src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     krf              = _mm256_set1_ps(fr->ic->k_rf);
 130     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 131     crf              = _mm256_set1_ps(fr->ic->c_rf);
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     vftab            = kernel_data->table_vdw->data;
 137     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 142     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 143     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 144     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152     j_coord_offsetE = 0;
 153     j_coord_offsetF = 0;
 154     j_coord_offsetG = 0;
 155     j_coord_offsetH = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 182
 183         fix0             = _mm256_setzero_ps();
 184         fiy0             = _mm256_setzero_ps();
 185         fiz0             = _mm256_setzero_ps();
 186         fix1             = _mm256_setzero_ps();
 187         fiy1             = _mm256_setzero_ps();
 188         fiz1             = _mm256_setzero_ps();
 189         fix2             = _mm256_setzero_ps();
 190         fiy2             = _mm256_setzero_ps();
 191         fiz2             = _mm256_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_ps();
 195         vvdwsum          = _mm256_setzero_ps();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             jnrE             = jjnr[jidx+4];
 207             jnrF             = jjnr[jidx+5];
 208             jnrG             = jjnr[jidx+6];
 209             jnrH             = jjnr[jidx+7];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214             j_coord_offsetE  = DIM*jnrE;
 215             j_coord_offsetF  = DIM*jnrF;
 216             j_coord_offsetG  = DIM*jnrG;
 217             j_coord_offsetH  = DIM*jnrH;
 218
 219             /* load j atom coordinates */
 220             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 221                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 222                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 223                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 224                                                  &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_ps(ix0,jx0);
 228             dy00             = _mm256_sub_ps(iy0,jy0);
 229             dz00             = _mm256_sub_ps(iz0,jz0);
 230             dx10             = _mm256_sub_ps(ix1,jx0);
 231             dy10             = _mm256_sub_ps(iy1,jy0);
 232             dz10             = _mm256_sub_ps(iz1,jz0);
 233             dx20             = _mm256_sub_ps(ix2,jx0);
 234             dy20             = _mm256_sub_ps(iy2,jy0);
 235             dz20             = _mm256_sub_ps(iz2,jz0);
 236
 237             /* Calculate squared distance and things based on it */
 238             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 239             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 240             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 241
 242             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 243             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 244             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 245
 246             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 247             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 248             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 249
 250             /* Load parameters for j particles */
 251             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 252                                                                  charge+jnrC+0,charge+jnrD+0,
 253                                                                  charge+jnrE+0,charge+jnrF+0,
 254                                                                  charge+jnrG+0,charge+jnrH+0);
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256             vdwjidx0B        = 2*vdwtype[jnrB+0];
 257             vdwjidx0C        = 2*vdwtype[jnrC+0];
 258             vdwjidx0D        = 2*vdwtype[jnrD+0];
 259             vdwjidx0E        = 2*vdwtype[jnrE+0];
 260             vdwjidx0F        = 2*vdwtype[jnrF+0];
 261             vdwjidx0G        = 2*vdwtype[jnrG+0];
 262             vdwjidx0H        = 2*vdwtype[jnrH+0];
 263
 264             fjx0             = _mm256_setzero_ps();
 265             fjy0             = _mm256_setzero_ps();
 266             fjz0             = _mm256_setzero_ps();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             r00              = _mm256_mul_ps(rsq00,rinv00);
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq00             = _mm256_mul_ps(iq0,jq0);
 276             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 277                                             vdwioffsetptr0+vdwjidx0B,
 278                                             vdwioffsetptr0+vdwjidx0C,
 279                                             vdwioffsetptr0+vdwjidx0D,
 280                                             vdwioffsetptr0+vdwjidx0E,
 281                                             vdwioffsetptr0+vdwjidx0F,
 282                                             vdwioffsetptr0+vdwjidx0G,
 283                                             vdwioffsetptr0+vdwjidx0H,
 284                                             &c6_00,&c12_00);
 285
 286             /* Calculate table index by multiplying r with table scale and truncate to integer */
 287             rt               = _mm256_mul_ps(r00,vftabscale);
 288             vfitab           = _mm256_cvttps_epi32(rt);
 289             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 290             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 291             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 292             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 293             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 294             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 295
 296             /* REACTION-FIELD ELECTROSTATICS */
