src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 104     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 105     __m256i          vfitab;
 106     __m128i          vfitab_lo,vfitab_hi;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m256           dummy_mask,cutoff_mask;
 111     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 112     __m256           one     = _mm256_set1_ps(1.0);
 113     __m256           two     = _mm256_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     vftab            = kernel_data->table_elec->data;
 132     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 137     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 138     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181         fix1             = _mm256_setzero_ps();
 182         fiy1             = _mm256_setzero_ps();
 183         fiz1             = _mm256_setzero_ps();
 184         fix2             = _mm256_setzero_ps();
 185         fiy2             = _mm256_setzero_ps();
 186         fiz2             = _mm256_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190         vvdwsum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 240
 241             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 245                                                                  charge+jnrC+0,charge+jnrD+0,
 246                                                                  charge+jnrE+0,charge+jnrF+0,
 247                                                                  charge+jnrG+0,charge+jnrH+0);
 248             vdwjidx0A        = 2*vdwtype[jnrA+0];
 249             vdwjidx0B        = 2*vdwtype[jnrB+0];
 250             vdwjidx0C        = 2*vdwtype[jnrC+0];
 251             vdwjidx0D        = 2*vdwtype[jnrD+0];
 252             vdwjidx0E        = 2*vdwtype[jnrE+0];
 253             vdwjidx0F        = 2*vdwtype[jnrF+0];
 254             vdwjidx0G        = 2*vdwtype[jnrG+0];
 255             vdwjidx0H        = 2*vdwtype[jnrH+0];
 256
 257             fjx0             = _mm256_setzero_ps();
 258             fjy0             = _mm256_setzero_ps();
 259             fjz0             = _mm256_setzero_ps();
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             r00              = _mm256_mul_ps(rsq00,rinv00);
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq00             = _mm256_mul_ps(iq0,jq0);
 269             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 270                                             vdwioffsetptr0+vdwjidx0B,
 271                                             vdwioffsetptr0+vdwjidx0C,
 272                                             vdwioffsetptr0+vdwjidx0D,
 273                                             vdwioffsetptr0+vdwjidx0E,
 274                                             vdwioffsetptr0+vdwjidx0F,
 275                                             vdwioffsetptr0+vdwjidx0G,
 276                                             vdwioffsetptr0+vdwjidx0H,
 277                                             &c6_00,&c12_00);
 278
 279             /* Calculate table index by multiplying r with table scale and truncate to integer */
 280             rt               = _mm256_mul_ps(r00,vftabscale);
 281             vfitab           = _mm256_cvttps_epi32(rt);
 282             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 283             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 284             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 285             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 286             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 287             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 288
 289             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 290             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 292             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 294             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 296             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 298             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 299             Heps             = _mm256_mul_ps(vfeps,H);
 300             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 301             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 302             velec            = _mm256_mul_ps(qq00,VV);
 303             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 304             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 305
 306             /* LENNARD-JONES DISPERSION/REPULSION */
 307
 308             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 309             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 310             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 311             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 312             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm256_add_ps(velecsum,velec);
 316             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 317
 318             fscal            = _mm256_add_ps(felec,fvdw);
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm256_mul_ps(fscal,dx00);
 322             ty               = _mm256_mul_ps(fscal,dy00);
 323             tz               = _mm256_mul_ps(fscal,dz00);
 324
 325             /* Update vectorial force */
 326             fix0             = _mm256_add_ps(fix0,tx);
 327             fiy0             = _mm256_add_ps(fiy0,ty);
 328             fiz0             = _mm256_add_ps(fiz0,tz);
 329
 330             fjx0             = _mm256_add_ps(fjx0,tx);
 331             fjy0             = _mm256_add_ps(fjy0,ty);
 332             fjz0             = _mm256_add_ps(fjz0,tz);
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             r10              = _mm256_mul_ps(rsq10,rinv10);
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq10             = _mm256_mul_ps(iq1,jq0);
 342
 343             /* Calculate table index by multiplying r with table scale and truncate to integer */
 344             rt               = _mm256_mul_ps(r10,vftabscale);
 345             vfitab           = _mm256_cvttps_epi32(rt);
 346             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 347             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 348             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 349             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 350             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 351             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 352
 353             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 354             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 355                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 356             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 357                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 358             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 359                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 360             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 361                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 362             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 363             Heps             = _mm256_mul_ps(vfeps,H);
 364             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 365             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 366             velec            = _mm256_mul_ps(qq10,VV);
 367             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 368             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velecsum         = _mm256_add_ps(velecsum,velec);
 372
 373             fscal            = felec;
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm256_mul_ps(fscal,dx10);
 377             ty               = _mm256_mul_ps(fscal,dy10);
