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) 2012,2013,2014, 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_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_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     krf              = _mm256_set1_ps(fr->ic->k_rf);
 128     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 129     crf              = _mm256_set1_ps(fr->ic->c_rf);
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_vdw->data;
 135     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 140     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 141     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 180
 181         fix0             = _mm256_setzero_ps();
 182         fiy0             = _mm256_setzero_ps();
 183         fiz0             = _mm256_setzero_ps();
 184         fix1             = _mm256_setzero_ps();
 185         fiy1             = _mm256_setzero_ps();
 186         fiz1             = _mm256_setzero_ps();
 187         fix2             = _mm256_setzero_ps();
 188         fiy2             = _mm256_setzero_ps();
 189         fiz2             = _mm256_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm256_setzero_ps();
 193         vvdwsum          = _mm256_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             jnrE             = jjnr[jidx+4];
 205             jnrF             = jjnr[jidx+5];
 206             jnrG             = jjnr[jidx+6];
 207             jnrH             = jjnr[jidx+7];
 208             j_coord_offsetA  = DIM*jnrA;
 209             j_coord_offsetB  = DIM*jnrB;
 210             j_coord_offsetC  = DIM*jnrC;
 211             j_coord_offsetD  = DIM*jnrD;
 212             j_coord_offsetE  = DIM*jnrE;
 213             j_coord_offsetF  = DIM*jnrF;
 214             j_coord_offsetG  = DIM*jnrG;
 215             j_coord_offsetH  = DIM*jnrH;
 216
 217             /* load j atom coordinates */
 218             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 219                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 220                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 221                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 222                                                  &jx0,&jy0,&jz0);
 223
 224             /* Calculate displacement vector */
 225             dx00             = _mm256_sub_ps(ix0,jx0);
 226             dy00             = _mm256_sub_ps(iy0,jy0);
 227             dz00             = _mm256_sub_ps(iz0,jz0);
 228             dx10             = _mm256_sub_ps(ix1,jx0);
 229             dy10             = _mm256_sub_ps(iy1,jy0);
 230             dz10             = _mm256_sub_ps(iz1,jz0);
 231             dx20             = _mm256_sub_ps(ix2,jx0);
 232             dy20             = _mm256_sub_ps(iy2,jy0);
 233             dz20             = _mm256_sub_ps(iz2,jz0);
 234
 235             /* Calculate squared distance and things based on it */
 236             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 237             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 238             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 239
 240             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 241             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 242             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 243
 244             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 245             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 246             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 247
 248             /* Load parameters for j particles */
 249             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 250                                                                  charge+jnrC+0,charge+jnrD+0,
 251                                                                  charge+jnrE+0,charge+jnrF+0,
 252                                                                  charge+jnrG+0,charge+jnrH+0);
 253             vdwjidx0A        = 2*vdwtype[jnrA+0];
 254             vdwjidx0B        = 2*vdwtype[jnrB+0];
 255             vdwjidx0C        = 2*vdwtype[jnrC+0];
 256             vdwjidx0D        = 2*vdwtype[jnrD+0];
 257             vdwjidx0E        = 2*vdwtype[jnrE+0];
 258             vdwjidx0F        = 2*vdwtype[jnrF+0];
 259             vdwjidx0G        = 2*vdwtype[jnrG+0];
 260             vdwjidx0H        = 2*vdwtype[jnrH+0];
 261
 262             fjx0             = _mm256_setzero_ps();
 263             fjy0             = _mm256_setzero_ps();
 264             fjz0             = _mm256_setzero_ps();
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             r00              = _mm256_mul_ps(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm256_mul_ps(iq0,jq0);
 274             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 275                                             vdwioffsetptr0+vdwjidx0B,
 276                                             vdwioffsetptr0+vdwjidx0C,
 277                                             vdwioffsetptr0+vdwjidx0D,
 278                                             vdwioffsetptr0+vdwjidx0E,
 279                                             vdwioffsetptr0+vdwjidx0F,
 280                                             vdwioffsetptr0+vdwjidx0G,
 281                                             vdwioffsetptr0+vdwjidx0H,
 282                                             &c6_00,&c12_00);
 283
 284             /* Calculate table index by multiplying r with table scale and truncate to integer */
 285             rt               = _mm256_mul_ps(r00,vftabscale);
 286             vfitab           = _mm256_cvttps_epi32(rt);
 287             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 288             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 289             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 290             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 291             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 292             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 293
 294             /* REACTION-FIELD ELECTROSTATICS */
 295             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 296             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 297
