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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 104     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 105     __m256i          vfitab;
 106     __m128i          vfitab_lo,vfitab_hi;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m256           dummy_mask,cutoff_mask;
 111     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 112     __m256           one     = _mm256_set1_ps(1.0);
 113     __m256           two     = _mm256_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     vftab            = kernel_data->table_vdw->data;
 132     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 137     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 138     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181         fix1             = _mm256_setzero_ps();
 182         fiy1             = _mm256_setzero_ps();
 183         fiz1             = _mm256_setzero_ps();
 184         fix2             = _mm256_setzero_ps();
 185         fiy2             = _mm256_setzero_ps();
 186         fiz2             = _mm256_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190         vvdwsum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 240
 241             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 242             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                                  charge+jnrC+0,charge+jnrD+0,
 248                                                                  charge+jnrE+0,charge+jnrF+0,
 249                                                                  charge+jnrG+0,charge+jnrH+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251             vdwjidx0B        = 2*vdwtype[jnrB+0];
 252             vdwjidx0C        = 2*vdwtype[jnrC+0];
 253             vdwjidx0D        = 2*vdwtype[jnrD+0];
 254             vdwjidx0E        = 2*vdwtype[jnrE+0];
 255             vdwjidx0F        = 2*vdwtype[jnrF+0];
 256             vdwjidx0G        = 2*vdwtype[jnrG+0];
 257             vdwjidx0H        = 2*vdwtype[jnrH+0];
 258
 259             fjx0             = _mm256_setzero_ps();
 260             fjy0             = _mm256_setzero_ps();
 261             fjz0             = _mm256_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r00              = _mm256_mul_ps(rsq00,rinv00);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq00             = _mm256_mul_ps(iq0,jq0);
 271             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 272                                             vdwioffsetptr0+vdwjidx0B,
 273                                             vdwioffsetptr0+vdwjidx0C,
 274                                             vdwioffsetptr0+vdwjidx0D,
 275                                             vdwioffsetptr0+vdwjidx0E,
 276                                             vdwioffsetptr0+vdwjidx0F,
 277                                             vdwioffsetptr0+vdwjidx0G,
 278                                             vdwioffsetptr0+vdwjidx0H,
 279                                             &c6_00,&c12_00);
 280
 281             /* Calculate table index by multiplying r with table scale and truncate to integer */
 282             rt               = _mm256_mul_ps(r00,vftabscale);
 283             vfitab           = _mm256_cvttps_epi32(rt);
 284             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 285             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 286             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 287             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 288             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 289             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 290
 291             /* COULOMB ELECTROSTATICS */
 292             velec            = _mm256_mul_ps(qq00,rinv00);
 293             felec            = _mm256_mul_ps(velec,rinvsq00);
 294
 295             /* CUBIC SPLINE TABLE DISPERSION */
 296             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 298             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 299                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 300             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 301                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 302             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 303                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 304             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 305             Heps             = _mm256_mul_ps(vfeps,H);
 306             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 307             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 308             vvdw6            = _mm256_mul_ps(c6_00,VV);
 309             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 310             fvdw6            = _mm256_mul_ps(c6_00,FF);
 311
 312             /* CUBIC SPLINE TABLE REPULSION */
 313             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 314             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 315             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 316                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 317             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 318                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 319             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 320                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 321             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 323             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 324             Heps             = _mm256_mul_ps(vfeps,H);
 325             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 326             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 327             vvdw12           = _mm256_mul_ps(c12_00,VV);
 328             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 329             fvdw12           = _mm256_mul_ps(c12_00,FF);
 330             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 331             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velecsum         = _mm256_add_ps(velecsum,velec);
 335             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 336
 337             fscal            = _mm256_add_ps(felec,fvdw);
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm256_mul_ps(fscal,dx00);
 341             ty               = _mm256_mul_ps(fscal,dy00);
 342             tz               = _mm256_mul_ps(fscal,dz00);
 343
 344             /* Update vectorial force */
 345             fix0             = _mm256_add_ps(fix0,tx);
 346             fiy0             = _mm256_add_ps(fiy0,ty);
 347             fiz0             = _mm256_add_ps(fiz0,tz);