 378             tz               = _mm256_mul_ps(fscal,dz10);
 379
 380             /* Update vectorial force */
 381             fix1             = _mm256_add_ps(fix1,tx);
 382             fiy1             = _mm256_add_ps(fiy1,ty);
 383             fiz1             = _mm256_add_ps(fiz1,tz);
 384
 385             fjx0             = _mm256_add_ps(fjx0,tx);
 386             fjy0             = _mm256_add_ps(fjy0,ty);
 387             fjz0             = _mm256_add_ps(fjz0,tz);
 388
 389             /**************************
 390              * CALCULATE INTERACTIONS *
 391              **************************/
 392
 393             r20              = _mm256_mul_ps(rsq20,rinv20);
 394
 395             /* Compute parameters for interactions between i and j atoms */
 396             qq20             = _mm256_mul_ps(iq2,jq0);
 397
 398             /* Calculate table index by multiplying r with table scale and truncate to integer */
 399             rt               = _mm256_mul_ps(r20,vftabscale);
 400             vfitab           = _mm256_cvttps_epi32(rt);
 401             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 402             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 403             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 404             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 405             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 406             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 407
 408             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 409             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 410                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 411             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 412                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 413             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 414                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 415             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 416                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 417             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 418             Heps             = _mm256_mul_ps(vfeps,H);
 419             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 420             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 421             velec            = _mm256_mul_ps(qq20,VV);
 422             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 423             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velecsum         = _mm256_add_ps(velecsum,velec);
 427
 428             fscal            = felec;
 429
 430             /* Calculate temporary vectorial force */
 431             tx               = _mm256_mul_ps(fscal,dx20);
 432             ty               = _mm256_mul_ps(fscal,dy20);
 433             tz               = _mm256_mul_ps(fscal,dz20);
 434
 435             /* Update vectorial force */
 436             fix2             = _mm256_add_ps(fix2,tx);
 437             fiy2             = _mm256_add_ps(fiy2,ty);
 438             fiz2             = _mm256_add_ps(fiz2,tz);
 439
 440             fjx0             = _mm256_add_ps(fjx0,tx);
 441             fjy0             = _mm256_add_ps(fjy0,ty);
 442             fjz0             = _mm256_add_ps(fjz0,tz);
 443
 444             fjptrA             = f+j_coord_offsetA;
 445             fjptrB             = f+j_coord_offsetB;
 446             fjptrC             = f+j_coord_offsetC;
 447             fjptrD             = f+j_coord_offsetD;
 448             fjptrE             = f+j_coord_offsetE;
 449             fjptrF             = f+j_coord_offsetF;
 450             fjptrG             = f+j_coord_offsetG;
 451             fjptrH             = f+j_coord_offsetH;
 452
 453             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 454
 455             /* Inner loop uses 145 flops */
 456         }
 457
 458         if(jidx<j_index_end)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrlistA         = jjnr[jidx];
 463             jnrlistB         = jjnr[jidx+1];
 464             jnrlistC         = jjnr[jidx+2];
 465             jnrlistD         = jjnr[jidx+3];
 466             jnrlistE         = jjnr[jidx+4];
 467             jnrlistF         = jjnr[jidx+5];
 468             jnrlistG         = jjnr[jidx+6];
 469             jnrlistH         = jjnr[jidx+7];
 470             /* Sign of each element will be negative for non-real atoms.
 471              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 472              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 473              */
 474             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 475                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 476
 477             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 478             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 479             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 480             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 481             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 482             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 483             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 484             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 485             j_coord_offsetA  = DIM*jnrA;
 486             j_coord_offsetB  = DIM*jnrB;
 487             j_coord_offsetC  = DIM*jnrC;
 488             j_coord_offsetD  = DIM*jnrD;
 489             j_coord_offsetE  = DIM*jnrE;
 490             j_coord_offsetF  = DIM*jnrF;
 491             j_coord_offsetG  = DIM*jnrG;
 492             j_coord_offsetH  = DIM*jnrH;
 493
 494             /* load j atom coordinates */
 495             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 496                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 497                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 498                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 499                                                  &jx0,&jy0,&jz0);
 500
 501             /* Calculate displacement vector */
 502             dx00             = _mm256_sub_ps(ix0,jx0);
 503             dy00             = _mm256_sub_ps(iy0,jy0);
 504             dz00             = _mm256_sub_ps(iz0,jz0);
 505             dx10             = _mm256_sub_ps(ix1,jx0);
 506             dy10             = _mm256_sub_ps(iy1,jy0);
 507             dz10             = _mm256_sub_ps(iz1,jz0);
 508             dx20             = _mm256_sub_ps(ix2,jx0);
 509             dy20             = _mm256_sub_ps(iy2,jy0);
 510             dz20             = _mm256_sub_ps(iz2,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 514             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 515             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 516
 517             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 518             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 519             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 520
 521             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 522
 523             /* Load parameters for j particles */
 524             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 525                                                                  charge+jnrC+0,charge+jnrD+0,
 526                                                                  charge+jnrE+0,charge+jnrF+0,
 527                                                                  charge+jnrG+0,charge+jnrH+0);
 528             vdwjidx0A        = 2*vdwtype[jnrA+0];
 529             vdwjidx0B        = 2*vdwtype[jnrB+0];
 530             vdwjidx0C        = 2*vdwtype[jnrC+0];
 531             vdwjidx0D        = 2*vdwtype[jnrD+0];
 532             vdwjidx0E        = 2*vdwtype[jnrE+0];
 533             vdwjidx0F        = 2*vdwtype[jnrF+0];
