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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  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           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_elec->data;
 134     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 139     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 140     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 141     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149     j_coord_offsetE = 0;
 150     j_coord_offsetF = 0;
 151     j_coord_offsetG = 0;
 152     j_coord_offsetH = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183         fix1             = _mm256_setzero_ps();
 184         fiy1             = _mm256_setzero_ps();
 185         fiz1             = _mm256_setzero_ps();
 186         fix2             = _mm256_setzero_ps();
 187         fiy2             = _mm256_setzero_ps();
 188         fiz2             = _mm256_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_ps();
 192         vvdwsum          = _mm256_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             jnrE             = jjnr[jidx+4];
 204             jnrF             = jjnr[jidx+5];
 205             jnrG             = jjnr[jidx+6];
 206             jnrH             = jjnr[jidx+7];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211             j_coord_offsetE  = DIM*jnrE;
 212             j_coord_offsetF  = DIM*jnrF;
 213             j_coord_offsetG  = DIM*jnrG;
 214             j_coord_offsetH  = DIM*jnrH;
 215
 216             /* load j atom coordinates */
 217             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 219                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 220                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 221                                                  &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm256_sub_ps(ix0,jx0);
 225             dy00             = _mm256_sub_ps(iy0,jy0);
 226             dz00             = _mm256_sub_ps(iz0,jz0);
 227             dx10             = _mm256_sub_ps(ix1,jx0);
 228             dy10             = _mm256_sub_ps(iy1,jy0);
 229             dz10             = _mm256_sub_ps(iz1,jz0);
 230             dx20             = _mm256_sub_ps(ix2,jx0);
 231             dy20             = _mm256_sub_ps(iy2,jy0);
 232             dz20             = _mm256_sub_ps(iz2,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 236             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 237             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 238
 239             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 240             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 241             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 242
 243             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                                  charge+jnrC+0,charge+jnrD+0,
 248                                                                  charge+jnrE+0,charge+jnrF+0,
 249                                                                  charge+jnrG+0,charge+jnrH+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251             vdwjidx0B        = 2*vdwtype[jnrB+0];
 252             vdwjidx0C        = 2*vdwtype[jnrC+0];
 253             vdwjidx0D        = 2*vdwtype[jnrD+0];
 254             vdwjidx0E        = 2*vdwtype[jnrE+0];
 255             vdwjidx0F        = 2*vdwtype[jnrF+0];
 256             vdwjidx0G        = 2*vdwtype[jnrG+0];
 257             vdwjidx0H        = 2*vdwtype[jnrH+0];
 258
 259             fjx0             = _mm256_setzero_ps();
 260             fjy0             = _mm256_setzero_ps();
 261             fjz0             = _mm256_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r00              = _mm256_mul_ps(rsq00,rinv00);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq00             = _mm256_mul_ps(iq0,jq0);
 271             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 272                                             vdwioffsetptr0+vdwjidx0B,
 273                                             vdwioffsetptr0+vdwjidx0C,
 274                                             vdwioffsetptr0+vdwjidx0D,
 275                                             vdwioffsetptr0+vdwjidx0E,
 276                                             vdwioffsetptr0+vdwjidx0F,
 277                                             vdwioffsetptr0+vdwjidx0G,
 278                                             vdwioffsetptr0+vdwjidx0H,
 279                                             &c6_00,&c12_00);
 280
 281             /* Calculate table index by multiplying r with table scale and truncate to integer */
 282             rt               = _mm256_mul_ps(r00,vftabscale);
 283             vfitab           = _mm256_cvttps_epi32(rt);
 284             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 285             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 286             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 287             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 288             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 289             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 290
 291             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 292             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 294             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 296             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 298             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 299                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 300             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 301             Heps             = _mm256_mul_ps(vfeps,H);
 302             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 303             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 304             velec            = _mm256_mul_ps(qq00,VV);
 305             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 306             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 307
 308             /* LENNARD-JONES DISPERSION/REPULSION */
 309
 310             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 311             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 312             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 313             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 314             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm256_add_ps(velecsum,velec);
 318             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 319
 320             fscal            = _mm256_add_ps(felec,fvdw);
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm256_mul_ps(fscal,dx00);
 324             ty               = _mm256_mul_ps(fscal,dy00);