 298             /* CUBIC SPLINE TABLE DISPERSION */
 299             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 300                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 301             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 302                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 303             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 304                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 305             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 307             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 308             Heps             = _mm256_mul_ps(vfeps,H);
 309             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 310             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 311             vvdw6            = _mm256_mul_ps(c6_00,VV);
 312             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 313             fvdw6            = _mm256_mul_ps(c6_00,FF);
 314
 315             /* CUBIC SPLINE TABLE REPULSION */
 316             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 317             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 318             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 320             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 322             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 324             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 326             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 327             Heps             = _mm256_mul_ps(vfeps,H);
 328             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 329             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 330             vvdw12           = _mm256_mul_ps(c12_00,VV);
 331             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 332             fvdw12           = _mm256_mul_ps(c12_00,FF);
 333             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 334             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velecsum         = _mm256_add_ps(velecsum,velec);
 338             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 339
 340             fscal            = _mm256_add_ps(felec,fvdw);
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm256_mul_ps(fscal,dx00);
 344             ty               = _mm256_mul_ps(fscal,dy00);
 345             tz               = _mm256_mul_ps(fscal,dz00);
 346
 347             /* Update vectorial force */
 348             fix0             = _mm256_add_ps(fix0,tx);
 349             fiy0             = _mm256_add_ps(fiy0,ty);
 350             fiz0             = _mm256_add_ps(fiz0,tz);
 351
 352             fjx0             = _mm256_add_ps(fjx0,tx);
 353             fjy0             = _mm256_add_ps(fjy0,ty);
 354             fjz0             = _mm256_add_ps(fjz0,tz);
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq10             = _mm256_mul_ps(iq1,jq0);
 362
 363             /* REACTION-FIELD ELECTROSTATICS */
 364             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 365             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velecsum         = _mm256_add_ps(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm256_mul_ps(fscal,dx10);
 374             ty               = _mm256_mul_ps(fscal,dy10);
 375             tz               = _mm256_mul_ps(fscal,dz10);
 376
 377             /* Update vectorial force */
 378             fix1             = _mm256_add_ps(fix1,tx);
 379             fiy1             = _mm256_add_ps(fiy1,ty);
 380             fiz1             = _mm256_add_ps(fiz1,tz);
 381
 382             fjx0             = _mm256_add_ps(fjx0,tx);
 383             fjy0             = _mm256_add_ps(fjy0,ty);
 384             fjz0             = _mm256_add_ps(fjz0,tz);
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             /* Compute parameters for interactions between i and j atoms */
 391             qq20             = _mm256_mul_ps(iq2,jq0);
 392
 393             /* REACTION-FIELD ELECTROSTATICS */
 394             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 395             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velecsum         = _mm256_add_ps(velecsum,velec);
 399
 400             fscal            = felec;
 401
 402             /* Calculate temporary vectorial force */
 403             tx               = _mm256_mul_ps(fscal,dx20);
 404             ty               = _mm256_mul_ps(fscal,dy20);
 405             tz               = _mm256_mul_ps(fscal,dz20);
 406
 407             /* Update vectorial force */
 408             fix2             = _mm256_add_ps(fix2,tx);
 409             fiy2             = _mm256_add_ps(fiy2,ty);
 410             fiz2             = _mm256_add_ps(fiz2,tz);
 411
 412             fjx0             = _mm256_add_ps(fjx0,tx);
 413             fjy0             = _mm256_add_ps(fjy0,ty);
 414             fjz0             = _mm256_add_ps(fjz0,tz);
 415
 416             fjptrA             = f+j_coord_offsetA;
 417             fjptrB             = f+j_coord_offsetB;
 418             fjptrC             = f+j_coord_offsetC;
 419             fjptrD             = f+j_coord_offsetD;
 420             fjptrE             = f+j_coord_offsetE;
 421             fjptrF             = f+j_coord_offsetF;
 422             fjptrG             = f+j_coord_offsetG;
 423             fjptrH             = f+j_coord_offsetH;
 424
 425             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 426
 427             /* Inner loop uses 134 flops */
 428         }
 429
 430         if(jidx<j_index_end)
 431         {
 432
 433             /* Get j neighbor index, and coordinate index */
 434             jnrlistA         = jjnr[jidx];
 435             jnrlistB         = jjnr[jidx+1];
 436             jnrlistC         = jjnr[jidx+2];
 437             jnrlistD         = jjnr[jidx+3];
 438             jnrlistE         = jjnr[jidx+4];
 439             jnrlistF         = jjnr[jidx+5];
 440             jnrlistG         = jjnr[jidx+6];
 441             jnrlistH         = jjnr[jidx+7];
 442             /* Sign of each element will be negative for non-real atoms.
