src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 145     rcutoff_scalar   = fr->rcoulomb;
 146     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 147     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153     j_coord_offsetC = 0;
 154     j_coord_offsetD = 0;
 155     j_coord_offsetE = 0;
 156     j_coord_offsetF = 0;
 157     j_coord_offsetG = 0;
 158     j_coord_offsetH = 0;
 159
 160     outeriter        = 0;
 161     inneriter        = 0;
 162
 163     for(iidx=0;iidx<4*DIM;iidx++)
 164     {
 165         scratch[iidx] = 0.0;
 166     }
 167
 168     /* Start outer loop over neighborlists */
 169     for(iidx=0; iidx<nri; iidx++)
 170     {
 171         /* Load shift vector for this list */
 172         i_shift_offset   = DIM*shiftidx[iidx];
 173
 174         /* Load limits for loop over neighbors */
 175         j_index_start    = jindex[iidx];
 176         j_index_end      = jindex[iidx+1];
 177
 178         /* Get outer coordinate index */
 179         inr              = iinr[iidx];
 180         i_coord_offset   = DIM*inr;
 181
 182         /* Load i particle coords and add shift vector */
 183         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 184                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 185
 186         fix0             = _mm256_setzero_ps();
 187         fiy0             = _mm256_setzero_ps();
 188         fiz0             = _mm256_setzero_ps();
 189         fix1             = _mm256_setzero_ps();
 190         fiy1             = _mm256_setzero_ps();
 191         fiz1             = _mm256_setzero_ps();
 192         fix2             = _mm256_setzero_ps();
 193         fiy2             = _mm256_setzero_ps();
 194         fiz2             = _mm256_setzero_ps();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm256_setzero_ps();
 198         vvdwsum          = _mm256_setzero_ps();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             jnrE             = jjnr[jidx+4];
 210             jnrF             = jjnr[jidx+5];
 211             jnrG             = jjnr[jidx+6];
 212             jnrH             = jjnr[jidx+7];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217             j_coord_offsetE  = DIM*jnrE;
 218             j_coord_offsetF  = DIM*jnrF;
 219             j_coord_offsetG  = DIM*jnrG;
 220             j_coord_offsetH  = DIM*jnrH;
 221
 222             /* load j atom coordinates */
 223             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 224                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 225                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 226                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 227                                                  &jx0,&jy0,&jz0);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm256_sub_ps(ix0,jx0);
 231             dy00             = _mm256_sub_ps(iy0,jy0);
 232             dz00             = _mm256_sub_ps(iz0,jz0);
 233             dx10             = _mm256_sub_ps(ix1,jx0);
 234             dy10             = _mm256_sub_ps(iy1,jy0);
 235             dz10             = _mm256_sub_ps(iz1,jz0);
 236             dx20             = _mm256_sub_ps(ix2,jx0);
 237             dy20             = _mm256_sub_ps(iy2,jy0);
 238             dz20             = _mm256_sub_ps(iz2,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 243             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 244
 245             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 246             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 247             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 248
 249             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 250             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 251             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 252
 253             /* Load parameters for j particles */
 254             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 255                                                                  charge+jnrC+0,charge+jnrD+0,
 256                                                                  charge+jnrE+0,charge+jnrF+0,
 257                                                                  charge+jnrG+0,charge+jnrH+0);
 258             vdwjidx0A        = 2*vdwtype[jnrA+0];
 259             vdwjidx0B        = 2*vdwtype[jnrB+0];
 260             vdwjidx0C        = 2*vdwtype[jnrC+0];
 261             vdwjidx0D        = 2*vdwtype[jnrD+0];
 262             vdwjidx0E        = 2*vdwtype[jnrE+0];
 263             vdwjidx0F        = 2*vdwtype[jnrF+0];
 264             vdwjidx0G        = 2*vdwtype[jnrG+0];
 265             vdwjidx0H        = 2*vdwtype[jnrH+0];
 266
 267             fjx0             = _mm256_setzero_ps();
 268             fjy0             = _mm256_setzero_ps();
 269             fjz0             = _mm256_setzero_ps();
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 276             {
 277
 278             r00              = _mm256_mul_ps(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq00             = _mm256_mul_ps(iq0,jq0);
 282             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 283                                             vdwioffsetptr0+vdwjidx0B,
 284                                             vdwioffsetptr0+vdwjidx0C,
 285                                             vdwioffsetptr0+vdwjidx0D,
 286                                             vdwioffsetptr0+vdwjidx0E,
 287                                             vdwioffsetptr0+vdwjidx0F,
 288                                             vdwioffsetptr0+vdwjidx0G,
 289                                             vdwioffsetptr0+vdwjidx0H,
 290                                             &c6_00,&c12_00);
 291
 292             /* Calculate table index by multiplying r with table scale and truncate to integer */
 293             rt               = _mm256_mul_ps(r00,vftabscale);
 294             vfitab           = _mm256_cvttps_epi32(rt);
 295             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 296             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 297             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 298             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 299             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 300             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 301
 302             /* REACTION-FIELD ELECTROSTATICS */
 303             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 304             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 305
