src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256i          vfitab;
 105     __m128i          vfitab_lo,vfitab_hi;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256           dummy_mask,cutoff_mask;
 110     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 111     __m256           one     = _mm256_set1_ps(1.0);
 112     __m256           two     = _mm256_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_ps(fr->ic->epsfac);
 125     charge           = mdatoms->chargeA;
 126     krf              = _mm256_set1_ps(fr->ic->k_rf);
 127     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 128     crf              = _mm256_set1_ps(fr->ic->c_rf);
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_vdw->data;
 134     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 139     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 140     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 141     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 142
 143     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 144     rcutoff_scalar   = fr->ic->rcoulomb;
 145     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 146     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154     j_coord_offsetE = 0;
 155     j_coord_offsetF = 0;
 156     j_coord_offsetG = 0;
 157     j_coord_offsetH = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     for(iidx=0;iidx<4*DIM;iidx++)
 163     {
 164         scratch[iidx] = 0.0;
 165     }
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 183                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 184
 185         fix0             = _mm256_setzero_ps();
 186         fiy0             = _mm256_setzero_ps();
 187         fiz0             = _mm256_setzero_ps();
 188         fix1             = _mm256_setzero_ps();
 189         fiy1             = _mm256_setzero_ps();
 190         fiz1             = _mm256_setzero_ps();
 191         fix2             = _mm256_setzero_ps();
 192         fiy2             = _mm256_setzero_ps();
 193         fiz2             = _mm256_setzero_ps();
 194
 195         /* Reset potential sums */
 196         velecsum         = _mm256_setzero_ps();
 197         vvdwsum          = _mm256_setzero_ps();
 198
 199         /* Start inner kernel loop */
 200         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 201         {
 202
 203             /* Get j neighbor index, and coordinate index */
 204             jnrA             = jjnr[jidx];
 205             jnrB             = jjnr[jidx+1];
 206             jnrC             = jjnr[jidx+2];
 207             jnrD             = jjnr[jidx+3];
 208             jnrE             = jjnr[jidx+4];
 209             jnrF             = jjnr[jidx+5];
 210             jnrG             = jjnr[jidx+6];
 211             jnrH             = jjnr[jidx+7];
 212             j_coord_offsetA  = DIM*jnrA;
 213             j_coord_offsetB  = DIM*jnrB;
 214             j_coord_offsetC  = DIM*jnrC;
 215             j_coord_offsetD  = DIM*jnrD;
 216             j_coord_offsetE  = DIM*jnrE;
 217             j_coord_offsetF  = DIM*jnrF;
 218             j_coord_offsetG  = DIM*jnrG;
 219             j_coord_offsetH  = DIM*jnrH;
 220
 221             /* load j atom coordinates */
 222             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 224                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 225                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 226                                                  &jx0,&jy0,&jz0);
 227
 228             /* Calculate displacement vector */
 229             dx00             = _mm256_sub_ps(ix0,jx0);
 230             dy00             = _mm256_sub_ps(iy0,jy0);
 231             dz00             = _mm256_sub_ps(iz0,jz0);
 232             dx10             = _mm256_sub_ps(ix1,jx0);
 233             dy10             = _mm256_sub_ps(iy1,jy0);
 234             dz10             = _mm256_sub_ps(iz1,jz0);
 235             dx20             = _mm256_sub_ps(ix2,jx0);
 236             dy20             = _mm256_sub_ps(iy2,jy0);
 237             dz20             = _mm256_sub_ps(iz2,jz0);
 238
 239             /* Calculate squared distance and things based on it */
 240             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 241             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 242             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 243
 244             rinv00           = avx256_invsqrt_f(rsq00);
 245             rinv10           = avx256_invsqrt_f(rsq10);
 246             rinv20           = avx256_invsqrt_f(rsq20);
 247
 248             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 249             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 250             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 251
 252             /* Load parameters for j particles */
 253             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 254                                                                  charge+jnrC+0,charge+jnrD+0,
 255                                                                  charge+jnrE+0,charge+jnrF+0,
 256                                                                  charge+jnrG+0,charge+jnrH+0);
 257             vdwjidx0A        = 2*vdwtype[jnrA+0];
 258             vdwjidx0B        = 2*vdwtype[jnrB+0];
 259             vdwjidx0C        = 2*vdwtype[jnrC+0];
 260             vdwjidx0D        = 2*vdwtype[jnrD+0];
 261             vdwjidx0E        = 2*vdwtype[jnrE+0];
 262             vdwjidx0F        = 2*vdwtype[jnrF+0];
 263             vdwjidx0G        = 2*vdwtype[jnrG+0];
 264             vdwjidx0H        = 2*vdwtype[jnrH+0];
 265
 266             fjx0             = _mm256_setzero_ps();
 267             fjy0             = _mm256_setzero_ps();
 268             fjz0             = _mm256_setzero_ps();
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 275             {
 276
 277             r00              = _mm256_mul_ps(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq00             = _mm256_mul_ps(iq0,jq0);
 281             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 282                                             vdwioffsetptr0+vdwjidx0B,
 283                                             vdwioffsetptr0+vdwjidx0C,
 284                                             vdwioffsetptr0+vdwjidx0D,
 285                                             vdwioffsetptr0+vdwjidx0E,
 286                                             vdwioffsetptr0+vdwjidx0F,
 287                                             vdwioffsetptr0+vdwjidx0G,
 288                                             vdwioffsetptr0+vdwjidx0H,
 289                                             &c6_00,&c12_00);
 290
 291             /* Calculate table index by multiplying r with table scale and truncate to integer */
 292             rt               = _mm256_mul_ps(r00,vftabscale);
 293             vfitab           = _mm256_cvttps_epi32(rt);