 534             vdwjidx0G        = 2*vdwtype[jnrG+0];
 535             vdwjidx0H        = 2*vdwtype[jnrH+0];
 536
 537             fjx0             = _mm256_setzero_ps();
 538             fjy0             = _mm256_setzero_ps();
 539             fjz0             = _mm256_setzero_ps();
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             r00              = _mm256_mul_ps(rsq00,rinv00);
 546             r00              = _mm256_andnot_ps(dummy_mask,r00);
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq00             = _mm256_mul_ps(iq0,jq0);
 550             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 551                                             vdwioffsetptr0+vdwjidx0B,
 552                                             vdwioffsetptr0+vdwjidx0C,
 553                                             vdwioffsetptr0+vdwjidx0D,
 554                                             vdwioffsetptr0+vdwjidx0E,
 555                                             vdwioffsetptr0+vdwjidx0F,
 556                                             vdwioffsetptr0+vdwjidx0G,
 557                                             vdwioffsetptr0+vdwjidx0H,
 558                                             &c6_00,&c12_00);
 559
 560             /* Calculate table index by multiplying r with table scale and truncate to integer */
 561             rt               = _mm256_mul_ps(r00,vftabscale);
 562             vfitab           = _mm256_cvttps_epi32(rt);
 563             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 564             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 565             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 566             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 567             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 568             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 569
 570             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 571             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 572                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 573             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 574                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 575             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 576                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 577             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 578                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 579             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 580             Heps             = _mm256_mul_ps(vfeps,H);
 581             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 582             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 583             velec            = _mm256_mul_ps(qq00,VV);
 584             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 585             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 586
 587             /* LENNARD-JONES DISPERSION/REPULSION */
 588
 589             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 590             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 591             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 592             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 593             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 594
 595             /* Update potential sum for this i atom from the interaction with this j atom. */
 596             velec            = _mm256_andnot_ps(dummy_mask,velec);
 597             velecsum         = _mm256_add_ps(velecsum,velec);
 598             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 599             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 600
 601             fscal            = _mm256_add_ps(felec,fvdw);
 602
 603             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 604
 605             /* Calculate temporary vectorial force */
 606             tx               = _mm256_mul_ps(fscal,dx00);
 607             ty               = _mm256_mul_ps(fscal,dy00);
 608             tz               = _mm256_mul_ps(fscal,dz00);
 609
 610             /* Update vectorial force */
 611             fix0             = _mm256_add_ps(fix0,tx);
 612             fiy0             = _mm256_add_ps(fiy0,ty);
 613             fiz0             = _mm256_add_ps(fiz0,tz);
 614
 615             fjx0             = _mm256_add_ps(fjx0,tx);
 616             fjy0             = _mm256_add_ps(fjy0,ty);
 617             fjz0             = _mm256_add_ps(fjz0,tz);
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             r10              = _mm256_mul_ps(rsq10,rinv10);
 624             r10              = _mm256_andnot_ps(dummy_mask,r10);
 625
 626             /* Compute parameters for interactions between i and j atoms */
 627             qq10             = _mm256_mul_ps(iq1,jq0);
 628
 629             /* Calculate table index by multiplying r with table scale and truncate to integer */
 630             rt               = _mm256_mul_ps(r10,vftabscale);
 631             vfitab           = _mm256_cvttps_epi32(rt);
 632             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 633             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 634             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 635             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 636             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 637             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 638
 639             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 640             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 641                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 642             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 643                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 644             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 645                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 646             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 647                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 648             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 649             Heps             = _mm256_mul_ps(vfeps,H);
 650             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 651             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 652             velec            = _mm256_mul_ps(qq10,VV);
 653             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 654             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             velec            = _mm256_andnot_ps(dummy_mask,velec);
 658             velecsum         = _mm256_add_ps(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm256_mul_ps(fscal,dx10);
 666             ty               = _mm256_mul_ps(fscal,dy10);
 667             tz               = _mm256_mul_ps(fscal,dz10);
 668
 669             /* Update vectorial force */
 670             fix1             = _mm256_add_ps(fix1,tx);
 671             fiy1             = _mm256_add_ps(fiy1,ty);
 672             fiz1             = _mm256_add_ps(fiz1,tz);
 673
 674             fjx0             = _mm256_add_ps(fjx0,tx);
 675             fjy0             = _mm256_add_ps(fjy0,ty);
 676             fjz0             = _mm256_add_ps(fjz0,tz);
 677
 678             /**************************
 679              * CALCULATE INTERACTIONS *
 680              **************************/
 681
 682             r20              = _mm256_mul_ps(rsq20,rinv20);
 683             r20              = _mm256_andnot_ps(dummy_mask,r20);
 684
 685             /* Compute parameters for interactions between i and j atoms */
 686             qq20             = _mm256_mul_ps(iq2,jq0);
 687
 688             /* Calculate table index by multiplying r with table scale and truncate to integer */
