src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256i          vfitab;
 109     __m128i          vfitab_lo,vfitab_hi;
 110     __m128i          ifour       = _mm_set1_epi32(4);
 111     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 112     real             *vftab;
 113     __m256           dummy_mask,cutoff_mask;
 114     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 115     __m256           one     = _mm256_set1_ps(1.0);
 116     __m256           two     = _mm256_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_vdw->data;
 135     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 140     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 141     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm256_setzero_ps();
 182         fiy0             = _mm256_setzero_ps();
 183         fiz0             = _mm256_setzero_ps();
 184         fix1             = _mm256_setzero_ps();
 185         fiy1             = _mm256_setzero_ps();
 186         fiz1             = _mm256_setzero_ps();
 187         fix2             = _mm256_setzero_ps();
 188         fiy2             = _mm256_setzero_ps();
 189         fiz2             = _mm256_setzero_ps();
 190         fix3             = _mm256_setzero_ps();
 191         fiy3             = _mm256_setzero_ps();
 192         fiz3             = _mm256_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_ps();
 196         vvdwsum          = _mm256_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             jnrE             = jjnr[jidx+4];
 208             jnrF             = jjnr[jidx+5];
 209             jnrG             = jjnr[jidx+6];
 210             jnrH             = jjnr[jidx+7];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215             j_coord_offsetE  = DIM*jnrE;
 216             j_coord_offsetF  = DIM*jnrF;
 217             j_coord_offsetG  = DIM*jnrG;
 218             j_coord_offsetH  = DIM*jnrH;
 219
 220             /* load j atom coordinates */
 221             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 223                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 224                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 225                                                  &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm256_sub_ps(ix0,jx0);
 229             dy00             = _mm256_sub_ps(iy0,jy0);
 230             dz00             = _mm256_sub_ps(iz0,jz0);
 231             dx10             = _mm256_sub_ps(ix1,jx0);
 232             dy10             = _mm256_sub_ps(iy1,jy0);
 233             dz10             = _mm256_sub_ps(iz1,jz0);
 234             dx20             = _mm256_sub_ps(ix2,jx0);
 235             dy20             = _mm256_sub_ps(iy2,jy0);
 236             dz20             = _mm256_sub_ps(iz2,jz0);
 237             dx30             = _mm256_sub_ps(ix3,jx0);
 238             dy30             = _mm256_sub_ps(iy3,jy0);
 239             dz30             = _mm256_sub_ps(iz3,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 243             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 244             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 245             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 246
 247             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 248             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 249             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 250             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 251
 252             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 253             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 254             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 258                                                                  charge+jnrC+0,charge+jnrD+0,
 259                                                                  charge+jnrE+0,charge+jnrF+0,
 260                                                                  charge+jnrG+0,charge+jnrH+0);
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262             vdwjidx0B        = 2*vdwtype[jnrB+0];
 263             vdwjidx0C        = 2*vdwtype[jnrC+0];
 264             vdwjidx0D        = 2*vdwtype[jnrD+0];
 265             vdwjidx0E        = 2*vdwtype[jnrE+0];
 266             vdwjidx0F        = 2*vdwtype[jnrF+0];
 267             vdwjidx0G        = 2*vdwtype[jnrG+0];
 268             vdwjidx0H        = 2*vdwtype[jnrH+0];
 269
 270             fjx0             = _mm256_setzero_ps();
 271             fjy0             = _mm256_setzero_ps();
 272             fjz0             = _mm256_setzero_ps();
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             r00              = _mm256_mul_ps(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 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             /* CUBIC SPLINE TABLE DISPERSION */
 302             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 303                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 304             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 305                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 306             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 308             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 310             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 311             Heps             = _mm256_mul_ps(vfeps,H);
 312             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 313             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 314             vvdw6            = _mm256_mul_ps(c6_00,VV);
 315             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 316             fvdw6            = _mm256_mul_ps(c6_00,FF);
 317
 318             /* CUBIC SPLINE TABLE REPULSION */
