src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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_ElecRF_VdwCSTab_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->ic->epsfac);
 128     charge           = mdatoms->chargeA;
 129     krf              = _mm256_set1_ps(fr->ic->k_rf);
 130     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 131     crf              = _mm256_set1_ps(fr->ic->c_rf);
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     vftab            = kernel_data->table_vdw->data;
 137     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 142     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 143     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 144     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152     j_coord_offsetE = 0;
 153     j_coord_offsetF = 0;
 154     j_coord_offsetG = 0;
 155     j_coord_offsetH = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm256_setzero_ps();
 184         fiy0             = _mm256_setzero_ps();
 185         fiz0             = _mm256_setzero_ps();
 186         fix1             = _mm256_setzero_ps();
 187         fiy1             = _mm256_setzero_ps();
 188         fiz1             = _mm256_setzero_ps();
 189         fix2             = _mm256_setzero_ps();
 190         fiy2             = _mm256_setzero_ps();
 191         fiz2             = _mm256_setzero_ps();
 192         fix3             = _mm256_setzero_ps();
 193         fiy3             = _mm256_setzero_ps();
 194         fiz3             = _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             dx30             = _mm256_sub_ps(ix3,jx0);
 240             dy30             = _mm256_sub_ps(iy3,jy0);
 241             dz30             = _mm256_sub_ps(iz3,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 245             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 246             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 247             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 248
 249             rinv00           = avx256_invsqrt_f(rsq00);
 250             rinv10           = avx256_invsqrt_f(rsq10);
 251             rinv20           = avx256_invsqrt_f(rsq20);
 252             rinv30           = avx256_invsqrt_f(rsq30);
 253
 254             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 255             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 256             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 260                                                                  charge+jnrC+0,charge+jnrD+0,
 261                                                                  charge+jnrE+0,charge+jnrF+0,
 262                                                                  charge+jnrG+0,charge+jnrH+0);
 263             vdwjidx0A        = 2*vdwtype[jnrA+0];
 264             vdwjidx0B        = 2*vdwtype[jnrB+0];
 265             vdwjidx0C        = 2*vdwtype[jnrC+0];
 266             vdwjidx0D        = 2*vdwtype[jnrD+0];
 267             vdwjidx0E        = 2*vdwtype[jnrE+0];
 268             vdwjidx0F        = 2*vdwtype[jnrF+0];
 269             vdwjidx0G        = 2*vdwtype[jnrG+0];
 270             vdwjidx0H        = 2*vdwtype[jnrH+0];
 271
 272             fjx0             = _mm256_setzero_ps();
 273             fjy0             = _mm256_setzero_ps();
 274             fjz0             = _mm256_setzero_ps();
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             r00              = _mm256_mul_ps(rsq00,rinv00);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 284                                             vdwioffsetptr0+vdwjidx0B,
 285                                             vdwioffsetptr0+vdwjidx0C,
 286                                             vdwioffsetptr0+vdwjidx0D,
 287                                             vdwioffsetptr0+vdwjidx0E,
 288                                             vdwioffsetptr0+vdwjidx0F,
 289                                             vdwioffsetptr0+vdwjidx0G,
 290                                             vdwioffsetptr0+vdwjidx0H,
 291                                             &c6_00,&c12_00);
 292
 293             /* Calculate table index by multiplying r with table scale and truncate to integer */
 294             rt               = _mm256_mul_ps(r00,vftabscale);
 295             vfitab           = _mm256_cvttps_epi32(rt);
 296             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 297             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 298             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 299             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 300             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 301             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 302
 303             /* CUBIC SPLINE TABLE DISPERSION */
 304             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 305                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 306             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 308             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 310             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 312             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 313             Heps             = _mm256_mul_ps(vfeps,H);
 314             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 315             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 316             vvdw6            = _mm256_mul_ps(c6_00,VV);
 317             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 318             fvdw6            = _mm256_mul_ps(c6_00,FF);
 319
 320             /* CUBIC SPLINE TABLE REPULSION */