 306             /* CUBIC SPLINE TABLE DISPERSION */
 307             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 309             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 311             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 313             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 314                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 315             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 316             Heps             = _mm256_mul_ps(vfeps,H);
 317             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 318             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 319             vvdw6            = _mm256_mul_ps(c6_00,VV);
 320             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 321             fvdw6            = _mm256_mul_ps(c6_00,FF);
 322
 323             /* CUBIC SPLINE TABLE REPULSION */
 324             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 325             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 326             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 328             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 330             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 332             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 334             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 335             Heps             = _mm256_mul_ps(vfeps,H);
 336             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 337             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 338             vvdw12           = _mm256_mul_ps(c12_00,VV);
 339             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 340             fvdw12           = _mm256_mul_ps(c12_00,FF);
 341             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 342             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 343
 344             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _mm256_and_ps(velec,cutoff_mask);
 348             velecsum         = _mm256_add_ps(velecsum,velec);
 349             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 350             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 351
 352             fscal            = _mm256_add_ps(felec,fvdw);
 353
 354             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm256_mul_ps(fscal,dx00);
 358             ty               = _mm256_mul_ps(fscal,dy00);
 359             tz               = _mm256_mul_ps(fscal,dz00);
 360
 361             /* Update vectorial force */
 362             fix0             = _mm256_add_ps(fix0,tx);
 363             fiy0             = _mm256_add_ps(fiy0,ty);
 364             fiz0             = _mm256_add_ps(fiz0,tz);
 365
 366             fjx0             = _mm256_add_ps(fjx0,tx);
 367             fjy0             = _mm256_add_ps(fjy0,ty);
 368             fjz0             = _mm256_add_ps(fjz0,tz);
 369
 370             }
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 377             {
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq10             = _mm256_mul_ps(iq1,jq0);
 381
 382             /* REACTION-FIELD ELECTROSTATICS */
 383             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 384             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 385
 386             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velec            = _mm256_and_ps(velec,cutoff_mask);
 390             velecsum         = _mm256_add_ps(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm256_mul_ps(fscal,dx10);
 398             ty               = _mm256_mul_ps(fscal,dy10);
 399             tz               = _mm256_mul_ps(fscal,dz10);
 400
 401             /* Update vectorial force */
 402             fix1             = _mm256_add_ps(fix1,tx);
 403             fiy1             = _mm256_add_ps(fiy1,ty);
 404             fiz1             = _mm256_add_ps(fiz1,tz);
 405
 406             fjx0             = _mm256_add_ps(fjx0,tx);
 407             fjy0             = _mm256_add_ps(fjy0,ty);
 408             fjz0             = _mm256_add_ps(fjz0,tz);
 409
 410             }
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 417             {
 418
 419             /* Compute parameters for interactions between i and j atoms */
 420             qq20             = _mm256_mul_ps(iq2,jq0);
 421
 422             /* REACTION-FIELD ELECTROSTATICS */
 423             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 424             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 425
 426             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velec            = _mm256_and_ps(velec,cutoff_mask);
 430             velecsum         = _mm256_add_ps(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm256_mul_ps(fscal,dx20);
 438             ty               = _mm256_mul_ps(fscal,dy20);
 439             tz               = _mm256_mul_ps(fscal,dz20);
 440
 441             /* Update vectorial force */
 442             fix2             = _mm256_add_ps(fix2,tx);
 443             fiy2             = _mm256_add_ps(fiy2,ty);
 444             fiz2             = _mm256_add_ps(fiz2,tz);
 445
 446             fjx0             = _mm256_add_ps(fjx0,tx);
 447             fjy0             = _mm256_add_ps(fjy0,ty);
 448             fjz0             = _mm256_add_ps(fjz0,tz);
 449
 450             }
 451
 452             fjptrA             = f+j_coord_offsetA;
 453             fjptrB             = f+j_coord_offsetB;
 454             fjptrC             = f+j_coord_offsetC;
 455             fjptrD             = f+j_coord_offsetD;
 456             fjptrE             = f+j_coord_offsetE;
 457             fjptrF             = f+j_coord_offsetF;
 458             fjptrG             = f+j_coord_offsetG;
 459             fjptrH             = f+j_coord_offsetH;
 460
 461             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 462
 463             /* Inner loop uses 147 flops */
 464         }
 465
 466         if(jidx<j_index_end)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrlistA         = jjnr[jidx];
 471             jnrlistB         = jjnr[jidx+1];
 472             jnrlistC         = jjnr[jidx+2];
 473             jnrlistD         = jjnr[jidx+3];
 474             jnrlistE         = jjnr[jidx+4];
 475             jnrlistF         = jjnr[jidx+5];
 476             jnrlistG         = jjnr[jidx+6];
 477             jnrlistH         = jjnr[jidx+7];
 478             /* Sign of each element will be negative for non-real atoms.