 689             rt               = _mm256_mul_ps(r20,vftabscale);
 690             vfitab           = _mm256_cvttps_epi32(rt);
 691             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 692             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 693             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 694             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 695             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 696             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 697
 698             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 699             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 700                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 701             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 702                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 703             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 704                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 705             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 706                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 707             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 708             Heps             = _mm256_mul_ps(vfeps,H);
 709             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 710             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 711             velec            = _mm256_mul_ps(qq20,VV);
 712             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 713             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 714
 715             /* Update potential sum for this i atom from the interaction with this j atom. */
 716             velec            = _mm256_andnot_ps(dummy_mask,velec);
 717             velecsum         = _mm256_add_ps(velecsum,velec);
 718
 719             fscal            = felec;
 720
 721             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 722
 723             /* Calculate temporary vectorial force */
 724             tx               = _mm256_mul_ps(fscal,dx20);
 725             ty               = _mm256_mul_ps(fscal,dy20);
 726             tz               = _mm256_mul_ps(fscal,dz20);
 727
 728             /* Update vectorial force */
 729             fix2             = _mm256_add_ps(fix2,tx);
 730             fiy2             = _mm256_add_ps(fiy2,ty);
 731             fiz2             = _mm256_add_ps(fiz2,tz);
 732
 733             fjx0             = _mm256_add_ps(fjx0,tx);
 734             fjy0             = _mm256_add_ps(fjy0,ty);
 735             fjz0             = _mm256_add_ps(fjz0,tz);
 736
 737             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 738             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 739             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 740             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 741             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 742             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 743             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 744             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 745
 746             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 747
 748             /* Inner loop uses 148 flops */
 749         }
 750
 751         /* End of innermost loop */
 752
 753         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 754                                                  f+i_coord_offset,fshift+i_shift_offset);
 755
 756         ggid                        = gid[iidx];
 757         /* Update potential energies */
 758         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 759         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 760
 761         /* Increment number of inner iterations */
 762         inneriter                  += j_index_end - j_index_start;
 763
 764         /* Outer loop uses 20 flops */
 765     }
 766
 767     /* Increment number of outer iterations */
 768     outeriter        += nri;
 769
 770     /* Update outer/inner flops */
 771
 772     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 773 }
 774 /*
 775  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_single
 776  * Electrostatics interaction: CubicSplineTable
 777  * VdW interaction:            LennardJones
 778  * Geometry:                   Water3-Particle
 779  * Calculate force/pot:        Force
 780  */
 781 void
 782 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_single
 783                     (t_nblist                    * gmx_restrict       nlist,
 784                      rvec                        * gmx_restrict          xx,
 785                      rvec                        * gmx_restrict          ff,
 786                      t_forcerec                  * gmx_restrict          fr,
 787                      t_mdatoms                   * gmx_restrict     mdatoms,
 788                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 789                      t_nrnb                      * gmx_restrict        nrnb)
 790 {
 791     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 792      * just 0 for non-waters.
 793      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 794      * jnr indices corresponding to data put in the four positions in the SIMD register.
 795      */
 796     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 797     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 798     int              jnrA,jnrB,jnrC,jnrD;
 799     int              jnrE,jnrF,jnrG,jnrH;
 800     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 801     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 802     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 803     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 804     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 805     real             rcutoff_scalar;
 806     real             *shiftvec,*fshift,*x,*f;
 807     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 808     real             scratch[4*DIM];
 809     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 810     real *           vdwioffsetptr0;
 811     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 812     real *           vdwioffsetptr1;
 813     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 814     real *           vdwioffsetptr2;
 815     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 816     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 817     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 818     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 819     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 820     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 821     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 822     real             *charge;
 823     int              nvdwtype;
 824     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 825     int              *vdwtype;
 826     real             *vdwparam;
 827     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 828     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 829     __m256i          vfitab;
 830     __m128i          vfitab_lo,vfitab_hi;
 831     __m128i          ifour       = _mm_set1_epi32(4);
 832     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 833     real             *vftab;
 834     __m256           dummy_mask,cutoff_mask;
 835     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 836     __m256           one     = _mm256_set1_ps(1.0);
 837     __m256           two     = _mm256_set1_ps(2.0);
 838     x                = xx[0];
 839     f                = ff[0];
 840
 841     nri              = nlist->nri;
 842     iinr             = nlist->iinr;
 843     jindex           = nlist->jindex;
