[0];
 795     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 796     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 797     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 798     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 799
 800     /* Avoid stupid compiler warnings */
 801     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 802     j_coord_offsetA = 0;
 803     j_coord_offsetB = 0;
 804     j_coord_offsetC = 0;
 805     j_coord_offsetD = 0;
 806     j_coord_offsetE = 0;
 807     j_coord_offsetF = 0;
 808     j_coord_offsetG = 0;
 809     j_coord_offsetH = 0;
 810
 811     outeriter        = 0;
 812     inneriter        = 0;
 813
 814     for(iidx=0;iidx<4*DIM;iidx++)
 815     {
 816         scratch[iidx] = 0.0;
 817     }
 818
 819     /* Start outer loop over neighborlists */
 820     for(iidx=0; iidx<nri; iidx++)
 821     {
 822         /* Load shift vector for this list */
 823         i_shift_offset   = DIM*shiftidx[iidx];
 824
 825         /* Load limits for loop over neighbors */
 826         j_index_start    = jindex[iidx];
 827         j_index_end      = jindex[iidx+1];
 828
 829         /* Get outer coordinate index */
 830         inr              = iinr[iidx];
 831         i_coord_offset   = DIM*inr;
 832
 833         /* Load i particle coords and add shift vector */
 834         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 835                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 836
 837         fix0             = _mm256_setzero_ps();
 838         fiy0             = _mm256_setzero_ps();
 839         fiz0             = _mm256_setzero_ps();
 840         fix1             = _mm256_setzero_ps();
 841         fiy1             = _mm256_setzero_ps();
 842         fiz1             = _mm256_setzero_ps();
 843         fix2             = _mm256_setzero_ps();
 844         fiy2             = _mm256_setzero_ps();
 845         fiz2             = _mm256_setzero_ps();
 846
 847         /* Start inner kernel loop */
 848         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 849         {
 850
 851             /* Get j neighbor index, and coordinate index */
 852             jnrA             = jjnr[jidx];
 853             jnrB             = jjnr[jidx+1];
 854             jnrC             = jjnr[jidx+2];
 855             jnrD             = jjnr[jidx+3];
 856             jnrE             = jjnr[jidx+4];
 857             jnrF             = jjnr[jidx+5];
 858             jnrG             = jjnr[jidx+6];
 859             jnrH             = jjnr[jidx+7];
 860             j_coord_offsetA  = DIM*jnrA;
 861             j_coord_offsetB  = DIM*jnrB;
 862             j_coord_offsetC  = DIM*jnrC;
 863             j_coord_offsetD  = DIM*jnrD;
 864             j_coord_offsetE  = DIM*jnrE;
 865             j_coord_offsetF  = DIM*jnrF;
 866             j_coord_offsetG  = DIM*jnrG;
 867             j_coord_offsetH  = DIM*jnrH;
 868
 869             /* load j atom coordinates */
 870             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 871                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 872                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 873                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 874                                                  &jx0,&jy0,&jz0);
 875
 876             /* Calculate displacement vector */
 877             dx00             = _mm256_sub_ps(ix0,jx0);
 878             dy00             = _mm256_sub_ps(iy0,jy0);
 879             dz00             = _mm256_sub_ps(iz0,jz0);
 880             dx10             = _mm256_sub_ps(ix1,jx0);
 881             dy10             = _mm256_sub_ps(iy1,jy0);
 882             dz10             = _mm256_sub_ps(iz1,jz0);
 883             dx20             = _mm256_sub_ps(ix2,jx0);
 884             dy20             = _mm256_sub_ps(iy2,jy0);
 885             dz20             = _mm256_sub_ps(iz2,jz0);
 886
 887             /* Calculate squared distance and things based on it */
 888             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 889             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 890             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 891
 892             rinv00           = avx256_invsqrt_f(rsq00);
 893             rinv10           = avx256_invsqrt_f(rsq10);
 894             rinv20           = avx256_invsqrt_f(rsq20);
 895