 479              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 480              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 481              */
 482             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 483                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 484
 485             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 486             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 487             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 488             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 489             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 490             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 491             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 492             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 493             j_coord_offsetA  = DIM*jnrA;
 494             j_coord_offsetB  = DIM*jnrB;
 495             j_coord_offsetC  = DIM*jnrC;
 496             j_coord_offsetD  = DIM*jnrD;
 497             j_coord_offsetE  = DIM*jnrE;
 498             j_coord_offsetF  = DIM*jnrF;
 499             j_coord_offsetG  = DIM*jnrG;
 500             j_coord_offsetH  = DIM*jnrH;
 501
 502             /* load j atom coordinates */
 503             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 504                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 505                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 506                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 507                                                  &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm256_sub_ps(ix0,jx0);
 511             dy00             = _mm256_sub_ps(iy0,jy0);
 512             dz00             = _mm256_sub_ps(iz0,jz0);
 513             dx10             = _mm256_sub_ps(ix1,jx0);
 514             dy10             = _mm256_sub_ps(iy1,jy0);
 515             dz10             = _mm256_sub_ps(iz1,jz0);
 516             dx20             = _mm256_sub_ps(ix2,jx0);
 517             dy20             = _mm256_sub_ps(iy2,jy0);
 518             dz20             = _mm256_sub_ps(iz2,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 522             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 523             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 524
 525             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 526             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 527             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 528
 529             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 530             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 531             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 532
 533             /* Load parameters for j particles */
 534             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 535                                                                  charge+jnrC+0,charge+jnrD+0,
 536                                                                  charge+jnrE+0,charge+jnrF+0,
 537                                                                  charge+jnrG+0,charge+jnrH+0);
 538             vdwjidx0A        = 2*vdwtype[jnrA+0];
 539             vdwjidx0B        = 2*vdwtype[jnrB+0];
 540             vdwjidx0C        = 2*vdwtype[jnrC+0];
 541             vdwjidx0D        = 2*vdwtype[jnrD+0];
 542             vdwjidx0E        = 2*vdwtype[jnrE+0];
 543             vdwjidx0F        = 2*vdwtype[jnrF+0];
 544             vdwjidx0G        = 2*vdwtype[jnrG+0];
 545             vdwjidx0H        = 2*vdwtype[jnrH+0];
 546
 547             fjx0             = _mm256_setzero_ps();
 548             fjy0             = _mm256_setzero_ps();
 549             fjz0             = _mm256_setzero_ps();
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 556             {
 557
 558             r00              = _mm256_mul_ps(rsq00,rinv00);
 559             r00              = _mm256_andnot_ps(dummy_mask,r00);
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq00             = _mm256_mul_ps(iq0,jq0);
 563             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 564                                             vdwioffsetptr0+vdwjidx0B,
 565                                             vdwioffsetptr0+vdwjidx0C,
 566                                             vdwioffsetptr0+vdwjidx0D,
 567                                             vdwioffsetptr0+vdwjidx0E,
 568                                             vdwioffsetptr0+vdwjidx0F,
 569                                             vdwioffsetptr0+vdwjidx0G,
 570                                             vdwioffsetptr0+vdwjidx0H,
 571                                             &c6_00,&c12_00);
 572
 573             /* Calculate table index by multiplying r with table scale and truncate to integer */
 574             rt               = _mm256_mul_ps(r00,vftabscale);
 575             vfitab           = _mm256_cvttps_epi32(rt);
 576             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 577             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 578             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 579             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 580             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 581             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 582
 583             /* REACTION-FIELD ELECTROSTATICS */
 584             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 585             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 586
 587             /* CUBIC SPLINE TABLE DISPERSION */
 588             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 589                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 590             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 591                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 592             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 593                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 594             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 595                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 596             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 597             Heps             = _mm256_mul_ps(vfeps,H);
 598             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 599             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 600             vvdw6            = _mm256_mul_ps(c6_00,VV);
 601             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 602             fvdw6            = _mm256_mul_ps(c6_00,FF);
 603
 604             /* CUBIC SPLINE TABLE REPULSION */
 605             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 606             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 607             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 608                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 609             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 610                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 611             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 612                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 613             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 614                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 615             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 616             Heps             = _mm256_mul_ps(vfeps,H);
 617             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 618             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 619             vvdw12           = _mm256_mul_ps(c12_00,VV);