 294             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 295             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 296             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 297             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 298             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 299             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 300
 301             /* REACTION-FIELD ELECTROSTATICS */
 302             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 303             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 304
 305             /* CUBIC SPLINE TABLE DISPERSION */
 306             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 308             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 310             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 312             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 314             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 315             Heps             = _mm256_mul_ps(vfeps,H);
 316             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 317             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 318             vvdw6            = _mm256_mul_ps(c6_00,VV);
 319             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 320             fvdw6            = _mm256_mul_ps(c6_00,FF);
 321
 322             /* CUBIC SPLINE TABLE REPULSION */
 323             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 324             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 325             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 327             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 329             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 331             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 332                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 333             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 334             Heps             = _mm256_mul_ps(vfeps,H);
 335             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 336             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 337             vvdw12           = _mm256_mul_ps(c12_00,VV);
 338             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 339             fvdw12           = _mm256_mul_ps(c12_00,FF);
 340             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 341             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 342
 343             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm256_and_ps(velec,cutoff_mask);
 347             velecsum         = _mm256_add_ps(velecsum,velec);
 348             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 349             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 350
 351             fscal            = _mm256_add_ps(felec,fvdw);
 352
 353             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm256_mul_ps(fscal,dx00);
 357             ty               = _mm256_mul_ps(fscal,dy00);
 358             tz               = _mm256_mul_ps(fscal,dz00);
 359
 360             /* Update vectorial force */
 361             fix0             = _mm256_add_ps(fix0,tx);
 362             fiy0             = _mm256_add_ps(fiy0,ty);
 363             fiz0             = _mm256_add_ps(fiz0,tz);
 364
 365             fjx0             = _mm256_add_ps(fjx0,tx);
 366             fjy0             = _mm256_add_ps(fjy0,ty);
 367             fjz0             = _mm256_add_ps(fjz0,tz);
 368
 369             }
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 376             {
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq10             = _mm256_mul_ps(iq1,jq0);
 380
 381             /* REACTION-FIELD ELECTROSTATICS */
 382             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 383             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 384
 385             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm256_and_ps(velec,cutoff_mask);
 389             velecsum         = _mm256_add_ps(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm256_mul_ps(fscal,dx10);
 397             ty               = _mm256_mul_ps(fscal,dy10);
 398             tz               = _mm256_mul_ps(fscal,dz10);
 399
 400             /* Update vectorial force */
 401             fix1             = _mm256_add_ps(fix1,tx);
 402             fiy1             = _mm256_add_ps(fiy1,ty);
 403             fiz1             = _mm256_add_ps(fiz1,tz);
 404
 405             fjx0             = _mm256_add_ps(fjx0,tx);
 406             fjy0             = _mm256_add_ps(fjy0,ty);
 407             fjz0             = _mm256_add_ps(fjz0,tz);
 408
 409             }
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 416             {
 417
 418             /* Compute parameters for interactions between i and j atoms */
 419             qq20             = _mm256_mul_ps(iq2,jq0);
 420
 421             /* REACTION-FIELD ELECTROSTATICS */
 422             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 423             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 424
 425             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velec            = _mm256_and_ps(velec,cutoff_mask);
 429             velecsum         = _mm256_add_ps(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 434
 435             /* Calculate temporary vectorial force */
 436             tx               = _mm256_mul_ps(fscal,dx20);
 437             ty               = _mm256_mul_ps(fscal,dy20);
 438             tz               = _mm256_mul_ps(fscal,dz20);
 439
 440             /* Update vectorial force */
 441             fix2             = _mm256_add_ps(fix2,tx);
 442             fiy2             = _mm256_add_ps(fiy2,ty);
 443             fiz2             = _mm256_add_ps(fiz2,tz);
 444
 445             fjx0             = _mm256_add_ps(fjx0,tx);
 446             fjy0             = _mm256_add_ps(fjy0,ty);
 447             fjz0             = _mm256_add_ps(fjz0,tz);
 448
 449             }
 450
 451             fjptrA             = f+j_coord_offsetA;
 452             fjptrB             = f+j_coord_offsetB;
 453             fjptrC             = f+j_coord_offsetC;
 454             fjptrD             = f+j_coord_offsetD;
 455             fjptrE             = f+j_coord_offsetE;
 456             fjptrF             = f+j_coord_offsetF;
 457             fjptrG             = f+j_coord_offsetG;
 458             fjptrH             = f+j_coord_offsetH;
 459
 460             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 461
 462             /* Inner loop uses 147 flops */
 463         }
 464
 465         if(jidx<j_index_end)
 466         {
 467
 468             /* Get j neighbor index, and coordinate index */
 469             jnrlistA         = jjnr[jidx];
 470             jnrlistB         = jjnr[jidx+1];
 471             jnrlistC         = jjnr[jidx+2];
 472             jnrlistD         = jjnr[jidx+3];
 473             jnrlistE         = jjnr[jidx+4];
 474             jnrlistF         = jjnr[jidx+5];
 475             jnrlistG         = jjnr[jidx+6];
 476             jnrlistH         = jjnr[jidx+7];
 477             /* Sign of each element will be negative for non-real atoms.