 348
 349             fjx0             = _mm256_add_ps(fjx0,tx);
 350             fjy0             = _mm256_add_ps(fjy0,ty);
 351             fjz0             = _mm256_add_ps(fjz0,tz);
 352
 353             /**************************
 354              * CALCULATE INTERACTIONS *
 355              **************************/
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq10             = _mm256_mul_ps(iq1,jq0);
 359
 360             /* COULOMB ELECTROSTATICS */
 361             velec            = _mm256_mul_ps(qq10,rinv10);
 362             felec            = _mm256_mul_ps(velec,rinvsq10);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velecsum         = _mm256_add_ps(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm256_mul_ps(fscal,dx10);
 371             ty               = _mm256_mul_ps(fscal,dy10);
 372             tz               = _mm256_mul_ps(fscal,dz10);
 373
 374             /* Update vectorial force */
 375             fix1             = _mm256_add_ps(fix1,tx);
 376             fiy1             = _mm256_add_ps(fiy1,ty);
 377             fiz1             = _mm256_add_ps(fiz1,tz);
 378
 379             fjx0             = _mm256_add_ps(fjx0,tx);
 380             fjy0             = _mm256_add_ps(fjy0,ty);
 381             fjz0             = _mm256_add_ps(fjz0,tz);
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq20             = _mm256_mul_ps(iq2,jq0);
 389
 390             /* COULOMB ELECTROSTATICS */
 391             velec            = _mm256_mul_ps(qq20,rinv20);
 392             felec            = _mm256_mul_ps(velec,rinvsq20);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velecsum         = _mm256_add_ps(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm256_mul_ps(fscal,dx20);
 401             ty               = _mm256_mul_ps(fscal,dy20);
 402             tz               = _mm256_mul_ps(fscal,dz20);
 403
 404             /* Update vectorial force */
 405             fix2             = _mm256_add_ps(fix2,tx);
 406             fiy2             = _mm256_add_ps(fiy2,ty);
 407             fiz2             = _mm256_add_ps(fiz2,tz);
 408
 409             fjx0             = _mm256_add_ps(fjx0,tx);
 410             fjy0             = _mm256_add_ps(fjy0,ty);
 411             fjz0             = _mm256_add_ps(fjz0,tz);
 412
 413             fjptrA             = f+j_coord_offsetA;
 414             fjptrB             = f+j_coord_offsetB;
 415             fjptrC             = f+j_coord_offsetC;
 416             fjptrD             = f+j_coord_offsetD;
 417             fjptrE             = f+j_coord_offsetE;
 418             fjptrF             = f+j_coord_offsetF;
 419             fjptrG             = f+j_coord_offsetG;
 420             fjptrH             = f+j_coord_offsetH;
 421
 422             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 423
 424             /* Inner loop uses 119 flops */
 425         }
 426
 427         if(jidx<j_index_end)
 428         {
 429
 430             /* Get j neighbor index, and coordinate index */
 431             jnrlistA         = jjnr[jidx];
 432             jnrlistB         = jjnr[jidx+1];
 433             jnrlistC         = jjnr[jidx+2];
 434             jnrlistD         = jjnr[jidx+3];
 435             jnrlistE         = jjnr[jidx+4];
 436             jnrlistF         = jjnr[jidx+5];
 437             jnrlistG         = jjnr[jidx+6];
 438             jnrlistH         = jjnr[jidx+7];
 439             /* Sign of each element will be negative for non-real atoms.
 440              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 441              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 442              */
 443             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 444                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 445
 446             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 447             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 448             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 449             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 450             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 451             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 452             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 453             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 454             j_coord_offsetA  = DIM*jnrA;
 455             j_coord_offsetB  = DIM*jnrB;
 456             j_coord_offsetC  = DIM*jnrC;
 457             j_coord_offsetD  = DIM*jnrD;
 458             j_coord_offsetE  = DIM*jnrE;
 459             j_coord_offsetF  = DIM*jnrF;
 460             j_coord_offsetG  = DIM*jnrG;
 461             j_coord_offsetH  = DIM*jnrH;
 462
 463             /* load j atom coordinates */
 464             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 465                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 466                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 467                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 468                                                  &jx0,&jy0,&jz0);
 469
 470             /* Calculate displacement vector */
 471             dx00             = _mm256_sub_ps(ix0,jx0);
 472             dy00             = _mm256_sub_ps(iy0,jy0);
 473             dz00             = _mm256_sub_ps(iz0,jz0);
 474             dx10             = _mm256_sub_ps(ix1,jx0);
 475             dy10             = _mm256_sub_ps(iy1,jy0);
 476             dz10             = _mm256_sub_ps(iz1,jz0);
 477             dx20             = _mm256_sub_ps(ix2,jx0);
 478             dy20             = _mm256_sub_ps(iy2,jy0);
 479             dz20             = _mm256_sub_ps(iz2,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 483             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 484             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 485
 486             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 487             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 488             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 489
 490             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 491             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 492             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 496                                                                  charge+jnrC+0,charge+jnrD+0,
 497                                                                  charge+jnrE+0,charge+jnrF+0,
 498                                                                  charge+jnrG+0,charge+jnrH+0);
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501             vdwjidx0C        = 2*vdwtype[jnrC+0];
 502             vdwjidx0D        = 2*vdwtype[jnrD+0];
 503             vdwjidx0E        = 2*vdwtype[jnrE+0];