 620             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 621             fvdw12           = _mm256_mul_ps(c12_00,FF);
 622             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 623             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 624
 625             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm256_and_ps(velec,cutoff_mask);
 629             velec            = _mm256_andnot_ps(dummy_mask,velec);
 630             velecsum         = _mm256_add_ps(velecsum,velec);
 631             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 632             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 633             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 634
 635             fscal            = _mm256_add_ps(felec,fvdw);
 636
 637             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 638
 639             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm256_mul_ps(fscal,dx00);
 643             ty               = _mm256_mul_ps(fscal,dy00);
 644             tz               = _mm256_mul_ps(fscal,dz00);
 645
 646             /* Update vectorial force */
 647             fix0             = _mm256_add_ps(fix0,tx);
 648             fiy0             = _mm256_add_ps(fiy0,ty);
 649             fiz0             = _mm256_add_ps(fiz0,tz);
 650
 651             fjx0             = _mm256_add_ps(fjx0,tx);
 652             fjy0             = _mm256_add_ps(fjy0,ty);
 653             fjz0             = _mm256_add_ps(fjz0,tz);
 654
 655             }
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 662             {
 663
 664             /* Compute parameters for interactions between i and j atoms */
 665             qq10             = _mm256_mul_ps(iq1,jq0);
 666
 667             /* REACTION-FIELD ELECTROSTATICS */
 668             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 669             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 670
 671             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm256_and_ps(velec,cutoff_mask);
 675             velec            = _mm256_andnot_ps(dummy_mask,velec);
 676             velecsum         = _mm256_add_ps(velecsum,velec);
 677
 678             fscal            = felec;
 679
 680             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 681
 682             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 683
 684             /* Calculate temporary vectorial force */
 685             tx               = _mm256_mul_ps(fscal,dx10);
 686             ty               = _mm256_mul_ps(fscal,dy10);
 687             tz               = _mm256_mul_ps(fscal,dz10);
 688
 689             /* Update vectorial force */
 690             fix1             = _mm256_add_ps(fix1,tx);
 691             fiy1             = _mm256_add_ps(fiy1,ty);
 692             fiz1             = _mm256_add_ps(fiz1,tz);
 693
 694             fjx0             = _mm256_add_ps(fjx0,tx);
 695             fjy0             = _mm256_add_ps(fjy0,ty);
 696             fjz0             = _mm256_add_ps(fjz0,tz);
 697
 698             }
 699
 700             /**************************
 701              * CALCULATE INTERACTIONS *
 702              **************************/
 703
 704             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 705             {
 706
 707             /* Compute parameters for interactions between i and j atoms */
 708             qq20             = _mm256_mul_ps(iq2,jq0);
 709
 710             /* REACTION-FIELD ELECTROSTATICS */
 711             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 712             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 713
 714             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 715
 716             /* Update potential sum for this i atom from the interaction with this j atom. */
 717             velec            = _mm256_and_ps(velec,cutoff_mask);
 718             velec            = _mm256_andnot_ps(dummy_mask,velec);
 719             velecsum         = _mm256_add_ps(velecsum,velec);
 720
 721             fscal            = felec;
 722
 723             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 724
 725             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 726
 727             /* Calculate temporary vectorial force */
 728             tx               = _mm256_mul_ps(fscal,dx20);
 729             ty               = _mm256_mul_ps(fscal,dy20);
 730             tz               = _mm256_mul_ps(fscal,dz20);
 731
 732             /* Update vectorial force */
 733             fix2             = _mm256_add_ps(fix2,tx);
 734             fiy2             = _mm256_add_ps(fiy2,ty);
 735             fiz2             = _mm256_add_ps(fiz2,tz);
 736
 737             fjx0             = _mm256_add_ps(fjx0,tx);
 738             fjy0             = _mm256_add_ps(fjy0,ty);
 739             fjz0             = _mm256_add_ps(fjz0,tz);
 740
 741             }
 742
 743             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 744             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 745             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 746             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 747             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 748             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 749             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 750             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 751
 752             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 753
 754             /* Inner loop uses 148 flops */
 755         }
 756
 757         /* End of innermost loop */
 758
 759         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 760                                                  f+i_coord_offset,fshift+i_shift_offset);
 761
 762         ggid                        = gid[iidx];
 763         /* Update potential energies */
 764         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 765         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 766
 767         /* Increment number of inner iterations */
 768         inneriter                  += j_index_end - j_index_start;
 769
 770         /* Outer loop uses 20 flops */
 771     }
 772
 773     /* Increment number of outer iterations */
 774     outeriter        += nri;
 775
 776     /* Update outer/inner flops */
 777
 778     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 779 }
 780 /*
 781  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_single
 782  * Electrostatics interaction: ReactionField
 783  * VdW interaction:            CubicSplineTable
 784  * Geometry:                   Water3-Particle
 785  * Calculate force/pot:        Force
 786  */
 787 void
 788 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_single
 789                     (t_nblist                    * gmx_restrict       nlist,
 790                      rvec                        * gmx_restrict          xx,
 791                      rvec                        * gmx_restrict          ff,
 792                      t_forcerec                  * gmx_restrict          fr,
 793                      t_mdatoms                   * gmx_restrict     mdatoms,
 794                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 795                      t_nrnb                      * gmx_restrict        nrnb)
 796 {
 797     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 798      * just 0 for non-waters.
 799      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 800      * jnr indices corresponding to data put in the four positions in the SIMD register.