 325             tz               = _mm256_mul_ps(fscal,dz00);
 326
 327             /* Update vectorial force */
 328             fix0             = _mm256_add_ps(fix0,tx);
 329             fiy0             = _mm256_add_ps(fiy0,ty);
 330             fiz0             = _mm256_add_ps(fiz0,tz);
 331
 332             fjx0             = _mm256_add_ps(fjx0,tx);
 333             fjy0             = _mm256_add_ps(fjy0,ty);
 334             fjz0             = _mm256_add_ps(fjz0,tz);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             r10              = _mm256_mul_ps(rsq10,rinv10);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq10             = _mm256_mul_ps(iq1,jq0);
 344
 345             /* Calculate table index by multiplying r with table scale and truncate to integer */
 346             rt               = _mm256_mul_ps(r10,vftabscale);
 347             vfitab           = _mm256_cvttps_epi32(rt);
 348             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 349             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 350             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 351             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 352             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 353             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 354
 355             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 356             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 357                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 358             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 359                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 360             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 361                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 362             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 364             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 365             Heps             = _mm256_mul_ps(vfeps,H);
 366             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 367             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 368             velec            = _mm256_mul_ps(qq10,VV);
 369             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 370             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm256_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm256_mul_ps(fscal,dx10);
 379             ty               = _mm256_mul_ps(fscal,dy10);
 380             tz               = _mm256_mul_ps(fscal,dz10);
 381
 382             /* Update vectorial force */
 383             fix1             = _mm256_add_ps(fix1,tx);
 384             fiy1             = _mm256_add_ps(fiy1,ty);
 385             fiz1             = _mm256_add_ps(fiz1,tz);
 386
 387             fjx0             = _mm256_add_ps(fjx0,tx);
 388             fjy0             = _mm256_add_ps(fjy0,ty);
 389             fjz0             = _mm256_add_ps(fjz0,tz);
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             r20              = _mm256_mul_ps(rsq20,rinv20);
 396
 397             /* Compute parameters for interactions between i and j atoms */
 398             qq20             = _mm256_mul_ps(iq2,jq0);
 399
 400             /* Calculate table index by multiplying r with table scale and truncate to integer */
 401             rt               = _mm256_mul_ps(r20,vftabscale);
 402             vfitab           = _mm256_cvttps_epi32(rt);
 403             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 404             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 405             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 406             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 407             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 408             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 409
 410             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 411             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 412                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 413             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 414                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 415             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 416                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 417             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 418                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 419             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 420             Heps             = _mm256_mul_ps(vfeps,H);
 421             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 422             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 423             velec            = _mm256_mul_ps(qq20,VV);
 424             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 425             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velecsum         = _mm256_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm256_mul_ps(fscal,dx20);
 434             ty               = _mm256_mul_ps(fscal,dy20);
 435             tz               = _mm256_mul_ps(fscal,dz20);
 436
 437             /* Update vectorial force */
 438             fix2             = _mm256_add_ps(fix2,tx);
 439             fiy2             = _mm256_add_ps(fiy2,ty);
 440             fiz2             = _mm256_add_ps(fiz2,tz);
 441
 442             fjx0             = _mm256_add_ps(fjx0,tx);
 443             fjy0             = _mm256_add_ps(fjy0,ty);
 444             fjz0             = _mm256_add_ps(fjz0,tz);
 445
 446             fjptrA             = f+j_coord_offsetA;
 447             fjptrB             = f+j_coord_offsetB;
 448             fjptrC             = f+j_coord_offsetC;
 449             fjptrD             = f+j_coord_offsetD;
 450             fjptrE             = f+j_coord_offsetE;
 451             fjptrF             = f+j_coord_offsetF;
 452             fjptrG             = f+j_coord_offsetG;
 453             fjptrH             = f+j_coord_offsetH;
 454
 455             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 456
 457             /* Inner loop uses 145 flops */
 458         }
 459
 460         if(jidx<j_index_end)
 461         {
 462
 463             /* Get j neighbor index, and coordinate index */
 464             jnrlistA         = jjnr[jidx];
 465             jnrlistB         = jjnr[jidx+1];
 466             jnrlistC         = jjnr[jidx+2];
 467             jnrlistD         = jjnr[jidx+3];
 468             jnrlistE         = jjnr[jidx+4];
 469             jnrlistF         = jjnr[jidx+5];
 470             jnrlistG         = jjnr[jidx+6];
 471             jnrlistH         = jjnr[jidx+7];
 472             /* Sign of each element will be negative for non-real atoms.