 443              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 444              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 445              */
 446             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 447                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 448
 449             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 450             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 451             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 452             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 453             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 454             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 455             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 456             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 457             j_coord_offsetA  = DIM*jnrA;
 458             j_coord_offsetB  = DIM*jnrB;
 459             j_coord_offsetC  = DIM*jnrC;
 460             j_coord_offsetD  = DIM*jnrD;
 461             j_coord_offsetE  = DIM*jnrE;
 462             j_coord_offsetF  = DIM*jnrF;
 463             j_coord_offsetG  = DIM*jnrG;
 464             j_coord_offsetH  = DIM*jnrH;
 465
 466             /* load j atom coordinates */
 467             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 468                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 469                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 470                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 471                                                  &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm256_sub_ps(ix0,jx0);
 475             dy00             = _mm256_sub_ps(iy0,jy0);
 476             dz00             = _mm256_sub_ps(iz0,jz0);
 477             dx10             = _mm256_sub_ps(ix1,jx0);
 478             dy10             = _mm256_sub_ps(iy1,jy0);
 479             dz10             = _mm256_sub_ps(iz1,jz0);
 480             dx20             = _mm256_sub_ps(ix2,jx0);
 481             dy20             = _mm256_sub_ps(iy2,jy0);
 482             dz20             = _mm256_sub_ps(iz2,jz0);
 483
 484             /* Calculate squared distance and things based on it */
 485             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 486             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 487             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 488
 489             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 490             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 491             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 492
 493             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 494             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 495             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 496
 497             /* Load parameters for j particles */
 498             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 499                                                                  charge+jnrC+0,charge+jnrD+0,
 500                                                                  charge+jnrE+0,charge+jnrF+0,
 501                                                                  charge+jnrG+0,charge+jnrH+0);
 502             vdwjidx0A        = 2*vdwtype[jnrA+0];
 503             vdwjidx0B        = 2*vdwtype[jnrB+0];
 504             vdwjidx0C        = 2*vdwtype[jnrC+0];
 505             vdwjidx0D        = 2*vdwtype[jnrD+0];
 506             vdwjidx0E        = 2*vdwtype[jnrE+0];
 507             vdwjidx0F        = 2*vdwtype[jnrF+0];
 508             vdwjidx0G        = 2*vdwtype[jnrG+0];
 509             vdwjidx0H        = 2*vdwtype[jnrH+0];
 510
 511             fjx0             = _mm256_setzero_ps();
 512             fjy0             = _mm256_setzero_ps();
 513             fjz0             = _mm256_setzero_ps();
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r00              = _mm256_mul_ps(rsq00,rinv00);
 520             r00              = _mm256_andnot_ps(dummy_mask,r00);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             qq00             = _mm256_mul_ps(iq0,jq0);
 524             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 525                                             vdwioffsetptr0+vdwjidx0B,
 526                                             vdwioffsetptr0+vdwjidx0C,
 527                                             vdwioffsetptr0+vdwjidx0D,
 528                                             vdwioffsetptr0+vdwjidx0E,
 529                                             vdwioffsetptr0+vdwjidx0F,
 530                                             vdwioffsetptr0+vdwjidx0G,
 531                                             vdwioffsetptr0+vdwjidx0H,
 532                                             &c6_00,&c12_00);
 533
 534             /* Calculate table index by multiplying r with table scale and truncate to integer */
 535             rt               = _mm256_mul_ps(r00,vftabscale);
 536             vfitab           = _mm256_cvttps_epi32(rt);
 537             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 538             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 539             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 540             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 541             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 542             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 543
 544             /* REACTION-FIELD ELECTROSTATICS */
 545             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 546             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 547
 548             /* CUBIC SPLINE TABLE DISPERSION */
 549             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 550                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 551             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 552                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 553             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 554                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 555             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 556                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 557             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 558             Heps             = _mm256_mul_ps(vfeps,H);
 559             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 560             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 561             vvdw6            = _mm256_mul_ps(c6_00,VV);
 562             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 563             fvdw6            = _mm256_mul_ps(c6_00,FF);
 564
 565             /* CUBIC SPLINE TABLE REPULSION */
