src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     real *           vdwioffsetptr3;
  79     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  80     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  81     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  82     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  84     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  85     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  86     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  93     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  94     __m256i          vfitab;
  95     __m128i          vfitab_lo,vfitab_hi;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m256           dummy_mask,cutoff_mask;
 100     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 101     __m256           one     = _mm256_set1_ps(1.0);
 102     __m256           two     = _mm256_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm256_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_vdw->data;
 121     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 126     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 127     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 128     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136     j_coord_offsetE = 0;
 137     j_coord_offsetF = 0;
 138     j_coord_offsetG = 0;
 139     j_coord_offsetH = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix0             = _mm256_setzero_ps();
 168         fiy0             = _mm256_setzero_ps();
 169         fiz0             = _mm256_setzero_ps();
 170         fix1             = _mm256_setzero_ps();
 171         fiy1             = _mm256_setzero_ps();
 172         fiz1             = _mm256_setzero_ps();
 173         fix2             = _mm256_setzero_ps();
 174         fiy2             = _mm256_setzero_ps();
 175         fiz2             = _mm256_setzero_ps();
 176         fix3             = _mm256_setzero_ps();
 177         fiy3             = _mm256_setzero_ps();
 178         fiz3             = _mm256_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_ps();
 182         vvdwsum          = _mm256_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             jnrE             = jjnr[jidx+4];
 194             jnrF             = jjnr[jidx+5];
 195             jnrG             = jjnr[jidx+6];
 196             jnrH             = jjnr[jidx+7];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201             j_coord_offsetE  = DIM*jnrE;
 202             j_coord_offsetF  = DIM*jnrF;
 203             j_coord_offsetG  = DIM*jnrG;
 204             j_coord_offsetH  = DIM*jnrH;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 210                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_ps(ix0,jx0);
 215             dy00             = _mm256_sub_ps(iy0,jy0);
 216             dz00             = _mm256_sub_ps(iz0,jz0);
 217             dx10             = _mm256_sub_ps(ix1,jx0);
 218             dy10             = _mm256_sub_ps(iy1,jy0);
 219             dz10             = _mm256_sub_ps(iz1,jz0);
 220             dx20             = _mm256_sub_ps(ix2,jx0);
 221             dy20             = _mm256_sub_ps(iy2,jy0);
 222             dz20             = _mm256_sub_ps(iz2,jz0);
 223             dx30             = _mm256_sub_ps(ix3,jx0);
 224             dy30             = _mm256_sub_ps(iy3,jy0);
 225             dz30             = _mm256_sub_ps(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 229             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 230             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 231             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 232
 233             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 234             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 235             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 236             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 237
 238             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 239             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 240             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                                  charge+jnrC+0,charge+jnrD+0,
 245                                                                  charge+jnrE+0,charge+jnrF+0,
 246                                                                  charge+jnrG+0,charge+jnrH+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251             vdwjidx0E        = 2*vdwtype[jnrE+0];
 252             vdwjidx0F        = 2*vdwtype[jnrF+0];
 253             vdwjidx0G        = 2*vdwtype[jnrG+0];
 254             vdwjidx0H        = 2*vdwtype[jnrH+0];
 255
 256             fjx0             = _mm256_setzero_ps();
 257             fjy0             = _mm256_setzero_ps();
 258             fjz0             = _mm256_setzero_ps();
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             r00              = _mm256_mul_ps(rsq00,rinv00);
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 268                                             vdwioffsetptr0+vdwjidx0B,
 269                                             vdwioffsetptr0+vdwjidx0C,
 270                                             vdwioffsetptr0+vdwjidx0D,
 271                                             vdwioffsetptr0+vdwjidx0E,
 272                                             vdwioffsetptr0+vdwjidx0F,
 273                                             vdwioffsetptr0+vdwjidx0G,
 274                                             vdwioffsetptr0+vdwjidx0H,
 275                                             &c6_00,&c12_00);
 276
 277             /* Calculate table index by multiplying r with table scale and truncate to integer */
 278             rt               = _mm256_mul_ps(r00,vftabscale);
 279             vfitab           = _mm256_cvttps_epi32(rt);
 280             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 281             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 282             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 283             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 284             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 285             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 286
 287             /* CUBIC SPLINE TABLE DISPERSION */
 288             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 289                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 290             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 292             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 294             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 296             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 297             Heps             = _mm256_mul_ps(vfeps,H);