 297             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 298             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 299
 300             /* CUBIC SPLINE TABLE DISPERSION */
 301             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 302                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 303             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 304                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 305             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 307             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 309             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 310             Heps             = _mm256_mul_ps(vfeps,H);
 311             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 312             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 313             vvdw6            = _mm256_mul_ps(c6_00,VV);
 314             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 315             fvdw6            = _mm256_mul_ps(c6_00,FF);
 316
 317             /* CUBIC SPLINE TABLE REPULSION */
 318             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 319             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 320             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 322             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 324             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 326             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 328             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 329             Heps             = _mm256_mul_ps(vfeps,H);
 330             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 331             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 332             vvdw12           = _mm256_mul_ps(c12_00,VV);
 333             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 334             fvdw12           = _mm256_mul_ps(c12_00,FF);
 335             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 336             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velecsum         = _mm256_add_ps(velecsum,velec);
 340             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 341
 342             fscal            = _mm256_add_ps(felec,fvdw);
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm256_mul_ps(fscal,dx00);
 346             ty               = _mm256_mul_ps(fscal,dy00);
 347             tz               = _mm256_mul_ps(fscal,dz00);
 348
 349             /* Update vectorial force */
 350             fix0             = _mm256_add_ps(fix0,tx);
 351             fiy0             = _mm256_add_ps(fiy0,ty);
 352             fiz0             = _mm256_add_ps(fiz0,tz);
 353
 354             fjx0             = _mm256_add_ps(fjx0,tx);
 355             fjy0             = _mm256_add_ps(fjy0,ty);
 356             fjz0             = _mm256_add_ps(fjz0,tz);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq10             = _mm256_mul_ps(iq1,jq0);
 364
 365             /* REACTION-FIELD ELECTROSTATICS */
 366             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 367             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velecsum         = _mm256_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm256_mul_ps(fscal,dx10);
 376             ty               = _mm256_mul_ps(fscal,dy10);
 377             tz               = _mm256_mul_ps(fscal,dz10);
 378
 379             /* Update vectorial force */
 380             fix1             = _mm256_add_ps(fix1,tx);
 381             fiy1             = _mm256_add_ps(fiy1,ty);
 382             fiz1             = _mm256_add_ps(fiz1,tz);
 383
 384             fjx0             = _mm256_add_ps(fjx0,tx);
 385             fjy0             = _mm256_add_ps(fjy0,ty);
 386             fjz0             = _mm256_add_ps(fjz0,tz);
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq20             = _mm256_mul_ps(iq2,jq0);
 394
 395             /* REACTION-FIELD ELECTROSTATICS */
 396             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 397             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velecsum         = _mm256_add_ps(velecsum,velec);
 401
 402             fscal            = felec;
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm256_mul_ps(fscal,dx20);
 406             ty               = _mm256_mul_ps(fscal,dy20);
 407             tz               = _mm256_mul_ps(fscal,dz20);
 408
 409             /* Update vectorial force */
 410             fix2             = _mm256_add_ps(fix2,tx);
 411             fiy2             = _mm256_add_ps(fiy2,ty);
 412             fiz2             = _mm256_add_ps(fiz2,tz);
 413
 414             fjx0             = _mm256_add_ps(fjx0,tx);
 415             fjy0             = _mm256_add_ps(fjy0,ty);
 416             fjz0             = _mm256_add_ps(fjz0,tz);
 417
 418             fjptrA             = f+j_coord_offsetA;
 419             fjptrB             = f+j_coord_offsetB;
 420             fjptrC             = f+j_coord_offsetC;
 421             fjptrD             = f+j_coord_offsetD;
 422             fjptrE             = f+j_coord_offsetE;
 423             fjptrF             = f+j_coord_offsetF;
 424             fjptrG             = f+j_coord_offsetG;
 425             fjptrH             = f+j_coord_offsetH;
 426
 427             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 428
 429             /* Inner loop uses 134 flops */
 430         }
 431
 432         if(jidx<j_index_end)
 433         {
 434
 435             /* Get j neighbor index, and coordinate index */
 436             jnrlistA         = jjnr[jidx];
 437             jnrlistB         = jjnr[jidx+1];
 438             jnrlistC         = jjnr[jidx+2];
 439             jnrlistD         = jjnr[jidx+3];
 440             jnrlistE         = jjnr[jidx+4];
 441             jnrlistF         = jjnr[jidx+5];
 442             jnrlistG         = jjnr[jidx+6];
 443             jnrlistH         = jjnr[jidx+7];
 444             /* Sign of each element will be negative for non-real atoms.