 473              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 474              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 475              */
 476             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 477                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 478
 479             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 480             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 481             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 482             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 483             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 484             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 485             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 486             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 487             j_coord_offsetA  = DIM*jnrA;
 488             j_coord_offsetB  = DIM*jnrB;
 489             j_coord_offsetC  = DIM*jnrC;
 490             j_coord_offsetD  = DIM*jnrD;
 491             j_coord_offsetE  = DIM*jnrE;
 492             j_coord_offsetF  = DIM*jnrF;
 493             j_coord_offsetG  = DIM*jnrG;
 494             j_coord_offsetH  = DIM*jnrH;
 495
 496             /* load j atom coordinates */
 497             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 498                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 499                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 500                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 501                                                  &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _mm256_sub_ps(ix0,jx0);
 505             dy00             = _mm256_sub_ps(iy0,jy0);
 506             dz00             = _mm256_sub_ps(iz0,jz0);
 507             dx10             = _mm256_sub_ps(ix1,jx0);
 508             dy10             = _mm256_sub_ps(iy1,jy0);
 509             dz10             = _mm256_sub_ps(iz1,jz0);
 510             dx20             = _mm256_sub_ps(ix2,jx0);
 511             dy20             = _mm256_sub_ps(iy2,jy0);
 512             dz20             = _mm256_sub_ps(iz2,jz0);
 513
 514             /* Calculate squared distance and things based on it */
 515             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 516             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 517             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 518
 519             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 520             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 521             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 522
 523             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 524
 525             /* Load parameters for j particles */
 526             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 527                                                                  charge+jnrC+0,charge+jnrD+0,
 528                                                                  charge+jnrE+0,charge+jnrF+0,
 529                                                                  charge+jnrG+0,charge+jnrH+0);
 530             vdwjidx0A        = 2*vdwtype[jnrA+0];
 531             vdwjidx0B        = 2*vdwtype[jnrB+0];
 532             vdwjidx0C        = 2*vdwtype[jnrC+0];
 533             vdwjidx0D        = 2*vdwtype[jnrD+0];
 534             vdwjidx0E        = 2*vdwtype[jnrE+0];
 535             vdwjidx0F        = 2*vdwtype[jnrF+0];
 536             vdwjidx0G        = 2*vdwtype[jnrG+0];
 537             vdwjidx0H        = 2*vdwtype[jnrH+0];
 538
 539             fjx0             = _mm256_setzero_ps();
 540             fjy0             = _mm256_setzero_ps();
 541             fjz0             = _mm256_setzero_ps();
 542
 543             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             r00              = _mm256_mul_ps(rsq00,rinv00);
 548             r00              = _mm256_andnot_ps(dummy_mask,r00);
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq00             = _mm256_mul_ps(iq0,jq0);
 552             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 553                                             vdwioffsetptr0+vdwjidx0B,
 554                                             vdwioffsetptr0+vdwjidx0C,
 555                                             vdwioffsetptr0+vdwjidx0D,
 556                                             vdwioffsetptr0+vdwjidx0E,
 557                                             vdwioffsetptr0+vdwjidx0F,
 558                                             vdwioffsetptr0+vdwjidx0G,
 559                                             vdwioffsetptr0+vdwjidx0H,
 560                                             &c6_00,&c12_00);
 561
 562             /* Calculate table index by multiplying r with table scale and truncate to integer */
 563             rt               = _mm256_mul_ps(r00,vftabscale);
 564             vfitab           = _mm256_cvttps_epi32(rt);
 565             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 566             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 567             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 568             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 569             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 570             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 571
 572             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 573             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 574                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 575             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 576                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 577             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 578                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 579             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 580                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 581             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 582             Heps             = _mm256_mul_ps(vfeps,H);
 583             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 584             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 585             velec            = _mm256_mul_ps(qq00,VV);
 586             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 587             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 588
 589             /* LENNARD-JONES DISPERSION/REPULSION */
 590
 591             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 592             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 593             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 594             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 595             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 596
 597             /* Update potential sum for this i atom from the interaction with this j atom. */
 598             velec            = _mm256_andnot_ps(dummy_mask,velec);
 599             velecsum         = _mm256_add_ps(velecsum,velec);
 600             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 601             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 602
 603             fscal            = _mm256_add_ps(felec,fvdw);
 604
 605             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm256_mul_ps(fscal,dx00);
 609             ty               = _mm256_mul_ps(fscal,dy00);
 610             tz               = _mm256_mul_ps(fscal,dz00);
 611
 612             /* Update vectorial force */
 613             fix0             = _mm256_add_ps(fix0,tx);
 614             fiy0             = _mm256_add_ps(fiy0,ty);
 615             fiz0             = _mm256_add_ps(fiz0,tz);
 616
 617             fjx0             = _mm256_add_ps(fjx0,tx);
 618             fjy0             = _mm256_add_ps(fjy0,ty);