 566             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 567             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 568             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 569                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 570             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 571                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 572             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 573                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 574             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 576             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 577             Heps             = _mm256_mul_ps(vfeps,H);
 578             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 579             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 580             vvdw12           = _mm256_mul_ps(c12_00,VV);
 581             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 582             fvdw12           = _mm256_mul_ps(c12_00,FF);
 583             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 584             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm256_andnot_ps(dummy_mask,velec);
 588             velecsum         = _mm256_add_ps(velecsum,velec);
 589             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 590             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 591
 592             fscal            = _mm256_add_ps(felec,fvdw);
 593
 594             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm256_mul_ps(fscal,dx00);
 598             ty               = _mm256_mul_ps(fscal,dy00);
 599             tz               = _mm256_mul_ps(fscal,dz00);
 600
 601             /* Update vectorial force */
 602             fix0             = _mm256_add_ps(fix0,tx);
 603             fiy0             = _mm256_add_ps(fiy0,ty);
 604             fiz0             = _mm256_add_ps(fiz0,tz);
 605
 606             fjx0             = _mm256_add_ps(fjx0,tx);
 607             fjy0             = _mm256_add_ps(fjy0,ty);
 608             fjz0             = _mm256_add_ps(fjz0,tz);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq10             = _mm256_mul_ps(iq1,jq0);
 616
 617             /* REACTION-FIELD ELECTROSTATICS */
 618             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 619             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm256_andnot_ps(dummy_mask,velec);
 623             velecsum         = _mm256_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 628
 629             /* Calculate temporary vectorial force */
 630             tx               = _mm256_mul_ps(fscal,dx10);
 631             ty               = _mm256_mul_ps(fscal,dy10);
 632             tz               = _mm256_mul_ps(fscal,dz10);
 633
 634             /* Update vectorial force */
 635             fix1             = _mm256_add_ps(fix1,tx);
 636             fiy1             = _mm256_add_ps(fiy1,ty);
 637             fiz1             = _mm256_add_ps(fiz1,tz);
 638
 639             fjx0             = _mm256_add_ps(fjx0,tx);
 640             fjy0             = _mm256_add_ps(fjy0,ty);
 641             fjz0             = _mm256_add_ps(fjz0,tz);
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             /* Compute parameters for interactions between i and j atoms */
 648             qq20             = _mm256_mul_ps(iq2,jq0);
 649
 650             /* REACTION-FIELD ELECTROSTATICS */
 651             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 652             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 653
 654             /* Update potential sum for this i atom from the interaction with this j atom. */
 655             velec            = _mm256_andnot_ps(dummy_mask,velec);
 656             velecsum         = _mm256_add_ps(velecsum,velec);
 657
 658             fscal            = felec;
 659
 660             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 661
 662             /* Calculate temporary vectorial force */
 663             tx               = _mm256_mul_ps(fscal,dx20);
 664             ty               = _mm256_mul_ps(fscal,dy20);
 665             tz               = _mm256_mul_ps(fscal,dz20);
 666
 667             /* Update vectorial force */
 668             fix2             = _mm256_add_ps(fix2,tx);
 669             fiy2             = _mm256_add_ps(fiy2,ty);
 670             fiz2             = _mm256_add_ps(fiz2,tz);
 671
 672             fjx0             = _mm256_add_ps(fjx0,tx);
 673             fjy0             = _mm256_add_ps(fjy0,ty);
 674             fjz0             = _mm256_add_ps(fjz0,tz);
 675
 676             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 677             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 678             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 679             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 680             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 681             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 682             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 683             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 684
 685             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 686
 687             /* Inner loop uses 135 flops */
 688         }
 689
 690         /* End of innermost loop */
 691
 692         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 693                                                  f+i_coord_offset,fshift+i_shift_offset);
 694
 695         ggid                        = gid[iidx];
 696         /* Update potential energies */
 697         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 698         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 699
 700         /* Increment number of inner iterations */
 701         inneriter                  += j_index_end - j_index_start;
 702
 703         /* Outer loop uses 20 flops */
 704     }
 705
 706     /* Increment number of outer iterations */
 707     outeriter        += nri;
 708
 709     /* Update outer/inner flops */
 710
 711     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*135);
 712 }
 713 /*
 714  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_single
 715  * Electrostatics interaction: ReactionField
 716  * VdW interaction:            CubicSplineTable
 717  * Geometry:                   Water3-Particle
 718  * Calculate force/pot:        Force
 719  */
 720 void
 721 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_single
 722                     (t_nblist                    * gmx_restrict       nlist,
 723                      rvec                        * gmx_restrict          xx,
 724                      rvec                        * gmx_restrict          ff,
 725                      t_forcerec                  * gmx_restrict          fr,
 726                      t_mdatoms                   * gmx_restrict     mdatoms,
 727                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 728                      t_nrnb                      * gmx_restrict        nrnb)
 729 {
 730     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 731      * just 0 for non-waters.
 732      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 733      * jnr indices corresponding to data put in the four positions in the SIMD register.