 298             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 299             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 300             vvdw6            = _mm256_mul_ps(c6_00,VV);
 301             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 302             fvdw6            = _mm256_mul_ps(c6_00,FF);
 303
 304             /* CUBIC SPLINE TABLE REPULSION */
 305             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 306             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 307             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 309             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 311             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 313             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 314                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 315             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 316             Heps             = _mm256_mul_ps(vfeps,H);
 317             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 318             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 319             vvdw12           = _mm256_mul_ps(c12_00,VV);
 320             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 321             fvdw12           = _mm256_mul_ps(c12_00,FF);
 322             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 323             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 327
 328             fscal            = fvdw;
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm256_mul_ps(fscal,dx00);
 332             ty               = _mm256_mul_ps(fscal,dy00);
 333             tz               = _mm256_mul_ps(fscal,dz00);
 334
 335             /* Update vectorial force */
 336             fix0             = _mm256_add_ps(fix0,tx);
 337             fiy0             = _mm256_add_ps(fiy0,ty);
 338             fiz0             = _mm256_add_ps(fiz0,tz);
 339
 340             fjx0             = _mm256_add_ps(fjx0,tx);
 341             fjy0             = _mm256_add_ps(fjy0,ty);
 342             fjz0             = _mm256_add_ps(fjz0,tz);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq10             = _mm256_mul_ps(iq1,jq0);
 350
 351             /* COULOMB ELECTROSTATICS */
 352             velec            = _mm256_mul_ps(qq10,rinv10);
 353             felec            = _mm256_mul_ps(velec,rinvsq10);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm256_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_ps(fscal,dx10);
 362             ty               = _mm256_mul_ps(fscal,dy10);
 363             tz               = _mm256_mul_ps(fscal,dz10);
 364
 365             /* Update vectorial force */
 366             fix1             = _mm256_add_ps(fix1,tx);
 367             fiy1             = _mm256_add_ps(fiy1,ty);
 368             fiz1             = _mm256_add_ps(fiz1,tz);
 369
 370             fjx0             = _mm256_add_ps(fjx0,tx);
 371             fjy0             = _mm256_add_ps(fjy0,ty);
 372             fjz0             = _mm256_add_ps(fjz0,tz);
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq20             = _mm256_mul_ps(iq2,jq0);
 380
 381             /* COULOMB ELECTROSTATICS */
 382             velec            = _mm256_mul_ps(qq20,rinv20);
 383             felec            = _mm256_mul_ps(velec,rinvsq20);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velecsum         = _mm256_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm256_mul_ps(fscal,dx20);
 392             ty               = _mm256_mul_ps(fscal,dy20);
 393             tz               = _mm256_mul_ps(fscal,dz20);
 394
 395             /* Update vectorial force */
 396             fix2             = _mm256_add_ps(fix2,tx);
 397             fiy2             = _mm256_add_ps(fiy2,ty);
 398             fiz2             = _mm256_add_ps(fiz2,tz);
 399
 400             fjx0             = _mm256_add_ps(fjx0,tx);
 401             fjy0             = _mm256_add_ps(fjy0,ty);
 402             fjz0             = _mm256_add_ps(fjz0,tz);
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             /* Compute parameters for interactions between i and j atoms */
 409             qq30             = _mm256_mul_ps(iq3,jq0);
 410
 411             /* COULOMB ELECTROSTATICS */
 412             velec            = _mm256_mul_ps(qq30,rinv30);
 413             felec            = _mm256_mul_ps(velec,rinvsq30);
 414
 415             /* Update potential sum for this i atom from the interaction with this j atom. */
 416             velecsum         = _mm256_add_ps(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             /* Calculate temporary vectorial force */
 421             tx               = _mm256_mul_ps(fscal,dx30);
 422             ty               = _mm256_mul_ps(fscal,dy30);
 423             tz               = _mm256_mul_ps(fscal,dz30);
 424
 425             /* Update vectorial force */
 426             fix3             = _mm256_add_ps(fix3,tx);
 427             fiy3             = _mm256_add_ps(fiy3,ty);
 428             fiz3             = _mm256_add_ps(fiz3,tz);
 429
 430             fjx0             = _mm256_add_ps(fjx0,tx);
 431             fjy0             = _mm256_add_ps(fjy0,ty);
 432             fjz0             = _mm256_add_ps(fjz0,tz);
 433
 434             fjptrA             = f+j_coord_offsetA;
 435             fjptrB             = f+j_coord_offsetB;
 436             fjptrC             = f+j_coord_offsetC;
 437             fjptrD             = f+j_coord_offsetD;
 438             fjptrE             = f+j_coord_offsetE;
 439             fjptrF             = f+j_coord_offsetF;
 440             fjptrG             = f+j_coord_offsetG;
 441             fjptrH             = f+j_coord_offsetH;
 442
 443             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 444
 445             /* Inner loop uses 140 flops */
 446         }
 447
 448         if(jidx<j_index_end)
 449         {
 450
 451             /* Get j neighbor index, and coordinate index */
 452             jnrlistA         = jjnr[jidx];
 453             jnrlistB         = jjnr[jidx+1];
 454             jnrlistC         = jjnr[jidx+2];
 455             jnrlistD         = jjnr[jidx+3];
 456             jnrlistE         = jjnr[jidx+4];
 457             jnrlistF         = jjnr[jidx+5];
 458             jnrlistG         = jjnr[jidx+6];
 459             jnrlistH         = jjnr[jidx+7];
 460             /* Sign of each element will be negative for non-real atoms.