 896             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 897             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 898             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 899
 900             /* Load parameters for j particles */
 901             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 902                                                                  charge+jnrC+0,charge+jnrD+0,
 903                                                                  charge+jnrE+0,charge+jnrF+0,
 904                                                                  charge+jnrG+0,charge+jnrH+0);
 905             vdwjidx0A        = 2*vdwtype[jnrA+0];
 906             vdwjidx0B        = 2*vdwtype[jnrB+0];
 907             vdwjidx0C        = 2*vdwtype[jnrC+0];
 908             vdwjidx0D        = 2*vdwtype[jnrD+0];
 909             vdwjidx0E        = 2*vdwtype[jnrE+0];
 910             vdwjidx0F        = 2*vdwtype[jnrF+0];
 911             vdwjidx0G        = 2*vdwtype[jnrG+0];
 912             vdwjidx0H        = 2*vdwtype[jnrH+0];
 913
 914             fjx0             = _mm256_setzero_ps();
 915             fjy0             = _mm256_setzero_ps();
 916             fjz0             = _mm256_setzero_ps();
 917
 918             /**************************
 919              * CALCULATE INTERACTIONS *
 920              **************************/
 921
 922             r00              = _mm256_mul_ps(rsq00,rinv00);
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             qq00             = _mm256_mul_ps(iq0,jq0);
 926             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 927                                             vdwioffsetptr0+vdwjidx0B,
 928                                             vdwioffsetptr0+vdwjidx0C,
 929                                             vdwioffsetptr0+vdwjidx0D,
 930                                             vdwioffsetptr0+vdwjidx0E,
 931                                             vdwioffsetptr0+vdwjidx0F,
 932                                             vdwioffsetptr0+vdwjidx0G,
 933                                             vdwioffsetptr0+vdwjidx0H,
 934                                             &c6_00,&c12_00);
 935
 936             /* Calculate table index by multiplying r with table scale and truncate to integer */
 937             rt               = _mm256_mul_ps(r00,vftabscale);
 938             vfitab           = _mm256_cvttps_epi32(rt);
 939             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 940             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 941             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 942             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 943             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 944             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 945
 946             /* COULOMB ELECTROSTATICS */
 947             velec            = _mm256_mul_ps(qq00,rinv00);
 948             felec            = _mm256_mul_ps(velec,rinvsq00);
 949
 950             /* CUBIC SPLINE TABLE DISPERSION */
 951             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 952                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 953             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 954                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 955             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 956                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 957             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 958                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 959             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 960             Heps             = _mm256_mul_ps(vfeps,H);
 961             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 962             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 963             fvdw6            = _mm256_mul_ps(c6_00,FF);
 964
 965             /* CUBIC SPLINE TABLE REPULSION */
 966             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 967             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 968             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 969                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 970             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 971                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 972             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 973                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 974             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 975                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 976             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 977             Heps             = _mm256_mul_ps(vfeps,H);