 619             fjz0             = _mm256_add_ps(fjz0,tz);
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             r10              = _mm256_mul_ps(rsq10,rinv10);
 626             r10              = _mm256_andnot_ps(dummy_mask,r10);
 627
 628             /* Compute parameters for interactions between i and j atoms */
 629             qq10             = _mm256_mul_ps(iq1,jq0);
 630
 631             /* Calculate table index by multiplying r with table scale and truncate to integer */
 632             rt               = _mm256_mul_ps(r10,vftabscale);
 633             vfitab           = _mm256_cvttps_epi32(rt);
 634             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 635             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 636             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 637             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 638             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 639             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 640
 641             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 642             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 643                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 644             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 645                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 646             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 647                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 648             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 649                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 650             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 651             Heps             = _mm256_mul_ps(vfeps,H);
 652             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 653             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 654             velec            = _mm256_mul_ps(qq10,VV);
 655             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 656             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 657
 658             /* Update potential sum for this i atom from the interaction with this j atom. */
 659             velec            = _mm256_andnot_ps(dummy_mask,velec);
 660             velecsum         = _mm256_add_ps(velecsum,velec);
 661
 662             fscal            = felec;
 663
 664             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm256_mul_ps(fscal,dx10);
 668             ty               = _mm256_mul_ps(fscal,dy10);
 669             tz               = _mm256_mul_ps(fscal,dz10);
 670
 671             /* Update vectorial force */
 672             fix1             = _mm256_add_ps(fix1,tx);
 673             fiy1             = _mm256_add_ps(fiy1,ty);
 674             fiz1             = _mm256_add_ps(fiz1,tz);
 675
 676             fjx0             = _mm256_add_ps(fjx0,tx);
 677             fjy0             = _mm256_add_ps(fjy0,ty);
 678             fjz0             = _mm256_add_ps(fjz0,tz);
 679
 680             /**************************
 681              * CALCULATE INTERACTIONS *
 682              **************************/
 683
 684             r20              = _mm256_mul_ps(rsq20,rinv20);
 685             r20              = _mm256_andnot_ps(dummy_mask,r20);
 686
 687             /* Compute parameters for interactions between i and j atoms */
 688             qq20             = _mm256_mul_ps(iq2,jq0);
 689
 690             /* Calculate table index by multiplying r with table scale and truncate to integer */
 691             rt               = _mm256_mul_ps(r20,vftabscale);
 692             vfitab           = _mm256_cvttps_epi32(rt);
 693             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 694             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 695             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 696             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 697             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 698             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 699
 700             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 701             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 702                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 703             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 704                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 705             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 706                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 707             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 708                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 709             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 710             Heps             = _mm256_mul_ps(vfeps,H);
 711             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 712             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 713             velec            = _mm256_mul_ps(qq20,VV);
 714             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 715             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 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,dx20);
 727             ty               = _mm256_mul_ps(fscal,dy20);
 728             tz               = _mm256_mul_ps(fscal,dz20);
 729
 730             /* Update vectorial force */
 731             fix2             = _mm256_add_ps(fix2,tx);
 732             fiy2             = _mm256_add_ps(fiy2,ty);
 733             fiz2             = _mm256_add_ps(fiz2,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 148 flops */
 751         }
 752
 753         /* End of innermost loop */
 754
 755         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 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 20 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_W3_VF,outeriter*20 + inneriter*148);
 775 }
 776 /*
 777  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_single
 778  * Electrostatics interaction: CubicSplineTable
 779  * VdW interaction:            LennardJones
 780  * Geometry:                   Water3-Particle
 781  * Calculate force/pot:        Force
 782  */
 783 void
 784 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 819     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 820     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 821     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 822     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 823     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 824     real             *charge;
 825     int              nvdwtype;
 826     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 827     int              *vdwtype;
 828     real             *vdwparam;
 829     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 830     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 831     __m256i          vfitab;
 832     __m128i          vfitab_lo,vfitab_hi;
 833     __m128i          ifour       = _mm_set1_epi32(4);
 834     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 835     real             *vftab;
 836     __m256           dummy_mask,cutoff_mask;
 837     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 838     __m256           one     = _mm256_set1_ps(1.0);
 839     __m256           two     = _mm256_set1_ps(2.0);
 840     x                = xx[0];
 841     f                = ff[0];
 842
 843     nri              = nlist->nri;
 844     iinr             = nlist->iinr;
 845     jindex           = nlist->jindex;
 846     jjnr             = nlist->jjnr;
 847     shiftidx         = nlist->shift;
 848     gid              = nlist->gid;
 849     shiftvec         = fr->shift_vec[0];
 850     fshift           = fr->fshift[0];
 851     facel            = _mm256_set1_ps(fr->epsfac);