 321             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 322             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 323             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 325             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 327             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 329             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 331             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 332             Heps             = _mm256_mul_ps(vfeps,H);
 333             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 334             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 335             vvdw12           = _mm256_mul_ps(c12_00,VV);
 336             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 337             fvdw12           = _mm256_mul_ps(c12_00,FF);
 338             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 339             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 343
 344             fscal            = fvdw;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm256_mul_ps(fscal,dx00);
 348             ty               = _mm256_mul_ps(fscal,dy00);
 349             tz               = _mm256_mul_ps(fscal,dz00);
 350
 351             /* Update vectorial force */
 352             fix0             = _mm256_add_ps(fix0,tx);
 353             fiy0             = _mm256_add_ps(fiy0,ty);
 354             fiz0             = _mm256_add_ps(fiz0,tz);
 355
 356             fjx0             = _mm256_add_ps(fjx0,tx);
 357             fjy0             = _mm256_add_ps(fjy0,ty);
 358             fjz0             = _mm256_add_ps(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq10             = _mm256_mul_ps(iq1,jq0);
 366
 367             /* REACTION-FIELD ELECTROSTATICS */
 368             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 369             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velecsum         = _mm256_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_ps(fscal,dx10);
 378             ty               = _mm256_mul_ps(fscal,dy10);
 379             tz               = _mm256_mul_ps(fscal,dz10);
 380
 381             /* Update vectorial force */
 382             fix1             = _mm256_add_ps(fix1,tx);
 383             fiy1             = _mm256_add_ps(fiy1,ty);
 384             fiz1             = _mm256_add_ps(fiz1,tz);
 385
 386             fjx0             = _mm256_add_ps(fjx0,tx);
 387             fjy0             = _mm256_add_ps(fjy0,ty);
 388             fjz0             = _mm256_add_ps(fjz0,tz);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             /* Compute parameters for interactions between i and j atoms */
 395             qq20             = _mm256_mul_ps(iq2,jq0);
 396
 397             /* REACTION-FIELD ELECTROSTATICS */
 398             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 399             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velecsum         = _mm256_add_ps(velecsum,velec);
 403
 404             fscal            = felec;
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm256_mul_ps(fscal,dx20);
 408             ty               = _mm256_mul_ps(fscal,dy20);
 409             tz               = _mm256_mul_ps(fscal,dz20);
 410
 411             /* Update vectorial force */
 412             fix2             = _mm256_add_ps(fix2,tx);
 413             fiy2             = _mm256_add_ps(fiy2,ty);
 414             fiz2             = _mm256_add_ps(fiz2,tz);
 415
 416             fjx0             = _mm256_add_ps(fjx0,tx);
 417             fjy0             = _mm256_add_ps(fjy0,ty);
 418             fjz0             = _mm256_add_ps(fjz0,tz);
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq30             = _mm256_mul_ps(iq3,jq0);
 426
 427             /* REACTION-FIELD ELECTROSTATICS */
 428             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 429             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velecsum         = _mm256_add_ps(velecsum,velec);
 433
 434             fscal            = felec;
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm256_mul_ps(fscal,dx30);
 438             ty               = _mm256_mul_ps(fscal,dy30);
 439             tz               = _mm256_mul_ps(fscal,dz30);
 440
 441             /* Update vectorial force */
 442             fix3             = _mm256_add_ps(fix3,tx);
 443             fiy3             = _mm256_add_ps(fiy3,ty);
 444             fiz3             = _mm256_add_ps(fiz3,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             fjptrA             = f+j_coord_offsetA;
 451             fjptrB             = f+j_coord_offsetB;
 452             fjptrC             = f+j_coord_offsetC;
 453             fjptrD             = f+j_coord_offsetD;
 454             fjptrE             = f+j_coord_offsetE;
 455             fjptrF             = f+j_coord_offsetF;
 456             fjptrG             = f+j_coord_offsetG;
 457             fjptrH             = f+j_coord_offsetH;
 458
 459             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 460
 461             /* Inner loop uses 155 flops */
 462         }
 463
 464         if(jidx<j_index_end)
 465         {
 466
 467             /* Get j neighbor index, and coordinate index */
 468             jnrlistA         = jjnr[jidx];
 469             jnrlistB         = jjnr[jidx+1];
 470             jnrlistC         = jjnr[jidx+2];
 471             jnrlistD         = jjnr[jidx+3];
 472             jnrlistE         = jjnr[jidx+4];
 473             jnrlistF         = jjnr[jidx+5];
 474             jnrlistG         = jjnr[jidx+6];
 475             jnrlistH         = jjnr[jidx+7];
 476             /* Sign of each element will be negative for non-real atoms.