 319             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 320             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 321             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 323             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 325             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 327             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 329             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 330             Heps             = _mm256_mul_ps(vfeps,H);
 331             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 332             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 333             vvdw12           = _mm256_mul_ps(c12_00,VV);
 334             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 335             fvdw12           = _mm256_mul_ps(c12_00,FF);
 336             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 337             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 341
 342             fscal            = fvdw;
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm256_mul_ps(fscal,dx00);
 346             ty               = _mm256_mul_ps(fscal,dy00);
 347             tz               = _mm256_mul_ps(fscal,dz00);
 348
 349             /* Update vectorial force */
 350             fix0             = _mm256_add_ps(fix0,tx);
 351             fiy0             = _mm256_add_ps(fiy0,ty);
 352             fiz0             = _mm256_add_ps(fiz0,tz);
 353
 354             fjx0             = _mm256_add_ps(fjx0,tx);
 355             fjy0             = _mm256_add_ps(fjy0,ty);
 356             fjz0             = _mm256_add_ps(fjz0,tz);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq10             = _mm256_mul_ps(iq1,jq0);
 364
 365             /* COULOMB ELECTROSTATICS */
 366             velec            = _mm256_mul_ps(qq10,rinv10);
 367             felec            = _mm256_mul_ps(velec,rinvsq10);
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velecsum         = _mm256_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm256_mul_ps(fscal,dx10);
 376             ty               = _mm256_mul_ps(fscal,dy10);
 377             tz               = _mm256_mul_ps(fscal,dz10);
 378
 379             /* Update vectorial force */
 380             fix1             = _mm256_add_ps(fix1,tx);
 381             fiy1             = _mm256_add_ps(fiy1,ty);
 382             fiz1             = _mm256_add_ps(fiz1,tz);
 383
 384             fjx0             = _mm256_add_ps(fjx0,tx);
 385             fjy0             = _mm256_add_ps(fjy0,ty);
 386             fjz0             = _mm256_add_ps(fjz0,tz);
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq20             = _mm256_mul_ps(iq2,jq0);
 394
 395             /* COULOMB ELECTROSTATICS */
 396             velec            = _mm256_mul_ps(qq20,rinv20);
 397             felec            = _mm256_mul_ps(velec,rinvsq20);
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velecsum         = _mm256_add_ps(velecsum,velec);
 401
 402             fscal            = felec;
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm256_mul_ps(fscal,dx20);
 406             ty               = _mm256_mul_ps(fscal,dy20);
 407             tz               = _mm256_mul_ps(fscal,dz20);
 408
 409             /* Update vectorial force */
 410             fix2             = _mm256_add_ps(fix2,tx);
 411             fiy2             = _mm256_add_ps(fiy2,ty);
 412             fiz2             = _mm256_add_ps(fiz2,tz);
 413
 414             fjx0             = _mm256_add_ps(fjx0,tx);
 415             fjy0             = _mm256_add_ps(fjy0,ty);
 416             fjz0             = _mm256_add_ps(fjz0,tz);
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             /* Compute parameters for interactions between i and j atoms */
 423             qq30             = _mm256_mul_ps(iq3,jq0);
 424
 425             /* COULOMB ELECTROSTATICS */
 426             velec            = _mm256_mul_ps(qq30,rinv30);
 427             felec            = _mm256_mul_ps(velec,rinvsq30);
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velecsum         = _mm256_add_ps(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm256_mul_ps(fscal,dx30);
 436             ty               = _mm256_mul_ps(fscal,dy30);
 437             tz               = _mm256_mul_ps(fscal,dz30);
 438
 439             /* Update vectorial force */
 440             fix3             = _mm256_add_ps(fix3,tx);
 441             fiy3             = _mm256_add_ps(fiy3,ty);
 442             fiz3             = _mm256_add_ps(fiz3,tz);
 443
 444             fjx0             = _mm256_add_ps(fjx0,tx);
 445             fjy0             = _mm256_add_ps(fjy0,ty);
 446             fjz0             = _mm256_add_ps(fjz0,tz);
 447
 448             fjptrA             = f+j_coord_offsetA;
 449             fjptrB             = f+j_coord_offsetB;
 450             fjptrC             = f+j_coord_offsetC;
 451             fjptrD             = f+j_coord_offsetD;
 452             fjptrE             = f+j_coord_offsetE;
 453             fjptrF             = f+j_coord_offsetF;
 454             fjptrG             = f+j_coord_offsetG;
 455             fjptrH             = f+j_coord_offsetH;
 456
 457             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 458
 459             /* Inner loop uses 140 flops */
 460         }
 461
 462         if(jidx<j_index_end)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrlistA         = jjnr[jidx];
 467             jnrlistB         = jjnr[jidx+1];
 468             jnrlistC         = jjnr[jidx+2];
 469             jnrlistD         = jjnr[jidx+3];
 470             jnrlistE         = jjnr[jidx+4];
 471             jnrlistF         = jjnr[jidx+5];
 472             jnrlistG         = jjnr[jidx+6];
 473             jnrlistH         = jjnr[jidx+7];
 474             /* Sign of each element will be negative for non-real atoms.