 445              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 446              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 447              */
 448             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 449                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 450
 451             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 452             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 453             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 454             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 455             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 456             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 457             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 458             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 459             j_coord_offsetA  = DIM*jnrA;
 460             j_coord_offsetB  = DIM*jnrB;
 461             j_coord_offsetC  = DIM*jnrC;
 462             j_coord_offsetD  = DIM*jnrD;
 463             j_coord_offsetE  = DIM*jnrE;
 464             j_coord_offsetF  = DIM*jnrF;
 465             j_coord_offsetG  = DIM*jnrG;
 466             j_coord_offsetH  = DIM*jnrH;
 467
 468             /* load j atom coordinates */
 469             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 470                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 471                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 472                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 473                                                  &jx0,&jy0,&jz0);
 474
 475             /* Calculate displacement vector */
 476             dx00             = _mm256_sub_ps(ix0,jx0);
 477             dy00             = _mm256_sub_ps(iy0,jy0);
 478             dz00             = _mm256_sub_ps(iz0,jz0);
 479             dx10             = _mm256_sub_ps(ix1,jx0);
 480             dy10             = _mm256_sub_ps(iy1,jy0);
 481             dz10             = _mm256_sub_ps(iz1,jz0);
 482             dx20             = _mm256_sub_ps(ix2,jx0);
 483             dy20             = _mm256_sub_ps(iy2,jy0);
 484             dz20             = _mm256_sub_ps(iz2,jz0);
 485
 486             /* Calculate squared distance and things based on it */
 487             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 488             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 489             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 490
 491             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 492             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 493             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 494
 495             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 496             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 497             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 498
 499             /* Load parameters for j particles */
 500             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 501                                                                  charge+jnrC+0,charge+jnrD+0,
 502                                                                  charge+jnrE+0,charge+jnrF+0,
 503                                                                  charge+jnrG+0,charge+jnrH+0);
 504             vdwjidx0A        = 2*vdwtype[jnrA+0];
 505             vdwjidx0B        = 2*vdwtype[jnrB+0];
 506             vdwjidx0C        = 2*vdwtype[jnrC+0];
 507             vdwjidx0D        = 2*vdwtype[jnrD+0];
 508             vdwjidx0E        = 2*vdwtype[jnrE+0];
 509             vdwjidx0F        = 2*vdwtype[jnrF+0];
 510             vdwjidx0G        = 2*vdwtype[jnrG+0];
 511             vdwjidx0H        = 2*vdwtype[jnrH+0];
 512
 513             fjx0             = _mm256_setzero_ps();
 514             fjy0             = _mm256_setzero_ps();
 515             fjz0             = _mm256_setzero_ps();
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             r00              = _mm256_mul_ps(rsq00,rinv00);
 522             r00              = _mm256_andnot_ps(dummy_mask,r00);
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             qq00             = _mm256_mul_ps(iq0,jq0);
 526             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 527                                             vdwioffsetptr0+vdwjidx0B,
 528                                             vdwioffsetptr0+vdwjidx0C,
 529                                             vdwioffsetptr0+vdwjidx0D,
 530                                             vdwioffsetptr0+vdwjidx0E,
 531                                             vdwioffsetptr0+vdwjidx0F,
 532                                             vdwioffsetptr0+vdwjidx0G,
 533                                             vdwioffsetptr0+vdwjidx0H,
 534                                             &c6_00,&c12_00);
 535
 536             /* Calculate table index by multiplying r with table scale and truncate to integer */
 537             rt               = _mm256_mul_ps(r00,vftabscale);
 538             vfitab           = _mm256_cvttps_epi32(rt);
 539             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 540             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 541             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 542             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 543             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 544             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 545
 546             /* REACTION-FIELD ELECTROSTATICS */
 547             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 548             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 549
 550             /* CUBIC SPLINE TABLE DISPERSION */
 551             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 552                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 553             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 554                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 555             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 556                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 557             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 558                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 559             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 560             Heps             = _mm256_mul_ps(vfeps,H);
 561             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 562             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 563             vvdw6            = _mm256_mul_ps(c6_00,VV);
 564             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 565             fvdw6            = _mm256_mul_ps(c6_00,FF);
 566
 567             /* CUBIC SPLINE TABLE REPULSION */