 461              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 462              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 463              */
 464             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 465                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 466
 467             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 468             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 469             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 470             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 471             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 472             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 473             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 474             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 475             j_coord_offsetA  = DIM*jnrA;
 476             j_coord_offsetB  = DIM*jnrB;
 477             j_coord_offsetC  = DIM*jnrC;
 478             j_coord_offsetD  = DIM*jnrD;
 479             j_coord_offsetE  = DIM*jnrE;
 480             j_coord_offsetF  = DIM*jnrF;
 481             j_coord_offsetG  = DIM*jnrG;
 482             j_coord_offsetH  = DIM*jnrH;
 483
 484             /* load j atom coordinates */
 485             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 486                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 487                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 488                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 489                                                  &jx0,&jy0,&jz0);
 490
 491             /* Calculate displacement vector */
 492             dx00             = _mm256_sub_ps(ix0,jx0);
 493             dy00             = _mm256_sub_ps(iy0,jy0);
 494             dz00             = _mm256_sub_ps(iz0,jz0);
 495             dx10             = _mm256_sub_ps(ix1,jx0);
 496             dy10             = _mm256_sub_ps(iy1,jy0);
 497             dz10             = _mm256_sub_ps(iz1,jz0);
 498             dx20             = _mm256_sub_ps(ix2,jx0);
 499             dy20             = _mm256_sub_ps(iy2,jy0);
 500             dz20             = _mm256_sub_ps(iz2,jz0);
 501             dx30             = _mm256_sub_ps(ix3,jx0);
 502             dy30             = _mm256_sub_ps(iy3,jy0);
 503             dz30             = _mm256_sub_ps(iz3,jz0);
 504
 505             /* Calculate squared distance and things based on it */
 506             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 507             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 508             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 509             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 510
 511             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 512             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 513             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 514             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 515
 516             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 517             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 518             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 519
 520             /* Load parameters for j particles */
 521             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 522                                                                  charge+jnrC+0,charge+jnrD+0,
 523                                                                  charge+jnrE+0,charge+jnrF+0,
 524                                                                  charge+jnrG+0,charge+jnrH+0);
 525             vdwjidx0A        = 2*vdwtype[jnrA+0];
 526             vdwjidx0B        = 2*vdwtype[jnrB+0];
 527             vdwjidx0C        = 2*vdwtype[jnrC+0];
 528             vdwjidx0D        = 2*vdwtype[jnrD+0];
 529             vdwjidx0E        = 2*vdwtype[jnrE+0];
 530             vdwjidx0F        = 2*vdwtype[jnrF+0];
 531             vdwjidx0G        = 2*vdwtype[jnrG+0];
 532             vdwjidx0H        = 2*vdwtype[jnrH+0];
 533
 534             fjx0             = _mm256_setzero_ps();
 535             fjy0             = _mm256_setzero_ps();
 536             fjz0             = _mm256_setzero_ps();
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             r00              = _mm256_mul_ps(rsq00,rinv00);
 543             r00              = _mm256_andnot_ps(dummy_mask,r00);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 547                                             vdwioffsetptr0+vdwjidx0B,
 548                                             vdwioffsetptr0+vdwjidx0C,
 549                                             vdwioffsetptr0+vdwjidx0D,
 550                                             vdwioffsetptr0+vdwjidx0E,
 551                                             vdwioffsetptr0+vdwjidx0F,
 552                                             vdwioffsetptr0+vdwjidx0G,
 553                                             vdwioffsetptr0+vdwjidx0H,
 554                                             &c6_00,&c12_00);
 555
 556             /* Calculate table index by multiplying r with table scale and truncate to integer */
 557             rt               = _mm256_mul_ps(r00,vftabscale);
 558             vfitab           = _mm256_cvttps_epi32(rt);
 559             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 560             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 561             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 562             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 563             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 564             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 565
 566             /* CUBIC SPLINE TABLE DISPERSION */
 567             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 568                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 569             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 570                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 571             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 572                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 573             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 574                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 575             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 576             Heps             = _mm256_mul_ps(vfeps,H);
 577             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 578             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 579             vvdw6            = _mm256_mul_ps(c6_00,VV);
 580             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 581             fvdw6            = _mm256_mul_ps(c6_00,FF);
 582
 583             /* CUBIC SPLINE TABLE REPULSION */
 584             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 585             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 586             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 587                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 588             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 589                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 590             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 591                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 592             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 593                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 594             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 595             Heps             = _mm256_mul_ps(vfeps,H);