 801      */
 802     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 803     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 804     int              jnrA,jnrB,jnrC,jnrD;
 805     int              jnrE,jnrF,jnrG,jnrH;
 806     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 807     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 808     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 809     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 810     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 811     real             rcutoff_scalar;
 812     real             *shiftvec,*fshift,*x,*f;
 813     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 814     real             scratch[4*DIM];
 815     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 816     real *           vdwioffsetptr0;
 817     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 818     real *           vdwioffsetptr1;
 819     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 820     real *           vdwioffsetptr2;
 821     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 822     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 823     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 824     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 825     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 826     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 827     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 828     real             *charge;
 829     int              nvdwtype;
 830     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 831     int              *vdwtype;
 832     real             *vdwparam;
 833     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 834     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 835     __m256i          vfitab;
 836     __m128i          vfitab_lo,vfitab_hi;
 837     __m128i          ifour       = _mm_set1_epi32(4);
 838     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 839     real             *vftab;
 840     __m256           dummy_mask,cutoff_mask;
 841     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 842     __m256           one     = _mm256_set1_ps(1.0);
 843     __m256           two     = _mm256_set1_ps(2.0);
 844     x                = xx[0];
 845     f                = ff[0];
 846
 847     nri              = nlist->nri;
 848     iinr             = nlist->iinr;
 849     jindex           = nlist->jindex;
 850     jjnr             = nlist->jjnr;
 851     shiftidx         = nlist->shift;
 852     gid              = nlist->gid;
 853     shiftvec         = fr->shift_vec[0];
 854     fshift           = fr->fshift[0];
 855     facel            = _mm256_set1_ps(fr->epsfac);
 856     charge           = mdatoms->chargeA;
 857     krf              = _mm256_set1_ps(fr->ic->k_rf);
 858     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 859     crf              = _mm256_set1_ps(fr->ic->c_rf);
 860     nvdwtype         = fr->ntype;
 861     vdwparam         = fr->nbfp;
 862     vdwtype          = mdatoms->typeA;
 863
 864     vftab            = kernel_data->table_vdw->data;
 865     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 866
 867     /* Setup water-specific parameters */
 868     inr              = nlist->iinr[0];
 869     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 870     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 871     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 872     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 873
 874     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 875     rcutoff_scalar   = fr->rcoulomb;
 876     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 877     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 878
 879     /* Avoid stupid compiler warnings */
 880     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 881     j_coord_offsetA = 0;
 882     j_coord_offsetB = 0;
 883     j_coord_offsetC = 0;
 884     j_coord_offsetD = 0;
 885     j_coord_offsetE = 0;
 886     j_coord_offsetF = 0;
 887     j_coord_offsetG = 0;
 888     j_coord_offsetH = 0;
 889
 890     outeriter        = 0;
 891     inneriter        = 0;
 892
 893     for(iidx=0;iidx<4*DIM;iidx++)
 894     {
 895         scratch[iidx] = 0.0;
 896     }
 897
 898     /* Start outer loop over neighborlists */
 899     for(iidx=0; iidx<nri; iidx++)
 900     {
 901         /* Load shift vector for this list */
 902         i_shift_offset   = DIM*shiftidx[iidx];
 903
 904         /* Load limits for loop over neighbors */
 905         j_index_start    = jindex[iidx];
 906         j_index_end      = jindex[iidx+1];
 907
 908         /* Get outer coordinate index */
 909         inr              = iinr[iidx];
 910         i_coord_offset   = DIM*inr;
 911
 912         /* Load i particle coords and add shift vector */
 913         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 914                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 915
 916         fix0             = _mm256_setzero_ps();
 917         fiy0             = _mm256_setzero_ps();
 918         fiz0             = _mm256_setzero_ps();
 919         fix1             = _mm256_setzero_ps();
 920         fiy1             = _mm256_setzero_ps();
 921         fiz1             = _mm256_setzero_ps();
 922         fix2             = _mm256_setzero_ps();
 923         fiy2             = _mm256_setzero_ps();
 924         fiz2             = _mm256_setzero_ps();
 925
 926         /* Start inner kernel loop */
 927         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 928         {
 929
 930             /* Get j neighbor index, and coordinate index */
 931             jnrA             = jjnr[jidx];
 932             jnrB             = jjnr[jidx+1];
 933             jnrC             = jjnr[jidx+2];
 934             jnrD             = jjnr[jidx+3];
 935             jnrE             = jjnr[jidx+4];
 936             jnrF             = jjnr[jidx+5];
 937             jnrG             = jjnr[jidx+6];
 938             jnrH             = jjnr[jidx+7];
 939             j_coord_offsetA  = DIM*jnrA;
 940             j_coord_offsetB  = DIM*jnrB;
 941             j_coord_offsetC  = DIM*jnrC;
 942             j_coord_offsetD  = DIM*jnrD;
 943             j_coord_offsetE  = DIM*jnrE;
 944             j_coord_offsetF  = DIM*jnrF;
 945             j_coord_offsetG  = DIM*jnrG;
 946             j_coord_offsetH  = DIM*jnrH;
 947
 948             /* load j atom coordinates */
 949             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 950                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 951                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 952                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 953                                                  &jx0,&jy0,&jz0);
 954
 955             /* Calculate displacement vector */
 956             dx00             = _mm256_sub_ps(ix0,jx0);
 957             dy00             = _mm256_sub_ps(iy0,jy0);
 958             dz00             = _mm256_sub_ps(iz0,jz0);
 959             dx10             = _mm256_sub_ps(ix1,jx0);
 960             dy10             = _mm256_sub_ps(iy1,jy0);
 961             dz10             = _mm256_sub_ps(iz1,jz0);
 962             dx20             = _mm256_sub_ps(ix2,jx0);
 963             dy20             = _mm256_sub_ps(iy2,jy0);
 964             dz20             = _mm256_sub_ps(iz2,jz0);
 965
 966             /* Calculate squared distance and things based on it */