 852     charge           = mdatoms->chargeA;
 853     nvdwtype         = fr->ntype;
 854     vdwparam         = fr->nbfp;
 855     vdwtype          = mdatoms->typeA;
 856
 857     vftab            = kernel_data->table_elec->data;
 858     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 859
 860     /* Setup water-specific parameters */
 861     inr              = nlist->iinr[0];
 862     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 863     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 864     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 865     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 866
 867     /* Avoid stupid compiler warnings */
 868     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 869     j_coord_offsetA = 0;
 870     j_coord_offsetB = 0;
 871     j_coord_offsetC = 0;
 872     j_coord_offsetD = 0;
 873     j_coord_offsetE = 0;
 874     j_coord_offsetF = 0;
 875     j_coord_offsetG = 0;
 876     j_coord_offsetH = 0;
 877
 878     outeriter        = 0;
 879     inneriter        = 0;
 880
 881     for(iidx=0;iidx<4*DIM;iidx++)
 882     {
 883         scratch[iidx] = 0.0;
 884     }
 885
 886     /* Start outer loop over neighborlists */
 887     for(iidx=0; iidx<nri; iidx++)
 888     {
 889         /* Load shift vector for this list */
 890         i_shift_offset   = DIM*shiftidx[iidx];
 891
 892         /* Load limits for loop over neighbors */
 893         j_index_start    = jindex[iidx];
 894         j_index_end      = jindex[iidx+1];
 895
 896         /* Get outer coordinate index */
 897         inr              = iinr[iidx];
 898         i_coord_offset   = DIM*inr;
 899
 900         /* Load i particle coords and add shift vector */
 901         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 902                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 903
 904         fix0             = _mm256_setzero_ps();
 905         fiy0             = _mm256_setzero_ps();
 906         fiz0             = _mm256_setzero_ps();
 907         fix1             = _mm256_setzero_ps();
 908         fiy1             = _mm256_setzero_ps();
 909         fiz1             = _mm256_setzero_ps();
 910         fix2             = _mm256_setzero_ps();
 911         fiy2             = _mm256_setzero_ps();
 912         fiz2             = _mm256_setzero_ps();
 913
 914         /* Start inner kernel loop */
 915         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 916         {
 917
 918             /* Get j neighbor index, and coordinate index */
 919             jnrA             = jjnr[jidx];
 920             jnrB             = jjnr[jidx+1];
 921             jnrC             = jjnr[jidx+2];
 922             jnrD             = jjnr[jidx+3];
 923             jnrE             = jjnr[jidx+4];
 924             jnrF             = jjnr[jidx+5];
 925             jnrG             = jjnr[jidx+6];
 926             jnrH             = jjnr[jidx+7];
 927             j_coord_offsetA  = DIM*jnrA;
 928             j_coord_offsetB  = DIM*jnrB;
 929             j_coord_offsetC  = DIM*jnrC;
 930             j_coord_offsetD  = DIM*jnrD;
 931             j_coord_offsetE  = DIM*jnrE;
 932             j_coord_offsetF  = DIM*jnrF;
 933             j_coord_offsetG  = DIM*jnrG;
 934             j_coord_offsetH  = DIM*jnrH;
 935
 936             /* load j atom coordinates */
 937             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 938                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 939                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 940                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 941                                                  &jx0,&jy0,&jz0);
 942
 943             /* Calculate displacement vector */
 944             dx00             = _mm256_sub_ps(ix0,jx0);
 945             dy00             = _mm256_sub_ps(iy0,jy0);
 946             dz00             = _mm256_sub_ps(iz0,jz0);
 947             dx10             = _mm256_sub_ps(ix1,jx0);
 948             dy10             = _mm256_sub_ps(iy1,jy0);
 949             dz10             = _mm256_sub_ps(iz1,jz0);
 950             dx20             = _mm256_sub_ps(ix2,jx0);
 951             dy20             = _mm256_sub_ps(iy2,jy0);
 952             dz20             = _mm256_sub_ps(iz2,jz0);
 953
 954             /* Calculate squared distance and things based on it */
 955             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 956             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 957             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 958
 959             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 960             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 961             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 962
 963             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 964
 965             /* Load parameters for j particles */
 966             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 967                                                                  charge+jnrC+0,charge+jnrD+0,
 968                                                                  charge+jnrE+0,charge+jnrF+0,
 969                                                                  charge+jnrG+0,charge+jnrH+0);
 970             vdwjidx0A        = 2*vdwtype[jnrA+0];
 971             vdwjidx0B        = 2*vdwtype[jnrB+0];
 972             vdwjidx0C        = 2*vdwtype[jnrC+0];
 973             vdwjidx0D        = 2*vdwtype[jnrD+0];
 974             vdwjidx0E        = 2*vdwtype[jnrE+0];
 975             vdwjidx0F        = 2*vdwtype[jnrF+0];
 976             vdwjidx0G        = 2*vdwtype[jnrG+0];
 977             vdwjidx0H        = 2*vdwtype[jnrH+0];
 978
 979             fjx0             = _mm256_setzero_ps();
 980             fjy0             = _mm256_setzero_ps();
 981             fjz0             = _mm256_setzero_ps();
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             r00              = _mm256_mul_ps(rsq00,rinv00);
 988
 989             /* Compute parameters for interactions between i and j atoms */
 990             qq00             = _mm256_mul_ps(iq0,jq0);
 991             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 992                                             vdwioffsetptr0+vdwjidx0B,
 993                                             vdwioffsetptr0+vdwjidx0C,
 994                                             vdwioffsetptr0+vdwjidx0D,
 995                                             vdwioffsetptr0+vdwjidx0E,
 996                                             vdwioffsetptr0+vdwjidx0F,
 997                                             vdwioffsetptr0+vdwjidx0G,
 998                                             vdwioffsetptr0+vdwjidx0H,
 999                                             &c6_00,&c12_00);
1000
1001             /* Calculate table index by multiplying r with table scale and truncate to integer */
1002             rt               = _mm256_mul_ps(r00,vftabscale);
1003             vfitab           = _mm256_cvttps_epi32(rt);
1004             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1005             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1006             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1007             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1008             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1009             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1010
1011             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1012             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1013                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1014             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1015                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1016             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1017                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1018             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1019                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1020             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1021             Heps             = _mm256_mul_ps(vfeps,H);