 596             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 597             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 598             vvdw12           = _mm256_mul_ps(c12_00,VV);
 599             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 600             fvdw12           = _mm256_mul_ps(c12_00,FF);
 601             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 602             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 606             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 607
 608             fscal            = fvdw;
 609
 610             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm256_mul_ps(fscal,dx00);
 614             ty               = _mm256_mul_ps(fscal,dy00);
 615             tz               = _mm256_mul_ps(fscal,dz00);
 616
 617             /* Update vectorial force */
 618             fix0             = _mm256_add_ps(fix0,tx);
 619             fiy0             = _mm256_add_ps(fiy0,ty);
 620             fiz0             = _mm256_add_ps(fiz0,tz);
 621
 622             fjx0             = _mm256_add_ps(fjx0,tx);
 623             fjy0             = _mm256_add_ps(fjy0,ty);
 624             fjz0             = _mm256_add_ps(fjz0,tz);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq10             = _mm256_mul_ps(iq1,jq0);
 632
 633             /* COULOMB ELECTROSTATICS */
 634             velec            = _mm256_mul_ps(qq10,rinv10);
 635             felec            = _mm256_mul_ps(velec,rinvsq10);
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm256_andnot_ps(dummy_mask,velec);
 639             velecsum         = _mm256_add_ps(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm256_mul_ps(fscal,dx10);
 647             ty               = _mm256_mul_ps(fscal,dy10);
 648             tz               = _mm256_mul_ps(fscal,dz10);
 649
 650             /* Update vectorial force */
 651             fix1             = _mm256_add_ps(fix1,tx);
 652             fiy1             = _mm256_add_ps(fiy1,ty);
 653             fiz1             = _mm256_add_ps(fiz1,tz);
 654
 655             fjx0             = _mm256_add_ps(fjx0,tx);
 656             fjy0             = _mm256_add_ps(fjy0,ty);
 657             fjz0             = _mm256_add_ps(fjz0,tz);
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             /* Compute parameters for interactions between i and j atoms */
 664             qq20             = _mm256_mul_ps(iq2,jq0);
 665
 666             /* COULOMB ELECTROSTATICS */
 667             velec            = _mm256_mul_ps(qq20,rinv20);
 668             felec            = _mm256_mul_ps(velec,rinvsq20);
 669
 670             /* Update potential sum for this i atom from the interaction with this j atom. */
 671             velec            = _mm256_andnot_ps(dummy_mask,velec);
 672             velecsum         = _mm256_add_ps(velecsum,velec);
 673
 674             fscal            = felec;
 675
 676             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 677
 678             /* Calculate temporary vectorial force */
 679             tx               = _mm256_mul_ps(fscal,dx20);
 680             ty               = _mm256_mul_ps(fscal,dy20);
 681             tz               = _mm256_mul_ps(fscal,dz20);
 682
 683             /* Update vectorial force */
 684             fix2             = _mm256_add_ps(fix2,tx);
 685             fiy2             = _mm256_add_ps(fiy2,ty);
 686             fiz2             = _mm256_add_ps(fiz2,tz);
 687
 688             fjx0             = _mm256_add_ps(fjx0,tx);
 689             fjy0             = _mm256_add_ps(fjy0,ty);
 690             fjz0             = _mm256_add_ps(fjz0,tz);
 691
 692             /**************************
 693              * CALCULATE INTERACTIONS *
 694              **************************/
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq30             = _mm256_mul_ps(iq3,jq0);
 698
 699             /* COULOMB ELECTROSTATICS */
 700             velec            = _mm256_mul_ps(qq30,rinv30);
 701             felec            = _mm256_mul_ps(velec,rinvsq30);
 702
 703             /* Update potential sum for this i atom from the interaction with this j atom. */
 704             velec            = _mm256_andnot_ps(dummy_mask,velec);
 705             velecsum         = _mm256_add_ps(velecsum,velec);
 706
 707             fscal            = felec;
 708
 709             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 710
 711             /* Calculate temporary vectorial force */
 712             tx               = _mm256_mul_ps(fscal,dx30);
 713             ty               = _mm256_mul_ps(fscal,dy30);
 714             tz               = _mm256_mul_ps(fscal,dz30);
 715
 716             /* Update vectorial force */
 717             fix3             = _mm256_add_ps(fix3,tx);
 718             fiy3             = _mm256_add_ps(fiy3,ty);
 719             fiz3             = _mm256_add_ps(fiz3,tz);
 720
 721             fjx0             = _mm256_add_ps(fjx0,tx);
 722             fjy0             = _mm256_add_ps(fjy0,ty);
 723             fjz0             = _mm256_add_ps(fjz0,tz);
 724
 725             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 726             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 727             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 728             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 729             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 730             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 731             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 732             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 733
 734             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 735
 736             /* Inner loop uses 141 flops */
 737         }
 738
 739         /* End of innermost loop */
 740
 741         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 742                                                  f+i_coord_offset,fshift+i_shift_offset);
 743
 744         ggid                        = gid[iidx];
 745         /* Update potential energies */
 746         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 747         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 748
 749         /* Increment number of inner iterations */
 750         inneriter                  += j_index_end - j_index_start;
 751
 752         /* Outer loop uses 26 flops */
 753     }
 754
 755     /* Increment number of outer iterations */
 756     outeriter        += nri;
 757
 758     /* Update outer/inner flops */
 759
 760     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 761 }
 762 /*
 763  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_single
 764  * Electrostatics interaction: Coulomb
 765  * VdW interaction:            CubicSplineTable
 766  * Geometry:                   Water4-Particle
 767  * Calculate force/pot:        Force
 768  */
 769 void
 770 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_single
 771                     (t_nblist * gmx_restrict                nlist,
 772                      rvec * gmx_restrict                    xx,
 773                      rvec * gmx_restrict                    ff,
 774                      t_forcerec * gmx_restrict              fr,
 775                      t_mdatoms * gmx_restrict               mdatoms,
 776                      nb_kernel_data_t * gmx_restrict        kernel_data,
 777                      t_nrnb * gmx_restrict                  nrnb)
 778 {
 779     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 780      * just 0 for non-waters.
 781      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 782      * jnr indices corresponding to data put in the four positions in the SIMD register.