 734      */
 735     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 736     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 737     int              jnrA,jnrB,jnrC,jnrD;
 738     int              jnrE,jnrF,jnrG,jnrH;
 739     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 740     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 741     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 742     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 743     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 744     real             rcutoff_scalar;
 745     real             *shiftvec,*fshift,*x,*f;
 746     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 747     real             scratch[4*DIM];
 748     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 749     real *           vdwioffsetptr0;
 750     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 751     real *           vdwioffsetptr1;
 752     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 753     real *           vdwioffsetptr2;
 754     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 755     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 756     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 757     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 758     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 759     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 760     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 761     real             *charge;
 762     int              nvdwtype;
 763     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 764     int              *vdwtype;
 765     real             *vdwparam;
 766     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 767     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 768     __m256i          vfitab;
 769     __m128i          vfitab_lo,vfitab_hi;
 770     __m128i          ifour       = _mm_set1_epi32(4);
 771     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 772     real             *vftab;
 773     __m256           dummy_mask,cutoff_mask;
 774     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 775     __m256           one     = _mm256_set1_ps(1.0);
 776     __m256           two     = _mm256_set1_ps(2.0);
 777     x                = xx[0];
 778     f                = ff[0];
 779
 780     nri              = nlist->nri;
 781     iinr             = nlist->iinr;
 782     jindex           = nlist->jindex;
 783     jjnr             = nlist->jjnr;
 784     shiftidx         = nlist->shift;
 785     gid              = nlist->gid;
 786     shiftvec         = fr->shift_vec[0];
 787     fshift           = fr->fshift[0];
 788     facel            = _mm256_set1_ps(fr->epsfac);
 789     charge           = mdatoms->chargeA;
 790     krf              = _mm256_set1_ps(fr->ic->k_rf);
 791     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 792     crf              = _mm256_set1_ps(fr->ic->c_rf);
 793     nvdwtype         = fr->ntype;
 794     vdwparam         = fr->nbfp;
 795     vdwtype          = mdatoms->typeA;
 796
 797     vftab            = kernel_data->table_vdw->data;
 798     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 799
 800     /* Setup water-specific parameters */
 801     inr              = nlist->iinr[0];
 802     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 803     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 804     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 805     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 806
 807     /* Avoid stupid compiler warnings */
 808     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 809     j_coord_offsetA = 0;
 810     j_coord_offsetB = 0;
 811     j_coord_offsetC = 0;
 812     j_coord_offsetD = 0;
 813     j_coord_offsetE = 0;
 814     j_coord_offsetF = 0;
 815     j_coord_offsetG = 0;
 816     j_coord_offsetH = 0;
 817
 818     outeriter        = 0;
 819     inneriter        = 0;
 820
 821     for(iidx=0;iidx<4*DIM;iidx++)
 822     {
 823         scratch[iidx] = 0.0;
 824     }
 825
 826     /* Start outer loop over neighborlists */
 827     for(iidx=0; iidx<nri; iidx++)
 828     {
 829         /* Load shift vector for this list */
 830         i_shift_offset   = DIM*shiftidx[iidx];
 831
 832         /* Load limits for loop over neighbors */
 833         j_index_start    = jindex[iidx];
 834         j_index_end      = jindex[iidx+1];
 835
 836         /* Get outer coordinate index */
 837         inr              = iinr[iidx];
 838         i_coord_offset   = DIM*inr;
 839
 840         /* Load i particle coords and add shift vector */
 841         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 842                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 843
 844         fix0             = _mm256_setzero_ps();
 845         fiy0             = _mm256_setzero_ps();
 846         fiz0             = _mm256_setzero_ps();
 847         fix1             = _mm256_setzero_ps();
 848         fiy1             = _mm256_setzero_ps();
 849         fiz1             = _mm256_setzero_ps();
 850         fix2             = _mm256_setzero_ps();
 851         fiy2             = _mm256_setzero_ps();
 852         fiz2             = _mm256_setzero_ps();
 853
 854         /* Start inner kernel loop */
 855         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 856         {
 857
 858             /* Get j neighbor index, and coordinate index */
 859             jnrA             = jjnr[jidx];
 860             jnrB             = jjnr[jidx+1];
 861             jnrC             = jjnr[jidx+2];
 862             jnrD             = jjnr[jidx+3];
 863             jnrE             = jjnr[jidx+4];
 864             jnrF             = jjnr[jidx+5];
 865             jnrG             = jjnr[jidx+6];
 866             jnrH             = jjnr[jidx+7];
 867             j_coord_offsetA  = DIM*jnrA;
 868             j_coord_offsetB  = DIM*jnrB;
 869             j_coord_offsetC  = DIM*jnrC;
 870             j_coord_offsetD  = DIM*jnrD;
 871             j_coord_offsetE  = DIM*jnrE;
 872             j_coord_offsetF  = DIM*jnrF;
 873             j_coord_offsetG  = DIM*jnrG;
 874             j_coord_offsetH  = DIM*jnrH;
 875
 876             /* load j atom coordinates */
 877             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 878                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 879                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 880                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 881                                                  &jx0,&jy0,&jz0);
 882
 883             /* Calculate displacement vector */