 477              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 478              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 479              */
 480             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 481                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 482
 483             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 484             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 485             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 486             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 487             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 488             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 489             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 490             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 491             j_coord_offsetA  = DIM*jnrA;
 492             j_coord_offsetB  = DIM*jnrB;
 493             j_coord_offsetC  = DIM*jnrC;
 494             j_coord_offsetD  = DIM*jnrD;
 495             j_coord_offsetE  = DIM*jnrE;
 496             j_coord_offsetF  = DIM*jnrF;
 497             j_coord_offsetG  = DIM*jnrG;
 498             j_coord_offsetH  = DIM*jnrH;
 499
 500             /* load j atom coordinates */
 501             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 502                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 503                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 504                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 505                                                  &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm256_sub_ps(ix0,jx0);
 509             dy00             = _mm256_sub_ps(iy0,jy0);
 510             dz00             = _mm256_sub_ps(iz0,jz0);
 511             dx10             = _mm256_sub_ps(ix1,jx0);
 512             dy10             = _mm256_sub_ps(iy1,jy0);
 513             dz10             = _mm256_sub_ps(iz1,jz0);
 514             dx20             = _mm256_sub_ps(ix2,jx0);
 515             dy20             = _mm256_sub_ps(iy2,jy0);
 516             dz20             = _mm256_sub_ps(iz2,jz0);
 517             dx30             = _mm256_sub_ps(ix3,jx0);
 518             dy30             = _mm256_sub_ps(iy3,jy0);
 519             dz30             = _mm256_sub_ps(iz3,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 523             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 524             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 525             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 526
 527             rinv00           = avx256_invsqrt_f(rsq00);
 528             rinv10           = avx256_invsqrt_f(rsq10);
 529             rinv20           = avx256_invsqrt_f(rsq20);
 530             rinv30           = avx256_invsqrt_f(rsq30);
 531
 532             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 533             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 534             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 535
 536             /* Load parameters for j particles */
 537             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 538                                                                  charge+jnrC+0,charge+jnrD+0,
 539                                                                  charge+jnrE+0,charge+jnrF+0,
 540                                                                  charge+jnrG+0,charge+jnrH+0);
 541             vdwjidx0A        = 2*vdwtype[jnrA+0];
 542             vdwjidx0B        = 2*vdwtype[jnrB+0];
 543             vdwjidx0C        = 2*vdwtype[jnrC+0];
 544             vdwjidx0D        = 2*vdwtype[jnrD+0];
 545             vdwjidx0E        = 2*vdwtype[jnrE+0];
 546             vdwjidx0F        = 2*vdwtype[jnrF+0];
 547             vdwjidx0G        = 2*vdwtype[jnrG+0];
 548             vdwjidx0H        = 2*vdwtype[jnrH+0];
 549
 550             fjx0             = _mm256_setzero_ps();
 551             fjy0             = _mm256_setzero_ps();
 552             fjz0             = _mm256_setzero_ps();
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 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             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             /* CUBIC SPLINE TABLE DISPERSION */
 583             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 584                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 585             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 586                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 587             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 588                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 589             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 590                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 591             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 592             Heps             = _mm256_mul_ps(vfeps,H);
 593             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 594             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 595             vvdw6            = _mm256_mul_ps(c6_00,VV);
 596             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 597             fvdw6            = _mm256_mul_ps(c6_00,FF);
 598
 599             /* CUBIC SPLINE TABLE REPULSION */
 600             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 601             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 602             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 603                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 604             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 605                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 606             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 607                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 608             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 609                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 610             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 611             Heps             = _mm256_mul_ps(vfeps,H);
 612             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 613             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 614             vvdw12           = _mm256_mul_ps(c12_00,VV);
 615             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 616             fvdw12           = _mm256_mul_ps(c12_00,FF);
 617             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 618             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 622             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 623
 624             fscal            = fvdw;
 625
 626             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 627
 628             /* Calculate temporary vectorial force */
 629             tx               = _mm256_mul_ps(fscal,dx00);
 630             ty               = _mm256_mul_ps(fscal,dy00);