 475              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 476              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 477              */
 478             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 479                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 480
 481             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 482             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 483             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 484             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 485             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 486             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 487             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 488             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 489             j_coord_offsetA  = DIM*jnrA;
 490             j_coord_offsetB  = DIM*jnrB;
 491             j_coord_offsetC  = DIM*jnrC;
 492             j_coord_offsetD  = DIM*jnrD;
 493             j_coord_offsetE  = DIM*jnrE;
 494             j_coord_offsetF  = DIM*jnrF;
 495             j_coord_offsetG  = DIM*jnrG;
 496             j_coord_offsetH  = DIM*jnrH;
 497
 498             /* load j atom coordinates */
 499             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 500                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 501                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 502                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 503                                                  &jx0,&jy0,&jz0);
 504
 505             /* Calculate displacement vector */
 506             dx00             = _mm256_sub_ps(ix0,jx0);
 507             dy00             = _mm256_sub_ps(iy0,jy0);
 508             dz00             = _mm256_sub_ps(iz0,jz0);
 509             dx10             = _mm256_sub_ps(ix1,jx0);
 510             dy10             = _mm256_sub_ps(iy1,jy0);
 511             dz10             = _mm256_sub_ps(iz1,jz0);
 512             dx20             = _mm256_sub_ps(ix2,jx0);
 513             dy20             = _mm256_sub_ps(iy2,jy0);
 514             dz20             = _mm256_sub_ps(iz2,jz0);
 515             dx30             = _mm256_sub_ps(ix3,jx0);
 516             dy30             = _mm256_sub_ps(iy3,jy0);
 517             dz30             = _mm256_sub_ps(iz3,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             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 524
 525             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 526             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 527             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 528             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 529
 530             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 531             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 532             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 533
 534             /* Load parameters for j particles */
 535             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 536                                                                  charge+jnrC+0,charge+jnrD+0,
 537                                                                  charge+jnrE+0,charge+jnrF+0,
 538                                                                  charge+jnrG+0,charge+jnrH+0);
 539             vdwjidx0A        = 2*vdwtype[jnrA+0];
 540             vdwjidx0B        = 2*vdwtype[jnrB+0];
 541             vdwjidx0C        = 2*vdwtype[jnrC+0];
 542             vdwjidx0D        = 2*vdwtype[jnrD+0];
 543             vdwjidx0E        = 2*vdwtype[jnrE+0];
 544             vdwjidx0F        = 2*vdwtype[jnrF+0];
 545             vdwjidx0G        = 2*vdwtype[jnrG+0];
 546             vdwjidx0H        = 2*vdwtype[jnrH+0];
 547
 548             fjx0             = _mm256_setzero_ps();
 549             fjy0             = _mm256_setzero_ps();
 550             fjz0             = _mm256_setzero_ps();
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             r00              = _mm256_mul_ps(rsq00,rinv00);
 557             r00              = _mm256_andnot_ps(dummy_mask,r00);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 561                                             vdwioffsetptr0+vdwjidx0B,
 562                                             vdwioffsetptr0+vdwjidx0C,
 563                                             vdwioffsetptr0+vdwjidx0D,
 564                                             vdwioffsetptr0+vdwjidx0E,
 565                                             vdwioffsetptr0+vdwjidx0F,
 566                                             vdwioffsetptr0+vdwjidx0G,
 567                                             vdwioffsetptr0+vdwjidx0H,
 568                                             &c6_00,&c12_00);
 569
 570             /* Calculate table index by multiplying r with table scale and truncate to integer */
 571             rt               = _mm256_mul_ps(r00,vftabscale);
 572             vfitab           = _mm256_cvttps_epi32(rt);
 573             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 574             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 575             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 576             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 577             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 578             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 579
 580             /* CUBIC SPLINE TABLE DISPERSION */
 581             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 582                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 583             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 584                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 585             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 586                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 587             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 588                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 589             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 590             Heps             = _mm256_mul_ps(vfeps,H);
 591             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 592             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 593             vvdw6            = _mm256_mul_ps(c6_00,VV);
 594             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 595             fvdw6            = _mm256_mul_ps(c6_00,FF);
 596
 597             /* CUBIC SPLINE TABLE REPULSION */
 598             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 599             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 600             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 601                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 602             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 603                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 604             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 605                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 606             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 607                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 608             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 609             Heps             = _mm256_mul_ps(vfeps,H);
 610             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 611             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 612             vvdw12           = _mm256_mul_ps(c12_00,VV);
 613             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 614             fvdw12           = _mm256_mul_ps(c12_00,FF);