1022             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1023             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1024             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1025
1026             /* LENNARD-JONES DISPERSION/REPULSION */
1027
1028             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1029             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1030
1031             fscal            = _mm256_add_ps(felec,fvdw);
1032
1033             /* Calculate temporary vectorial force */
1034             tx               = _mm256_mul_ps(fscal,dx00);
1035             ty               = _mm256_mul_ps(fscal,dy00);
1036             tz               = _mm256_mul_ps(fscal,dz00);
1037
1038             /* Update vectorial force */
1039             fix0             = _mm256_add_ps(fix0,tx);
1040             fiy0             = _mm256_add_ps(fiy0,ty);
1041             fiz0             = _mm256_add_ps(fiz0,tz);
1042
1043             fjx0             = _mm256_add_ps(fjx0,tx);
1044             fjy0             = _mm256_add_ps(fjy0,ty);
1045             fjz0             = _mm256_add_ps(fjz0,tz);
1046
1047             /**************************
1048              * CALCULATE INTERACTIONS *
1049              **************************/
1050
1051             r10              = _mm256_mul_ps(rsq10,rinv10);
1052
1053             /* Compute parameters for interactions between i and j atoms */
1054             qq10             = _mm256_mul_ps(iq1,jq0);
1055
1056             /* Calculate table index by multiplying r with table scale and truncate to integer */
1057             rt               = _mm256_mul_ps(r10,vftabscale);
1058             vfitab           = _mm256_cvttps_epi32(rt);
1059             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1060             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1061             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1062             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1063             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1064             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1065
1066             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1067             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1068                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1069             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1070                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1071             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1072                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1073             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1074                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1075             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1076             Heps             = _mm256_mul_ps(vfeps,H);
1077             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1078             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1079             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1080
1081             fscal            = felec;
1082
1083             /* Calculate temporary vectorial force */
1084             tx               = _mm256_mul_ps(fscal,dx10);
1085             ty               = _mm256_mul_ps(fscal,dy10);
1086             tz               = _mm256_mul_ps(fscal,dz10);
1087
1088             /* Update vectorial force */
1089             fix1             = _mm256_add_ps(fix1,tx);
1090             fiy1             = _mm256_add_ps(fiy1,ty);
1091             fiz1             = _mm256_add_ps(fiz1,tz);
1092
1093             fjx0             = _mm256_add_ps(fjx0,tx);
1094             fjy0             = _mm256_add_ps(fjy0,ty);
1095             fjz0             = _mm256_add_ps(fjz0,tz);
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             r20              = _mm256_mul_ps(rsq20,rinv20);
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq20             = _mm256_mul_ps(iq2,jq0);
1105
1106             /* Calculate table index by multiplying r with table scale and truncate to integer */
1107             rt               = _mm256_mul_ps(r20,vftabscale);
1108             vfitab           = _mm256_cvttps_epi32(rt);
1109             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1110             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1111             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1112             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1113             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1114             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1115
1116             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1117             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1118                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1119             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1120                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1121             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1122                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1123             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1124                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1125             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1126             Heps             = _mm256_mul_ps(vfeps,H);
1127             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1128             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1129             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1130
1131             fscal            = felec;
1132
1133             /* Calculate temporary vectorial force */
1134             tx               = _mm256_mul_ps(fscal,dx20);
1135             ty               = _mm256_mul_ps(fscal,dy20);
1136             tz               = _mm256_mul_ps(fscal,dz20);
1137
1138             /* Update vectorial force */
1139             fix2             = _mm256_add_ps(fix2,tx);
1140             fiy2             = _mm256_add_ps(fiy2,ty);
1141             fiz2             = _mm256_add_ps(fiz2,tz);
1142
1143             fjx0             = _mm256_add_ps(fjx0,tx);
1144             fjy0             = _mm256_add_ps(fjy0,ty);
1145             fjz0             = _mm256_add_ps(fjz0,tz);
1146
1147             fjptrA             = f+j_coord_offsetA;
1148             fjptrB             = f+j_coord_offsetB;
1149             fjptrC             = f+j_coord_offsetC;
1150             fjptrD             = f+j_coord_offsetD;
1151             fjptrE             = f+j_coord_offsetE;
1152             fjptrF             = f+j_coord_offsetF;
1153             fjptrG             = f+j_coord_offsetG;
1154             fjptrH             = f+j_coord_offsetH;
1155
1156             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1157
1158             /* Inner loop uses 128 flops */
1159         }
1160
1161         if(jidx<j_index_end)
1162         {
1163
1164             /* Get j neighbor index, and coordinate index */
1165             jnrlistA         = jjnr[jidx];
1166             jnrlistB         = jjnr[jidx+1];
1167             jnrlistC         = jjnr[jidx+2];
1168             jnrlistD         = jjnr[jidx+3];
1169             jnrlistE         = jjnr[jidx+4];
1170             jnrlistF         = jjnr[jidx+5];
1171             jnrlistG         = jjnr[jidx+6];
1172             jnrlistH         = jjnr[jidx+7];
1173             /* Sign of each element will be negative for non-real atoms.