 631             tz               = _mm256_mul_ps(fscal,dz00);
 632
 633             /* Update vectorial force */
 634             fix0             = _mm256_add_ps(fix0,tx);
 635             fiy0             = _mm256_add_ps(fiy0,ty);
 636             fiz0             = _mm256_add_ps(fiz0,tz);
 637
 638             fjx0             = _mm256_add_ps(fjx0,tx);
 639             fjy0             = _mm256_add_ps(fjy0,ty);
 640             fjz0             = _mm256_add_ps(fjz0,tz);
 641
 642             /**************************
 643              * CALCULATE INTERACTIONS *
 644              **************************/
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq10             = _mm256_mul_ps(iq1,jq0);
 648
 649             /* REACTION-FIELD ELECTROSTATICS */
 650             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 651             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 652
 653             /* Update potential sum for this i atom from the interaction with this j atom. */
 654             velec            = _mm256_andnot_ps(dummy_mask,velec);
 655             velecsum         = _mm256_add_ps(velecsum,velec);
 656
 657             fscal            = felec;
 658
 659             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm256_mul_ps(fscal,dx10);
 663             ty               = _mm256_mul_ps(fscal,dy10);
 664             tz               = _mm256_mul_ps(fscal,dz10);
 665
 666             /* Update vectorial force */
 667             fix1             = _mm256_add_ps(fix1,tx);
 668             fiy1             = _mm256_add_ps(fiy1,ty);
 669             fiz1             = _mm256_add_ps(fiz1,tz);
 670
 671             fjx0             = _mm256_add_ps(fjx0,tx);
 672             fjy0             = _mm256_add_ps(fjy0,ty);
 673             fjz0             = _mm256_add_ps(fjz0,tz);
 674
 675             /**************************
 676              * CALCULATE INTERACTIONS *
 677              **************************/
 678
 679             /* Compute parameters for interactions between i and j atoms */
 680             qq20             = _mm256_mul_ps(iq2,jq0);
 681
 682             /* REACTION-FIELD ELECTROSTATICS */
 683             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 684             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 685
 686             /* Update potential sum for this i atom from the interaction with this j atom. */
 687             velec            = _mm256_andnot_ps(dummy_mask,velec);
 688             velecsum         = _mm256_add_ps(velecsum,velec);
 689
 690             fscal            = felec;
 691
 692             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm256_mul_ps(fscal,dx20);
 696             ty               = _mm256_mul_ps(fscal,dy20);
 697             tz               = _mm256_mul_ps(fscal,dz20);
 698
 699             /* Update vectorial force */
 700             fix2             = _mm256_add_ps(fix2,tx);
 701             fiy2             = _mm256_add_ps(fiy2,ty);
 702             fiz2             = _mm256_add_ps(fiz2,tz);
 703
 704             fjx0             = _mm256_add_ps(fjx0,tx);
 705             fjy0             = _mm256_add_ps(fjy0,ty);
 706             fjz0             = _mm256_add_ps(fjz0,tz);
 707
 708             /**************************
 709              * CALCULATE INTERACTIONS *
 710              **************************/
 711
 712             /* Compute parameters for interactions between i and j atoms */
 713             qq30             = _mm256_mul_ps(iq3,jq0);
 714
 715             /* REACTION-FIELD ELECTROSTATICS */
 716             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 717             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 718
 719             /* Update potential sum for this i atom from the interaction with this j atom. */
 720             velec            = _mm256_andnot_ps(dummy_mask,velec);
 721             velecsum         = _mm256_add_ps(velecsum,velec);
 722
 723             fscal            = felec;
 724
 725             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 726
 727             /* Calculate temporary vectorial force */
 728             tx               = _mm256_mul_ps(fscal,dx30);
 729             ty               = _mm256_mul_ps(fscal,dy30);
 730             tz               = _mm256_mul_ps(fscal,dz30);
 731
 732             /* Update vectorial force */
 733             fix3             = _mm256_add_ps(fix3,tx);
 734             fiy3             = _mm256_add_ps(fiy3,ty);
 735             fiz3             = _mm256_add_ps(fiz3,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 742             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 743             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 744             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 745             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 746             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 747             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 748             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 749
 750             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 751
 752             /* Inner loop uses 156 flops */
 753         }
 754
 755         /* End of innermost loop */
 756
 757         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 758                                                  f+i_coord_offset,fshift+i_shift_offset);
 759
 760         ggid                        = gid[iidx];
 761         /* Update potential energies */
 762         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 763         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 764
 765         /* Increment number of inner iterations */
 766         inneriter                  += j_index_end - j_index_start;
 767
 768         /* Outer loop uses 26 flops */
 769     }
 770
 771     /* Increment number of outer iterations */
 772     outeriter        += nri;
 773
 774     /* Update outer/inner flops */
 775
 776     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*156);
 777 }
 778 /*
 779  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_single
 780  * Electrostatics interaction: ReactionField
 781  * VdW interaction:            CubicSplineTable
 782  * Geometry:                   Water4-Particle
 783  * Calculate force/pot:        Force
 784  */
 785 void
 786 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_single
 787                     (t_nblist                    * gmx_restrict       nlist,
 788                      rvec                        * gmx_restrict          xx,
 789                      rvec                        * gmx_restrict          ff,
 790                      struct t_forcerec           * gmx_restrict          fr,
 791                      t_mdatoms                   * gmx_restrict     mdatoms,
 792                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 793                      t_nrnb                      * gmx_restrict        nrnb)
 794 {
 795     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 796      * just 0 for non-waters.