 568             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 569             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 570             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 571                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 572             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 573                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 574             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 576             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 578             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 579             Heps             = _mm256_mul_ps(vfeps,H);
 580             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 581             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 582             vvdw12           = _mm256_mul_ps(c12_00,VV);
 583             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 584             fvdw12           = _mm256_mul_ps(c12_00,FF);
 585             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 586             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             velec            = _mm256_andnot_ps(dummy_mask,velec);
 590             velecsum         = _mm256_add_ps(velecsum,velec);
 591             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 592             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 593
 594             fscal            = _mm256_add_ps(felec,fvdw);
 595
 596             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_ps(fscal,dx00);
 600             ty               = _mm256_mul_ps(fscal,dy00);
 601             tz               = _mm256_mul_ps(fscal,dz00);
 602
 603             /* Update vectorial force */
 604             fix0             = _mm256_add_ps(fix0,tx);
 605             fiy0             = _mm256_add_ps(fiy0,ty);
 606             fiz0             = _mm256_add_ps(fiz0,tz);
 607
 608             fjx0             = _mm256_add_ps(fjx0,tx);
 609             fjy0             = _mm256_add_ps(fjy0,ty);
 610             fjz0             = _mm256_add_ps(fjz0,tz);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq10             = _mm256_mul_ps(iq1,jq0);
 618
 619             /* REACTION-FIELD ELECTROSTATICS */
 620             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 621             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             velec            = _mm256_andnot_ps(dummy_mask,velec);
 625             velecsum         = _mm256_add_ps(velecsum,velec);
 626
 627             fscal            = felec;
 628
 629             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm256_mul_ps(fscal,dx10);
 633             ty               = _mm256_mul_ps(fscal,dy10);
 634             tz               = _mm256_mul_ps(fscal,dz10);
 635
 636             /* Update vectorial force */
 637             fix1             = _mm256_add_ps(fix1,tx);
 638             fiy1             = _mm256_add_ps(fiy1,ty);
 639             fiz1             = _mm256_add_ps(fiz1,tz);
 640
 641             fjx0             = _mm256_add_ps(fjx0,tx);
 642             fjy0             = _mm256_add_ps(fjy0,ty);
 643             fjz0             = _mm256_add_ps(fjz0,tz);
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq20             = _mm256_mul_ps(iq2,jq0);
 651
 652             /* REACTION-FIELD ELECTROSTATICS */
 653             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 654             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             velec            = _mm256_andnot_ps(dummy_mask,velec);
 658             velecsum         = _mm256_add_ps(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm256_mul_ps(fscal,dx20);
 666             ty               = _mm256_mul_ps(fscal,dy20);
 667             tz               = _mm256_mul_ps(fscal,dz20);
 668
 669             /* Update vectorial force */
 670             fix2             = _mm256_add_ps(fix2,tx);
 671             fiy2             = _mm256_add_ps(fiy2,ty);
 672             fiz2             = _mm256_add_ps(fiz2,tz);
 673
 674             fjx0             = _mm256_add_ps(fjx0,tx);
 675             fjy0             = _mm256_add_ps(fjy0,ty);
 676             fjz0             = _mm256_add_ps(fjz0,tz);
 677
 678             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 679             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 680             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 681             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 682             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 683             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 684             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 685             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 686
 687             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 688
 689             /* Inner loop uses 135 flops */
 690         }
 691
 692         /* End of innermost loop */
 693
 694         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 695                                                  f+i_coord_offset,fshift+i_shift_offset);
 696
 697         ggid                        = gid[iidx];
 698         /* Update potential energies */
 699         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 700         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 701
 702         /* Increment number of inner iterations */
 703         inneriter                  += j_index_end - j_index_start;
 704
 705         /* Outer loop uses 20 flops */
 706     }
 707
 708     /* Increment number of outer iterations */
 709     outeriter        += nri;
 710
 711     /* Update outer/inner flops */
 712
 713     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*135);
 714 }
 715 /*
 716  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_single
 717  * Electrostatics interaction: ReactionField
 718  * VdW interaction:            CubicSplineTable
 719  * Geometry:                   Water3-Particle
 720  * Calculate force/pot:        Force
 721  */
 722 void
 723 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_single
 724                     (t_nblist                    * gmx_restrict       nlist,
 725                      rvec                        * gmx_restrict          xx,
 726                      rvec                        * gmx_restrict          ff,
 727                      t_forcerec                  * gmx_restrict          fr,
 728                      t_mdatoms                   * gmx_restrict     mdatoms,
 729                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 730                      t_nrnb                      * gmx_restrict        nrnb)
 731 {
 732     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 733      * just 0 for non-waters.
 734      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 735      * jnr indices corresponding to data put in the four positions in the SIMD register.