 504             vdwjidx0F        = 2*vdwtype[jnrF+0];
 505             vdwjidx0G        = 2*vdwtype[jnrG+0];
 506             vdwjidx0H        = 2*vdwtype[jnrH+0];
 507
 508             fjx0             = _mm256_setzero_ps();
 509             fjy0             = _mm256_setzero_ps();
 510             fjz0             = _mm256_setzero_ps();
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             r00              = _mm256_mul_ps(rsq00,rinv00);
 517             r00              = _mm256_andnot_ps(dummy_mask,r00);
 518
 519             /* Compute parameters for interactions between i and j atoms */
 520             qq00             = _mm256_mul_ps(iq0,jq0);
 521             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 522                                             vdwioffsetptr0+vdwjidx0B,
 523                                             vdwioffsetptr0+vdwjidx0C,
 524                                             vdwioffsetptr0+vdwjidx0D,
 525                                             vdwioffsetptr0+vdwjidx0E,
 526                                             vdwioffsetptr0+vdwjidx0F,
 527                                             vdwioffsetptr0+vdwjidx0G,
 528                                             vdwioffsetptr0+vdwjidx0H,
 529                                             &c6_00,&c12_00);
 530
 531             /* Calculate table index by multiplying r with table scale and truncate to integer */
 532             rt               = _mm256_mul_ps(r00,vftabscale);
 533             vfitab           = _mm256_cvttps_epi32(rt);
 534             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 535             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 536             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 537             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 538             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 539             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 540
 541             /* COULOMB ELECTROSTATICS */
 542             velec            = _mm256_mul_ps(qq00,rinv00);
 543             felec            = _mm256_mul_ps(velec,rinvsq00);
 544
 545             /* CUBIC SPLINE TABLE DISPERSION */
 546             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 547                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 548             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 549                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 550             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 551                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 552             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 553                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 554             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 555             Heps             = _mm256_mul_ps(vfeps,H);
 556             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 557             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 558             vvdw6            = _mm256_mul_ps(c6_00,VV);
 559             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 560             fvdw6            = _mm256_mul_ps(c6_00,FF);
 561
 562             /* CUBIC SPLINE TABLE REPULSION */
 563             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 564             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 565             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 566                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 567             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 568                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 569             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 570                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 571             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 572                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 573             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 574             Heps             = _mm256_mul_ps(vfeps,H);
 575             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 576             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 577             vvdw12           = _mm256_mul_ps(c12_00,VV);
 578             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 579             fvdw12           = _mm256_mul_ps(c12_00,FF);
 580             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 581             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm256_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm256_add_ps(velecsum,velec);
 586             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 587             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 588
 589             fscal            = _mm256_add_ps(felec,fvdw);
 590
 591             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm256_mul_ps(fscal,dx00);
 595             ty               = _mm256_mul_ps(fscal,dy00);
 596             tz               = _mm256_mul_ps(fscal,dz00);
 597
 598             /* Update vectorial force */
 599             fix0             = _mm256_add_ps(fix0,tx);
 600             fiy0             = _mm256_add_ps(fiy0,ty);
 601             fiz0             = _mm256_add_ps(fiz0,tz);
 602
 603             fjx0             = _mm256_add_ps(fjx0,tx);
 604             fjy0             = _mm256_add_ps(fjy0,ty);
 605             fjz0             = _mm256_add_ps(fjz0,tz);
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq10             = _mm256_mul_ps(iq1,jq0);
 613
 614             /* COULOMB ELECTROSTATICS */
 615             velec            = _mm256_mul_ps(qq10,rinv10);
 616             felec            = _mm256_mul_ps(velec,rinvsq10);
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm256_andnot_ps(dummy_mask,velec);
 620             velecsum         = _mm256_add_ps(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm256_mul_ps(fscal,dx10);
 628             ty               = _mm256_mul_ps(fscal,dy10);
 629             tz               = _mm256_mul_ps(fscal,dz10);
 630
 631             /* Update vectorial force */
 632             fix1             = _mm256_add_ps(fix1,tx);
 633             fiy1             = _mm256_add_ps(fiy1,ty);
 634             fiz1             = _mm256_add_ps(fiz1,tz);
 635
 636             fjx0             = _mm256_add_ps(fjx0,tx);
 637             fjy0             = _mm256_add_ps(fjy0,ty);
 638             fjz0             = _mm256_add_ps(fjz0,tz);
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             /* Compute parameters for interactions between i and j atoms */
 645             qq20             = _mm256_mul_ps(iq2,jq0);
 646
 647             /* COULOMB ELECTROSTATICS */