 797      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 798      * jnr indices corresponding to data put in the four positions in the SIMD register.
 799      */
 800     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 801     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 802     int              jnrA,jnrB,jnrC,jnrD;
 803     int              jnrE,jnrF,jnrG,jnrH;
 804     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 805     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 806     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 807     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 808     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 809     real             rcutoff_scalar;
 810     real             *shiftvec,*fshift,*x,*f;
 811     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 812     real             scratch[4*DIM];
 813     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 814     real *           vdwioffsetptr0;
 815     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 816     real *           vdwioffsetptr1;
 817     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 818     real *           vdwioffsetptr2;
 819     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 820     real *           vdwioffsetptr3;
 821     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 828     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 829     real             *charge;
 830     int              nvdwtype;
 831     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 832     int              *vdwtype;
 833     real             *vdwparam;
 834     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 835     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 836     __m256i          vfitab;
 837     __m128i          vfitab_lo,vfitab_hi;
 838     __m128i          ifour       = _mm_set1_epi32(4);
 839     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 840     real             *vftab;
 841     __m256           dummy_mask,cutoff_mask;
 842     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 843     __m256           one     = _mm256_set1_ps(1.0);
 844     __m256           two     = _mm256_set1_ps(2.0);
 845     x                = xx[0];
 846     f                = ff[0];
 847
 848     nri              = nlist->nri;
 849     iinr             = nlist->iinr;
 850     jindex           = nlist->jindex;
 851     jjnr             = nlist->jjnr;
 852     shiftidx         = nlist->shift;
 853     gid              = nlist->gid;
 854     shiftvec         = fr->shift_vec[0];
 855     fshift           = fr->fshift[0];
 856     facel            = _mm256_set1_ps(fr->ic->epsfac);
 857     charge           = mdatoms->chargeA;
 858     krf              = _mm256_set1_ps(fr->ic->k_rf);
 859     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 860     crf              = _mm256_set1_ps(fr->ic->c_rf);
 861     nvdwtype         = fr->ntype;
 862     vdwparam         = fr->nbfp;
 863     vdwtype          = mdatoms->typeA;
 864
 865     vftab            = kernel_data->table_vdw->data;
 866     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 867
 868     /* Setup water-specific parameters */
 869     inr              = nlist->iinr[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     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 873     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 874
 875     /* Avoid stupid compiler warnings */
 876     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 877     j_coord_offsetA = 0;
 878     j_coord_offsetB = 0;
 879     j_coord_offsetC = 0;
 880     j_coord_offsetD = 0;
 881     j_coord_offsetE = 0;
 882     j_coord_offsetF = 0;
 883     j_coord_offsetG = 0;
 884     j_coord_offsetH = 0;
 885
 886     outeriter        = 0;
 887     inneriter        = 0;
 888
 889     for(iidx=0;iidx<4*DIM;iidx++)
 890     {
 891         scratch[iidx] = 0.0;
 892     }
 893
 894     /* Start outer loop over neighborlists */
 895     for(iidx=0; iidx<nri; iidx++)
 896     {
 897         /* Load shift vector for this list */
 898         i_shift_offset   = DIM*shiftidx[iidx];
 899
 900         /* Load limits for loop over neighbors */
 901         j_index_start    = jindex[iidx];
 902         j_index_end      = jindex[iidx+1];
 903
 904         /* Get outer coordinate index */
 905         inr              = iinr[iidx];
 906         i_coord_offset   = DIM*inr;
 907
 908         /* Load i particle coords and add shift vector */
 909         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 910                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 911
 912         fix0             = _mm256_setzero_ps();
 913         fiy0             = _mm256_setzero_ps();
 914         fiz0             = _mm256_setzero_ps();
 915         fix1             = _mm256_setzero_ps();
 916         fiy1             = _mm256_setzero_ps();
 917         fiz1             = _mm256_setzero_ps();
 918         fix2             = _mm256_setzero_ps();
 919         fiy2             = _mm256_setzero_ps();
 920         fiz2             = _mm256_setzero_ps();
 921         fix3             = _mm256_setzero_ps();
 922         fiy3             = _mm256_setzero_ps();
 923         fiz3             = _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             dx30             = _mm256_sub_ps(ix3,jx0);
 965             dy30             = _mm256_sub_ps(iy3,jy0);
 966             dz30             = _mm256_sub_ps(iz3,jz0);
 967
 968             /* Calculate squared distance and things based on it */
 969             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 970             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 971             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 972             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 973