 844     jjnr             = nlist->jjnr;
 845     shiftidx         = nlist->shift;
 846     gid              = nlist->gid;
 847     shiftvec         = fr->shift_vec[0];
 848     fshift           = fr->fshift[0];
 849     facel            = _mm256_set1_ps(fr->epsfac);
 850     charge           = mdatoms->chargeA;
 851     nvdwtype         = fr->ntype;
 852     vdwparam         = fr->nbfp;
 853     vdwtype          = mdatoms->typeA;
 854
 855     vftab            = kernel_data->table_elec->data;
 856     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 857
 858     /* Setup water-specific parameters */
 859     inr              = nlist->iinr[0];
 860     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 861     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 862     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 863     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 864
 865     /* Avoid stupid compiler warnings */
 866     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 867     j_coord_offsetA = 0;
 868     j_coord_offsetB = 0;
 869     j_coord_offsetC = 0;
 870     j_coord_offsetD = 0;
 871     j_coord_offsetE = 0;
 872     j_coord_offsetF = 0;
 873     j_coord_offsetG = 0;
 874     j_coord_offsetH = 0;
 875
 876     outeriter        = 0;
 877     inneriter        = 0;
 878
 879     for(iidx=0;iidx<4*DIM;iidx++)
 880     {
 881         scratch[iidx] = 0.0;
 882     }
 883
 884     /* Start outer loop over neighborlists */
 885     for(iidx=0; iidx<nri; iidx++)
 886     {
 887         /* Load shift vector for this list */
 888         i_shift_offset   = DIM*shiftidx[iidx];
 889
 890         /* Load limits for loop over neighbors */
 891         j_index_start    = jindex[iidx];
 892         j_index_end      = jindex[iidx+1];
 893
 894         /* Get outer coordinate index */
 895         inr              = iinr[iidx];
 896         i_coord_offset   = DIM*inr;
 897
 898         /* Load i particle coords and add shift vector */
 899         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 900                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 901
 902         fix0             = _mm256_setzero_ps();
 903         fiy0             = _mm256_setzero_ps();
 904         fiz0             = _mm256_setzero_ps();
 905         fix1             = _mm256_setzero_ps();
 906         fiy1             = _mm256_setzero_ps();
 907         fiz1             = _mm256_setzero_ps();
 908         fix2             = _mm256_setzero_ps();
 909         fiy2             = _mm256_setzero_ps();
 910         fiz2             = _mm256_setzero_ps();
 911
 912         /* Start inner kernel loop */
 913         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrA             = jjnr[jidx];
 918             jnrB             = jjnr[jidx+1];
 919             jnrC             = jjnr[jidx+2];
 920             jnrD             = jjnr[jidx+3];
 921             jnrE             = jjnr[jidx+4];
 922             jnrF             = jjnr[jidx+5];
 923             jnrG             = jjnr[jidx+6];
 924             jnrH             = jjnr[jidx+7];
 925             j_coord_offsetA  = DIM*jnrA;
 926             j_coord_offsetB  = DIM*jnrB;
 927             j_coord_offsetC  = DIM*jnrC;
 928             j_coord_offsetD  = DIM*jnrD;
 929             j_coord_offsetE  = DIM*jnrE;
 930             j_coord_offsetF  = DIM*jnrF;
 931             j_coord_offsetG  = DIM*jnrG;
 932             j_coord_offsetH  = DIM*jnrH;
 933
 934             /* load j atom coordinates */
 935             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 936                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 937                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 938                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 939                                                  &jx0,&jy0,&jz0);
 940
 941             /* Calculate displacement vector */
 942             dx00             = _mm256_sub_ps(ix0,jx0);
 943             dy00             = _mm256_sub_ps(iy0,jy0);
 944             dz00             = _mm256_sub_ps(iz0,jz0);
 945             dx10             = _mm256_sub_ps(ix1,jx0);
 946             dy10             = _mm256_sub_ps(iy1,jy0);
 947             dz10             = _mm256_sub_ps(iz1,jz0);
 948             dx20             = _mm256_sub_ps(ix2,jx0);
 949             dy20             = _mm256_sub_ps(iy2,jy0);
 950             dz20             = _mm256_sub_ps(iz2,jz0);
 951
 952             /* Calculate squared distance and things based on it */
 953             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 954             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 955             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 956
 957             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 958             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 959             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 960
 961             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 962
 963             /* Load parameters for j particles */
 964             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 965                                                                  charge+jnrC+0,charge+jnrD+0,
 966                                                                  charge+jnrE+0,charge+jnrF+0,
 967                                                                  charge+jnrG+0,charge+jnrH+0);
 968             vdwjidx0A        = 2*vdwtype[jnrA+0];
 969             vdwjidx0B        = 2*vdwtype[jnrB+0];
 970             vdwjidx0C        = 2*vdwtype[jnrC+0];
 971             vdwjidx0D        = 2*vdwtype[jnrD+0];
 972             vdwjidx0E        = 2*vdwtype[jnrE+0];
 973             vdwjidx0F        = 2*vdwtype[jnrF+0];
 974             vdwjidx0G        = 2*vdwtype[jnrG+0];
 975             vdwjidx0H        = 2*vdwtype[jnrH+0];
 976
 977             fjx0             = _mm256_setzero_ps();
 978             fjy0             = _mm256_setzero_ps();
 979             fjz0             = _mm256_setzero_ps();
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             r00              = _mm256_mul_ps(rsq00,rinv00);
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq00             = _mm256_mul_ps(iq0,jq0);
 989             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 990                                             vdwioffsetptr0+vdwjidx0B,
 991                                             vdwioffsetptr0+vdwjidx0C,
 992                                             vdwioffsetptr0+vdwjidx0D,
 993                                             vdwioffsetptr0+vdwjidx0E,
 994                                             vdwioffsetptr0+vdwjidx0F,
 995                                             vdwioffsetptr0+vdwjidx0G,
 996                                             vdwioffsetptr0+vdwjidx0H,
 997                                             &c6_00,&c12_00);
 998
 999             /* Calculate table index by multiplying r with table scale and truncate to integer */
1000             rt               = _mm256_mul_ps(r00,vftabscale);
1001             vfitab           = _mm256_cvttps_epi32(rt);
1002             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1003             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1004             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1005             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1006             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1007             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1008