 736      */
 737     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 738     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 739     int              jnrA,jnrB,jnrC,jnrD;
 740     int              jnrE,jnrF,jnrG,jnrH;
 741     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 742     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 743     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 744     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 745     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 746     real             rcutoff_scalar;
 747     real             *shiftvec,*fshift,*x,*f;
 748     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 749     real             scratch[4*DIM];
 750     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 751     real *           vdwioffsetptr0;
 752     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 753     real *           vdwioffsetptr1;
 754     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 755     real *           vdwioffsetptr2;
 756     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 757     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 758     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 759     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 760     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 761     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 762     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 763     real             *charge;
 764     int              nvdwtype;
 765     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 766     int              *vdwtype;
 767     real             *vdwparam;
 768     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 769     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 770     __m256i          vfitab;
 771     __m128i          vfitab_lo,vfitab_hi;
 772     __m128i          ifour       = _mm_set1_epi32(4);
 773     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 774     real             *vftab;
 775     __m256           dummy_mask,cutoff_mask;
 776     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 777     __m256           one     = _mm256_set1_ps(1.0);
 778     __m256           two     = _mm256_set1_ps(2.0);
 779     x                = xx[0];
 780     f                = ff[0];
 781
 782     nri              = nlist->nri;
 783     iinr             = nlist->iinr;
 784     jindex           = nlist->jindex;
 785     jjnr             = nlist->jjnr;
 786     shiftidx         = nlist->shift;
 787     gid              = nlist->gid;
 788     shiftvec         = fr->shift_vec[0];
 789     fshift           = fr->fshift[0];
 790     facel            = _mm256_set1_ps(fr->epsfac);
 791     charge           = mdatoms->chargeA;
 792     krf              = _mm256_set1_ps(fr->ic->k_rf);
 793     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 794     crf              = _mm256_set1_ps(fr->ic->c_rf);
 795     nvdwtype         = fr->ntype;
 796     vdwparam         = fr->nbfp;
 797     vdwtype          = mdatoms->typeA;
 798
 799     vftab            = kernel_data->table_vdw->data;
 800     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 801
 802     /* Setup water-specific parameters */
 803     inr              = nlist->iinr[0];
 804     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 805     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 806     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 807     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 808
 809     /* Avoid stupid compiler warnings */
 810     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 811     j_coord_offsetA = 0;
 812     j_coord_offsetB = 0;
 813     j_coord_offsetC = 0;
 814     j_coord_offsetD = 0;
 815     j_coord_offsetE = 0;
 816     j_coord_offsetF = 0;
 817     j_coord_offsetG = 0;
 818     j_coord_offsetH = 0;
 819
 820     outeriter        = 0;
 821     inneriter        = 0;
 822
 823     for(iidx=0;iidx<4*DIM;iidx++)
 824     {
 825         scratch[iidx] = 0.0;
 826     }
 827
 828     /* Start outer loop over neighborlists */
 829     for(iidx=0; iidx<nri; iidx++)
 830     {
 831         /* Load shift vector for this list */
 832         i_shift_offset   = DIM*shiftidx[iidx];
 833
 834         /* Load limits for loop over neighbors */
 835         j_index_start    = jindex[iidx];
 836         j_index_end      = jindex[iidx+1];
 837
 838         /* Get outer coordinate index */
 839         inr              = iinr[iidx];
 840         i_coord_offset   = DIM*inr;
 841
 842         /* Load i particle coords and add shift vector */
 843         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 844                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 845
 846         fix0             = _mm256_setzero_ps();
 847         fiy0             = _mm256_setzero_ps();
 848         fiz0             = _mm256_setzero_ps();
 849         fix1             = _mm256_setzero_ps();
 850         fiy1             = _mm256_setzero_ps();
 851         fiz1             = _mm256_setzero_ps();
 852         fix2             = _mm256_setzero_ps();
 853         fiy2             = _mm256_setzero_ps();
 854         fiz2             = _mm256_setzero_ps();
 855
 856         /* Start inner kernel loop */
 857         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 858         {
 859
 860             /* Get j neighbor index, and coordinate index */
 861             jnrA             = jjnr[jidx];
 862             jnrB             = jjnr[jidx+1];
 863             jnrC             = jjnr[jidx+2];
 864             jnrD             = jjnr[jidx+3];
 865             jnrE             = jjnr[jidx+4];
 866             jnrF             = jjnr[jidx+5];
 867             jnrG             = jjnr[jidx+6];
 868             jnrH             = jjnr[jidx+7];
 869             j_coord_offsetA  = DIM*jnrA;
 870             j_coord_offsetB  = DIM*jnrB;
 871             j_coord_offsetC  = DIM*jnrC;
 872             j_coord_offsetD  = DIM*jnrD;
 873             j_coord_offsetE  = DIM*jnrE;
 874             j_coord_offsetF  = DIM*jnrF;
 875             j_coord_offsetG  = DIM*jnrG;
 876             j_coord_offsetH  = DIM*jnrH;
 877
 878             /* load j atom coordinates */
 879             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 880                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 881                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 882                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 883                                                  &jx0,&jy0,&jz0);
 884
 885             /* Calculate displacement vector */
 886             dx00             = _mm256_sub_ps(ix0,jx0);
 887             dy00             = _mm256_sub_ps(iy0,jy0);
 888             dz00             = _mm256_sub_ps(iz0,jz0);
 889             dx10             = _mm256_sub_ps(ix1,jx0);
 890             dy10             = _mm256_sub_ps(iy1,jy0);
 891             dz10             = _mm256_sub_ps(iz1,jz0);
 892             dx20             = _mm256_sub_ps(ix2,jx0);
 893             dy20             = _mm256_sub_ps(iy2,jy0);