 648             velec            = _mm256_mul_ps(qq20,rinv20);
 649             felec            = _mm256_mul_ps(velec,rinvsq20);
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             velec            = _mm256_andnot_ps(dummy_mask,velec);
 653             velecsum         = _mm256_add_ps(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm256_mul_ps(fscal,dx20);
 661             ty               = _mm256_mul_ps(fscal,dy20);
 662             tz               = _mm256_mul_ps(fscal,dz20);
 663
 664             /* Update vectorial force */
 665             fix2             = _mm256_add_ps(fix2,tx);
 666             fiy2             = _mm256_add_ps(fiy2,ty);
 667             fiz2             = _mm256_add_ps(fiz2,tz);
 668
 669             fjx0             = _mm256_add_ps(fjx0,tx);
 670             fjy0             = _mm256_add_ps(fjy0,ty);
 671             fjz0             = _mm256_add_ps(fjz0,tz);
 672
 673             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 674             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 675             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 676             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 677             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 678             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 679             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 680             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 681
 682             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 683
 684             /* Inner loop uses 120 flops */
 685         }
 686
 687         /* End of innermost loop */
 688
 689         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 690                                                  f+i_coord_offset,fshift+i_shift_offset);
 691
 692         ggid                        = gid[iidx];
 693         /* Update potential energies */
 694         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 695         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 696
 697         /* Increment number of inner iterations */
 698         inneriter                  += j_index_end - j_index_start;
 699
 700         /* Outer loop uses 20 flops */
 701     }
 702
 703     /* Increment number of outer iterations */
 704     outeriter        += nri;
 705
 706     /* Update outer/inner flops */
 707
 708     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 709 }
 710 /*
 711  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_single
 712  * Electrostatics interaction: Coulomb
 713  * VdW interaction:            CubicSplineTable
 714  * Geometry:                   Water3-Particle
 715  * Calculate force/pot:        Force
 716  */
 717 void
 718 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_single
 719                     (t_nblist                    * gmx_restrict       nlist,
 720                      rvec                        * gmx_restrict          xx,
 721                      rvec                        * gmx_restrict          ff,
 722                      t_forcerec                  * gmx_restrict          fr,
 723                      t_mdatoms                   * gmx_restrict     mdatoms,
 724                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 725                      t_nrnb                      * gmx_restrict        nrnb)
 726 {
 727     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 728      * just 0 for non-waters.
 729      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 730      * jnr indices corresponding to data put in the four positions in the SIMD register.
 731      */
 732     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 733     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 734     int              jnrA,jnrB,jnrC,jnrD;
 735     int              jnrE,jnrF,jnrG,jnrH;
 736     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 737     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 738     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 739     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 740     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 741     real             rcutoff_scalar;
 742     real             *shiftvec,*fshift,*x,*f;
 743     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 744     real             scratch[4*DIM];
 745     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 746     real *           vdwioffsetptr0;
 747     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 748     real *           vdwioffsetptr1;
 749     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 750     real *           vdwioffsetptr2;
 751     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 752     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 753     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 754     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 755     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 756     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 757     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 758     real             *charge;
 759     int              nvdwtype;
 760     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 761     int              *vdwtype;
 762     real             *vdwparam;
 763     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 764     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 765     __m256i          vfitab;
 766     __m128i          vfitab_lo,vfitab_hi;
 767     __m128i          ifour       = _mm_set1_epi32(4);
 768     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 769     real             *vftab;
 770     __m256           dummy_mask,cutoff_mask;
 771     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 772     __m256           one     = _mm256_set1_ps(1.0);
 773     __m256           two     = _mm256_set1_ps(2.0);
 774     x                = xx[0];
 775     f                = ff[0];
 776
 777     nri              = nlist->nri;
 778     iinr             = nlist->iinr;
 779     jindex           = nlist->jindex;
 780     jjnr             = nlist->jjnr;
 781     shiftidx         = nlist->shift;
 782     gid              = nlist->gid;
 783     shiftvec         = fr->shift_vec[0];
 784     fshift           = fr->fshift[0];
 785     facel            = _mm256_set1_ps(fr->epsfac);
 786     charge           = mdatoms->chargeA;
 787     nvdwtype         = fr->ntype;
 788     vdwparam         = fr->nbfp;
 789     vdwtype          = mdatoms->typeA;
 790
 791     vftab            = kernel_data->table_vdw->data;
 792     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 793
 794     /* Setup water-specific parameters */
 795     inr              = nlist->iinr[0];