 478              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 479              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 480              */
 481             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 482                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 483
 484             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 485             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 486             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 487             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 488             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 489             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 490             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 491             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 492             j_coord_offsetA  = DIM*jnrA;
 493             j_coord_offsetB  = DIM*jnrB;
 494             j_coord_offsetC  = DIM*jnrC;
 495             j_coord_offsetD  = DIM*jnrD;
 496             j_coord_offsetE  = DIM*jnrE;
 497             j_coord_offsetF  = DIM*jnrF;
 498             j_coord_offsetG  = DIM*jnrG;
 499             j_coord_offsetH  = DIM*jnrH;
 500
 501             /* load j atom coordinates */
 502             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 504                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 505                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 506                                                  &jx0,&jy0,&jz0);
 507
 508             /* Calculate displacement vector */
 509             dx00             = _mm256_sub_ps(ix0,jx0);
 510             dy00             = _mm256_sub_ps(iy0,jy0);
 511             dz00             = _mm256_sub_ps(iz0,jz0);
 512             dx10             = _mm256_sub_ps(ix1,jx0);
 513             dy10             = _mm256_sub_ps(iy1,jy0);
 514             dz10             = _mm256_sub_ps(iz1,jz0);
 515             dx20             = _mm256_sub_ps(ix2,jx0);
 516             dy20             = _mm256_sub_ps(iy2,jy0);
 517             dz20             = _mm256_sub_ps(iz2,jz0);
 518
 519             /* Calculate squared distance and things based on it */
 520             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 521             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 522             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 523
 524             rinv00           = avx256_invsqrt_f(rsq00);
 525             rinv10           = avx256_invsqrt_f(rsq10);
 526             rinv20           = avx256_invsqrt_f(rsq20);
 527
 528             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 529             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 530             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 531
 532             /* Load parameters for j particles */
 533             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 534                                                                  charge+jnrC+0,charge+jnrD+0,
 535                                                                  charge+jnrE+0,charge+jnrF+0,
 536                                                                  charge+jnrG+0,charge+jnrH+0);
 537             vdwjidx0A        = 2*vdwtype[jnrA+0];
 538             vdwjidx0B        = 2*vdwtype[jnrB+0];
 539             vdwjidx0C        = 2*vdwtype[jnrC+0];
 540             vdwjidx0D        = 2*vdwtype[jnrD+0];
 541             vdwjidx0E        = 2*vdwtype[jnrE+0];
 542             vdwjidx0F        = 2*vdwtype[jnrF+0];
 543             vdwjidx0G        = 2*vdwtype[jnrG+0];
 544             vdwjidx0H        = 2*vdwtype[jnrH+0];
 545
 546             fjx0             = _mm256_setzero_ps();
 547             fjy0             = _mm256_setzero_ps();
 548             fjz0             = _mm256_setzero_ps();
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 555             {
 556
 557             r00              = _mm256_mul_ps(rsq00,rinv00);
 558             r00              = _mm256_andnot_ps(dummy_mask,r00);
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             qq00             = _mm256_mul_ps(iq0,jq0);
 562             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 563                                             vdwioffsetptr0+vdwjidx0B,
 564                                             vdwioffsetptr0+vdwjidx0C,
 565                                             vdwioffsetptr0+vdwjidx0D,
 566                                             vdwioffsetptr0+vdwjidx0E,
 567                                             vdwioffsetptr0+vdwjidx0F,
 568                                             vdwioffsetptr0+vdwjidx0G,
 569                                             vdwioffsetptr0+vdwjidx0H,
 570                                             &c6_00,&c12_00);
 571
 572             /* Calculate table index by multiplying r with table scale and truncate to integer */
 573             rt               = _mm256_mul_ps(r00,vftabscale);
 574             vfitab           = _mm256_cvttps_epi32(rt);
 575             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 576             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 577             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 578             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 579             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 580             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 581
 582             /* REACTION-FIELD ELECTROSTATICS */
 583             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 584             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 585
 586             /* CUBIC SPLINE TABLE DISPERSION */
 587             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 588                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 589             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 590                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 591             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 592                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 593             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 594                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 595             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 596             Heps             = _mm256_mul_ps(vfeps,H);
 597             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 598             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 599             vvdw6            = _mm256_mul_ps(c6_00,VV);
 600             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 601             fvdw6            = _mm256_mul_ps(c6_00,FF);
 602
 603             /* CUBIC SPLINE TABLE REPULSION */
 604             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 605             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 606             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 607                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 608             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 609                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 610             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 611                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 612             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 613                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 614             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 615             Heps             = _mm256_mul_ps(vfeps,H);
 616             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 617             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 618             vvdw12           = _mm256_mul_ps(c12_00,VV);