 894             dz20             = _mm256_sub_ps(iz2,jz0);
 895
 896             /* Calculate squared distance and things based on it */
 897             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 898             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 899             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 900
 901             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 902             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 903             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 904
 905             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 906             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 907             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 908
 909             /* Load parameters for j particles */
 910             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 911                                                                  charge+jnrC+0,charge+jnrD+0,
 912                                                                  charge+jnrE+0,charge+jnrF+0,
 913                                                                  charge+jnrG+0,charge+jnrH+0);
 914             vdwjidx0A        = 2*vdwtype[jnrA+0];
 915             vdwjidx0B        = 2*vdwtype[jnrB+0];
 916             vdwjidx0C        = 2*vdwtype[jnrC+0];
 917             vdwjidx0D        = 2*vdwtype[jnrD+0];
 918             vdwjidx0E        = 2*vdwtype[jnrE+0];
 919             vdwjidx0F        = 2*vdwtype[jnrF+0];
 920             vdwjidx0G        = 2*vdwtype[jnrG+0];
 921             vdwjidx0H        = 2*vdwtype[jnrH+0];
 922
 923             fjx0             = _mm256_setzero_ps();
 924             fjy0             = _mm256_setzero_ps();
 925             fjz0             = _mm256_setzero_ps();
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             r00              = _mm256_mul_ps(rsq00,rinv00);
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq00             = _mm256_mul_ps(iq0,jq0);
 935             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 936                                             vdwioffsetptr0+vdwjidx0B,
 937                                             vdwioffsetptr0+vdwjidx0C,
 938                                             vdwioffsetptr0+vdwjidx0D,
 939                                             vdwioffsetptr0+vdwjidx0E,
 940                                             vdwioffsetptr0+vdwjidx0F,
 941                                             vdwioffsetptr0+vdwjidx0G,
 942                                             vdwioffsetptr0+vdwjidx0H,
 943                                             &c6_00,&c12_00);
 944
 945             /* Calculate table index by multiplying r with table scale and truncate to integer */
 946             rt               = _mm256_mul_ps(r00,vftabscale);
 947             vfitab           = _mm256_cvttps_epi32(rt);
 948             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 949             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 950             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 951             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 952             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 953             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 954
 955             /* REACTION-FIELD ELECTROSTATICS */
 956             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 957
 958             /* CUBIC SPLINE TABLE DISPERSION */
 959             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 960                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 961             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 962                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 963             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 964                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 965             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 966                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 967             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 968             Heps             = _mm256_mul_ps(vfeps,H);
 969             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 970             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 971             fvdw6            = _mm256_mul_ps(c6_00,FF);
 972
 973             /* CUBIC SPLINE TABLE REPULSION */
 974             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 975             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 976             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 977                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 978             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 979                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 980             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 981                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 982             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 983                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 984             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 985             Heps             = _mm256_mul_ps(vfeps,H);
 986             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 987             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 988             fvdw12           = _mm256_mul_ps(c12_00,FF);
 989             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 990
 991             fscal            = _mm256_add_ps(felec,fvdw);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_ps(fscal,dx00);
 995             ty               = _mm256_mul_ps(fscal,dy00);
 996             tz               = _mm256_mul_ps(fscal,dz00);
 997
 998             /* Update vectorial force */
 999             fix0             = _mm256_add_ps(fix0,tx);
1000             fiy0             = _mm256_add_ps(fiy0,ty);
1001             fiz0             = _mm256_add_ps(fiz0,tz);
1002
1003             fjx0             = _mm256_add_ps(fjx0,tx);
1004             fjy0             = _mm256_add_ps(fjy0,ty);
1005             fjz0             = _mm256_add_ps(fjz0,tz);
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             qq10             = _mm256_mul_ps(iq1,jq0);
1013
1014             /* REACTION-FIELD ELECTROSTATICS */
1015             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1016
1017             fscal            = felec;
1018
1019             /* Calculate temporary vectorial force */
1020             tx               = _mm256_mul_ps(fscal,dx10);
1021             ty               = _mm256_mul_ps(fscal,dy10);
1022             tz               = _mm256_mul_ps(fscal,dz10);
1023
1024             /* Update vectorial force */
1025             fix1             = _mm256_add_ps(fix1,tx);
1026             fiy1             = _mm256_add_ps(fiy1,ty);
1027             fiz1             = _mm256_add_ps(fiz1,tz);
1028
1029             fjx0             = _mm256_add_ps(fjx0,tx);
1030             fjy0             = _mm256_add_ps(fjy0,ty);
1031             fjz0             = _mm256_add_ps(fjz0,tz);