 967             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 968             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 969             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 970
 971             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 972             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 973             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 974
 975             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 976             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 977             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 978
 979             /* Load parameters for j particles */
 980             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 981                                                                  charge+jnrC+0,charge+jnrD+0,
 982                                                                  charge+jnrE+0,charge+jnrF+0,
 983                                                                  charge+jnrG+0,charge+jnrH+0);
 984             vdwjidx0A        = 2*vdwtype[jnrA+0];
 985             vdwjidx0B        = 2*vdwtype[jnrB+0];
 986             vdwjidx0C        = 2*vdwtype[jnrC+0];
 987             vdwjidx0D        = 2*vdwtype[jnrD+0];
 988             vdwjidx0E        = 2*vdwtype[jnrE+0];
 989             vdwjidx0F        = 2*vdwtype[jnrF+0];
 990             vdwjidx0G        = 2*vdwtype[jnrG+0];
 991             vdwjidx0H        = 2*vdwtype[jnrH+0];
 992
 993             fjx0             = _mm256_setzero_ps();
 994             fjy0             = _mm256_setzero_ps();
 995             fjz0             = _mm256_setzero_ps();
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1002             {
1003
1004             r00              = _mm256_mul_ps(rsq00,rinv00);
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq00             = _mm256_mul_ps(iq0,jq0);
1008             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1009                                             vdwioffsetptr0+vdwjidx0B,
1010                                             vdwioffsetptr0+vdwjidx0C,
1011                                             vdwioffsetptr0+vdwjidx0D,
1012                                             vdwioffsetptr0+vdwjidx0E,
1013                                             vdwioffsetptr0+vdwjidx0F,
1014                                             vdwioffsetptr0+vdwjidx0G,
1015                                             vdwioffsetptr0+vdwjidx0H,
1016                                             &c6_00,&c12_00);
1017
1018             /* Calculate table index by multiplying r with table scale and truncate to integer */
1019             rt               = _mm256_mul_ps(r00,vftabscale);
1020             vfitab           = _mm256_cvttps_epi32(rt);
1021             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1022             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1023             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1024             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1025             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1026             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1027
1028             /* REACTION-FIELD ELECTROSTATICS */
1029             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1030
1031             /* CUBIC SPLINE TABLE DISPERSION */
1032             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1033                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1034             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1035                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1036             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1037                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1038             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1039                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1040             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1041             Heps             = _mm256_mul_ps(vfeps,H);
1042             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1043             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1044             fvdw6            = _mm256_mul_ps(c6_00,FF);
1045
1046             /* CUBIC SPLINE TABLE REPULSION */
1047             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1048             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1049             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1050                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1051             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1052                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1053             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1054                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1055             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1056                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1057             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1058             Heps             = _mm256_mul_ps(vfeps,H);
1059             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1060             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1061             fvdw12           = _mm256_mul_ps(c12_00,FF);
1062             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1063
1064             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1065
1066             fscal            = _mm256_add_ps(felec,fvdw);
1067
1068             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1069
1070             /* Calculate temporary vectorial force */
1071             tx               = _mm256_mul_ps(fscal,dx00);
1072             ty               = _mm256_mul_ps(fscal,dy00);
1073             tz               = _mm256_mul_ps(fscal,dz00);
1074
1075             /* Update vectorial force */
1076             fix0             = _mm256_add_ps(fix0,tx);
1077             fiy0             = _mm256_add_ps(fiy0,ty);
1078             fiz0             = _mm256_add_ps(fiz0,tz);
1079
1080             fjx0             = _mm256_add_ps(fjx0,tx);
1081             fjy0             = _mm256_add_ps(fjy0,ty);
1082             fjz0             = _mm256_add_ps(fjz0,tz);
1083
1084             }
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1091             {
1092
1093             /* Compute parameters for interactions between i and j atoms */
1094             qq10             = _mm256_mul_ps(iq1,jq0);
1095
1096             /* REACTION-FIELD ELECTROSTATICS */
1097             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1098
1099             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1100
1101             fscal            = felec;
1102
1103             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1104
1105             /* Calculate temporary vectorial force */
1106             tx               = _mm256_mul_ps(fscal,dx10);
1107             ty               = _mm256_mul_ps(fscal,dy10);
1108             tz               = _mm256_mul_ps(fscal,dz10);
1109
1110             /* Update vectorial force */
1111             fix1             = _mm256_add_ps(fix1,tx);
1112             fiy1             = _mm256_add_ps(fiy1,ty);
1113             fiz1             = _mm256_add_ps(fiz1,tz);
1114
1115             fjx0             = _mm256_add_ps(fjx0,tx);
1116             fjy0             = _mm256_add_ps(fjy0,ty);