 619             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 620             fvdw12           = _mm256_mul_ps(c12_00,FF);
 621             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 622             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 623
 624             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm256_and_ps(velec,cutoff_mask);
 628             velec            = _mm256_andnot_ps(dummy_mask,velec);
 629             velecsum         = _mm256_add_ps(velecsum,velec);
 630             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 631             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 632             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 633
 634             fscal            = _mm256_add_ps(felec,fvdw);
 635
 636             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 637
 638             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 639
 640             /* Calculate temporary vectorial force */
 641             tx               = _mm256_mul_ps(fscal,dx00);
 642             ty               = _mm256_mul_ps(fscal,dy00);
 643             tz               = _mm256_mul_ps(fscal,dz00);
 644
 645             /* Update vectorial force */
 646             fix0             = _mm256_add_ps(fix0,tx);
 647             fiy0             = _mm256_add_ps(fiy0,ty);
 648             fiz0             = _mm256_add_ps(fiz0,tz);
 649
 650             fjx0             = _mm256_add_ps(fjx0,tx);
 651             fjy0             = _mm256_add_ps(fjy0,ty);
 652             fjz0             = _mm256_add_ps(fjz0,tz);
 653
 654             }
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 661             {
 662
 663             /* Compute parameters for interactions between i and j atoms */
 664             qq10             = _mm256_mul_ps(iq1,jq0);
 665
 666             /* REACTION-FIELD ELECTROSTATICS */
 667             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 668             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 669
 670             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 671
 672             /* Update potential sum for this i atom from the interaction with this j atom. */
 673             velec            = _mm256_and_ps(velec,cutoff_mask);
 674             velec            = _mm256_andnot_ps(dummy_mask,velec);
 675             velecsum         = _mm256_add_ps(velecsum,velec);
 676
 677             fscal            = felec;
 678
 679             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 680
 681             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm256_mul_ps(fscal,dx10);
 685             ty               = _mm256_mul_ps(fscal,dy10);
 686             tz               = _mm256_mul_ps(fscal,dz10);
 687
 688             /* Update vectorial force */
 689             fix1             = _mm256_add_ps(fix1,tx);
 690             fiy1             = _mm256_add_ps(fiy1,ty);
 691             fiz1             = _mm256_add_ps(fiz1,tz);
 692
 693             fjx0             = _mm256_add_ps(fjx0,tx);
 694             fjy0             = _mm256_add_ps(fjy0,ty);
 695             fjz0             = _mm256_add_ps(fjz0,tz);
 696
 697             }
 698
 699             /**************************
 700              * CALCULATE INTERACTIONS *
 701              **************************/
 702
 703             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 704             {
 705
 706             /* Compute parameters for interactions between i and j atoms */
 707             qq20             = _mm256_mul_ps(iq2,jq0);
 708
 709             /* REACTION-FIELD ELECTROSTATICS */
 710             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 711             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 712
 713             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 714
 715             /* Update potential sum for this i atom from the interaction with this j atom. */
 716             velec            = _mm256_and_ps(velec,cutoff_mask);
 717             velec            = _mm256_andnot_ps(dummy_mask,velec);
 718             velecsum         = _mm256_add_ps(velecsum,velec);
 719
 720             fscal            = felec;
 721
 722             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 723
 724             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 725
 726             /* Calculate temporary vectorial force */
 727             tx               = _mm256_mul_ps(fscal,dx20);
 728             ty               = _mm256_mul_ps(fscal,dy20);
 729             tz               = _mm256_mul_ps(fscal,dz20);
 730
 731             /* Update vectorial force */
 732             fix2             = _mm256_add_ps(fix2,tx);
 733             fiy2             = _mm256_add_ps(fiy2,ty);
 734             fiz2             = _mm256_add_ps(fiz2,tz);
 735
 736             fjx0             = _mm256_add_ps(fjx0,tx);
 737             fjy0             = _mm256_add_ps(fjy0,ty);
 738             fjz0             = _mm256_add_ps(fjz0,tz);
 739
 740             }
 741
 742             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 743             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 744             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 745             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 746             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 747             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 748             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 749             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 750
 751             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 752
 753             /* Inner loop uses 148 flops */
 754         }
 755
 756         /* End of innermost loop */
 757
 758         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 759                                                  f+i_coord_offset,fshift+i_shift_offset);
 760
 761         ggid                        = gid[iidx];
 762         /* Update potential energies */
 763         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 764         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 765
 766         /* Increment number of inner iterations */
 767         inneriter                  += j_index_end - j_index_start;
 768
 769         /* Outer loop uses 20 flops */
 770     }
 771
 772     /* Increment number of outer iterations */
 773     outeriter        += nri;
 774
 775     /* Update outer/inner flops */
 776
 777     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 778 }
 779 /*
 780  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_single
 781  * Electrostatics interaction: ReactionField
 782  * VdW interaction:            CubicSplineTable
 783  * Geometry:                   Water3-Particle
 784  * Calculate force/pot:        Force
 785  */
 786 void
 787 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_single
 788                     (t_nblist                    * gmx_restrict       nlist,
 789                      rvec                        * gmx_restrict          xx,
 790                      rvec                        * gmx_restrict          ff,
 791                      struct t_forcerec           * gmx_restrict          fr,
 792                      t_mdatoms                   * gmx_restrict     mdatoms,
 793                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 794                      t_nrnb                      * gmx_restrict        nrnb)
 795 {
 796     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 797      * just 0 for non-waters.
 798      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 799      * jnr indices corresponding to data put in the four positions in the SIMD register.