1032
1033             /**************************
1034              * CALCULATE INTERACTIONS *
1035              **************************/
1036
1037             /* Compute parameters for interactions between i and j atoms */
1038             qq20             = _mm256_mul_ps(iq2,jq0);
1039
1040             /* REACTION-FIELD ELECTROSTATICS */
1041             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1042
1043             fscal            = felec;
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm256_mul_ps(fscal,dx20);
1047             ty               = _mm256_mul_ps(fscal,dy20);
1048             tz               = _mm256_mul_ps(fscal,dz20);
1049
1050             /* Update vectorial force */
1051             fix2             = _mm256_add_ps(fix2,tx);
1052             fiy2             = _mm256_add_ps(fiy2,ty);
1053             fiz2             = _mm256_add_ps(fiz2,tz);
1054
1055             fjx0             = _mm256_add_ps(fjx0,tx);
1056             fjy0             = _mm256_add_ps(fjy0,ty);
1057             fjz0             = _mm256_add_ps(fjz0,tz);
1058
1059             fjptrA             = f+j_coord_offsetA;
1060             fjptrB             = f+j_coord_offsetB;
1061             fjptrC             = f+j_coord_offsetC;
1062             fjptrD             = f+j_coord_offsetD;
1063             fjptrE             = f+j_coord_offsetE;
1064             fjptrF             = f+j_coord_offsetF;
1065             fjptrG             = f+j_coord_offsetG;
1066             fjptrH             = f+j_coord_offsetH;
1067
1068             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1069
1070             /* Inner loop uses 111 flops */
1071         }
1072
1073         if(jidx<j_index_end)
1074         {
1075
1076             /* Get j neighbor index, and coordinate index */
1077             jnrlistA         = jjnr[jidx];
1078             jnrlistB         = jjnr[jidx+1];
1079             jnrlistC         = jjnr[jidx+2];
1080             jnrlistD         = jjnr[jidx+3];
1081             jnrlistE         = jjnr[jidx+4];
1082             jnrlistF         = jjnr[jidx+5];
1083             jnrlistG         = jjnr[jidx+6];
1084             jnrlistH         = jjnr[jidx+7];
1085             /* Sign of each element will be negative for non-real atoms.
1086              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1087              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1088              */
1089             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1090                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1091
1092             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1093             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1094             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1095             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1096             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1097             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1098             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1099             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1100             j_coord_offsetA  = DIM*jnrA;
1101             j_coord_offsetB  = DIM*jnrB;
1102             j_coord_offsetC  = DIM*jnrC;
1103             j_coord_offsetD  = DIM*jnrD;
1104             j_coord_offsetE  = DIM*jnrE;
1105             j_coord_offsetF  = DIM*jnrF;
1106             j_coord_offsetG  = DIM*jnrG;
1107             j_coord_offsetH  = DIM*jnrH;
1108
1109             /* load j atom coordinates */
1110             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1111                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1112                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1113                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1114                                                  &jx0,&jy0,&jz0);
1115
1116             /* Calculate displacement vector */
1117             dx00             = _mm256_sub_ps(ix0,jx0);
1118             dy00             = _mm256_sub_ps(iy0,jy0);
1119             dz00             = _mm256_sub_ps(iz0,jz0);
1120             dx10             = _mm256_sub_ps(ix1,jx0);
1121             dy10             = _mm256_sub_ps(iy1,jy0);
1122             dz10             = _mm256_sub_ps(iz1,jz0);
1123             dx20             = _mm256_sub_ps(ix2,jx0);
1124             dy20             = _mm256_sub_ps(iy2,jy0);
1125             dz20             = _mm256_sub_ps(iz2,jz0);
1126
1127             /* Calculate squared distance and things based on it */
1128             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1129             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1130             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1131
1132             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1133             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1134             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1135
1136             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1137             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1138             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1139
1140             /* Load parameters for j particles */
1141             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1142                                                                  charge+jnrC+0,charge+jnrD+0,
1143                                                                  charge+jnrE+0,charge+jnrF+0,
1144                                                                  charge+jnrG+0,charge+jnrH+0);
1145             vdwjidx0A        = 2*vdwtype[jnrA+0];
1146             vdwjidx0B        = 2*vdwtype[jnrB+0];
1147             vdwjidx0C        = 2*vdwtype[jnrC+0];
1148             vdwjidx0D        = 2*vdwtype[jnrD+0];
1149             vdwjidx0E        = 2*vdwtype[jnrE+0];
1150             vdwjidx0F        = 2*vdwtype[jnrF+0];
1151             vdwjidx0G        = 2*vdwtype[jnrG+0];
1152             vdwjidx0H        = 2*vdwtype[jnrH+0];
1153
1154             fjx0             = _mm256_setzero_ps();
1155             fjy0             = _mm256_setzero_ps();
1156             fjz0             = _mm256_setzero_ps();
1157
1158             /**************************
1159              * CALCULATE INTERACTIONS *
1160              **************************/
1161
1162             r00              = _mm256_mul_ps(rsq00,rinv00);
1163             r00              = _mm256_andnot_ps(dummy_mask,r00);
1164
1165             /* Compute parameters for interactions between i and j atoms */
1166             qq00             = _mm256_mul_ps(iq0,jq0);
1167             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1168                                             vdwioffsetptr0+vdwjidx0B,
1169                                             vdwioffsetptr0+vdwjidx0C,
1170                                             vdwioffsetptr0+vdwjidx0D,
1171                                             vdwioffsetptr0+vdwjidx0E,
1172                                             vdwioffsetptr0+vdwjidx0F,