 796     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 797     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 798     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 799     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 800
 801     /* Avoid stupid compiler warnings */
 802     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 803     j_coord_offsetA = 0;
 804     j_coord_offsetB = 0;
 805     j_coord_offsetC = 0;
 806     j_coord_offsetD = 0;
 807     j_coord_offsetE = 0;
 808     j_coord_offsetF = 0;
 809     j_coord_offsetG = 0;
 810     j_coord_offsetH = 0;
 811
 812     outeriter        = 0;
 813     inneriter        = 0;
 814
 815     for(iidx=0;iidx<4*DIM;iidx++)
 816     {
 817         scratch[iidx] = 0.0;
 818     }
 819
 820     /* Start outer loop over neighborlists */
 821     for(iidx=0; iidx<nri; iidx++)
 822     {
 823         /* Load shift vector for this list */
 824         i_shift_offset   = DIM*shiftidx[iidx];
 825
 826         /* Load limits for loop over neighbors */
 827         j_index_start    = jindex[iidx];
 828         j_index_end      = jindex[iidx+1];
 829
 830         /* Get outer coordinate index */
 831         inr              = iinr[iidx];
 832         i_coord_offset   = DIM*inr;
 833
 834         /* Load i particle coords and add shift vector */
 835         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 836                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 837
 838         fix0             = _mm256_setzero_ps();
 839         fiy0             = _mm256_setzero_ps();
 840         fiz0             = _mm256_setzero_ps();
 841         fix1             = _mm256_setzero_ps();
 842         fiy1             = _mm256_setzero_ps();
 843         fiz1             = _mm256_setzero_ps();
 844         fix2             = _mm256_setzero_ps();
 845         fiy2             = _mm256_setzero_ps();
 846         fiz2             = _mm256_setzero_ps();
 847
 848         /* Start inner kernel loop */
 849         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 850         {
 851
 852             /* Get j neighbor index, and coordinate index */
 853             jnrA             = jjnr[jidx];
 854             jnrB             = jjnr[jidx+1];
 855             jnrC             = jjnr[jidx+2];
 856             jnrD             = jjnr[jidx+3];
 857             jnrE             = jjnr[jidx+4];
 858             jnrF             = jjnr[jidx+5];
 859             jnrG             = jjnr[jidx+6];
 860             jnrH             = jjnr[jidx+7];
 861             j_coord_offsetA  = DIM*jnrA;
 862             j_coord_offsetB  = DIM*jnrB;
 863             j_coord_offsetC  = DIM*jnrC;
 864             j_coord_offsetD  = DIM*jnrD;
 865             j_coord_offsetE  = DIM*jnrE;
 866             j_coord_offsetF  = DIM*jnrF;
 867             j_coord_offsetG  = DIM*jnrG;
 868             j_coord_offsetH  = DIM*jnrH;
 869
 870             /* load j atom coordinates */
 871             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 872                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 873                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 874                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 875                                                  &jx0,&jy0,&jz0);
 876
 877             /* Calculate displacement vector */
 878             dx00             = _mm256_sub_ps(ix0,jx0);
 879             dy00             = _mm256_sub_ps(iy0,jy0);
 880             dz00             = _mm256_sub_ps(iz0,jz0);
 881             dx10             = _mm256_sub_ps(ix1,jx0);
 882             dy10             = _mm256_sub_ps(iy1,jy0);
 883             dz10             = _mm256_sub_ps(iz1,jz0);
 884             dx20             = _mm256_sub_ps(ix2,jx0);
 885             dy20             = _mm256_sub_ps(iy2,jy0);
 886             dz20             = _mm256_sub_ps(iz2,jz0);
 887
 888             /* Calculate squared distance and things based on it */
 889             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 890             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 891             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 892
 893             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 894             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 895             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 896
 897             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 898             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 899             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 900
 901             /* Load parameters for j particles */
 902             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 903                                                                  charge+jnrC+0,charge+jnrD+0,
 904                                                                  charge+jnrE+0,charge+jnrF+0,
 905                                                                  charge+jnrG+0,charge+jnrH+0);
 906             vdwjidx0A        = 2*vdwtype[jnrA+0];
 907             vdwjidx0B        = 2*vdwtype[jnrB+0];
 908             vdwjidx0C        = 2*vdwtype[jnrC+0];
 909             vdwjidx0D        = 2*vdwtype[jnrD+0];
 910             vdwjidx0E        = 2*vdwtype[jnrE+0];
 911             vdwjidx0F        = 2*vdwtype[jnrF+0];
 912             vdwjidx0G        = 2*vdwtype[jnrG+0];
 913             vdwjidx0H        = 2*vdwtype[jnrH+0];
 914
 915             fjx0             = _mm256_setzero_ps();
 916             fjy0             = _mm256_setzero_ps();
 917             fjz0             = _mm256_setzero_ps();
 918
 919             /**************************
 920              * CALCULATE INTERACTIONS *
 921              **************************/
 922
 923             r00              = _mm256_mul_ps(rsq00,rinv00);
 924
 925             /* Compute parameters for interactions between i and j atoms */
 926             qq00             = _mm256_mul_ps(iq0,jq0);
 927             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 928                                             vdwioffsetptr0+vdwjidx0B,
 929                                             vdwioffsetptr0+vdwjidx0C,
 930                                             vdwioffsetptr0+vdwjidx0D,
 931                                             vdwioffsetptr0+vdwjidx0E,
 932                                             vdwioffsetptr0+vdwjidx0F,
 933                                             vdwioffsetptr0+vdwjidx0G,
 934                                             vdwioffsetptr0+vdwjidx0H,
 935                                             &c6_00,&c12_00);
 936
 937             /* Calculate table index by multiplying r with table scale and truncate to integer */
 938             rt               = _mm256_mul_ps(r00,vftabscale);
 939             vfitab           = _mm256_cvttps_epi32(rt);