1117             fjz0             = _mm256_add_ps(fjz0,tz);
1118
1119             }
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1126             {
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             qq20             = _mm256_mul_ps(iq2,jq0);
1130
1131             /* REACTION-FIELD ELECTROSTATICS */
1132             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1133
1134             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1135
1136             fscal            = felec;
1137
1138             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm256_mul_ps(fscal,dx20);
1142             ty               = _mm256_mul_ps(fscal,dy20);
1143             tz               = _mm256_mul_ps(fscal,dz20);
1144
1145             /* Update vectorial force */
1146             fix2             = _mm256_add_ps(fix2,tx);
1147             fiy2             = _mm256_add_ps(fiy2,ty);
1148             fiz2             = _mm256_add_ps(fiz2,tz);
1149
1150             fjx0             = _mm256_add_ps(fjx0,tx);
1151             fjy0             = _mm256_add_ps(fjy0,ty);
1152             fjz0             = _mm256_add_ps(fjz0,tz);
1153
1154             }
1155
1156             fjptrA             = f+j_coord_offsetA;
1157             fjptrB             = f+j_coord_offsetB;
1158             fjptrC             = f+j_coord_offsetC;
1159             fjptrD             = f+j_coord_offsetD;
1160             fjptrE             = f+j_coord_offsetE;
1161             fjptrF             = f+j_coord_offsetF;
1162             fjptrG             = f+j_coord_offsetG;
1163             fjptrH             = f+j_coord_offsetH;
1164
1165             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1166
1167             /* Inner loop uses 120 flops */
1168         }
1169
1170         if(jidx<j_index_end)
1171         {
1172
1173             /* Get j neighbor index, and coordinate index */
1174             jnrlistA         = jjnr[jidx];
1175             jnrlistB         = jjnr[jidx+1];
1176             jnrlistC         = jjnr[jidx+2];
1177             jnrlistD         = jjnr[jidx+3];
1178             jnrlistE         = jjnr[jidx+4];
1179             jnrlistF         = jjnr[jidx+5];
1180             jnrlistG         = jjnr[jidx+6];
1181             jnrlistH         = jjnr[jidx+7];
1182             /* Sign of each element will be negative for non-real atoms.
1183              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1184              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1185              */
1186             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1187                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1188
1189             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1190             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1191             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1192             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1193             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1194             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1195             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1196             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1197             j_coord_offsetA  = DIM*jnrA;
1198             j_coord_offsetB  = DIM*jnrB;
1199             j_coord_offsetC  = DIM*jnrC;
1200             j_coord_offsetD  = DIM*jnrD;
1201             j_coord_offsetE  = DIM*jnrE;
1202             j_coord_offsetF  = DIM*jnrF;
1203             j_coord_offsetG  = DIM*jnrG;
1204             j_coord_offsetH  = DIM*jnrH;
1205
1206             /* load j atom coordinates */
1207             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1209                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1210                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1211                                                  &jx0,&jy0,&jz0);
1212
1213             /* Calculate displacement vector */
1214             dx00             = _mm256_sub_ps(ix0,jx0);
1215             dy00             = _mm256_sub_ps(iy0,jy0);
1216             dz00             = _mm256_sub_ps(iz0,jz0);
1217             dx10             = _mm256_sub_ps(ix1,jx0);
1218             dy10             = _mm256_sub_ps(iy1,jy0);
1219             dz10             = _mm256_sub_ps(iz1,jz0);
1220             dx20             = _mm256_sub_ps(ix2,jx0);
1221             dy20             = _mm256_sub_ps(iy2,jy0);
1222             dz20             = _mm256_sub_ps(iz2,jz0);
1223
1224             /* Calculate squared distance and things based on it */
1225             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1226             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1227             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1228
1229             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1230             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1231             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1232
1233             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1234             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1235             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1236
1237             /* Load parameters for j particles */
1238             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1239                                                                  charge+jnrC+0,charge+jnrD+0,
1240                                                                  charge+jnrE+0,charge+jnrF+0,
1241                                                                  charge+jnrG+0,charge+jnrH+0);
1242             vdwjidx0A        = 2*vdwtype[jnrA+0];
1243             vdwjidx0B        = 2*vdwtype[jnrB+0];
1244             vdwjidx0C        = 2*vdwtype[jnrC+0];
1245             vdwjidx0D        = 2*vdwtype[jnrD+0];
1246             vdwjidx0E        = 2*vdwtype[jnrE+0];
1247             vdwjidx0F        = 2*vdwtype[jnrF+0];
1248             vdwjidx0G        = 2*vdwtype[jnrG+0];
1249             vdwjidx0H        = 2*vdwtype[jnrH+0];
1250
1251             fjx0             = _mm256_setzero_ps();
1252             fjy0             = _mm256_setzero_ps();
1253             fjz0             = _mm256_setzero_ps();
1254
1255             /**************************
1256              * CALCULATE INTERACTIONS *
1257              **************************/
1258
1259             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1260             {
1261
1262             r00              = _mm256_mul_ps(rsq00,rinv00);
1263             r00              = _mm256_andnot_ps(dummy_mask,r00);
1264
1265             /* Compute parameters for interactions between i and j atoms */
1266             qq00             = _mm256_mul_ps(iq0,jq0);
1267             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1268                                             vdwioffsetptr0+vdwjidx0B,
1269                                             vdwioffsetptr0+vdwjidx0C,
1270                                             vdwioffsetptr0+vdwjidx0D,
1271                                             vdwioffsetptr0+vdwjidx0E,
1272                                             vdwioffsetptr0+vdwjidx0F,
1273                                             vdwioffsetptr0+vdwjidx0G,
1274                                             vdwioffsetptr0+vdwjidx0H,
1275                                             &c6_00,&c12_00);
1276
1277             /* Calculate table index by multiplying r with table scale and truncate to integer */