 615             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 616             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 620             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 621
 622             fscal            = fvdw;
 623
 624             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm256_mul_ps(fscal,dx00);
 628             ty               = _mm256_mul_ps(fscal,dy00);
 629             tz               = _mm256_mul_ps(fscal,dz00);
 630
 631             /* Update vectorial force */
 632             fix0             = _mm256_add_ps(fix0,tx);
 633             fiy0             = _mm256_add_ps(fiy0,ty);
 634             fiz0             = _mm256_add_ps(fiz0,tz);
 635
 636             fjx0             = _mm256_add_ps(fjx0,tx);
 637             fjy0             = _mm256_add_ps(fjy0,ty);
 638             fjz0             = _mm256_add_ps(fjz0,tz);
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             /* Compute parameters for interactions between i and j atoms */
 645             qq10             = _mm256_mul_ps(iq1,jq0);
 646
 647             /* COULOMB ELECTROSTATICS */
 648             velec            = _mm256_mul_ps(qq10,rinv10);
 649             felec            = _mm256_mul_ps(velec,rinvsq10);
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             velec            = _mm256_andnot_ps(dummy_mask,velec);
 653             velecsum         = _mm256_add_ps(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm256_mul_ps(fscal,dx10);
 661             ty               = _mm256_mul_ps(fscal,dy10);
 662             tz               = _mm256_mul_ps(fscal,dz10);
 663
 664             /* Update vectorial force */
 665             fix1             = _mm256_add_ps(fix1,tx);
 666             fiy1             = _mm256_add_ps(fiy1,ty);
 667             fiz1             = _mm256_add_ps(fiz1,tz);
 668
 669             fjx0             = _mm256_add_ps(fjx0,tx);
 670             fjy0             = _mm256_add_ps(fjy0,ty);
 671             fjz0             = _mm256_add_ps(fjz0,tz);
 672
 673             /**************************
 674              * CALCULATE INTERACTIONS *
 675              **************************/
 676
 677             /* Compute parameters for interactions between i and j atoms */
 678             qq20             = _mm256_mul_ps(iq2,jq0);
 679
 680             /* COULOMB ELECTROSTATICS */
 681             velec            = _mm256_mul_ps(qq20,rinv20);
 682             felec            = _mm256_mul_ps(velec,rinvsq20);
 683
 684             /* Update potential sum for this i atom from the interaction with this j atom. */
 685             velec            = _mm256_andnot_ps(dummy_mask,velec);
 686             velecsum         = _mm256_add_ps(velecsum,velec);
 687
 688             fscal            = felec;
 689
 690             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 691
 692             /* Calculate temporary vectorial force */
 693             tx               = _mm256_mul_ps(fscal,dx20);
 694             ty               = _mm256_mul_ps(fscal,dy20);
 695             tz               = _mm256_mul_ps(fscal,dz20);
 696
 697             /* Update vectorial force */
 698             fix2             = _mm256_add_ps(fix2,tx);
 699             fiy2             = _mm256_add_ps(fiy2,ty);
 700             fiz2             = _mm256_add_ps(fiz2,tz);
 701
 702             fjx0             = _mm256_add_ps(fjx0,tx);
 703             fjy0             = _mm256_add_ps(fjy0,ty);
 704             fjz0             = _mm256_add_ps(fjz0,tz);
 705
 706             /**************************
 707              * CALCULATE INTERACTIONS *
 708              **************************/
 709
 710             /* Compute parameters for interactions between i and j atoms */
 711             qq30             = _mm256_mul_ps(iq3,jq0);
 712
 713             /* COULOMB ELECTROSTATICS */
 714             velec            = _mm256_mul_ps(qq30,rinv30);
 715             felec            = _mm256_mul_ps(velec,rinvsq30);
 716
 717             /* Update potential sum for this i atom from the interaction with this j atom. */
 718             velec            = _mm256_andnot_ps(dummy_mask,velec);
 719             velecsum         = _mm256_add_ps(velecsum,velec);
 720
 721             fscal            = felec;
 722
 723             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 724
 725             /* Calculate temporary vectorial force */
 726             tx               = _mm256_mul_ps(fscal,dx30);
 727             ty               = _mm256_mul_ps(fscal,dy30);
 728             tz               = _mm256_mul_ps(fscal,dz30);
 729
 730             /* Update vectorial force */
 731             fix3             = _mm256_add_ps(fix3,tx);
 732             fiy3             = _mm256_add_ps(fiy3,ty);
 733             fiz3             = _mm256_add_ps(fiz3,tz);
 734
 735             fjx0             = _mm256_add_ps(fjx0,tx);
 736             fjy0             = _mm256_add_ps(fjy0,ty);
 737             fjz0             = _mm256_add_ps(fjz0,tz);
 738
 739             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 740             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 741             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 742             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 743             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 744             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 745             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 746             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 747
 748             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 749
 750             /* Inner loop uses 141 flops */
 751         }
 752
 753         /* End of innermost loop */
 754
 755         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 756                                                  f+i_coord_offset,fshift+i_shift_offset);
 757
 758         ggid                        = gid[iidx];
 759         /* Update potential energies */
 760         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 761         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 762
 763         /* Increment number of inner iterations */
 764         inneriter                  += j_index_end - j_index_start;
 765
 766         /* Outer loop uses 26 flops */
 767     }
 768
 769     /* Increment number of outer iterations */
 770     outeriter        += nri;
 771
 772     /* Update outer/inner flops */
 773
 774     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 775 }
 776 /*
 777  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_single
 778  * Electrostatics interaction: Coulomb
 779  * VdW interaction:            CubicSplineTable
 780  * Geometry:                   Water4-Particle
 781  * Calculate force/pot:        Force
 782  */
 783 void
 784 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_single
 785                     (t_nblist                    * gmx_restrict       nlist,
 786                      rvec                        * gmx_restrict          xx,
 787                      rvec                        * gmx_restrict          ff,
 788                      t_forcerec                  * gmx_restrict          fr,
 789                      t_mdatoms                   * gmx_restrict     mdatoms,
 790                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 791                      t_nrnb                      * gmx_restrict        nrnb)
 792 {
 793     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 794      * just 0 for non-waters.
 795      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 796      * jnr indices corresponding to data put in the four positions in the SIMD register.