 940             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 941             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 942             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 943             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 944             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 945             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 946
 947             /* COULOMB ELECTROSTATICS */
 948             velec            = _mm256_mul_ps(qq00,rinv00);
 949             felec            = _mm256_mul_ps(velec,rinvsq00);
 950
 951             /* CUBIC SPLINE TABLE DISPERSION */
 952             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 953                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 954             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 955                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 956             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 957                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 958             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 959                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 960             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 961             Heps             = _mm256_mul_ps(vfeps,H);
 962             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 963             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 964             fvdw6            = _mm256_mul_ps(c6_00,FF);
 965
 966             /* CUBIC SPLINE TABLE REPULSION */
 967             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 968             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 969             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 970                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 971             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 972                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 973             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 974                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 975             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 976                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 977             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 978             Heps             = _mm256_mul_ps(vfeps,H);
 979             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 980             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 981             fvdw12           = _mm256_mul_ps(c12_00,FF);
 982             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 983
 984             fscal            = _mm256_add_ps(felec,fvdw);
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm256_mul_ps(fscal,dx00);
 988             ty               = _mm256_mul_ps(fscal,dy00);
 989             tz               = _mm256_mul_ps(fscal,dz00);
 990
 991             /* Update vectorial force */
 992             fix0             = _mm256_add_ps(fix0,tx);
 993             fiy0             = _mm256_add_ps(fiy0,ty);
 994             fiz0             = _mm256_add_ps(fiz0,tz);
 995
 996             fjx0             = _mm256_add_ps(fjx0,tx);
 997             fjy0             = _mm256_add_ps(fjy0,ty);
 998             fjz0             = _mm256_add_ps(fjz0,tz);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             /* Compute parameters for interactions between i and j atoms */
1005             qq10             = _mm256_mul_ps(iq1,jq0);
1006
1007             /* COULOMB ELECTROSTATICS */
1008             velec            = _mm256_mul_ps(qq10,rinv10);
1009             felec            = _mm256_mul_ps(velec,rinvsq10);
1010
1011             fscal            = felec;
1012
1013             /* Calculate temporary vectorial force */
1014             tx               = _mm256_mul_ps(fscal,dx10);
1015             ty               = _mm256_mul_ps(fscal,dy10);
1016             tz               = _mm256_mul_ps(fscal,dz10);
1017
1018             /* Update vectorial force */
1019             fix1             = _mm256_add_ps(fix1,tx);
1020             fiy1             = _mm256_add_ps(fiy1,ty);
1021             fiz1             = _mm256_add_ps(fiz1,tz);
1022
1023             fjx0             = _mm256_add_ps(fjx0,tx);
1024             fjy0             = _mm256_add_ps(fjy0,ty);
1025             fjz0             = _mm256_add_ps(fjz0,tz);
1026
1027             /**************************
1028              * CALCULATE INTERACTIONS *
1029              **************************/
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq20             = _mm256_mul_ps(iq2,jq0);
1033
1034             /* COULOMB ELECTROSTATICS */
1035             velec            = _mm256_mul_ps(qq20,rinv20);
1036             felec            = _mm256_mul_ps(velec,rinvsq20);
1037
1038             fscal            = felec;
1039
1040             /* Calculate temporary vectorial force */
1041             tx               = _mm256_mul_ps(fscal,dx20);
1042             ty               = _mm256_mul_ps(fscal,dy20);
1043             tz               = _mm256_mul_ps(fscal,dz20);
1044
1045             /* Update vectorial force */
1046             fix2             = _mm256_add_ps(fix2,tx);
1047             fiy2             = _mm256_add_ps(fiy2,ty);
1048             fiz2             = _mm256_add_ps(fiz2,tz);
1049
1050             fjx0             = _mm256_add_ps(fjx0,tx);
1051             fjy0             = _mm256_add_ps(fjy0,ty);
1052             fjz0             = _mm256_add_ps(fjz0,tz);
1053
1054             fjptrA             = f+j_coord_offsetA;
1055             fjptrB             = f+j_coord_offsetB;
1056             fjptrC             = f+j_coord_offsetC;
1057             fjptrD             = f+j_coord_offsetD;
1058             fjptrE             = f+j_coord_offsetE;
1059             fjptrF             = f+j_coord_offsetF;
1060             fjptrG             = f+j_coord_offsetG;
1061             fjptrH             = f+j_coord_offsetH;
1062
1063             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1064
1065             /* Inner loop uses 108 flops */
1066         }
1067
1068         if(jidx<j_index_end)
1069         {
1070
1071             /* Get j neighbor index, and coordinate index */
1072             jnrlistA         = jjnr[jidx];
1073             jnrlistB         = jjnr[jidx+1];
1074             jnrlistC         = jjnr[jidx+2];
1075             jnrlistD         = jjnr[jidx+3];
1076             jnrlistE         = jjnr[jidx+4];
1077             jnrlistF         = jjnr[jidx+5];
1078             jnrlistG         = jjnr[jidx+6];
1079             jnrlistH         = jjnr[jidx+7];
1080             /* Sign of each element will be negative for non-real atoms.