 800      */
 801     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 802     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 803     int              jnrA,jnrB,jnrC,jnrD;
 804     int              jnrE,jnrF,jnrG,jnrH;
 805     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 806     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 807     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 808     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 809     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 810     real             rcutoff_scalar;
 811     real             *shiftvec,*fshift,*x,*f;
 812     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 813     real             scratch[4*DIM];
 814     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 815     real *           vdwioffsetptr0;
 816     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 817     real *           vdwioffsetptr1;
 818     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 819     real *           vdwioffsetptr2;
 820     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 821     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 822     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 823     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 824     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 825     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 826     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 827     real             *charge;
 828     int              nvdwtype;
 829     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 830     int              *vdwtype;
 831     real             *vdwparam;
 832     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 833     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 834     __m256i          vfitab;
 835     __m128i          vfitab_lo,vfitab_hi;
 836     __m128i          ifour       = _mm_set1_epi32(4);
 837     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 838     real             *vftab;
 839     __m256           dummy_mask,cutoff_mask;
 840     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 841     __m256           one     = _mm256_set1_ps(1.0);
 842     __m256           two     = _mm256_set1_ps(2.0);
 843     x                = xx[0];
 844     f                = ff[0];
 845
 846     nri              = nlist->nri;
 847     iinr             = nlist->iinr;
 848     jindex           = nlist->jindex;
 849     jjnr             = nlist->jjnr;
 850     shiftidx         = nlist->shift;
 851     gid              = nlist->gid;
 852     shiftvec         = fr->shift_vec[0];
 853     fshift           = fr->fshift[0];
 854     facel            = _mm256_set1_ps(fr->ic->epsfac);
 855     charge           = mdatoms->chargeA;
 856     krf              = _mm256_set1_ps(fr->ic->k_rf);
 857     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 858     crf              = _mm256_set1_ps(fr->ic->c_rf);
 859     nvdwtype         = fr->ntype;
 860     vdwparam         = fr->nbfp;
 861     vdwtype          = mdatoms->typeA;
 862
 863     vftab            = kernel_data->table_vdw->data;
 864     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 865
 866     /* Setup water-specific parameters */
 867     inr              = nlist->iinr[0];
 868     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 869     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 870     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 871     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 872
 873     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 874     rcutoff_scalar   = fr->ic->rcoulomb;
 875     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 876     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 877
 878     /* Avoid stupid compiler warnings */
 879     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 880     j_coord_offsetA = 0;
 881     j_coord_offsetB = 0;
 882     j_coord_offsetC = 0;
 883     j_coord_offsetD = 0;
 884     j_coord_offsetE = 0;
 885     j_coord_offsetF = 0;
 886     j_coord_offsetG = 0;
 887     j_coord_offsetH = 0;
 888
 889     outeriter        = 0;
 890     inneriter        = 0;
 891
 892     for(iidx=0;iidx<4*DIM;iidx++)
 893     {
 894         scratch[iidx] = 0.0;
 895     }
 896
 897     /* Start outer loop over neighborlists */
 898     for(iidx=0; iidx<nri; iidx++)
 899     {
 900         /* Load shift vector for this list */
 901         i_shift_offset   = DIM*shiftidx[iidx];
 902
 903         /* Load limits for loop over neighbors */
 904         j_index_start    = jindex[iidx];
 905         j_index_end      = jindex[iidx+1];
 906
 907         /* Get outer coordinate index */
 908         inr              = iinr[iidx];
 909         i_coord_offset   = DIM*inr;
 910
 911         /* Load i particle coords and add shift vector */
 912         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 913                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 914
 915         fix0             = _mm256_setzero_ps();
 916         fiy0             = _mm256_setzero_ps();
 917         fiz0             = _mm256_setzero_ps();
 918         fix1             = _mm256_setzero_ps();
 919         fiy1             = _mm256_setzero_ps();
 920         fiz1             = _mm256_setzero_ps();
 921         fix2             = _mm256_setzero_ps();
 922         fiy2             = _mm256_setzero_ps();
 923         fiz2             = _mm256_setzero_ps();
 924
 925         /* Start inner kernel loop */
 926         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 927         {
 928
 929             /* Get j neighbor index, and coordinate index */
 930             jnrA             = jjnr[jidx];
 931             jnrB             = jjnr[jidx+1];
 932             jnrC             = jjnr[jidx+2];
 933             jnrD             = jjnr[jidx+3];
 934             jnrE             = jjnr[jidx+4];
 935             jnrF             = jjnr[jidx+5];
 936             jnrG             = jjnr[jidx+6];
 937             jnrH             = jjnr[jidx+7];
 938             j_coord_offsetA  = DIM*jnrA;
 939             j_coord_offsetB  = DIM*jnrB;
 940             j_coord_offsetC  = DIM*jnrC;
 941             j_coord_offsetD  = DIM*jnrD;
 942             j_coord_offsetE  = DIM*jnrE;
 943             j_coord_offsetF  = DIM*jnrF;
 944             j_coord_offsetG  = DIM*jnrG;
 945             j_coord_offsetH  = DIM*jnrH;
 946
 947             /* load j atom coordinates */
 948             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 949                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 950                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 951                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 952                                                  &jx0,&jy0,&jz0);
 953
 954             /* Calculate displacement vector */
 955             dx00             = _mm256_sub_ps(ix0,jx0);
 956             dy00             = _mm256_sub_ps(iy0,jy0);
 957             dz00             = _mm256_sub_ps(iz0,jz0);
 958             dx10             = _mm256_sub_ps(ix1,jx0);
 959             dy10             = _mm256_sub_ps(iy1,jy0);
 960             dz10             = _mm256_sub_ps(iz1,jz0);
 961             dx20             = _mm256_sub_ps(ix2,jx0);
 962             dy20             = _mm256_sub_ps(iy2,jy0);
 963             dz20             = _mm256_sub_ps(iz2,jz0);
 964
 965             /* Calculate squared distance and things based on it */