 978             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 979             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 980             fvdw12           = _mm256_mul_ps(c12_00,FF);
 981             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 982
 983             fscal            = _mm256_add_ps(felec,fvdw);
 984
 985             /* Calculate temporary vectorial force */
 986             tx               = _mm256_mul_ps(fscal,dx00);
 987             ty               = _mm256_mul_ps(fscal,dy00);
 988             tz               = _mm256_mul_ps(fscal,dz00);
 989
 990             /* Update vectorial force */
 991             fix0             = _mm256_add_ps(fix0,tx);
 992             fiy0             = _mm256_add_ps(fiy0,ty);
 993             fiz0             = _mm256_add_ps(fiz0,tz);
 994
 995             fjx0             = _mm256_add_ps(fjx0,tx);
 996             fjy0             = _mm256_add_ps(fjy0,ty);
 997             fjz0             = _mm256_add_ps(fjz0,tz);
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq10             = _mm256_mul_ps(iq1,jq0);
1005
1006             /* COULOMB ELECTROSTATICS */
1007             velec            = _mm256_mul_ps(qq10,rinv10);
1008             felec            = _mm256_mul_ps(velec,rinvsq10);
1009
1010             fscal            = felec;
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm256_mul_ps(fscal,dx10);
1014             ty               = _mm256_mul_ps(fscal,dy10);
1015             tz               = _mm256_mul_ps(fscal,dz10);
1016
1017             /* Update vectorial force */
1018             fix1             = _mm256_add_ps(fix1,tx);
1019             fiy1             = _mm256_add_ps(fiy1,ty);
1020             fiz1             = _mm256_add_ps(fiz1,tz);
1021
1022             fjx0             = _mm256_add_ps(fjx0,tx);
1023             fjy0             = _mm256_add_ps(fjy0,ty);
1024             fjz0             = _mm256_add_ps(fjz0,tz);
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             qq20             = _mm256_mul_ps(iq2,jq0);
1032
1033             /* COULOMB ELECTROSTATICS */
1034             velec            = _mm256_mul_ps(qq20,rinv20);
1035             felec            = _mm256_mul_ps(velec,rinvsq20);
1036
1037             fscal            = felec;
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm256_mul_ps(fscal,dx20);
1041             ty               = _mm256_mul_ps(fscal,dy20);
1042             tz               = _mm256_mul_ps(fscal,dz20);
1043
1044             /* Update vectorial force */
1045             fix2             = _mm256_add_ps(fix2,tx);
1046             fiy2             = _mm256_add_ps(fiy2,ty);
1047             fiz2             = _mm256_add_ps(fiz2,tz);
1048
1049             fjx0             = _mm256_add_ps(fjx0,tx);
1050             fjy0             = _mm256_add_ps(fjy0,ty);
1051             fjz0             = _mm256_add_ps(fjz0,tz);
1052
1053             fjptrA             = f+j_coord_offsetA;
1054             fjptrB             = f+j_coord_offsetB;
1055             fjptrC             = f+j_coord_offsetC;
1056             fjptrD             = f+j_coord_offsetD;
1057             fjptrE             = f+j_coord_offsetE;
1058             fjptrF             = f+j_coord_offsetF;
1059             fjptrG             = f+j_coord_offsetG;
1060             fjptrH             = f+j_coord_offsetH;
1061
1062             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1063
1064             /* Inner loop uses 108 flops */
1065         }
1066
1067         if(jidx<j_index_end)
1068         {
1069
1070             /* Get j neighbor index, and coordinate index */
1071             jnrlistA         = jjnr[jidx];
1072             jnrlistB         = jjnr[jidx+1];
1073             jnrlistC         = jjnr[jidx+2];
1074             jnrlistD         = jjnr[jidx+3];
1075             jnrlistE         = jjnr[jidx+4];
1076             jnrlistF         = jjnr[jidx+5];
1077             jnrlistG         = jjnr[jidx+6];
1078             jnrlistH         = jjnr[jidx+7];
1079             /* Sign of each element will be negative for non-real atoms.