 974             rinv00           = avx256_invsqrt_f(rsq00);
 975             rinv10           = avx256_invsqrt_f(rsq10);
 976             rinv20           = avx256_invsqrt_f(rsq20);
 977             rinv30           = avx256_invsqrt_f(rsq30);
 978
 979             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 980             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 981             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 982
 983             /* Load parameters for j particles */
 984             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 985                                                                  charge+jnrC+0,charge+jnrD+0,
 986                                                                  charge+jnrE+0,charge+jnrF+0,
 987                                                                  charge+jnrG+0,charge+jnrH+0);
 988             vdwjidx0A        = 2*vdwtype[jnrA+0];
 989             vdwjidx0B        = 2*vdwtype[jnrB+0];
 990             vdwjidx0C        = 2*vdwtype[jnrC+0];
 991             vdwjidx0D        = 2*vdwtype[jnrD+0];
 992             vdwjidx0E        = 2*vdwtype[jnrE+0];
 993             vdwjidx0F        = 2*vdwtype[jnrF+0];
 994             vdwjidx0G        = 2*vdwtype[jnrG+0];
 995             vdwjidx0H        = 2*vdwtype[jnrH+0];
 996
 997             fjx0             = _mm256_setzero_ps();
 998             fjy0             = _mm256_setzero_ps();
 999             fjz0             = _mm256_setzero_ps();
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             r00              = _mm256_mul_ps(rsq00,rinv00);
1006
1007             /* Compute parameters for interactions between i and j atoms */
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             /* CUBIC SPLINE TABLE DISPERSION */
1029             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1030                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1031             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1032                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1033             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1034                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1035             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1036                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1037             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1038             Heps             = _mm256_mul_ps(vfeps,H);
1039             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1040             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1041             fvdw6            = _mm256_mul_ps(c6_00,FF);
1042
1043             /* CUBIC SPLINE TABLE REPULSION */
1044             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1045             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1046             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1047                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1048             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1049                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1050             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1051                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1052             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1053                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1054             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1055             Heps             = _mm256_mul_ps(vfeps,H);
1056             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1057             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1058             fvdw12           = _mm256_mul_ps(c12_00,FF);
1059             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1060
1061             fscal            = fvdw;
1062
1063             /* Calculate temporary vectorial force */
1064             tx               = _mm256_mul_ps(fscal,dx00);
1065             ty               = _mm256_mul_ps(fscal,dy00);
1066             tz               = _mm256_mul_ps(fscal,dz00);
1067
1068             /* Update vectorial force */
1069             fix0             = _mm256_add_ps(fix0,tx);
1070             fiy0             = _mm256_add_ps(fiy0,ty);
1071             fiz0             = _mm256_add_ps(fiz0,tz);
1072
1073             fjx0             = _mm256_add_ps(fjx0,tx);
1074             fjy0             = _mm256_add_ps(fjy0,ty);
1075             fjz0             = _mm256_add_ps(fjz0,tz);
1076
1077             /**************************
1078              * CALCULATE INTERACTIONS *
1079              **************************/
1080
1081             /* Compute parameters for interactions between i and j atoms */
1082             qq10             = _mm256_mul_ps(iq1,jq0);
1083
1084             /* REACTION-FIELD ELECTROSTATICS */
1085             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1086
1087             fscal            = felec;
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm256_mul_ps(fscal,dx10);
1091             ty               = _mm256_mul_ps(fscal,dy10);
1092             tz               = _mm256_mul_ps(fscal,dz10);
1093
1094             /* Update vectorial force */
1095             fix1             = _mm256_add_ps(fix1,tx);
1096             fiy1             = _mm256_add_ps(fiy1,ty);
1097             fiz1             = _mm256_add_ps(fiz1,tz);
1098
1099             fjx0             = _mm256_add_ps(fjx0,tx);
1100             fjy0             = _mm256_add_ps(fjy0,ty);
1101             fjz0             = _mm256_add_ps(fjz0,tz);
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             /* Compute parameters for interactions between i and j atoms */
1108             qq20             = _mm256_mul_ps(iq2,jq0);
1109
1110             /* REACTION-FIELD ELECTROSTATICS */