 797      */
 798     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 799     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 800     int              jnrA,jnrB,jnrC,jnrD;
 801     int              jnrE,jnrF,jnrG,jnrH;
 802     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 803     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 804     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 805     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 806     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 807     real             rcutoff_scalar;
 808     real             *shiftvec,*fshift,*x,*f;
 809     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 810     real             scratch[4*DIM];
 811     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 812     real *           vdwioffsetptr0;
 813     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 814     real *           vdwioffsetptr1;
 815     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 816     real *           vdwioffsetptr2;
 817     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 818     real *           vdwioffsetptr3;
 819     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 820     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 821     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 822     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 823     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 824     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 825     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 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->epsfac);
 855     charge           = mdatoms->chargeA;
 856     nvdwtype         = fr->ntype;
 857     vdwparam         = fr->nbfp;
 858     vdwtype          = mdatoms->typeA;
 859
 860     vftab            = kernel_data->table_vdw->data;
 861     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 862
 863     /* Setup water-specific parameters */
 864     inr              = nlist->iinr[0];
 865     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 866     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 867     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 868     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 869
 870     /* Avoid stupid compiler warnings */
 871     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 872     j_coord_offsetA = 0;
 873     j_coord_offsetB = 0;
 874     j_coord_offsetC = 0;
 875     j_coord_offsetD = 0;
 876     j_coord_offsetE = 0;
 877     j_coord_offsetF = 0;
 878     j_coord_offsetG = 0;
 879     j_coord_offsetH = 0;
 880
 881     outeriter        = 0;
 882     inneriter        = 0;
 883
 884     for(iidx=0;iidx<4*DIM;iidx++)
 885     {
 886         scratch[iidx] = 0.0;
 887     }
 888
 889     /* Start outer loop over neighborlists */
 890     for(iidx=0; iidx<nri; iidx++)
 891     {
 892         /* Load shift vector for this list */
 893         i_shift_offset   = DIM*shiftidx[iidx];
 894
 895         /* Load limits for loop over neighbors */
 896         j_index_start    = jindex[iidx];
 897         j_index_end      = jindex[iidx+1];
 898
 899         /* Get outer coordinate index */
 900         inr              = iinr[iidx];
 901         i_coord_offset   = DIM*inr;
 902
 903         /* Load i particle coords and add shift vector */
 904         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 905                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 906
 907         fix0             = _mm256_setzero_ps();
 908         fiy0             = _mm256_setzero_ps();
 909         fiz0             = _mm256_setzero_ps();
 910         fix1             = _mm256_setzero_ps();
 911         fiy1             = _mm256_setzero_ps();
 912         fiz1             = _mm256_setzero_ps();
 913         fix2             = _mm256_setzero_ps();
 914         fiy2             = _mm256_setzero_ps();
 915         fiz2             = _mm256_setzero_ps();
 916         fix3             = _mm256_setzero_ps();
 917         fiy3             = _mm256_setzero_ps();
 918         fiz3             = _mm256_setzero_ps();
 919
 920         /* Start inner kernel loop */
 921         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 922         {
 923
 924             /* Get j neighbor index, and coordinate index */
 925             jnrA             = jjnr[jidx];
 926             jnrB             = jjnr[jidx+1];
 927             jnrC             = jjnr[jidx+2];
 928             jnrD             = jjnr[jidx+3];
 929             jnrE             = jjnr[jidx+4];
 930             jnrF             = jjnr[jidx+5];
 931             jnrG             = jjnr[jidx+6];
 932             jnrH             = jjnr[jidx+7];
 933             j_coord_offsetA  = DIM*jnrA;
 934             j_coord_offsetB  = DIM*jnrB;
 935             j_coord_offsetC  = DIM*jnrC;
 936             j_coord_offsetD  = DIM*jnrD;
 937             j_coord_offsetE  = DIM*jnrE;
 938             j_coord_offsetF  = DIM*jnrF;
 939             j_coord_offsetG  = DIM*jnrG;
 940             j_coord_offsetH  = DIM*jnrH;
 941
 942             /* load j atom coordinates */
 943             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 944                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 945                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 946                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 947                                                  &jx0,&jy0,&jz0);
 948
 949             /* Calculate displacement vector */
 950             dx00             = _mm256_sub_ps(ix0,jx0);
 951             dy00             = _mm256_sub_ps(iy0,jy0);
 952             dz00             = _mm256_sub_ps(iz0,jz0);
 953             dx10             = _mm256_sub_ps(ix1,jx0);
 954             dy10             = _mm256_sub_ps(iy1,jy0);
 955             dz10             = _mm256_sub_ps(iz1,jz0);
 956             dx20             = _mm256_sub_ps(ix2,jx0);
 957             dy20             = _mm256_sub_ps(iy2,jy0);
 958             dz20             = _mm256_sub_ps(iz2,jz0);
 959             dx30             = _mm256_sub_ps(ix3,jx0);
 960             dy30             = _mm256_sub_ps(iy3,jy0);
 961             dz30             = _mm256_sub_ps(iz3,jz0);
 962
 963             /* Calculate squared distance and things based on it */
 964             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 965             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 966             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 967             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 968
 969             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 970             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 971             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 972             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 973