 783      */
 784     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 785     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 786     int              jnrA,jnrB,jnrC,jnrD;
 787     int              jnrE,jnrF,jnrG,jnrH;
 788     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 789     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 790     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 791     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 792     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 793     real             rcutoff_scalar;
 794     real             *shiftvec,*fshift,*x,*f;
 795     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 796     real             scratch[4*DIM];
 797     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 798     real *           vdwioffsetptr0;
 799     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 800     real *           vdwioffsetptr1;
 801     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 802     real *           vdwioffsetptr2;
 803     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 804     real *           vdwioffsetptr3;
 805     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 806     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 807     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 808     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 809     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 810     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 811     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 812     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 813     real             *charge;
 814     int              nvdwtype;
 815     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 816     int              *vdwtype;
 817     real             *vdwparam;
 818     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 819     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 820     __m256i          vfitab;
 821     __m128i          vfitab_lo,vfitab_hi;
 822     __m128i          ifour       = _mm_set1_epi32(4);
 823     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 824     real             *vftab;
 825     __m256           dummy_mask,cutoff_mask;
 826     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 827     __m256           one     = _mm256_set1_ps(1.0);
 828     __m256           two     = _mm256_set1_ps(2.0);
 829     x                = xx[0];
 830     f                = ff[0];
 831
 832     nri              = nlist->nri;
 833     iinr             = nlist->iinr;
 834     jindex           = nlist->jindex;
 835     jjnr             = nlist->jjnr;
 836     shiftidx         = nlist->shift;
 837     gid              = nlist->gid;
 838     shiftvec         = fr->shift_vec[0];
 839     fshift           = fr->fshift[0];
 840     facel            = _mm256_set1_ps(fr->epsfac);
 841     charge           = mdatoms->chargeA;
 842     nvdwtype         = fr->ntype;
 843     vdwparam         = fr->nbfp;
 844     vdwtype          = mdatoms->typeA;
 845
 846     vftab            = kernel_data->table_vdw->data;
 847     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 848
 849     /* Setup water-specific parameters */
 850     inr              = nlist->iinr[0];
 851     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 852     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 853     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 854     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 855
 856     /* Avoid stupid compiler warnings */
 857     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 858     j_coord_offsetA = 0;
 859     j_coord_offsetB = 0;
 860     j_coord_offsetC = 0;
 861     j_coord_offsetD = 0;
 862     j_coord_offsetE = 0;
 863     j_coord_offsetF = 0;
 864     j_coord_offsetG = 0;
 865     j_coord_offsetH = 0;
 866
 867     outeriter        = 0;
 868     inneriter        = 0;
 869
 870     for(iidx=0;iidx<4*DIM;iidx++)
 871     {
 872         scratch[iidx] = 0.0;
 873     }
 874
 875     /* Start outer loop over neighborlists */
 876     for(iidx=0; iidx<nri; iidx++)
 877     {
 878         /* Load shift vector for this list */
 879         i_shift_offset   = DIM*shiftidx[iidx];
 880
 881         /* Load limits for loop over neighbors */
 882         j_index_start    = jindex[iidx];
 883         j_index_end      = jindex[iidx+1];
 884
 885         /* Get outer coordinate index */
 886         inr              = iinr[iidx];
 887         i_coord_offset   = DIM*inr;
 888
 889         /* Load i particle coords and add shift vector */
 890         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 891                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 892
 893         fix0             = _mm256_setzero_ps();
 894         fiy0             = _mm256_setzero_ps();
 895         fiz0             = _mm256_setzero_ps();
 896         fix1             = _mm256_setzero_ps();
 897         fiy1             = _mm256_setzero_ps();
 898         fiz1             = _mm256_setzero_ps();
 899         fix2             = _mm256_setzero_ps();
 900         fiy2             = _mm256_setzero_ps();
 901         fiz2             = _mm256_setzero_ps();
 902         fix3             = _mm256_setzero_ps();
 903         fiy3             = _mm256_setzero_ps();
 904         fiz3             = _mm256_setzero_ps();
 905
 906         /* Start inner kernel loop */
 907         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 908         {
 909
 910             /* Get j neighbor index, and coordinate index */
 911             jnrA             = jjnr[jidx];
 912             jnrB             = jjnr[jidx+1];
 913             jnrC             = jjnr[jidx+2];
 914             jnrD             = jjnr[jidx+3];
 915             jnrE             = jjnr[jidx+4];
 916             jnrF             = jjnr[jidx+5];
 917             jnrG             = jjnr[jidx+6];
 918             jnrH             = jjnr[jidx+7];
 919             j_coord_offsetA  = DIM*jnrA;
 920             j_coord_offsetB  = DIM*jnrB;
 921             j_coord_offsetC  = DIM*jnrC;
 922             j_coord_offsetD  = DIM*jnrD;
 923             j_coord_offsetE  = DIM*jnrE;
 924             j_coord_offsetF  = DIM*jnrF;
 925             j_coord_offsetG  = DIM*jnrG;
 926             j_coord_offsetH  = DIM*jnrH;
 927
 928             /* load j atom coordinates */
 929             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 930                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 931                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 932                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 933                                                  &jx0,&jy0,&jz0);
 934
 935             /* Calculate displacement vector */
 936             dx00             = _mm256_sub_ps(ix0,jx0);
 937             dy00             = _mm256_sub_ps(iy0,jy0);
 938             dz00             = _mm256_sub_ps(iz0,jz0);