1111             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1112
1113             fscal            = felec;
1114
1115             /* Calculate temporary vectorial force */
1116             tx               = _mm256_mul_ps(fscal,dx20);
1117             ty               = _mm256_mul_ps(fscal,dy20);
1118             tz               = _mm256_mul_ps(fscal,dz20);
1119
1120             /* Update vectorial force */
1121             fix2             = _mm256_add_ps(fix2,tx);
1122             fiy2             = _mm256_add_ps(fiy2,ty);
1123             fiz2             = _mm256_add_ps(fiz2,tz);
1124
1125             fjx0             = _mm256_add_ps(fjx0,tx);
1126             fjy0             = _mm256_add_ps(fjy0,ty);
1127             fjz0             = _mm256_add_ps(fjz0,tz);
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             /* Compute parameters for interactions between i and j atoms */
1134             qq30             = _mm256_mul_ps(iq3,jq0);
1135
1136             /* REACTION-FIELD ELECTROSTATICS */
1137             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1138
1139             fscal            = felec;
1140
1141             /* Calculate temporary vectorial force */
1142             tx               = _mm256_mul_ps(fscal,dx30);
1143             ty               = _mm256_mul_ps(fscal,dy30);
1144             tz               = _mm256_mul_ps(fscal,dz30);
1145
1146             /* Update vectorial force */
1147             fix3             = _mm256_add_ps(fix3,tx);
1148             fiy3             = _mm256_add_ps(fiy3,ty);
1149             fiz3             = _mm256_add_ps(fiz3,tz);
1150
1151             fjx0             = _mm256_add_ps(fjx0,tx);
1152             fjy0             = _mm256_add_ps(fjy0,ty);
1153             fjz0             = _mm256_add_ps(fjz0,tz);
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 132 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             dx30             = _mm256_sub_ps(ix3,jx0);
1223             dy30             = _mm256_sub_ps(iy3,jy0);
1224             dz30             = _mm256_sub_ps(iz3,jz0);
1225
1226             /* Calculate squared distance and things based on it */
1227             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1228             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1229             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1230             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1231
1232             rinv00           = avx256_invsqrt_f(rsq00);
1233             rinv10           = avx256_invsqrt_f(rsq10);
1234             rinv20           = avx256_invsqrt_f(rsq20);
1235             rinv30           = avx256_invsqrt_f(rsq30);
1236
1237             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1238             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1239             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1240
1241             /* Load parameters for j particles */
1242             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1243                                                                  charge+jnrC+0,charge+jnrD+0,
1244                                                                  charge+jnrE+0,charge+jnrF+0,
1245                                                                  charge+jnrG+0,charge+jnrH+0);
1246             vdwjidx0A        = 2*vdwtype[jnrA+0];
1247             vdwjidx0B        = 2*vdwtype[jnrB+0];
1248             vdwjidx0C        = 2*vdwtype[jnrC+0];
1249             vdwjidx0D        = 2*vdwtype[jnrD+0];
1250             vdwjidx0E        = 2*vdwtype[jnrE+0];
1251             vdwjidx0F        = 2*vdwtype[jnrF+0];
1252             vdwjidx0G        = 2*vdwtype[jnrG+0];
1253             vdwjidx0H        = 2*vdwtype[jnrH+0];
1254
1255             fjx0             = _mm256_setzero_ps();
1256             fjy0             = _mm256_setzero_ps();
1257             fjz0             = _mm256_setzero_ps();
1258
1259             /**************************
1260              * CALCULATE INTERACTIONS *
1261              **************************/
1262
1263             r00              = _mm256_mul_ps(rsq00,rinv00);
1264             r00              = _mm256_andnot_ps(dummy_mask,r00);
1265
1266             /* Compute parameters for interactions between i and j atoms */
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             /* CUBIC SPLINE TABLE DISPERSION */
1288             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1289                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1290             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1292             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1294             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1296             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1297             Heps             = _mm256_mul_ps(vfeps,H);
1298             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1299             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1300             fvdw6            = _mm256_mul_ps(c6_00,FF);
1301
1302             /* CUBIC SPLINE TABLE REPULSION */
1303             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1304             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1305             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1307             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1309             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1311             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1313             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1314             Heps             = _mm256_mul_ps(vfeps,H);
1315             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1316             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1317             fvdw12           = _mm256_mul_ps(c12_00,FF);