 974             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 975             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 976             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 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             r00              = _mm256_mul_ps(rsq00,rinv00);
1001
1002             /* Compute parameters for interactions between i and j atoms */
1003             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1004                                             vdwioffsetptr0+vdwjidx0B,
1005                                             vdwioffsetptr0+vdwjidx0C,
1006                                             vdwioffsetptr0+vdwjidx0D,
1007                                             vdwioffsetptr0+vdwjidx0E,
1008                                             vdwioffsetptr0+vdwjidx0F,
1009                                             vdwioffsetptr0+vdwjidx0G,
1010                                             vdwioffsetptr0+vdwjidx0H,
1011                                             &c6_00,&c12_00);
1012
1013             /* Calculate table index by multiplying r with table scale and truncate to integer */
1014             rt               = _mm256_mul_ps(r00,vftabscale);
1015             vfitab           = _mm256_cvttps_epi32(rt);
1016             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1017             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1018             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1019             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1020             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1021             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1022
1023             /* CUBIC SPLINE TABLE DISPERSION */
1024             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1025                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1026             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1027                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1028             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1029                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1030             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1031                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1032             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1033             Heps             = _mm256_mul_ps(vfeps,H);
1034             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1035             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1036             fvdw6            = _mm256_mul_ps(c6_00,FF);
1037
1038             /* CUBIC SPLINE TABLE REPULSION */
1039             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1040             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1041             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1042                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1043             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1044                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1045             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1046                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1047             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1048                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1049             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1050             Heps             = _mm256_mul_ps(vfeps,H);
1051             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1052             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1053             fvdw12           = _mm256_mul_ps(c12_00,FF);
1054             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1055
1056             fscal            = fvdw;
1057
1058             /* Calculate temporary vectorial force */
1059             tx               = _mm256_mul_ps(fscal,dx00);
1060             ty               = _mm256_mul_ps(fscal,dy00);
1061             tz               = _mm256_mul_ps(fscal,dz00);
1062
1063             /* Update vectorial force */
1064             fix0             = _mm256_add_ps(fix0,tx);
1065             fiy0             = _mm256_add_ps(fiy0,ty);
1066             fiz0             = _mm256_add_ps(fiz0,tz);
1067
1068             fjx0             = _mm256_add_ps(fjx0,tx);
1069             fjy0             = _mm256_add_ps(fjy0,ty);
1070             fjz0             = _mm256_add_ps(fjz0,tz);
1071
1072             /**************************
1073              * CALCULATE INTERACTIONS *
1074              **************************/
1075
1076             /* Compute parameters for interactions between i and j atoms */
1077             qq10             = _mm256_mul_ps(iq1,jq0);
1078
1079             /* COULOMB ELECTROSTATICS */
1080             velec            = _mm256_mul_ps(qq10,rinv10);
1081             felec            = _mm256_mul_ps(velec,rinvsq10);
1082
1083             fscal            = felec;
1084
1085             /* Calculate temporary vectorial force */
1086             tx               = _mm256_mul_ps(fscal,dx10);
1087             ty               = _mm256_mul_ps(fscal,dy10);
1088             tz               = _mm256_mul_ps(fscal,dz10);
1089
1090             /* Update vectorial force */
1091             fix1             = _mm256_add_ps(fix1,tx);
1092             fiy1             = _mm256_add_ps(fiy1,ty);
1093             fiz1             = _mm256_add_ps(fiz1,tz);
1094
1095             fjx0             = _mm256_add_ps(fjx0,tx);
1096             fjy0             = _mm256_add_ps(fjy0,ty);
1097             fjz0             = _mm256_add_ps(fjz0,tz);
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq20             = _mm256_mul_ps(iq2,jq0);
1105
1106             /* COULOMB ELECTROSTATICS */
1107             velec            = _mm256_mul_ps(qq20,rinv20);
1108             felec            = _mm256_mul_ps(velec,rinvsq20);
1109
1110             fscal            = felec;
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm256_mul_ps(fscal,dx20);
1114             ty               = _mm256_mul_ps(fscal,dy20);
1115             tz               = _mm256_mul_ps(fscal,dz20);
1116
1117             /* Update vectorial force */
1118             fix2             = _mm256_add_ps(fix2,tx);
1119             fiy2             = _mm256_add_ps(fiy2,ty);
1120             fiz2             = _mm256_add_ps(fiz2,tz);
1121
1122             fjx0             = _mm256_add_ps(fjx0,tx);
1123             fjy0             = _mm256_add_ps(fjy0,ty);
1124             fjz0             = _mm256_add_ps(fjz0,tz);
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             qq30             = _mm256_mul_ps(iq3,jq0);
1132
1133             /* COULOMB ELECTROSTATICS */
1134             velec            = _mm256_mul_ps(qq30,rinv30);
1135             felec            = _mm256_mul_ps(velec,rinvsq30);
1136
1137             fscal            = felec;