1174              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1175              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1176              */
1177             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1178                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1179
1180             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1181             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1182             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1183             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1184             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1185             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1186             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1187             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1188             j_coord_offsetA  = DIM*jnrA;
1189             j_coord_offsetB  = DIM*jnrB;
1190             j_coord_offsetC  = DIM*jnrC;
1191             j_coord_offsetD  = DIM*jnrD;
1192             j_coord_offsetE  = DIM*jnrE;
1193             j_coord_offsetF  = DIM*jnrF;
1194             j_coord_offsetG  = DIM*jnrG;
1195             j_coord_offsetH  = DIM*jnrH;
1196
1197             /* load j atom coordinates */
1198             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1199                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1200                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1201                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1202                                                  &jx0,&jy0,&jz0);
1203
1204             /* Calculate displacement vector */
1205             dx00             = _mm256_sub_ps(ix0,jx0);
1206             dy00             = _mm256_sub_ps(iy0,jy0);
1207             dz00             = _mm256_sub_ps(iz0,jz0);
1208             dx10             = _mm256_sub_ps(ix1,jx0);
1209             dy10             = _mm256_sub_ps(iy1,jy0);
1210             dz10             = _mm256_sub_ps(iz1,jz0);
1211             dx20             = _mm256_sub_ps(ix2,jx0);
1212             dy20             = _mm256_sub_ps(iy2,jy0);
1213             dz20             = _mm256_sub_ps(iz2,jz0);
1214
1215             /* Calculate squared distance and things based on it */
1216             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1217             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1218             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1219
1220             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1221             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1222             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1223
1224             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1225
1226             /* Load parameters for j particles */
1227             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1228                                                                  charge+jnrC+0,charge+jnrD+0,
1229                                                                  charge+jnrE+0,charge+jnrF+0,
1230                                                                  charge+jnrG+0,charge+jnrH+0);
1231             vdwjidx0A        = 2*vdwtype[jnrA+0];
1232             vdwjidx0B        = 2*vdwtype[jnrB+0];
1233             vdwjidx0C        = 2*vdwtype[jnrC+0];
1234             vdwjidx0D        = 2*vdwtype[jnrD+0];
1235             vdwjidx0E        = 2*vdwtype[jnrE+0];
1236             vdwjidx0F        = 2*vdwtype[jnrF+0];
1237             vdwjidx0G        = 2*vdwtype[jnrG+0];
1238             vdwjidx0H        = 2*vdwtype[jnrH+0];
1239
1240             fjx0             = _mm256_setzero_ps();
1241             fjy0             = _mm256_setzero_ps();
1242             fjz0             = _mm256_setzero_ps();
1243
1244             /**************************
1245              * CALCULATE INTERACTIONS *
1246              **************************/
1247
1248             r00              = _mm256_mul_ps(rsq00,rinv00);
1249             r00              = _mm256_andnot_ps(dummy_mask,r00);
1250
1251             /* Compute parameters for interactions between i and j atoms */
1252             qq00             = _mm256_mul_ps(iq0,jq0);
1253             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1254                                             vdwioffsetptr0+vdwjidx0B,
1255                                             vdwioffsetptr0+vdwjidx0C,
1256                                             vdwioffsetptr0+vdwjidx0D,
1257                                             vdwioffsetptr0+vdwjidx0E,
1258                                             vdwioffsetptr0+vdwjidx0F,
1259                                             vdwioffsetptr0+vdwjidx0G,
1260                                             vdwioffsetptr0+vdwjidx0H,
1261                                             &c6_00,&c12_00);
1262
1263             /* Calculate table index by multiplying r with table scale and truncate to integer */
1264             rt               = _mm256_mul_ps(r00,vftabscale);
1265             vfitab           = _mm256_cvttps_epi32(rt);
1266             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1267             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1268             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1269             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1270             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1271             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1272
1273             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1274             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1275                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1276             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1277                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1278             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1279                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1280             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1281                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1282             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1283             Heps             = _mm256_mul_ps(vfeps,H);
1284             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1285             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1286             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1287
1288             /* LENNARD-JONES DISPERSION/REPULSION */
1289
1290             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1291             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1292
1293             fscal            = _mm256_add_ps(felec,fvdw);
1294
1295             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1296
1297             /* Calculate temporary vectorial force */
1298             tx               = _mm256_mul_ps(fscal,dx00);
1299             ty               = _mm256_mul_ps(fscal,dy00);
1300             tz               = _mm256_mul_ps(fscal,dz00);
1301
1302             /* Update vectorial force */
1303             fix0             = _mm256_add_ps(fix0,tx);