 939             dx10             = _mm256_sub_ps(ix1,jx0);
 940             dy10             = _mm256_sub_ps(iy1,jy0);
 941             dz10             = _mm256_sub_ps(iz1,jz0);
 942             dx20             = _mm256_sub_ps(ix2,jx0);
 943             dy20             = _mm256_sub_ps(iy2,jy0);
 944             dz20             = _mm256_sub_ps(iz2,jz0);
 945             dx30             = _mm256_sub_ps(ix3,jx0);
 946             dy30             = _mm256_sub_ps(iy3,jy0);
 947             dz30             = _mm256_sub_ps(iz3,jz0);
 948
 949             /* Calculate squared distance and things based on it */
 950             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 951             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 952             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 953             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 954
 955             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 956             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 957             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 958             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 959
 960             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 961             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 962             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 963
 964             /* Load parameters for j particles */
 965             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 966                                                                  charge+jnrC+0,charge+jnrD+0,
 967                                                                  charge+jnrE+0,charge+jnrF+0,
 968                                                                  charge+jnrG+0,charge+jnrH+0);
 969             vdwjidx0A        = 2*vdwtype[jnrA+0];
 970             vdwjidx0B        = 2*vdwtype[jnrB+0];
 971             vdwjidx0C        = 2*vdwtype[jnrC+0];
 972             vdwjidx0D        = 2*vdwtype[jnrD+0];
 973             vdwjidx0E        = 2*vdwtype[jnrE+0];
 974             vdwjidx0F        = 2*vdwtype[jnrF+0];
 975             vdwjidx0G        = 2*vdwtype[jnrG+0];
 976             vdwjidx0H        = 2*vdwtype[jnrH+0];
 977
 978             fjx0             = _mm256_setzero_ps();
 979             fjy0             = _mm256_setzero_ps();
 980             fjz0             = _mm256_setzero_ps();
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             r00              = _mm256_mul_ps(rsq00,rinv00);
 987
 988             /* Compute parameters for interactions between i and j atoms */
 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,3);
1007             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1008
1009             /* CUBIC SPLINE TABLE DISPERSION */
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             fvdw6            = _mm256_mul_ps(c6_00,FF);
1023
1024             /* CUBIC SPLINE TABLE REPULSION */
1025             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1026             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1027             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1028                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1029             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1030                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1031             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1032                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1033             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1034                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1035             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1036             Heps             = _mm256_mul_ps(vfeps,H);
1037             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1038             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1039             fvdw12           = _mm256_mul_ps(c12_00,FF);
1040             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1041
1042             fscal            = fvdw;
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm256_mul_ps(fscal,dx00);
1046             ty               = _mm256_mul_ps(fscal,dy00);
1047             tz               = _mm256_mul_ps(fscal,dz00);
1048
1049             /* Update vectorial force */
1050             fix0             = _mm256_add_ps(fix0,tx);
1051             fiy0             = _mm256_add_ps(fiy0,ty);
1052             fiz0             = _mm256_add_ps(fiz0,tz);
1053
1054             fjx0             = _mm256_add_ps(fjx0,tx);
1055             fjy0             = _mm256_add_ps(fjy0,ty);
1056             fjz0             = _mm256_add_ps(fjz0,tz);
1057
1058             /**************************
1059              * CALCULATE INTERACTIONS *
1060              **************************/
1061
1062             /* Compute parameters for interactions between i and j atoms */
1063             qq10             = _mm256_mul_ps(iq1,jq0);
1064
1065             /* COULOMB ELECTROSTATICS */
1066             velec            = _mm256_mul_ps(qq10,rinv10);
1067             felec            = _mm256_mul_ps(velec,rinvsq10);
1068
1069             fscal            = felec;
1070
1071             /* Calculate temporary vectorial force */
1072             tx               = _mm256_mul_ps(fscal,dx10);
1073             ty               = _mm256_mul_ps(fscal,dy10);
1074             tz               = _mm256_mul_ps(fscal,dz10);
1075
1076             /* Update vectorial force */
1077             fix1             = _mm256_add_ps(fix1,tx);
1078             fiy1             = _mm256_add_ps(fiy1,ty);
1079             fiz1             = _mm256_add_ps(fiz1,tz);
1080
1081             fjx0             = _mm256_add_ps(fjx0,tx);
1082             fjy0             = _mm256_add_ps(fjy0,ty);
1083             fjz0             = _mm256_add_ps(fjz0,tz);
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             /* Compute parameters for interactions between i and j atoms */
1090             qq20             = _mm256_mul_ps(iq2,jq0);
1091
1092             /* COULOMB ELECTROSTATICS */
1093             velec            = _mm256_mul_ps(qq20,rinv20);
1094             felec            = _mm256_mul_ps(velec,rinvsq20);
1095
1096             fscal            = felec;
1097
1098             /* Calculate temporary vectorial force */
1099             tx               = _mm256_mul_ps(fscal,dx20);
1100             ty               = _mm256_mul_ps(fscal,dy20);
1101             tz               = _mm256_mul_ps(fscal,dz20);
1102
1103             /* Update vectorial force */
1104             fix2             = _mm256_add_ps(fix2,tx);
1105             fiy2             = _mm256_add_ps(fiy2,ty);
1106             fiz2             = _mm256_add_ps(fiz2,tz);
1107
1108             fjx0             = _mm256_add_ps(fjx0,tx);
1109             fjy0             = _mm256_add_ps(fjy0,ty);
1110             fjz0             = _mm256_add_ps(fjz0,tz);