 966             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 967             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 968             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 969
 970             rinv00           = avx256_invsqrt_f(rsq00);
 971             rinv10           = avx256_invsqrt_f(rsq10);
 972             rinv20           = avx256_invsqrt_f(rsq20);
 973
 974             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 975             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 976             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 977
 978             /* Load parameters for j particles */
 979             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 980                                                                  charge+jnrC+0,charge+jnrD+0,
 981                                                                  charge+jnrE+0,charge+jnrF+0,
 982                                                                  charge+jnrG+0,charge+jnrH+0);
 983             vdwjidx0A        = 2*vdwtype[jnrA+0];
 984             vdwjidx0B        = 2*vdwtype[jnrB+0];
 985             vdwjidx0C        = 2*vdwtype[jnrC+0];
 986             vdwjidx0D        = 2*vdwtype[jnrD+0];
 987             vdwjidx0E        = 2*vdwtype[jnrE+0];
 988             vdwjidx0F        = 2*vdwtype[jnrF+0];
 989             vdwjidx0G        = 2*vdwtype[jnrG+0];
 990             vdwjidx0H        = 2*vdwtype[jnrH+0];
 991
 992             fjx0             = _mm256_setzero_ps();
 993             fjy0             = _mm256_setzero_ps();
 994             fjz0             = _mm256_setzero_ps();
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1001             {
1002
1003             r00              = _mm256_mul_ps(rsq00,rinv00);
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             qq00             = _mm256_mul_ps(iq0,jq0);
1007             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1008                                             vdwioffsetptr0+vdwjidx0B,
1009                                             vdwioffsetptr0+vdwjidx0C,
1010                                             vdwioffsetptr0+vdwjidx0D,
1011                                             vdwioffsetptr0+vdwjidx0E,
1012                                             vdwioffsetptr0+vdwjidx0F,
1013                                             vdwioffsetptr0+vdwjidx0G,
1014                                             vdwioffsetptr0+vdwjidx0H,
1015                                             &c6_00,&c12_00);
1016
1017             /* Calculate table index by multiplying r with table scale and truncate to integer */
1018             rt               = _mm256_mul_ps(r00,vftabscale);
1019             vfitab           = _mm256_cvttps_epi32(rt);
1020             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1021             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1022             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1023             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1024             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1025             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1026
1027             /* REACTION-FIELD ELECTROSTATICS */
1028             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1029
1030             /* CUBIC SPLINE TABLE DISPERSION */
1031             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1032                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1033             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1034                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1035             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1036                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1037             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1038                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1039             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1040             Heps             = _mm256_mul_ps(vfeps,H);
1041             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1042             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1043             fvdw6            = _mm256_mul_ps(c6_00,FF);
1044
1045             /* CUBIC SPLINE TABLE REPULSION */
1046             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1047             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1048             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1049                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1050             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1051                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1052             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1053                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1054             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1055                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1056             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1057             Heps             = _mm256_mul_ps(vfeps,H);
1058             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1059             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1060             fvdw12           = _mm256_mul_ps(c12_00,FF);
1061             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1062
1063             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1064
1065             fscal            = _mm256_add_ps(felec,fvdw);
1066
1067             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm256_mul_ps(fscal,dx00);
1071             ty               = _mm256_mul_ps(fscal,dy00);
1072             tz               = _mm256_mul_ps(fscal,dz00);
1073
1074             /* Update vectorial force */
1075             fix0             = _mm256_add_ps(fix0,tx);
1076             fiy0             = _mm256_add_ps(fiy0,ty);
1077             fiz0             = _mm256_add_ps(fiz0,tz);
1078
1079             fjx0             = _mm256_add_ps(fjx0,tx);
1080             fjy0             = _mm256_add_ps(fjy0,ty);
1081             fjz0             = _mm256_add_ps(fjz0,tz);
1082
1083             }
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1090             {
1091
1092             /* Compute parameters for interactions between i and j atoms */
1093             qq10             = _mm256_mul_ps(iq1,jq0);
1094
1095             /* REACTION-FIELD ELECTROSTATICS */
1096             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1097
1098             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1099
1100             fscal            = felec;
1101
1102             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1103
1104             /* Calculate temporary vectorial force */
1105             tx               = _mm256_mul_ps(fscal,dx10);
1106             ty               = _mm256_mul_ps(fscal,dy10);
1107             tz               = _mm256_mul_ps(fscal,dz10);
1108
1109             /* Update vectorial force */
1110             fix1             = _mm256_add_ps(fix1,tx);
1111             fiy1             = _mm256_add_ps(fiy1,ty);
1112             fiz1             = _mm256_add_ps(fiz1,tz);
1113
1114             fjx0             = _mm256_add_ps(fjx0,tx);
1115             fjy0             = _mm256_add_ps(fjy0,ty);
1116             fjz0             = _mm256_add_ps(fjz0,tz);
1117
1118             }
1119
1120             /**************************