1081              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1082              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1083              */
1084             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1085                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1086
1087             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1088             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1089             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1090             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1091             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1092             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1093             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1094             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1095             j_coord_offsetA  = DIM*jnrA;
1096             j_coord_offsetB  = DIM*jnrB;
1097             j_coord_offsetC  = DIM*jnrC;
1098             j_coord_offsetD  = DIM*jnrD;
1099             j_coord_offsetE  = DIM*jnrE;
1100             j_coord_offsetF  = DIM*jnrF;
1101             j_coord_offsetG  = DIM*jnrG;
1102             j_coord_offsetH  = DIM*jnrH;
1103
1104             /* load j atom coordinates */
1105             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1106                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1107                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1108                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1109                                                  &jx0,&jy0,&jz0);
1110
1111             /* Calculate displacement vector */
1112             dx00             = _mm256_sub_ps(ix0,jx0);
1113             dy00             = _mm256_sub_ps(iy0,jy0);
1114             dz00             = _mm256_sub_ps(iz0,jz0);
1115             dx10             = _mm256_sub_ps(ix1,jx0);
1116             dy10             = _mm256_sub_ps(iy1,jy0);
1117             dz10             = _mm256_sub_ps(iz1,jz0);
1118             dx20             = _mm256_sub_ps(ix2,jx0);
1119             dy20             = _mm256_sub_ps(iy2,jy0);
1120             dz20             = _mm256_sub_ps(iz2,jz0);
1121
1122             /* Calculate squared distance and things based on it */
1123             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1124             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1125             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1126
1127             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1128             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1129             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1130
1131             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1132             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1133             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1134
1135             /* Load parameters for j particles */
1136             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1137                                                                  charge+jnrC+0,charge+jnrD+0,
1138                                                                  charge+jnrE+0,charge+jnrF+0,
1139                                                                  charge+jnrG+0,charge+jnrH+0);
1140             vdwjidx0A        = 2*vdwtype[jnrA+0];
1141             vdwjidx0B        = 2*vdwtype[jnrB+0];
1142             vdwjidx0C        = 2*vdwtype[jnrC+0];
1143             vdwjidx0D        = 2*vdwtype[jnrD+0];
1144             vdwjidx0E        = 2*vdwtype[jnrE+0];
1145             vdwjidx0F        = 2*vdwtype[jnrF+0];
1146             vdwjidx0G        = 2*vdwtype[jnrG+0];
1147             vdwjidx0H        = 2*vdwtype[jnrH+0];
1148
1149             fjx0             = _mm256_setzero_ps();
1150             fjy0             = _mm256_setzero_ps();
1151             fjz0             = _mm256_setzero_ps();
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             r00              = _mm256_mul_ps(rsq00,rinv00);
1158             r00              = _mm256_andnot_ps(dummy_mask,r00);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq00             = _mm256_mul_ps(iq0,jq0);
1162             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1163                                             vdwioffsetptr0+vdwjidx0B,
1164                                             vdwioffsetptr0+vdwjidx0C,
1165                                             vdwioffsetptr0+vdwjidx0D,
1166                                             vdwioffsetptr0+vdwjidx0E,
1167                                             vdwioffsetptr0+vdwjidx0F,
1168                                             vdwioffsetptr0+vdwjidx0G,
1169                                             vdwioffsetptr0+vdwjidx0H,
1170                                             &c6_00,&c12_00);
1171
1172             /* Calculate table index by multiplying r with table scale and truncate to integer */
1173             rt               = _mm256_mul_ps(r00,vftabscale);
1174             vfitab           = _mm256_cvttps_epi32(rt);
1175             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1176             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1177             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1178             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1179             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1180             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1181
1182             /* COULOMB ELECTROSTATICS */
1183             velec            = _mm256_mul_ps(qq00,rinv00);
1184             felec            = _mm256_mul_ps(velec,rinvsq00);
1185
1186             /* CUBIC SPLINE TABLE DISPERSION */