 884             dx00             = _mm256_sub_ps(ix0,jx0);
 885             dy00             = _mm256_sub_ps(iy0,jy0);
 886             dz00             = _mm256_sub_ps(iz0,jz0);
 887             dx10             = _mm256_sub_ps(ix1,jx0);
 888             dy10             = _mm256_sub_ps(iy1,jy0);
 889             dz10             = _mm256_sub_ps(iz1,jz0);
 890             dx20             = _mm256_sub_ps(ix2,jx0);
 891             dy20             = _mm256_sub_ps(iy2,jy0);
 892             dz20             = _mm256_sub_ps(iz2,jz0);
 893
 894             /* Calculate squared distance and things based on it */
 895             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 896             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 897             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 898
 899             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 900             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 901             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 902
 903             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 904             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 905             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 906
 907             /* Load parameters for j particles */
 908             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 909                                                                  charge+jnrC+0,charge+jnrD+0,
 910                                                                  charge+jnrE+0,charge+jnrF+0,
 911                                                                  charge+jnrG+0,charge+jnrH+0);
 912             vdwjidx0A        = 2*vdwtype[jnrA+0];
 913             vdwjidx0B        = 2*vdwtype[jnrB+0];
 914             vdwjidx0C        = 2*vdwtype[jnrC+0];
 915             vdwjidx0D        = 2*vdwtype[jnrD+0];
 916             vdwjidx0E        = 2*vdwtype[jnrE+0];
 917             vdwjidx0F        = 2*vdwtype[jnrF+0];
 918             vdwjidx0G        = 2*vdwtype[jnrG+0];
 919             vdwjidx0H        = 2*vdwtype[jnrH+0];
 920
 921             fjx0             = _mm256_setzero_ps();
 922             fjy0             = _mm256_setzero_ps();
 923             fjz0             = _mm256_setzero_ps();
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             r00              = _mm256_mul_ps(rsq00,rinv00);
 930
 931             /* Compute parameters for interactions between i and j atoms */
 932             qq00             = _mm256_mul_ps(iq0,jq0);
 933             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 934                                             vdwioffsetptr0+vdwjidx0B,
 935                                             vdwioffsetptr0+vdwjidx0C,
 936                                             vdwioffsetptr0+vdwjidx0D,
 937                                             vdwioffsetptr0+vdwjidx0E,
 938                                             vdwioffsetptr0+vdwjidx0F,
 939                                             vdwioffsetptr0+vdwjidx0G,
 940                                             vdwioffsetptr0+vdwjidx0H,
 941                                             &c6_00,&c12_00);
 942
 943             /* Calculate table index by multiplying r with table scale and truncate to integer */
 944             rt               = _mm256_mul_ps(r00,vftabscale);
 945             vfitab           = _mm256_cvttps_epi32(rt);
 946             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 947             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 948             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 949             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 950             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 951             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 952
 953             /* REACTION-FIELD ELECTROSTATICS */
 954             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 955
 956             /* CUBIC SPLINE TABLE DISPERSION */
 957             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 958                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 959             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 960                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 961             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 962                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 963             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 964                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 965             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 966             Heps             = _mm256_mul_ps(vfeps,H);
 967             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 968             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 969             fvdw6            = _mm256_mul_ps(c6_00,FF);
 970
 971             /* CUBIC SPLINE TABLE REPULSION */
 972             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 973             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 974             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 975                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 976             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 977                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 978             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 979                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 980             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 981                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 982             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 983             Heps             = _mm256_mul_ps(vfeps,H);
 984             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 985             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 986             fvdw12           = _mm256_mul_ps(c12_00,FF);
 987             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 988
 989             fscal            = _mm256_add_ps(felec,fvdw);
 990
 991             /* Calculate temporary vectorial force */
 992             tx               = _mm256_mul_ps(fscal,dx00);
 993             ty               = _mm256_mul_ps(fscal,dy00);
 994             tz               = _mm256_mul_ps(fscal,dz00);
 995
 996             /* Update vectorial force */
 997             fix0             = _mm256_add_ps(fix0,tx);
 998             fiy0             = _mm256_add_ps(fiy0,ty);
 999             fiz0             = _mm256_add_ps(fiz0,tz);