1080              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1081              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1082              */
1083             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1084                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1085
1086             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1087             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1088             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1089             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1090             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1091             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1092             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1093             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1094             j_coord_offsetA  = DIM*jnrA;
1095             j_coord_offsetB  = DIM*jnrB;
1096             j_coord_offsetC  = DIM*jnrC;
1097             j_coord_offsetD  = DIM*jnrD;
1098             j_coord_offsetE  = DIM*jnrE;
1099             j_coord_offsetF  = DIM*jnrF;
1100             j_coord_offsetG  = DIM*jnrG;
1101             j_coord_offsetH  = DIM*jnrH;
1102
1103             /* load j atom coordinates */
1104             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1105                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1106                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1107                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1108                                                  &jx0,&jy0,&jz0);
1109
1110             /* Calculate displacement vector */
1111             dx00             = _mm256_sub_ps(ix0,jx0);
1112             dy00             = _mm256_sub_ps(iy0,jy0);
1113             dz00             = _mm256_sub_ps(iz0,jz0);
1114             dx10             = _mm256_sub_ps(ix1,jx0);
1115             dy10             = _mm256_sub_ps(iy1,jy0);
1116             dz10             = _mm256_sub_ps(iz1,jz0);
1117             dx20             = _mm256_sub_ps(ix2,jx0);
1118             dy20             = _mm256_sub_ps(iy2,jy0);
1119             dz20             = _mm256_sub_ps(iz2,jz0);
1120
1121             /* Calculate squared distance and things based on it */
1122             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1123             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1124             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1125
1126             rinv00           = avx256_invsqrt_f(rsq00);
1127             rinv10           = avx256_invsqrt_f(rsq10);
1128             rinv20           = avx256_invsqrt_f(rsq20);
1129
1130             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1131             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1132             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1133
1134             /* Load parameters for j particles */
1135             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1136                                                                  charge+jnrC+0,charge+jnrD+0,
1137                                                                  charge+jnrE+0,charge+jnrF+0,
1138                                                                  charge+jnrG+0,charge+jnrH+0);
1139             vdwjidx0A        = 2*vdwtype[jnrA+0];
1140             vdwjidx0B        = 2*vdwtype[jnrB+0];
1141             vdwjidx0C        = 2*vdwtype[jnrC+0];
1142             vdwjidx0D        = 2*vdwtype[jnrD+0];
1143             vdwjidx0E        = 2*vdwtype[jnrE+0];
1144             vdwjidx0F        = 2*vdwtype[jnrF+0];
1145             vdwjidx0G        = 2*vdwtype[jnrG+0];
1146             vdwjidx0H        = 2*vdwtype[jnrH+0];
1147
1148             fjx0             = _mm256_setzero_ps();
1149             fjy0             = _mm256_setzero_ps();
1150             fjz0             = _mm256_setzero_ps();
1151
1152             /**************************
1153              * CALCULATE INTERACTIONS *
1154              **************************/
1155
1156             r00              = _mm256_mul_ps(rsq00,rinv00);
1157             r00              = _mm256_andnot_ps(dummy_mask,r00);
1158
1159             /* Compute parameters for interactions between i and j atoms */
1160             qq00             = _mm256_mul_ps(iq0,jq0);
1161             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1162                                             vdwioffsetptr0+vdwjidx0B,
1163                                             vdwioffsetptr0+vdwjidx0C,
1164                                             vdwioffsetptr0+vdwjidx0D,
1165                                             vdwioffsetptr0+vdwjidx0E,
1166                                             vdwioffsetptr0+vdwjidx0F,
1167                                             vdwioffsetptr0+vdwjidx0G,
1168                                             vdwioffsetptr0+vdwjidx0H,
1169                                             &c6_00,&c12_00);
1170
1171             /* Calculate table index by multiplying r with table scale and truncate to integer */
1172             rt               = _mm256_mul_ps(r00,vftabscale);
1173             vfitab           = _mm256_cvttps_epi32(rt);
1174             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1175             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1176             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1177             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1178             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1179             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1180
1181             /* COULOMB ELECTROSTATICS */
1182             velec            = _mm256_mul_ps(qq00,rinv00);
1183             felec            = _mm256_mul_ps(velec,rinvsq00);
1184
1185             /* CUBIC SPLINE TABLE DISPERSION */