1304             fiy0             = _mm256_add_ps(fiy0,ty);
1305             fiz0             = _mm256_add_ps(fiz0,tz);
1306
1307             fjx0             = _mm256_add_ps(fjx0,tx);
1308             fjy0             = _mm256_add_ps(fjy0,ty);
1309             fjz0             = _mm256_add_ps(fjz0,tz);
1310
1311             /**************************
1312              * CALCULATE INTERACTIONS *
1313              **************************/
1314
1315             r10              = _mm256_mul_ps(rsq10,rinv10);
1316             r10              = _mm256_andnot_ps(dummy_mask,r10);
1317
1318             /* Compute parameters for interactions between i and j atoms */
1319             qq10             = _mm256_mul_ps(iq1,jq0);
1320
1321             /* Calculate table index by multiplying r with table scale and truncate to integer */
1322             rt               = _mm256_mul_ps(r10,vftabscale);
1323             vfitab           = _mm256_cvttps_epi32(rt);
1324             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1325             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1326             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1327             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1328             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1329             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1330
1331             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1332             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1334             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1335                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1336             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1337                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1338             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1339                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1340             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1341             Heps             = _mm256_mul_ps(vfeps,H);
1342             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1343             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1344             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1345
1346             fscal            = felec;
1347
1348             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1349
1350             /* Calculate temporary vectorial force */
1351             tx               = _mm256_mul_ps(fscal,dx10);
1352             ty               = _mm256_mul_ps(fscal,dy10);
1353             tz               = _mm256_mul_ps(fscal,dz10);
1354
1355             /* Update vectorial force */
1356             fix1             = _mm256_add_ps(fix1,tx);
1357             fiy1             = _mm256_add_ps(fiy1,ty);
1358             fiz1             = _mm256_add_ps(fiz1,tz);
1359
1360             fjx0             = _mm256_add_ps(fjx0,tx);
1361             fjy0             = _mm256_add_ps(fjy0,ty);
1362             fjz0             = _mm256_add_ps(fjz0,tz);
1363
1364             /**************************
1365              * CALCULATE INTERACTIONS *
1366              **************************/
1367
1368             r20              = _mm256_mul_ps(rsq20,rinv20);
1369             r20              = _mm256_andnot_ps(dummy_mask,r20);
1370
1371             /* Compute parameters for interactions between i and j atoms */
1372             qq20             = _mm256_mul_ps(iq2,jq0);
1373
1374             /* Calculate table index by multiplying r with table scale and truncate to integer */
1375             rt               = _mm256_mul_ps(r20,vftabscale);
1376             vfitab           = _mm256_cvttps_epi32(rt);
1377             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1378             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1379             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1380             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1381             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1382             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1383
1384             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1385             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1386                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1387             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1388                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1389             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1390                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1391             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1392                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1393             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1394             Heps             = _mm256_mul_ps(vfeps,H);
1395             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1396             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1397             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1398
1399             fscal            = felec;
1400
1401             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1402
1403             /* Calculate temporary vectorial force */
1404             tx               = _mm256_mul_ps(fscal,dx20);
1405             ty               = _mm256_mul_ps(fscal,dy20);
1406             tz               = _mm256_mul_ps(fscal,dz20);
1407
1408             /* Update vectorial force */
1409             fix2             = _mm256_add_ps(fix2,tx);
1410             fiy2             = _mm256_add_ps(fiy2,ty);
1411             fiz2             = _mm256_add_ps(fiz2,tz);
1412
1413             fjx0             = _mm256_add_ps(fjx0,tx);
1414             fjy0             = _mm256_add_ps(fjy0,ty);
1415             fjz0             = _mm256_add_ps(fjz0,tz);
1416
1417             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1418             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1419             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1420             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1421             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1422             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1423             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1424             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1425
1426             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1427
1428             /* Inner loop uses 131 flops */
1429         }
1430
1431         /* End of innermost loop */
1432
1433         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1434                                                  f+i_coord_offset,fshift+i_shift_offset);
1435
1436         /* Increment number of inner iterations */
1437         inneriter                  += j_index_end - j_index_start;
1438
1439         /* Outer loop uses 18 flops */
1440     }
1441
1442     /* Increment number of outer iterations */
1443     outeriter        += nri;
1444
1445     /* Update outer/inner flops */
1446
1447     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1448 }