1187             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1188                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1189             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1190                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1191             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1193             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1195             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1196             Heps             = _mm256_mul_ps(vfeps,H);
1197             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1198             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1199             fvdw6            = _mm256_mul_ps(c6_00,FF);
1200
1201             /* CUBIC SPLINE TABLE REPULSION */
1202             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1203             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1204             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1205                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1206             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1207                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1208             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1210             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1212             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1213             Heps             = _mm256_mul_ps(vfeps,H);
1214             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1215             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1216             fvdw12           = _mm256_mul_ps(c12_00,FF);
1217             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1218
1219             fscal            = _mm256_add_ps(felec,fvdw);
1220
1221             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm256_mul_ps(fscal,dx00);
1225             ty               = _mm256_mul_ps(fscal,dy00);
1226             tz               = _mm256_mul_ps(fscal,dz00);
1227
1228             /* Update vectorial force */
1229             fix0             = _mm256_add_ps(fix0,tx);
1230             fiy0             = _mm256_add_ps(fiy0,ty);
1231             fiz0             = _mm256_add_ps(fiz0,tz);
1232
1233             fjx0             = _mm256_add_ps(fjx0,tx);
1234             fjy0             = _mm256_add_ps(fjy0,ty);
1235             fjz0             = _mm256_add_ps(fjz0,tz);
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             /* Compute parameters for interactions between i and j atoms */
1242             qq10             = _mm256_mul_ps(iq1,jq0);
1243
1244             /* COULOMB ELECTROSTATICS */
1245             velec            = _mm256_mul_ps(qq10,rinv10);
1246             felec            = _mm256_mul_ps(velec,rinvsq10);
1247
1248             fscal            = felec;
1249
1250             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1251
1252             /* Calculate temporary vectorial force */
1253             tx               = _mm256_mul_ps(fscal,dx10);
1254             ty               = _mm256_mul_ps(fscal,dy10);
1255             tz               = _mm256_mul_ps(fscal,dz10);
1256
1257             /* Update vectorial force */
1258             fix1             = _mm256_add_ps(fix1,tx);
1259             fiy1             = _mm256_add_ps(fiy1,ty);
1260             fiz1             = _mm256_add_ps(fiz1,tz);
1261
1262             fjx0             = _mm256_add_ps(fjx0,tx);
1263             fjy0             = _mm256_add_ps(fjy0,ty);
1264             fjz0             = _mm256_add_ps(fjz0,tz);
1265
1266             /**************************
1267              * CALCULATE INTERACTIONS *
1268              **************************/
1269
1270             /* Compute parameters for interactions between i and j atoms */
1271             qq20             = _mm256_mul_ps(iq2,jq0);
1272
1273             /* COULOMB ELECTROSTATICS */
1274             velec            = _mm256_mul_ps(qq20,rinv20);
1275             felec            = _mm256_mul_ps(velec,rinvsq20);
1276
1277             fscal            = felec;
1278
1279             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1280
1281             /* Calculate temporary vectorial force */
1282             tx               = _mm256_mul_ps(fscal,dx20);
1283             ty               = _mm256_mul_ps(fscal,dy20);
1284             tz               = _mm256_mul_ps(fscal,dz20);
1285
1286             /* Update vectorial force */
1287             fix2             = _mm256_add_ps(fix2,tx);
1288             fiy2             = _mm256_add_ps(fiy2,ty);
1289             fiz2             = _mm256_add_ps(fiz2,tz);
1290
1291             fjx0             = _mm256_add_ps(fjx0,tx);
1292             fjy0             = _mm256_add_ps(fjy0,ty);
1293             fjz0             = _mm256_add_ps(fjz0,tz);
1294
1295             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1296             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1297             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1298             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1299             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1300             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1301             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1302             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1303
1304             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1305
1306             /* Inner loop uses 109 flops */
1307         }
1308
1309         /* End of innermost loop */
1310
1311         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1312                                                  f+i_coord_offset,fshift+i_shift_offset);
1313
1314         /* Increment number of inner iterations */
1315         inneriter                  += j_index_end - j_index_start;
1316
1317         /* Outer loop uses 18 flops */
1318     }
1319
1320     /* Increment number of outer iterations */
1321     outeriter        += nri;
1322
1323     /* Update outer/inner flops */
1324
1325     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1326 }