1009             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1010             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1011                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1012             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1013                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1014             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1015                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1016             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1017                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1018             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1019             Heps             = _mm256_mul_ps(vfeps,H);
1020             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1021             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1022             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1023
1024             /* LENNARD-JONES DISPERSION/REPULSION */
1025
1026             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1027             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1028
1029             fscal            = _mm256_add_ps(felec,fvdw);
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm256_mul_ps(fscal,dx00);
1033             ty               = _mm256_mul_ps(fscal,dy00);
1034             tz               = _mm256_mul_ps(fscal,dz00);
1035
1036             /* Update vectorial force */
1037             fix0             = _mm256_add_ps(fix0,tx);
1038             fiy0             = _mm256_add_ps(fiy0,ty);
1039             fiz0             = _mm256_add_ps(fiz0,tz);
1040
1041             fjx0             = _mm256_add_ps(fjx0,tx);
1042             fjy0             = _mm256_add_ps(fjy0,ty);
1043             fjz0             = _mm256_add_ps(fjz0,tz);
1044
1045             /**************************
1046              * CALCULATE INTERACTIONS *
1047              **************************/
1048
1049             r10              = _mm256_mul_ps(rsq10,rinv10);
1050
1051             /* Compute parameters for interactions between i and j atoms */
1052             qq10             = _mm256_mul_ps(iq1,jq0);
1053
1054             /* Calculate table index by multiplying r with table scale and truncate to integer */
1055             rt               = _mm256_mul_ps(r10,vftabscale);
1056             vfitab           = _mm256_cvttps_epi32(rt);
1057             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1058             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1059             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1060             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1061             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1062             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1063
1064             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1065             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1066                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1067             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1068                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1069             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1070                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1071             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1072                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1073             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1074             Heps             = _mm256_mul_ps(vfeps,H);
1075             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1076             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1077             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1078
1079             fscal            = felec;
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm256_mul_ps(fscal,dx10);
1083             ty               = _mm256_mul_ps(fscal,dy10);
1084             tz               = _mm256_mul_ps(fscal,dz10);
1085
1086             /* Update vectorial force */
1087             fix1             = _mm256_add_ps(fix1,tx);
1088             fiy1             = _mm256_add_ps(fiy1,ty);
1089             fiz1             = _mm256_add_ps(fiz1,tz);
1090
1091             fjx0             = _mm256_add_ps(fjx0,tx);
1092             fjy0             = _mm256_add_ps(fjy0,ty);
1093             fjz0             = _mm256_add_ps(fjz0,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r20              = _mm256_mul_ps(rsq20,rinv20);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq20             = _mm256_mul_ps(iq2,jq0);
1103
1104             /* Calculate table index by multiplying r with table scale and truncate to integer */
1105             rt               = _mm256_mul_ps(r20,vftabscale);
1106             vfitab           = _mm256_cvttps_epi32(rt);
1107             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1108             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1109             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1110             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1111             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1112             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1113
1114             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1115             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1116                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1117             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1118                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1119             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1120                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1121             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1122                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1123             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1124             Heps             = _mm256_mul_ps(vfeps,H);
1125             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1126             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1127             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1128
1129             fscal            = felec;
1130
1131             /* Calculate temporary vectorial force */
1132             tx               = _mm256_mul_ps(fscal,dx20);
1133             ty               = _mm256_mul_ps(fscal,dy20);
1134             tz               = _mm256_mul_ps(fscal,dz20);
1135
1136             /* Update vectorial force */
1137             fix2             = _mm256_add_ps(fix2,tx);
1138             fiy2             = _mm256_add_ps(fiy2,ty);
1139             fiz2             = _mm256_add_ps(fiz2,tz);
1140
1141             fjx0             = _mm256_add_ps(fjx0,tx);
1142             fjy0             = _mm256_add_ps(fjy0,ty);
1143             fjz0             = _mm256_add_ps(fjz0,tz);
1144
1145             fjptrA             = f+j_coord_offsetA;
1146             fjptrB             = f+j_coord_offsetB;
1147             fjptrC             = f+j_coord_offsetC;
1148             fjptrD             = f+j_coord_offsetD;
1149             fjptrE             = f+j_coord_offsetE;
1150             fjptrF             = f+j_coord_offsetF;
1151             fjptrG             = f+j_coord_offsetG;
1152             fjptrH             = f+j_coord_offsetH;
1153
1154             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1155
1156             /* Inner loop uses 128 flops */
1157         }
1158
1159         if(jidx<j_index_end)
1160         {
1161
1162             /* Get j neighbor index, and coordinate index */
1163             jnrlistA         = jjnr[jidx];
1164             jnrlistB         = jjnr[jidx+1];
1165             jnrlistC         = jjnr[jidx+2];
1166             jnrlistD         = jjnr[jidx+3];
1167             jnrlistE         = jjnr[jidx+4];
1168             jnrlistF         = jjnr[jidx+5];
1169             jnrlistG         = jjnr[jidx+6];
1170             jnrlistH         = jjnr[jidx+7];
1171             /* Sign of each element will be negative for non-real atoms.