1111
1112             /**************************
1113              * CALCULATE INTERACTIONS *
1114              **************************/
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq30             = _mm256_mul_ps(iq3,jq0);
1118
1119             /* COULOMB ELECTROSTATICS */
1120             velec            = _mm256_mul_ps(qq30,rinv30);
1121             felec            = _mm256_mul_ps(velec,rinvsq30);
1122
1123             fscal            = felec;
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm256_mul_ps(fscal,dx30);
1127             ty               = _mm256_mul_ps(fscal,dy30);
1128             tz               = _mm256_mul_ps(fscal,dz30);
1129
1130             /* Update vectorial force */
1131             fix3             = _mm256_add_ps(fix3,tx);
1132             fiy3             = _mm256_add_ps(fiy3,ty);
1133             fiz3             = _mm256_add_ps(fiz3,tz);
1134
1135             fjx0             = _mm256_add_ps(fjx0,tx);
1136             fjy0             = _mm256_add_ps(fjy0,ty);
1137             fjz0             = _mm256_add_ps(fjz0,tz);
1138
1139             fjptrA             = f+j_coord_offsetA;
1140             fjptrB             = f+j_coord_offsetB;
1141             fjptrC             = f+j_coord_offsetC;
1142             fjptrD             = f+j_coord_offsetD;
1143             fjptrE             = f+j_coord_offsetE;
1144             fjptrF             = f+j_coord_offsetF;
1145             fjptrG             = f+j_coord_offsetG;
1146             fjptrH             = f+j_coord_offsetH;
1147
1148             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1149
1150             /* Inner loop uses 129 flops */
1151         }
1152
1153         if(jidx<j_index_end)
1154         {
1155
1156             /* Get j neighbor index, and coordinate index */
1157             jnrlistA         = jjnr[jidx];
1158             jnrlistB         = jjnr[jidx+1];
1159             jnrlistC         = jjnr[jidx+2];
1160             jnrlistD         = jjnr[jidx+3];
1161             jnrlistE         = jjnr[jidx+4];
1162             jnrlistF         = jjnr[jidx+5];
1163             jnrlistG         = jjnr[jidx+6];
1164             jnrlistH         = jjnr[jidx+7];
1165             /* Sign of each element will be negative for non-real atoms.
1166              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1167              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1168              */
1169             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1170                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1171
1172             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1173             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1174             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1175             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1176             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1177             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1178             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1179             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1180             j_coord_offsetA  = DIM*jnrA;
1181             j_coord_offsetB  = DIM*jnrB;
1182             j_coord_offsetC  = DIM*jnrC;
1183             j_coord_offsetD  = DIM*jnrD;
1184             j_coord_offsetE  = DIM*jnrE;
1185             j_coord_offsetF  = DIM*jnrF;
1186             j_coord_offsetG  = DIM*jnrG;
1187             j_coord_offsetH  = DIM*jnrH;
1188
1189             /* load j atom coordinates */
1190             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1191                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1192                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1193                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1194                                                  &jx0,&jy0,&jz0);
1195
1196             /* Calculate displacement vector */
1197             dx00             = _mm256_sub_ps(ix0,jx0);
1198             dy00             = _mm256_sub_ps(iy0,jy0);
1199             dz00             = _mm256_sub_ps(iz0,jz0);
1200             dx10             = _mm256_sub_ps(ix1,jx0);
1201             dy10             = _mm256_sub_ps(iy1,jy0);
1202             dz10             = _mm256_sub_ps(iz1,jz0);
1203             dx20             = _mm256_sub_ps(ix2,jx0);
1204             dy20             = _mm256_sub_ps(iy2,jy0);
1205             dz20             = _mm256_sub_ps(iz2,jz0);
1206             dx30             = _mm256_sub_ps(ix3,jx0);
1207             dy30             = _mm256_sub_ps(iy3,jy0);
1208             dz30             = _mm256_sub_ps(iz3,jz0);
1209
1210             /* Calculate squared distance and things based on it */
1211             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1212             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1213             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1214             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1215
1216             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1217             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1218             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1219             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1220
1221             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1222             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1223             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1224
1225             /* Load parameters for j particles */
1226             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1227                                                                  charge+jnrC+0,charge+jnrD+0,
1228                                                                  charge+jnrE+0,charge+jnrF+0,
1229                                                                  charge+jnrG+0,charge+jnrH+0);
1230             vdwjidx0A        = 2*vdwtype[jnrA+0];
1231             vdwjidx0B        = 2*vdwtype[jnrB+0];
1232             vdwjidx0C        = 2*vdwtype[jnrC+0];
1233             vdwjidx0D        = 2*vdwtype[jnrD+0];
1234             vdwjidx0E        = 2*vdwtype[jnrE+0];
1235             vdwjidx0F        = 2*vdwtype[jnrF+0];
1236             vdwjidx0G        = 2*vdwtype[jnrG+0];
1237             vdwjidx0H        = 2*vdwtype[jnrH+0];
1238
1239             fjx0             = _mm256_setzero_ps();
1240             fjy0             = _mm256_setzero_ps();
1241             fjz0             = _mm256_setzero_ps();
1242
1243             /**************************
1244              * CALCULATE INTERACTIONS *
1245              **************************/
1246
1247             r00              = _mm256_mul_ps(rsq00,rinv00);
1248             r00              = _mm256_andnot_ps(dummy_mask,r00);
1249
1250             /* Compute parameters for interactions between i and j atoms */
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,3);