1000
1001             fjx0             = _mm256_add_ps(fjx0,tx);
1002             fjy0             = _mm256_add_ps(fjy0,ty);
1003             fjz0             = _mm256_add_ps(fjz0,tz);
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             qq10             = _mm256_mul_ps(iq1,jq0);
1011
1012             /* REACTION-FIELD ELECTROSTATICS */
1013             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1014
1015             fscal            = felec;
1016
1017             /* Calculate temporary vectorial force */
1018             tx               = _mm256_mul_ps(fscal,dx10);
1019             ty               = _mm256_mul_ps(fscal,dy10);
1020             tz               = _mm256_mul_ps(fscal,dz10);
1021
1022             /* Update vectorial force */
1023             fix1             = _mm256_add_ps(fix1,tx);
1024             fiy1             = _mm256_add_ps(fiy1,ty);
1025             fiz1             = _mm256_add_ps(fiz1,tz);
1026
1027             fjx0             = _mm256_add_ps(fjx0,tx);
1028             fjy0             = _mm256_add_ps(fjy0,ty);
1029             fjz0             = _mm256_add_ps(fjz0,tz);
1030
1031             /**************************
1032              * CALCULATE INTERACTIONS *
1033              **************************/
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq20             = _mm256_mul_ps(iq2,jq0);
1037
1038             /* REACTION-FIELD ELECTROSTATICS */
1039             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1040
1041             fscal            = felec;
1042
1043             /* Calculate temporary vectorial force */
1044             tx               = _mm256_mul_ps(fscal,dx20);
1045             ty               = _mm256_mul_ps(fscal,dy20);
1046             tz               = _mm256_mul_ps(fscal,dz20);
1047
1048             /* Update vectorial force */
1049             fix2             = _mm256_add_ps(fix2,tx);
1050             fiy2             = _mm256_add_ps(fiy2,ty);
1051             fiz2             = _mm256_add_ps(fiz2,tz);
1052
1053             fjx0             = _mm256_add_ps(fjx0,tx);
1054             fjy0             = _mm256_add_ps(fjy0,ty);
1055             fjz0             = _mm256_add_ps(fjz0,tz);
1056
1057             fjptrA             = f+j_coord_offsetA;
1058             fjptrB             = f+j_coord_offsetB;
1059             fjptrC             = f+j_coord_offsetC;
1060             fjptrD             = f+j_coord_offsetD;
1061             fjptrE             = f+j_coord_offsetE;
1062             fjptrF             = f+j_coord_offsetF;
1063             fjptrG             = f+j_coord_offsetG;
1064             fjptrH             = f+j_coord_offsetH;
1065
1066             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1067
1068             /* Inner loop uses 111 flops */
1069         }
1070
1071         if(jidx<j_index_end)
1072         {
1073
1074             /* Get j neighbor index, and coordinate index */
1075             jnrlistA         = jjnr[jidx];
1076             jnrlistB         = jjnr[jidx+1];
1077             jnrlistC         = jjnr[jidx+2];
1078             jnrlistD         = jjnr[jidx+3];
1079             jnrlistE         = jjnr[jidx+4];
1080             jnrlistF         = jjnr[jidx+5];
1081             jnrlistG         = jjnr[jidx+6];
1082             jnrlistH         = jjnr[jidx+7];
1083             /* Sign of each element will be negative for non-real atoms.
1084              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1085              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1086              */
1087             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1088                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1089
1090             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1091             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1092             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1093             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1094             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1095             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1096             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1097             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1098             j_coord_offsetA  = DIM*jnrA;
1099             j_coord_offsetB  = DIM*jnrB;
1100             j_coord_offsetC  = DIM*jnrC;
1101             j_coord_offsetD  = DIM*jnrD;
1102             j_coord_offsetE  = DIM*jnrE;
1103             j_coord_offsetF  = DIM*jnrF;
1104             j_coord_offsetG  = DIM*jnrG;
1105             j_coord_offsetH  = DIM*jnrH;
1106
1107             /* load j atom coordinates */
1108             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1109                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1110                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1111                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1112                                                  &jx0,&jy0,&jz0);
1113
1114             /* Calculate displacement vector */
1115             dx00             = _mm256_sub_ps(ix0,jx0);
1116             dy00             = _mm256_sub_ps(iy0,jy0);
1117             dz00             = _mm256_sub_ps(iz0,jz0);
1118             dx10             = _mm256_sub_ps(ix1,jx0);
1119             dy10             = _mm256_sub_ps(iy1,jy0);
1120             dz10             = _mm256_sub_ps(iz1,jz0);
1121             dx20             = _mm256_sub_ps(ix2,jx0);
1122             dy20             = _mm256_sub_ps(iy2,jy0);
1123             dz20             = _mm256_sub_ps(iz2,jz0);
1124
1125             /* Calculate squared distance and things based on it */
1126             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1127             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1128             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1129
1130             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1131             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1132             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1133
1134             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1135             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1136             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1137
1138             /* Load parameters for j particles */
1139             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1140                                                                  charge+jnrC+0,charge+jnrD+0,
1141                                                                  charge+jnrE+0,charge+jnrF+0,