1121              * CALCULATE INTERACTIONS *
1122              **************************/
1123
1124             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1125             {
1126
1127             /* Compute parameters for interactions between i and j atoms */
1128             qq20             = _mm256_mul_ps(iq2,jq0);
1129
1130             /* REACTION-FIELD ELECTROSTATICS */
1131             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1132
1133             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm256_mul_ps(fscal,dx20);
1141             ty               = _mm256_mul_ps(fscal,dy20);
1142             tz               = _mm256_mul_ps(fscal,dz20);
1143
1144             /* Update vectorial force */
1145             fix2             = _mm256_add_ps(fix2,tx);
1146             fiy2             = _mm256_add_ps(fiy2,ty);
1147             fiz2             = _mm256_add_ps(fiz2,tz);
1148
1149             fjx0             = _mm256_add_ps(fjx0,tx);
1150             fjy0             = _mm256_add_ps(fjy0,ty);
1151             fjz0             = _mm256_add_ps(fjz0,tz);
1152
1153             }
1154
1155             fjptrA             = f+j_coord_offsetA;
1156             fjptrB             = f+j_coord_offsetB;
1157             fjptrC             = f+j_coord_offsetC;
1158             fjptrD             = f+j_coord_offsetD;
1159             fjptrE             = f+j_coord_offsetE;
1160             fjptrF             = f+j_coord_offsetF;
1161             fjptrG             = f+j_coord_offsetG;
1162             fjptrH             = f+j_coord_offsetH;
1163
1164             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1165
1166             /* Inner loop uses 120 flops */
1167         }
1168
1169         if(jidx<j_index_end)
1170         {
1171
1172             /* Get j neighbor index, and coordinate index */
1173             jnrlistA         = jjnr[jidx];
1174             jnrlistB         = jjnr[jidx+1];
1175             jnrlistC         = jjnr[jidx+2];
1176             jnrlistD         = jjnr[jidx+3];
1177             jnrlistE         = jjnr[jidx+4];
1178             jnrlistF         = jjnr[jidx+5];
1179             jnrlistG         = jjnr[jidx+6];
1180             jnrlistH         = jjnr[jidx+7];
1181             /* Sign of each element will be negative for non-real atoms.
1182              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1183              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1184              */
1185             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1186                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1187
1188             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1189             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1190             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1191             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1192             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1193             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1194             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1195             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1196             j_coord_offsetA  = DIM*jnrA;
1197             j_coord_offsetB  = DIM*jnrB;
1198             j_coord_offsetC  = DIM*jnrC;
1199             j_coord_offsetD  = DIM*jnrD;
1200             j_coord_offsetE  = DIM*jnrE;
1201             j_coord_offsetF  = DIM*jnrF;
1202             j_coord_offsetG  = DIM*jnrG;
1203             j_coord_offsetH  = DIM*jnrH;
1204
1205             /* load j atom coordinates */
1206             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1207                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1208                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1209                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1210                                                  &jx0,&jy0,&jz0);
1211
1212             /* Calculate displacement vector */
1213             dx00             = _mm256_sub_ps(ix0,jx0);
1214             dy00             = _mm256_sub_ps(iy0,jy0);
1215             dz00             = _mm256_sub_ps(iz0,jz0);
1216             dx10             = _mm256_sub_ps(ix1,jx0);
1217             dy10             = _mm256_sub_ps(iy1,jy0);
1218             dz10             = _mm256_sub_ps(iz1,jz0);
1219             dx20             = _mm256_sub_ps(ix2,jx0);
1220             dy20             = _mm256_sub_ps(iy2,jy0);
1221             dz20             = _mm256_sub_ps(iz2,jz0);
1222
1223             /* Calculate squared distance and things based on it */
1224             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1225             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1226             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1227
1228             rinv00           = avx256_invsqrt_f(rsq00);
1229             rinv10           = avx256_invsqrt_f(rsq10);
1230             rinv20           = avx256_invsqrt_f(rsq20);
1231
1232             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1233             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1234             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1235
1236             /* Load parameters for j particles */
1237             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1238                                                                  charge+jnrC+0,charge+jnrD+0,
1239                                                                  charge+jnrE+0,charge+jnrF+0,
1240                                                                  charge+jnrG+0,charge+jnrH+0);
1241             vdwjidx0A        = 2*vdwtype[jnrA+0];
1242             vdwjidx0B        = 2*vdwtype[jnrB+0];
1243             vdwjidx0C        = 2*vdwtype[jnrC+0];
1244             vdwjidx0D        = 2*vdwtype[jnrD+0];
1245             vdwjidx0E        = 2*vdwtype[jnrE+0];
1246             vdwjidx0F        = 2*vdwtype[jnrF+0];
1247             vdwjidx0G        = 2*vdwtype[jnrG+0];
1248             vdwjidx0H        = 2*vdwtype[jnrH+0];
1249
1250             fjx0             = _mm256_setzero_ps();
1251             fjy0             = _mm256_setzero_ps();
1252             fjz0             = _mm256_setzero_ps();
1253
1254             /**************************
1255              * CALCULATE INTERACTIONS *
1256              **************************/
1257
1258             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1259             {
1260
1261             r00              = _mm256_mul_ps(rsq00,rinv00);
1262             r00              = _mm256_andnot_ps(dummy_mask,r00);
1263
1264             /* Compute parameters for interactions between i and j atoms */
1265             qq00             = _mm256_mul_ps(iq0,jq0);
1266             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1267                                             vdwioffsetptr0+vdwjidx0B,
1268                                             vdwioffsetptr0+vdwjidx0C,
1269                                             vdwioffsetptr0+vdwjidx0D,
1270                                             vdwioffsetptr0+vdwjidx0E,
1271                                             vdwioffsetptr0+vdwjidx0F,
1272                                             vdwioffsetptr0+vdwjidx0G,
1273                                             vdwioffsetptr0+vdwjidx0H,
1274                                             &c6_00,&c12_00);
1275
1276             /* Calculate table index by multiplying r with table scale and truncate to integer */