1269             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1270
1271             /* CUBIC SPLINE TABLE DISPERSION */
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             fvdw6            = _mm256_mul_ps(c6_00,FF);
1285
1286             /* CUBIC SPLINE TABLE REPULSION */
1287             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1288             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1289             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1291             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1293             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1295             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1297             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1298             Heps             = _mm256_mul_ps(vfeps,H);
1299             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1300             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1301             fvdw12           = _mm256_mul_ps(c12_00,FF);
1302             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1303
1304             fscal            = fvdw;
1305
1306             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1307
1308             /* Calculate temporary vectorial force */
1309             tx               = _mm256_mul_ps(fscal,dx00);
1310             ty               = _mm256_mul_ps(fscal,dy00);
1311             tz               = _mm256_mul_ps(fscal,dz00);
1312
1313             /* Update vectorial force */
1314             fix0             = _mm256_add_ps(fix0,tx);
1315             fiy0             = _mm256_add_ps(fiy0,ty);
1316             fiz0             = _mm256_add_ps(fiz0,tz);
1317
1318             fjx0             = _mm256_add_ps(fjx0,tx);
1319             fjy0             = _mm256_add_ps(fjy0,ty);
1320             fjz0             = _mm256_add_ps(fjz0,tz);
1321
1322             /**************************
1323              * CALCULATE INTERACTIONS *
1324              **************************/
1325
1326             /* Compute parameters for interactions between i and j atoms */
1327             qq10             = _mm256_mul_ps(iq1,jq0);
1328
1329             /* COULOMB ELECTROSTATICS */
1330             velec            = _mm256_mul_ps(qq10,rinv10);
1331             felec            = _mm256_mul_ps(velec,rinvsq10);
1332
1333             fscal            = felec;
1334
1335             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1336
1337             /* Calculate temporary vectorial force */
1338             tx               = _mm256_mul_ps(fscal,dx10);
1339             ty               = _mm256_mul_ps(fscal,dy10);
1340             tz               = _mm256_mul_ps(fscal,dz10);
1341
1342             /* Update vectorial force */
1343             fix1             = _mm256_add_ps(fix1,tx);
1344             fiy1             = _mm256_add_ps(fiy1,ty);
1345             fiz1             = _mm256_add_ps(fiz1,tz);
1346
1347             fjx0             = _mm256_add_ps(fjx0,tx);
1348             fjy0             = _mm256_add_ps(fjy0,ty);
1349             fjz0             = _mm256_add_ps(fjz0,tz);
1350
1351             /**************************
1352              * CALCULATE INTERACTIONS *
1353              **************************/
1354
1355             /* Compute parameters for interactions between i and j atoms */
1356             qq20             = _mm256_mul_ps(iq2,jq0);
1357
1358             /* COULOMB ELECTROSTATICS */
1359             velec            = _mm256_mul_ps(qq20,rinv20);
1360             felec            = _mm256_mul_ps(velec,rinvsq20);
1361
1362             fscal            = felec;
1363
1364             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1365
1366             /* Calculate temporary vectorial force */
1367             tx               = _mm256_mul_ps(fscal,dx20);
1368             ty               = _mm256_mul_ps(fscal,dy20);
1369             tz               = _mm256_mul_ps(fscal,dz20);
1370
1371             /* Update vectorial force */
1372             fix2             = _mm256_add_ps(fix2,tx);
1373             fiy2             = _mm256_add_ps(fiy2,ty);
1374             fiz2             = _mm256_add_ps(fiz2,tz);
1375
1376             fjx0             = _mm256_add_ps(fjx0,tx);
1377             fjy0             = _mm256_add_ps(fjy0,ty);
1378             fjz0             = _mm256_add_ps(fjz0,tz);
1379
1380             /**************************
1381              * CALCULATE INTERACTIONS *
1382              **************************/
1383
1384             /* Compute parameters for interactions between i and j atoms */
1385             qq30             = _mm256_mul_ps(iq3,jq0);
1386
1387             /* COULOMB ELECTROSTATICS */
1388             velec            = _mm256_mul_ps(qq30,rinv30);
1389             felec            = _mm256_mul_ps(velec,rinvsq30);
1390
1391             fscal            = felec;
1392
1393             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1394
1395             /* Calculate temporary vectorial force */
1396             tx               = _mm256_mul_ps(fscal,dx30);
1397             ty               = _mm256_mul_ps(fscal,dy30);
1398             tz               = _mm256_mul_ps(fscal,dz30);
1399
1400             /* Update vectorial force */
1401             fix3             = _mm256_add_ps(fix3,tx);
1402             fiy3             = _mm256_add_ps(fiy3,ty);
1403             fiz3             = _mm256_add_ps(fiz3,tz);
1404
1405             fjx0             = _mm256_add_ps(fjx0,tx);
1406             fjy0             = _mm256_add_ps(fjy0,ty);
1407             fjz0             = _mm256_add_ps(fjz0,tz);
1408
1409             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1410             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1411             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1412             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1413             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1414             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1415             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1416             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1417
1418             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1419
1420             /* Inner loop uses 130 flops */
1421         }
1422
1423         /* End of innermost loop */
1424
1425         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1426                                                  f+i_coord_offset,fshift+i_shift_offset);
1427
1428         /* Increment number of inner iterations */
1429         inneriter                  += j_index_end - j_index_start;
1430
1431         /* Outer loop uses 24 flops */
1432     }
1433
1434     /* Increment number of outer iterations */
1435     outeriter        += nri;
1436
1437     /* Update outer/inner flops */
1438
1439     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1440 }