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm256_mul_ps(fscal,dx30);
1141             ty               = _mm256_mul_ps(fscal,dy30);
1142             tz               = _mm256_mul_ps(fscal,dz30);
1143
1144             /* Update vectorial force */
1145             fix3             = _mm256_add_ps(fix3,tx);
1146             fiy3             = _mm256_add_ps(fiy3,ty);
1147             fiz3             = _mm256_add_ps(fiz3,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             fjptrA             = f+j_coord_offsetA;
1154             fjptrB             = f+j_coord_offsetB;
1155             fjptrC             = f+j_coord_offsetC;
1156             fjptrD             = f+j_coord_offsetD;
1157             fjptrE             = f+j_coord_offsetE;
1158             fjptrF             = f+j_coord_offsetF;
1159             fjptrG             = f+j_coord_offsetG;
1160             fjptrH             = f+j_coord_offsetH;
1161
1162             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1163
1164             /* Inner loop uses 129 flops */
1165         }
1166
1167         if(jidx<j_index_end)
1168         {
1169
1170             /* Get j neighbor index, and coordinate index */
1171             jnrlistA         = jjnr[jidx];
1172             jnrlistB         = jjnr[jidx+1];
1173             jnrlistC         = jjnr[jidx+2];
1174             jnrlistD         = jjnr[jidx+3];
1175             jnrlistE         = jjnr[jidx+4];
1176             jnrlistF         = jjnr[jidx+5];
1177             jnrlistG         = jjnr[jidx+6];
1178             jnrlistH         = jjnr[jidx+7];
1179             /* Sign of each element will be negative for non-real atoms.
1180              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1181              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1182              */
1183             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1184                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1185
1186             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1187             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1188             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1189             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1190             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1191             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1192             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1193             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1194             j_coord_offsetA  = DIM*jnrA;
1195             j_coord_offsetB  = DIM*jnrB;
1196             j_coord_offsetC  = DIM*jnrC;
1197             j_coord_offsetD  = DIM*jnrD;
1198             j_coord_offsetE  = DIM*jnrE;
1199             j_coord_offsetF  = DIM*jnrF;
1200             j_coord_offsetG  = DIM*jnrG;
1201             j_coord_offsetH  = DIM*jnrH;
1202
1203             /* load j atom coordinates */
1204             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1206                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1207                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1208                                                  &jx0,&jy0,&jz0);
1209
1210             /* Calculate displacement vector */
1211             dx00             = _mm256_sub_ps(ix0,jx0);
1212             dy00             = _mm256_sub_ps(iy0,jy0);
1213             dz00             = _mm256_sub_ps(iz0,jz0);
1214             dx10             = _mm256_sub_ps(ix1,jx0);
1215             dy10             = _mm256_sub_ps(iy1,jy0);
1216             dz10             = _mm256_sub_ps(iz1,jz0);
1217             dx20             = _mm256_sub_ps(ix2,jx0);
1218             dy20             = _mm256_sub_ps(iy2,jy0);
1219             dz20             = _mm256_sub_ps(iz2,jz0);
1220             dx30             = _mm256_sub_ps(ix3,jx0);
1221             dy30             = _mm256_sub_ps(iy3,jy0);
1222             dz30             = _mm256_sub_ps(iz3,jz0);
1223
1224             /* Calculate squared distance and things based on it */
1225             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1226             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1227             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1228             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1229
1230             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1231             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1232             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1233             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1234
1235             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1236             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1237             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1238
1239             /* Load parameters for j particles */
1240             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1241                                                                  charge+jnrC+0,charge+jnrD+0,
1242                                                                  charge+jnrE+0,charge+jnrF+0,
1243                                                                  charge+jnrG+0,charge+jnrH+0);
1244             vdwjidx0A        = 2*vdwtype[jnrA+0];
1245             vdwjidx0B        = 2*vdwtype[jnrB+0];
1246             vdwjidx0C        = 2*vdwtype[jnrC+0];
1247             vdwjidx0D        = 2*vdwtype[jnrD+0];
1248             vdwjidx0E        = 2*vdwtype[jnrE+0];
1249             vdwjidx0F        = 2*vdwtype[jnrF+0];
1250             vdwjidx0G        = 2*vdwtype[jnrG+0];
1251             vdwjidx0H        = 2*vdwtype[jnrH+0];
1252
1253             fjx0             = _mm256_setzero_ps();
1254             fjy0             = _mm256_setzero_ps();
1255             fjz0             = _mm256_setzero_ps();
1256
1257             /**************************
1258              * CALCULATE INTERACTIONS *
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             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1266                                             vdwioffsetptr0+vdwjidx0B,
1267                                             vdwioffsetptr0+vdwjidx0C,
1268                                             vdwioffsetptr0+vdwjidx0D,
1269                                             vdwioffsetptr0+vdwjidx0E,
1270                                             vdwioffsetptr0+vdwjidx0F,
1271                                             vdwioffsetptr0+vdwjidx0G,
1272                                             vdwioffsetptr0+vdwjidx0H,
1273                                             &c6_00,&c12_00);
1274
1275             /* Calculate table index by multiplying r with table scale and truncate to integer */
1276             rt               = _mm256_mul_ps(r00,vftabscale);
1277             vfitab           = _mm256_cvttps_epi32(rt);