1173                                             vdwioffsetptr0+vdwjidx0G,
1174                                             vdwioffsetptr0+vdwjidx0H,
1175                                             &c6_00,&c12_00);
1176
1177             /* Calculate table index by multiplying r with table scale and truncate to integer */
1178             rt               = _mm256_mul_ps(r00,vftabscale);
1179             vfitab           = _mm256_cvttps_epi32(rt);
1180             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1181             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1182             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1183             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1184             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1185             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1186
1187             /* REACTION-FIELD ELECTROSTATICS */
1188             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1189
1190             /* CUBIC SPLINE TABLE DISPERSION */
1191             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1193             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1195             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1196                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1197             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1198                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1199             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1200             Heps             = _mm256_mul_ps(vfeps,H);
1201             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1202             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1203             fvdw6            = _mm256_mul_ps(c6_00,FF);
1204
1205             /* CUBIC SPLINE TABLE REPULSION */
1206             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1207             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1208             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1210             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1212             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1214             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1215                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1216             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1217             Heps             = _mm256_mul_ps(vfeps,H);
1218             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1219             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1220             fvdw12           = _mm256_mul_ps(c12_00,FF);
1221             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1222
1223             fscal            = _mm256_add_ps(felec,fvdw);
1224
1225             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1226
1227             /* Calculate temporary vectorial force */
1228             tx               = _mm256_mul_ps(fscal,dx00);
1229             ty               = _mm256_mul_ps(fscal,dy00);
1230             tz               = _mm256_mul_ps(fscal,dz00);
1231
1232             /* Update vectorial force */
1233             fix0             = _mm256_add_ps(fix0,tx);
1234             fiy0             = _mm256_add_ps(fiy0,ty);
1235             fiz0             = _mm256_add_ps(fiz0,tz);
1236
1237             fjx0             = _mm256_add_ps(fjx0,tx);
1238             fjy0             = _mm256_add_ps(fjy0,ty);
1239             fjz0             = _mm256_add_ps(fjz0,tz);
1240
1241             /**************************
1242              * CALCULATE INTERACTIONS *
1243              **************************/
1244
1245             /* Compute parameters for interactions between i and j atoms */
1246             qq10             = _mm256_mul_ps(iq1,jq0);
1247
1248             /* REACTION-FIELD ELECTROSTATICS */
1249             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1250
1251             fscal            = felec;
1252
1253             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1254
1255             /* Calculate temporary vectorial force */
1256             tx               = _mm256_mul_ps(fscal,dx10);
1257             ty               = _mm256_mul_ps(fscal,dy10);
1258             tz               = _mm256_mul_ps(fscal,dz10);
1259
1260             /* Update vectorial force */
1261             fix1             = _mm256_add_ps(fix1,tx);
1262             fiy1             = _mm256_add_ps(fiy1,ty);
1263             fiz1             = _mm256_add_ps(fiz1,tz);
1264
1265             fjx0             = _mm256_add_ps(fjx0,tx);
1266             fjy0             = _mm256_add_ps(fjy0,ty);
1267             fjz0             = _mm256_add_ps(fjz0,tz);
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             /* Compute parameters for interactions between i and j atoms */
1274             qq20             = _mm256_mul_ps(iq2,jq0);
1275
1276             /* REACTION-FIELD ELECTROSTATICS */
1277             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1278
1279             fscal            = felec;
1280
1281             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1282
1283             /* Calculate temporary vectorial force */
1284             tx               = _mm256_mul_ps(fscal,dx20);
1285             ty               = _mm256_mul_ps(fscal,dy20);
1286             tz               = _mm256_mul_ps(fscal,dz20);
1287
1288             /* Update vectorial force */
1289             fix2             = _mm256_add_ps(fix2,tx);
1290             fiy2             = _mm256_add_ps(fiy2,ty);
1291             fiz2             = _mm256_add_ps(fiz2,tz);
1292
1293             fjx0             = _mm256_add_ps(fjx0,tx);
1294             fjy0             = _mm256_add_ps(fjy0,ty);
1295             fjz0             = _mm256_add_ps(fjz0,tz);
1296
1297             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1298             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1299             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1300             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1301             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1302             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1303             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1304             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1305
1306             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1307
1308             /* Inner loop uses 112 flops */
1309         }
1310
1311         /* End of innermost loop */
1312
1313         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1314                                                  f+i_coord_offset,fshift+i_shift_offset);
1315
1316         /* Increment number of inner iterations */
1317         inneriter                  += j_index_end - j_index_start;
1318
1319         /* Outer loop uses 18 flops */
1320     }
1321
1322     /* Increment number of outer iterations */
1323     outeriter        += nri;
1324
1325     /* Update outer/inner flops */
1326
1327     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*112);
1328 }