1318             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1319
1320             fscal            = fvdw;
1321
1322             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1323
1324             /* Calculate temporary vectorial force */
1325             tx               = _mm256_mul_ps(fscal,dx00);
1326             ty               = _mm256_mul_ps(fscal,dy00);
1327             tz               = _mm256_mul_ps(fscal,dz00);
1328
1329             /* Update vectorial force */
1330             fix0             = _mm256_add_ps(fix0,tx);
1331             fiy0             = _mm256_add_ps(fiy0,ty);
1332             fiz0             = _mm256_add_ps(fiz0,tz);
1333
1334             fjx0             = _mm256_add_ps(fjx0,tx);
1335             fjy0             = _mm256_add_ps(fjy0,ty);
1336             fjz0             = _mm256_add_ps(fjz0,tz);
1337
1338             /**************************
1339              * CALCULATE INTERACTIONS *
1340              **************************/
1341
1342             /* Compute parameters for interactions between i and j atoms */
1343             qq10             = _mm256_mul_ps(iq1,jq0);
1344
1345             /* REACTION-FIELD ELECTROSTATICS */
1346             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1347
1348             fscal            = felec;
1349
1350             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1351
1352             /* Calculate temporary vectorial force */
1353             tx               = _mm256_mul_ps(fscal,dx10);
1354             ty               = _mm256_mul_ps(fscal,dy10);
1355             tz               = _mm256_mul_ps(fscal,dz10);
1356
1357             /* Update vectorial force */
1358             fix1             = _mm256_add_ps(fix1,tx);
1359             fiy1             = _mm256_add_ps(fiy1,ty);
1360             fiz1             = _mm256_add_ps(fiz1,tz);
1361
1362             fjx0             = _mm256_add_ps(fjx0,tx);
1363             fjy0             = _mm256_add_ps(fjy0,ty);
1364             fjz0             = _mm256_add_ps(fjz0,tz);
1365
1366             /**************************
1367              * CALCULATE INTERACTIONS *
1368              **************************/
1369
1370             /* Compute parameters for interactions between i and j atoms */
1371             qq20             = _mm256_mul_ps(iq2,jq0);
1372
1373             /* REACTION-FIELD ELECTROSTATICS */
1374             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1375
1376             fscal            = felec;
1377
1378             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1379
1380             /* Calculate temporary vectorial force */
1381             tx               = _mm256_mul_ps(fscal,dx20);
1382             ty               = _mm256_mul_ps(fscal,dy20);
1383             tz               = _mm256_mul_ps(fscal,dz20);
1384
1385             /* Update vectorial force */
1386             fix2             = _mm256_add_ps(fix2,tx);
1387             fiy2             = _mm256_add_ps(fiy2,ty);
1388             fiz2             = _mm256_add_ps(fiz2,tz);
1389
1390             fjx0             = _mm256_add_ps(fjx0,tx);
1391             fjy0             = _mm256_add_ps(fjy0,ty);
1392             fjz0             = _mm256_add_ps(fjz0,tz);
1393
1394             /**************************
1395              * CALCULATE INTERACTIONS *
1396              **************************/
1397
1398             /* Compute parameters for interactions between i and j atoms */
1399             qq30             = _mm256_mul_ps(iq3,jq0);
1400
1401             /* REACTION-FIELD ELECTROSTATICS */
1402             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1403
1404             fscal            = felec;
1405
1406             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1407
1408             /* Calculate temporary vectorial force */
1409             tx               = _mm256_mul_ps(fscal,dx30);
1410             ty               = _mm256_mul_ps(fscal,dy30);
1411             tz               = _mm256_mul_ps(fscal,dz30);
1412
1413             /* Update vectorial force */
1414             fix3             = _mm256_add_ps(fix3,tx);
1415             fiy3             = _mm256_add_ps(fiy3,ty);
1416             fiz3             = _mm256_add_ps(fiz3,tz);
1417
1418             fjx0             = _mm256_add_ps(fjx0,tx);
1419             fjy0             = _mm256_add_ps(fjy0,ty);
1420             fjz0             = _mm256_add_ps(fjz0,tz);
1421
1422             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1423             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1424             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1425             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1426             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1427             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1428             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1429             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1430
1431             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1432
1433             /* Inner loop uses 133 flops */
1434         }
1435
1436         /* End of innermost loop */
1437
1438         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1439                                                  f+i_coord_offset,fshift+i_shift_offset);
1440
1441         /* Increment number of inner iterations */
1442         inneriter                  += j_index_end - j_index_start;
1443
1444         /* Outer loop uses 24 flops */
1445     }
1446
1447     /* Increment number of outer iterations */
1448     outeriter        += nri;
1449
1450     /* Update outer/inner flops */
1451
1452     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*133);
1453 }