1278             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1279             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1280             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1281             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1282             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1283             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1284
1285             /* CUBIC SPLINE TABLE DISPERSION */
1286             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1287                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1288             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1289                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1290             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1292             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1294             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1295             Heps             = _mm256_mul_ps(vfeps,H);
1296             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1297             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1298             fvdw6            = _mm256_mul_ps(c6_00,FF);
1299
1300             /* CUBIC SPLINE TABLE REPULSION */
1301             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1302             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1303             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1304                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1305             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1307             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1309             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1311             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1312             Heps             = _mm256_mul_ps(vfeps,H);
1313             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1314             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1315             fvdw12           = _mm256_mul_ps(c12_00,FF);
1316             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1317
1318             fscal            = fvdw;
1319
1320             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1321
1322             /* Calculate temporary vectorial force */
1323             tx               = _mm256_mul_ps(fscal,dx00);
1324             ty               = _mm256_mul_ps(fscal,dy00);
1325             tz               = _mm256_mul_ps(fscal,dz00);
1326
1327             /* Update vectorial force */
1328             fix0             = _mm256_add_ps(fix0,tx);
1329             fiy0             = _mm256_add_ps(fiy0,ty);
1330             fiz0             = _mm256_add_ps(fiz0,tz);
1331
1332             fjx0             = _mm256_add_ps(fjx0,tx);
1333             fjy0             = _mm256_add_ps(fjy0,ty);
1334             fjz0             = _mm256_add_ps(fjz0,tz);
1335
1336             /**************************
1337              * CALCULATE INTERACTIONS *
1338              **************************/
1339
1340             /* Compute parameters for interactions between i and j atoms */
1341             qq10             = _mm256_mul_ps(iq1,jq0);
1342
1343             /* COULOMB ELECTROSTATICS */
1344             velec            = _mm256_mul_ps(qq10,rinv10);
1345             felec            = _mm256_mul_ps(velec,rinvsq10);
1346
1347             fscal            = felec;
1348
1349             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1350
1351             /* Calculate temporary vectorial force */
1352             tx               = _mm256_mul_ps(fscal,dx10);
1353             ty               = _mm256_mul_ps(fscal,dy10);
1354             tz               = _mm256_mul_ps(fscal,dz10);
1355
1356             /* Update vectorial force */
1357             fix1             = _mm256_add_ps(fix1,tx);
1358             fiy1             = _mm256_add_ps(fiy1,ty);
1359             fiz1             = _mm256_add_ps(fiz1,tz);
1360
1361             fjx0             = _mm256_add_ps(fjx0,tx);
1362             fjy0             = _mm256_add_ps(fjy0,ty);
1363             fjz0             = _mm256_add_ps(fjz0,tz);
1364
1365             /**************************
1366              * CALCULATE INTERACTIONS *
1367              **************************/
1368
1369             /* Compute parameters for interactions between i and j atoms */
1370             qq20             = _mm256_mul_ps(iq2,jq0);
1371
1372             /* COULOMB ELECTROSTATICS */
1373             velec            = _mm256_mul_ps(qq20,rinv20);
1374             felec            = _mm256_mul_ps(velec,rinvsq20);
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             /* COULOMB ELECTROSTATICS */
1402             velec            = _mm256_mul_ps(qq30,rinv30);
1403             felec            = _mm256_mul_ps(velec,rinvsq30);
1404
1405             fscal            = felec;
1406
1407             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1408
1409             /* Calculate temporary vectorial force */
1410             tx               = _mm256_mul_ps(fscal,dx30);
1411             ty               = _mm256_mul_ps(fscal,dy30);
1412             tz               = _mm256_mul_ps(fscal,dz30);
1413
1414             /* Update vectorial force */
1415             fix3             = _mm256_add_ps(fix3,tx);
1416             fiy3             = _mm256_add_ps(fiy3,ty);
1417             fiz3             = _mm256_add_ps(fiz3,tz);
1418
1419             fjx0             = _mm256_add_ps(fjx0,tx);
1420             fjy0             = _mm256_add_ps(fjy0,ty);
1421             fjz0             = _mm256_add_ps(fjz0,tz);
1422
1423             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1424             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1425             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1426             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1427             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1428             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1429             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1430             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1431
1432             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1433
1434             /* Inner loop uses 130 flops */
1435         }
1436
1437         /* End of innermost loop */
1438
1439         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1440                                                  f+i_coord_offset,fshift+i_shift_offset);
1441
1442         /* Increment number of inner iterations */
1443         inneriter                  += j_index_end - j_index_start;
1444
1445         /* Outer loop uses 24 flops */
1446     }
1447
1448     /* Increment number of outer iterations */
1449     outeriter        += nri;
1450
1451     /* Update outer/inner flops */
1452
1453     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1454 }