1277             rt               = _mm256_mul_ps(r00,vftabscale);
1278             vfitab           = _mm256_cvttps_epi32(rt);
1279             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1280             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1281             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1282             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1283             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1284             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1285
1286             /* REACTION-FIELD ELECTROSTATICS */
1287             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1288
1289             /* CUBIC SPLINE TABLE DISPERSION */
1290             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1292             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1294             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1296             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1298             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1299             Heps             = _mm256_mul_ps(vfeps,H);
1300             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1301             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1302             fvdw6            = _mm256_mul_ps(c6_00,FF);
1303
1304             /* CUBIC SPLINE TABLE REPULSION */
1305             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1306             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1307             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1309             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1311             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1313             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1314                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1315             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1316             Heps             = _mm256_mul_ps(vfeps,H);
1317             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1318             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1319             fvdw12           = _mm256_mul_ps(c12_00,FF);
1320             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1321
1322             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1323
1324             fscal            = _mm256_add_ps(felec,fvdw);
1325
1326             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1327
1328             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1329
1330             /* Calculate temporary vectorial force */
1331             tx               = _mm256_mul_ps(fscal,dx00);
1332             ty               = _mm256_mul_ps(fscal,dy00);
1333             tz               = _mm256_mul_ps(fscal,dz00);
1334
1335             /* Update vectorial force */
1336             fix0             = _mm256_add_ps(fix0,tx);
1337             fiy0             = _mm256_add_ps(fiy0,ty);
1338             fiz0             = _mm256_add_ps(fiz0,tz);
1339
1340             fjx0             = _mm256_add_ps(fjx0,tx);
1341             fjy0             = _mm256_add_ps(fjy0,ty);
1342             fjz0             = _mm256_add_ps(fjz0,tz);
1343
1344             }
1345
1346             /**************************
1347              * CALCULATE INTERACTIONS *
1348              **************************/
1349
1350             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1351             {
1352
1353             /* Compute parameters for interactions between i and j atoms */
1354             qq10             = _mm256_mul_ps(iq1,jq0);
1355
1356             /* REACTION-FIELD ELECTROSTATICS */
1357             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1358
1359             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1360
1361             fscal            = felec;
1362
1363             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1364
1365             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1366
1367             /* Calculate temporary vectorial force */
1368             tx               = _mm256_mul_ps(fscal,dx10);
1369             ty               = _mm256_mul_ps(fscal,dy10);
1370             tz               = _mm256_mul_ps(fscal,dz10);
1371
1372             /* Update vectorial force */
1373             fix1             = _mm256_add_ps(fix1,tx);
1374             fiy1             = _mm256_add_ps(fiy1,ty);
1375             fiz1             = _mm256_add_ps(fiz1,tz);
1376
1377             fjx0             = _mm256_add_ps(fjx0,tx);
1378             fjy0             = _mm256_add_ps(fjy0,ty);
1379             fjz0             = _mm256_add_ps(fjz0,tz);
1380
1381             }
1382
1383             /**************************
1384              * CALCULATE INTERACTIONS *
1385              **************************/
1386
1387             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1388             {
1389
1390             /* Compute parameters for interactions between i and j atoms */
1391             qq20             = _mm256_mul_ps(iq2,jq0);
1392
1393             /* REACTION-FIELD ELECTROSTATICS */
1394             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1395
1396             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1397
1398             fscal            = felec;
1399
1400             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1401
1402             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1403
1404             /* Calculate temporary vectorial force */
1405             tx               = _mm256_mul_ps(fscal,dx20);
1406             ty               = _mm256_mul_ps(fscal,dy20);
1407             tz               = _mm256_mul_ps(fscal,dz20);
1408
1409             /* Update vectorial force */
1410             fix2             = _mm256_add_ps(fix2,tx);
1411             fiy2             = _mm256_add_ps(fiy2,ty);
1412             fiz2             = _mm256_add_ps(fiz2,tz);
1413
1414             fjx0             = _mm256_add_ps(fjx0,tx);
1415             fjy0             = _mm256_add_ps(fjy0,ty);
1416             fjz0             = _mm256_add_ps(fjz0,tz);
1417
1418             }
1419
1420             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1421             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1422             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1423             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1424             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1425             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1426             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1427             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1428
1429             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1430
1431             /* Inner loop uses 121 flops */
1432         }
1433
1434         /* End of innermost loop */
1435
1436         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1437                                                  f+i_coord_offset,fshift+i_shift_offset);
1438
1439         /* Increment number of inner iterations */
1440         inneriter                  += j_index_end - j_index_start;
1441
1442         /* Outer loop uses 18 flops */
1443     }
1444
1445     /* Increment number of outer iterations */
1446     outeriter        += nri;
1447
1448     /* Update outer/inner flops */
1449
1450     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1451 }