1142                                                                  charge+jnrG+0,charge+jnrH+0);
1143             vdwjidx0A        = 2*vdwtype[jnrA+0];
1144             vdwjidx0B        = 2*vdwtype[jnrB+0];
1145             vdwjidx0C        = 2*vdwtype[jnrC+0];
1146             vdwjidx0D        = 2*vdwtype[jnrD+0];
1147             vdwjidx0E        = 2*vdwtype[jnrE+0];
1148             vdwjidx0F        = 2*vdwtype[jnrF+0];
1149             vdwjidx0G        = 2*vdwtype[jnrG+0];
1150             vdwjidx0H        = 2*vdwtype[jnrH+0];
1151
1152             fjx0             = _mm256_setzero_ps();
1153             fjy0             = _mm256_setzero_ps();
1154             fjz0             = _mm256_setzero_ps();
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             r00              = _mm256_mul_ps(rsq00,rinv00);
1161             r00              = _mm256_andnot_ps(dummy_mask,r00);
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq00             = _mm256_mul_ps(iq0,jq0);
1165             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1166                                             vdwioffsetptr0+vdwjidx0B,
1167                                             vdwioffsetptr0+vdwjidx0C,
1168                                             vdwioffsetptr0+vdwjidx0D,
1169                                             vdwioffsetptr0+vdwjidx0E,
1170                                             vdwioffsetptr0+vdwjidx0F,
1171                                             vdwioffsetptr0+vdwjidx0G,
1172                                             vdwioffsetptr0+vdwjidx0H,
1173                                             &c6_00,&c12_00);
1174
1175             /* Calculate table index by multiplying r with table scale and truncate to integer */
1176             rt               = _mm256_mul_ps(r00,vftabscale);
1177             vfitab           = _mm256_cvttps_epi32(rt);
1178             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1179             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1180             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1181             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1182             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1183             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1184
1185             /* REACTION-FIELD ELECTROSTATICS */
1186             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1187
1188             /* CUBIC SPLINE TABLE DISPERSION */
1189             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1190                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1191             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1193             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1195             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1196                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1197             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1198             Heps             = _mm256_mul_ps(vfeps,H);
1199             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1200             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1201             fvdw6            = _mm256_mul_ps(c6_00,FF);
1202
1203             /* CUBIC SPLINE TABLE REPULSION */
1204             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1205             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1206             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1207                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1208             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1210             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1212             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1214             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1215             Heps             = _mm256_mul_ps(vfeps,H);
1216             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1217             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1218             fvdw12           = _mm256_mul_ps(c12_00,FF);
1219             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1220
1221             fscal            = _mm256_add_ps(felec,fvdw);
1222
1223             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1224
1225             /* Calculate temporary vectorial force */
1226             tx               = _mm256_mul_ps(fscal,dx00);
1227             ty               = _mm256_mul_ps(fscal,dy00);
1228             tz               = _mm256_mul_ps(fscal,dz00);
1229
1230             /* Update vectorial force */
1231             fix0             = _mm256_add_ps(fix0,tx);
1232             fiy0             = _mm256_add_ps(fiy0,ty);
1233             fiz0             = _mm256_add_ps(fiz0,tz);
1234
1235             fjx0             = _mm256_add_ps(fjx0,tx);
1236             fjy0             = _mm256_add_ps(fjy0,ty);
1237             fjz0             = _mm256_add_ps(fjz0,tz);
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             /* Compute parameters for interactions between i and j atoms */
1244             qq10             = _mm256_mul_ps(iq1,jq0);
1245
1246             /* REACTION-FIELD ELECTROSTATICS */
1247             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1248
1249             fscal            = felec;
1250
1251             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1252
1253             /* Calculate temporary vectorial force */
1254             tx               = _mm256_mul_ps(fscal,dx10);
1255             ty               = _mm256_mul_ps(fscal,dy10);
1256             tz               = _mm256_mul_ps(fscal,dz10);
1257
1258             /* Update vectorial force */
1259             fix1             = _mm256_add_ps(fix1,tx);
1260             fiy1             = _mm256_add_ps(fiy1,ty);
1261             fiz1             = _mm256_add_ps(fiz1,tz);
1262
1263             fjx0             = _mm256_add_ps(fjx0,tx);
1264             fjy0             = _mm256_add_ps(fjy0,ty);
1265             fjz0             = _mm256_add_ps(fjz0,tz);
1266
1267             /**************************
1268              * CALCULATE INTERACTIONS *
1269              **************************/
1270
1271             /* Compute parameters for interactions between i and j atoms */
1272             qq20             = _mm256_mul_ps(iq2,jq0);
1273
1274             /* REACTION-FIELD ELECTROSTATICS */
1275             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
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 112 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*112);
1326 }