1172              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1173              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1174              */
1175             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1176                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1177
1178             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1179             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1180             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1181             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1182             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1183             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1184             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1185             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1186             j_coord_offsetA  = DIM*jnrA;
1187             j_coord_offsetB  = DIM*jnrB;
1188             j_coord_offsetC  = DIM*jnrC;
1189             j_coord_offsetD  = DIM*jnrD;
1190             j_coord_offsetE  = DIM*jnrE;
1191             j_coord_offsetF  = DIM*jnrF;
1192             j_coord_offsetG  = DIM*jnrG;
1193             j_coord_offsetH  = DIM*jnrH;
1194
1195             /* load j atom coordinates */
1196             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1197                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1198                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1199                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1200                                                  &jx0,&jy0,&jz0);
1201
1202             /* Calculate displacement vector */
1203             dx00             = _mm256_sub_ps(ix0,jx0);
1204             dy00             = _mm256_sub_ps(iy0,jy0);
1205             dz00             = _mm256_sub_ps(iz0,jz0);
1206             dx10             = _mm256_sub_ps(ix1,jx0);
1207             dy10             = _mm256_sub_ps(iy1,jy0);
1208             dz10             = _mm256_sub_ps(iz1,jz0);
1209             dx20             = _mm256_sub_ps(ix2,jx0);
1210             dy20             = _mm256_sub_ps(iy2,jy0);
1211             dz20             = _mm256_sub_ps(iz2,jz0);
1212
1213             /* Calculate squared distance and things based on it */
1214             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1215             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1216             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1217
1218             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1219             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1220             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1221
1222             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1223
1224             /* Load parameters for j particles */
1225             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1226                                                                  charge+jnrC+0,charge+jnrD+0,
1227                                                                  charge+jnrE+0,charge+jnrF+0,
1228                                                                  charge+jnrG+0,charge+jnrH+0);
1229             vdwjidx0A        = 2*vdwtype[jnrA+0];
1230             vdwjidx0B        = 2*vdwtype[jnrB+0];
1231             vdwjidx0C        = 2*vdwtype[jnrC+0];
1232             vdwjidx0D        = 2*vdwtype[jnrD+0];
1233             vdwjidx0E        = 2*vdwtype[jnrE+0];
1234             vdwjidx0F        = 2*vdwtype[jnrF+0];
1235             vdwjidx0G        = 2*vdwtype[jnrG+0];
1236             vdwjidx0H        = 2*vdwtype[jnrH+0];
1237
1238             fjx0             = _mm256_setzero_ps();
1239             fjy0             = _mm256_setzero_ps();
1240             fjz0             = _mm256_setzero_ps();
1241
1242             /**************************
1243              * CALCULATE INTERACTIONS *
1244              **************************/
1245
1246             r00              = _mm256_mul_ps(rsq00,rinv00);
1247             r00              = _mm256_andnot_ps(dummy_mask,r00);
1248
1249             /* Compute parameters for interactions between i and j atoms */
1250             qq00             = _mm256_mul_ps(iq0,jq0);
1251             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1252                                             vdwioffsetptr0+vdwjidx0B,
1253                                             vdwioffsetptr0+vdwjidx0C,
1254                                             vdwioffsetptr0+vdwjidx0D,
1255                                             vdwioffsetptr0+vdwjidx0E,
1256                                             vdwioffsetptr0+vdwjidx0F,
1257                                             vdwioffsetptr0+vdwjidx0G,
1258                                             vdwioffsetptr0+vdwjidx0H,
1259                                             &c6_00,&c12_00);
1260
1261             /* Calculate table index by multiplying r with table scale and truncate to integer */
1262             rt               = _mm256_mul_ps(r00,vftabscale);
1263             vfitab           = _mm256_cvttps_epi32(rt);
1264             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1265             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1266             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1267             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1268             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1269             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1270
1271             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1272             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1273                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1274             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1275                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1276             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1277                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1278             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1279                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1280             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1281             Heps             = _mm256_mul_ps(vfeps,H);
1282             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1283             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1284             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1285
1286             /* LENNARD-JONES DISPERSION/REPULSION */
1287
1288             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1289             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1290
1291             fscal            = _mm256_add_ps(felec,fvdw);
1292
1293             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1294
1295             /* Calculate temporary vectorial force */
1296             tx               = _mm256_mul_ps(fscal,dx00);
1297             ty               = _mm256_mul_ps(fscal,dy00);
1298             tz               = _mm256_mul_ps(fscal,dz00);
1299
1300             /* Update vectorial force */