1186             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1187                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1188             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1189                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1190             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1191                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1192             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1193                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1194             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1195             Heps             = _mm256_mul_ps(vfeps,H);
1196             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1197             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1198             fvdw6            = _mm256_mul_ps(c6_00,FF);
1199
1200             /* CUBIC SPLINE TABLE REPULSION */
1201             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1202             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1203             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1204                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1205             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1206                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1207             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1208                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1209             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1210                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1211             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1212             Heps             = _mm256_mul_ps(vfeps,H);
1213             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1214             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1215             fvdw12           = _mm256_mul_ps(c12_00,FF);
1216             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1217
1218             fscal            = _mm256_add_ps(felec,fvdw);
1219
1220             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1221
1222             /* Calculate temporary vectorial force */
1223             tx               = _mm256_mul_ps(fscal,dx00);
1224             ty               = _mm256_mul_ps(fscal,dy00);
1225             tz               = _mm256_mul_ps(fscal,dz00);
1226
1227             /* Update vectorial force */
1228             fix0             = _mm256_add_ps(fix0,tx);
1229             fiy0             = _mm256_add_ps(fiy0,ty);
1230             fiz0             = _mm256_add_ps(fiz0,tz);
1231
1232             fjx0             = _mm256_add_ps(fjx0,tx);
1233             fjy0             = _mm256_add_ps(fjy0,ty);
1234             fjz0             = _mm256_add_ps(fjz0,tz);
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq10             = _mm256_mul_ps(iq1,jq0);
1242
1243             /* COULOMB ELECTROSTATICS */
1244             velec            = _mm256_mul_ps(qq10,rinv10);
1245             felec            = _mm256_mul_ps(velec,rinvsq10);
1246
1247             fscal            = felec;
1248
1249             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1250
1251             /* Calculate temporary vectorial force */
1252             tx               = _mm256_mul_ps(fscal,dx10);
1253             ty               = _mm256_mul_ps(fscal,dy10);
1254             tz               = _mm256_mul_ps(fscal,dz10);
1255
1256             /* Update vectorial force */
1257             fix1             = _mm256_add_ps(fix1,tx);
1258             fiy1             = _mm256_add_ps(fiy1,ty);
1259             fiz1             = _mm256_add_ps(fiz1,tz);
1260
1261             fjx0             = _mm256_add_ps(fjx0,tx);
1262             fjy0             = _mm256_add_ps(fjy0,ty);
1263             fjz0             = _mm256_add_ps(fjz0,tz);
1264
1265             /**************************
1266              * CALCULATE INTERACTIONS *
1267              **************************/
1268
1269             /* Compute parameters for interactions between i and j atoms */
1270             qq20             = _mm256_mul_ps(iq2,jq0);
1271
1272             /* COULOMB ELECTROSTATICS */
1273             velec            = _mm256_mul_ps(qq20,rinv20);
1274             felec            = _mm256_mul_ps(velec,rinvsq20);
1275
1276             fscal            = felec;
1277
1278             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1279
1280             /* Calculate temporary vectorial force */
1281             tx               = _mm256_mul_ps(fscal,dx20);
1282             ty               = _mm256_mul_ps(fscal,dy20);
1283             tz               = _mm256_mul_ps(fscal,dz20);
1284
1285             /* Update vectorial force */
1286             fix2             = _mm256_add_ps(fix2,tx);
1287             fiy2             = _mm256_add_ps(fiy2,ty);
1288             fiz2             = _mm256_add_ps(fiz2,tz);
1289
1290             fjx0             = _mm256_add_ps(fjx0,tx);
1291             fjy0             = _mm256_add_ps(fjy0,ty);
1292             fjz0             = _mm256_add_ps(fjz0,tz);
1293
1294             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1295             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1296             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1297             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1298             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1299             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1300             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1301             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1302
1303             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1304
1305             /* Inner loop uses 109 flops */
1306         }
1307
1308         /* End of innermost loop */
1309
1310         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1311                                                  f+i_coord_offset,fshift+i_shift_offset);
1312
1313         /* Increment number of inner iterations */
1314         inneriter                  += j_index_end - j_index_start;
1315
1316         /* Outer loop uses 18 flops */
1317     }
1318
1319     /* Increment number of outer iterations */
1320     outeriter        += nri;
1321
1322     /* Update outer/inner flops */
1323
1324     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1325 }