1278             rt               = _mm256_mul_ps(r00,vftabscale);
1279             vfitab           = _mm256_cvttps_epi32(rt);
1280             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1281             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1282             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1283             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1284             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1285             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1286
1287             /* REACTION-FIELD ELECTROSTATICS */
1288             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1289
1290             /* CUBIC SPLINE TABLE DISPERSION */
1291             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1293             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1295             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1297             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1298                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1299             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1300             Heps             = _mm256_mul_ps(vfeps,H);
1301             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1302             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1303             fvdw6            = _mm256_mul_ps(c6_00,FF);
1304
1305             /* CUBIC SPLINE TABLE REPULSION */
1306             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1307             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1308             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1310             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1312             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1314             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1316             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1317             Heps             = _mm256_mul_ps(vfeps,H);
1318             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1319             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1320             fvdw12           = _mm256_mul_ps(c12_00,FF);
1321             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1322
1323             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1324
1325             fscal            = _mm256_add_ps(felec,fvdw);
1326
1327             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1328
1329             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1330
1331             /* Calculate temporary vectorial force */
1332             tx               = _mm256_mul_ps(fscal,dx00);
1333             ty               = _mm256_mul_ps(fscal,dy00);
1334             tz               = _mm256_mul_ps(fscal,dz00);
1335
1336             /* Update vectorial force */
1337             fix0             = _mm256_add_ps(fix0,tx);
1338             fiy0             = _mm256_add_ps(fiy0,ty);
1339             fiz0             = _mm256_add_ps(fiz0,tz);
1340
1341             fjx0             = _mm256_add_ps(fjx0,tx);
1342             fjy0             = _mm256_add_ps(fjy0,ty);
1343             fjz0             = _mm256_add_ps(fjz0,tz);
1344
1345             }
1346
1347             /**************************
1348              * CALCULATE INTERACTIONS *
1349              **************************/
1350
1351             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1352             {
1353
1354             /* Compute parameters for interactions between i and j atoms */
1355             qq10             = _mm256_mul_ps(iq1,jq0);
1356
1357             /* REACTION-FIELD ELECTROSTATICS */
1358             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1359
1360             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1361
1362             fscal            = felec;
1363
1364             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1365
1366             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1367
1368             /* Calculate temporary vectorial force */
1369             tx               = _mm256_mul_ps(fscal,dx10);
1370             ty               = _mm256_mul_ps(fscal,dy10);
1371             tz               = _mm256_mul_ps(fscal,dz10);
1372
1373             /* Update vectorial force */
1374             fix1             = _mm256_add_ps(fix1,tx);
1375             fiy1             = _mm256_add_ps(fiy1,ty);
1376             fiz1             = _mm256_add_ps(fiz1,tz);
1377
1378             fjx0             = _mm256_add_ps(fjx0,tx);
1379             fjy0             = _mm256_add_ps(fjy0,ty);
1380             fjz0             = _mm256_add_ps(fjz0,tz);
1381
1382             }
1383
1384             /**************************
1385              * CALCULATE INTERACTIONS *
1386              **************************/
1387
1388             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1389             {
1390
1391             /* Compute parameters for interactions between i and j atoms */
1392             qq20             = _mm256_mul_ps(iq2,jq0);
1393
1394             /* REACTION-FIELD ELECTROSTATICS */
1395             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1396
1397             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1398
1399             fscal            = felec;
1400
1401             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1402
1403             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1404
1405             /* Calculate temporary vectorial force */
1406             tx               = _mm256_mul_ps(fscal,dx20);
1407             ty               = _mm256_mul_ps(fscal,dy20);
1408             tz               = _mm256_mul_ps(fscal,dz20);
1409
1410             /* Update vectorial force */
1411             fix2             = _mm256_add_ps(fix2,tx);
1412             fiy2             = _mm256_add_ps(fiy2,ty);
1413             fiz2             = _mm256_add_ps(fiz2,tz);
1414
1415             fjx0             = _mm256_add_ps(fjx0,tx);
1416             fjy0             = _mm256_add_ps(fjy0,ty);
1417             fjz0             = _mm256_add_ps(fjz0,tz);
1418
1419             }
1420
1421             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1422             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1423             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1424             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1425             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1426             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1427             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1428             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1429
1430             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1431
1432             /* Inner loop uses 121 flops */
1433         }
1434
1435         /* End of innermost loop */
1436
1437         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1438                                                  f+i_coord_offset,fshift+i_shift_offset);
1439
1440         /* Increment number of inner iterations */
1441         inneriter                  += j_index_end - j_index_start;
1442
1443         /* Outer loop uses 18 flops */
1444     }
1445
1446     /* Increment number of outer iterations */
1447     outeriter        += nri;
1448
1449     /* Update outer/inner flops */
1450
1451     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1452 }