1301             fix0             = _mm256_add_ps(fix0,tx);
1302             fiy0             = _mm256_add_ps(fiy0,ty);
1303             fiz0             = _mm256_add_ps(fiz0,tz);
1304
1305             fjx0             = _mm256_add_ps(fjx0,tx);
1306             fjy0             = _mm256_add_ps(fjy0,ty);
1307             fjz0             = _mm256_add_ps(fjz0,tz);
1308
1309             /**************************
1310              * CALCULATE INTERACTIONS *
1311              **************************/
1312
1313             r10              = _mm256_mul_ps(rsq10,rinv10);
1314             r10              = _mm256_andnot_ps(dummy_mask,r10);
1315
1316             /* Compute parameters for interactions between i and j atoms */
1317             qq10             = _mm256_mul_ps(iq1,jq0);
1318
1319             /* Calculate table index by multiplying r with table scale and truncate to integer */
1320             rt               = _mm256_mul_ps(r10,vftabscale);
1321             vfitab           = _mm256_cvttps_epi32(rt);
1322             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1323             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1324             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1325             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1326             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1327             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1328
1329             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1330             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1332             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1334             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1335                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1336             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1337                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1338             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1339             Heps             = _mm256_mul_ps(vfeps,H);
1340             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1341             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1342             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1343
1344             fscal            = felec;
1345
1346             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1347
1348             /* Calculate temporary vectorial force */
1349             tx               = _mm256_mul_ps(fscal,dx10);
1350             ty               = _mm256_mul_ps(fscal,dy10);
1351             tz               = _mm256_mul_ps(fscal,dz10);
1352
1353             /* Update vectorial force */
1354             fix1             = _mm256_add_ps(fix1,tx);
1355             fiy1             = _mm256_add_ps(fiy1,ty);
1356             fiz1             = _mm256_add_ps(fiz1,tz);
1357
1358             fjx0             = _mm256_add_ps(fjx0,tx);
1359             fjy0             = _mm256_add_ps(fjy0,ty);
1360             fjz0             = _mm256_add_ps(fjz0,tz);
1361
1362             /**************************
1363              * CALCULATE INTERACTIONS *
1364              **************************/
1365
1366             r20              = _mm256_mul_ps(rsq20,rinv20);
1367             r20              = _mm256_andnot_ps(dummy_mask,r20);
1368
1369             /* Compute parameters for interactions between i and j atoms */
1370             qq20             = _mm256_mul_ps(iq2,jq0);
1371
1372             /* Calculate table index by multiplying r with table scale and truncate to integer */
1373             rt               = _mm256_mul_ps(r20,vftabscale);
1374             vfitab           = _mm256_cvttps_epi32(rt);
1375             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1376             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1377             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1378             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1379             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1380             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1381
1382             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1383             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1384                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1385             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1386                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1387             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1388                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1389             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1390                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1391             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1392             Heps             = _mm256_mul_ps(vfeps,H);
1393             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1394             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1395             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1396
1397             fscal            = felec;
1398
1399             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1400
1401             /* Calculate temporary vectorial force */
1402             tx               = _mm256_mul_ps(fscal,dx20);
1403             ty               = _mm256_mul_ps(fscal,dy20);
1404             tz               = _mm256_mul_ps(fscal,dz20);
1405
1406             /* Update vectorial force */
1407             fix2             = _mm256_add_ps(fix2,tx);
1408             fiy2             = _mm256_add_ps(fiy2,ty);
1409             fiz2             = _mm256_add_ps(fiz2,tz);
1410
1411             fjx0             = _mm256_add_ps(fjx0,tx);
1412             fjy0             = _mm256_add_ps(fjy0,ty);
1413             fjz0             = _mm256_add_ps(fjz0,tz);
1414
1415             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1416             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1417             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1418             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1419             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1420             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1421             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1422             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1423
1424             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1425
1426             /* Inner loop uses 131 flops */
1427         }
1428
1429         /* End of innermost loop */
1430
1431         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1432                                                  f+i_coord_offset,fshift+i_shift_offset);
1433
1434         /* Increment number of inner iterations */
1435         inneriter                  += j_index_end - j_index_start;
1436
1437         /* Outer loop uses 18 flops */
1438     }
1439
1440     /* Increment number of outer iterations */
1441     outeriter        += nri;
1442
1443     /* Update outer/inner flops */
1444
1445     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1446 }