src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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_ElecEw_VdwCSTab_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256i          vfitab;
  92     __m128i          vfitab_lo,vfitab_hi;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m256i          ewitab;
  97     __m128i          ewitab_lo,ewitab_hi;
  98     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  99     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 100     real             *ewtab;
 101     __m256           dummy_mask,cutoff_mask;
 102     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 103     __m256           one     = _mm256_set1_ps(1.0);
 104     __m256           two     = _mm256_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm256_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     vftab            = kernel_data->table_vdw->data;
 123     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 124
 125     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 126     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 127     beta2            = _mm256_mul_ps(beta,beta);
 128     beta3            = _mm256_mul_ps(beta,beta2);
 129
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 137     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 138     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181         fix1             = _mm256_setzero_ps();
 182         fiy1             = _mm256_setzero_ps();
 183         fiz1             = _mm256_setzero_ps();
 184         fix2             = _mm256_setzero_ps();
 185         fiy2             = _mm256_setzero_ps();
 186         fiz2             = _mm256_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190         vvdwsum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 240
 241             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 242             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 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,3);
 289             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 290
 291             /* EWALD ELECTROSTATICS */
 292
 293             /* Analytical PME correction */
 294             zeta2            = _mm256_mul_ps(beta2,rsq00);
 295             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 296             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 297             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 298             felec            = _mm256_mul_ps(qq00,felec);
 299             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 300             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 301             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 302             velec            = _mm256_mul_ps(qq00,velec);
 303
 304             /* CUBIC SPLINE TABLE DISPERSION */
 305             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 307             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 309             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 311             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 313             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 314             Heps             = _mm256_mul_ps(vfeps,H);
 315             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 316             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 317             vvdw6            = _mm256_mul_ps(c6_00,VV);
 318             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 319             fvdw6            = _mm256_mul_ps(c6_00,FF);
 320
 321             /* CUBIC SPLINE TABLE REPULSION */
 322             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 323             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 324             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 326             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 328             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 330             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 332             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 333             Heps             = _mm256_mul_ps(vfeps,H);
 334             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 335             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 336             vvdw12           = _mm256_mul_ps(c12_00,VV);
 337             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 338             fvdw12           = _mm256_mul_ps(c12_00,FF);
 339             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 340             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velecsum         = _mm256_add_ps(velecsum,velec);
 344             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 345
 346             fscal            = _mm256_add_ps(felec,fvdw);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_ps(fscal,dx00);
 350             ty               = _mm256_mul_ps(fscal,dy00);
 351             tz               = _mm256_mul_ps(fscal,dz00);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm256_add_ps(fix0,tx);
 355             fiy0             = _mm256_add_ps(fiy0,ty);
 356             fiz0             = _mm256_add_ps(fiz0,tz);
 357
 358             fjx0             = _mm256_add_ps(fjx0,tx);
 359             fjy0             = _mm256_add_ps(fjy0,ty);
 360             fjz0             = _mm256_add_ps(fjz0,tz);
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             r10              = _mm256_mul_ps(rsq10,rinv10);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq10             = _mm256_mul_ps(iq1,jq0);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Analytical PME correction */
 374             zeta2            = _mm256_mul_ps(beta2,rsq10);
 375             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 376             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 377             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 378             felec            = _mm256_mul_ps(qq10,felec);
 379             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 380             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 381             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 382             velec            = _mm256_mul_ps(qq10,velec);
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velecsum         = _mm256_add_ps(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             /* Calculate temporary vectorial force */
 390             tx               = _mm256_mul_ps(fscal,dx10);
 391             ty               = _mm256_mul_ps(fscal,dy10);
 392             tz               = _mm256_mul_ps(fscal,dz10);
 393
 394             /* Update vectorial force */
 395             fix1             = _mm256_add_ps(fix1,tx);
 396             fiy1             = _mm256_add_ps(fiy1,ty);
 397             fiz1             = _mm256_add_ps(fiz1,tz);
 398
 399             fjx0             = _mm256_add_ps(fjx0,tx);
 400             fjy0             = _mm256_add_ps(fjy0,ty);
 401             fjz0             = _mm256_add_ps(fjz0,tz);
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             r20              = _mm256_mul_ps(rsq20,rinv20);
 408
 409             /* Compute parameters for interactions between i and j atoms */
 410             qq20             = _mm256_mul_ps(iq2,jq0);
 411
 412             /* EWALD ELECTROSTATICS */
 413
 414             /* Analytical PME correction */
 415             zeta2            = _mm256_mul_ps(beta2,rsq20);
 416             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 417             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 418             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 419             felec            = _mm256_mul_ps(qq20,felec);
 420             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 421             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 422             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 423             velec            = _mm256_mul_ps(qq20,velec);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velecsum         = _mm256_add_ps(velecsum,velec);
 427
 428             fscal            = felec;
 429
 430             /* Calculate temporary vectorial force */
 431             tx               = _mm256_mul_ps(fscal,dx20);
 432             ty               = _mm256_mul_ps(fscal,dy20);
 433             tz               = _mm256_mul_ps(fscal,dz20);
 434
 435             /* Update vectorial force */
 436             fix2             = _mm256_add_ps(fix2,tx);
 437             fiy2             = _mm256_add_ps(fiy2,ty);
 438             fiz2             = _mm256_add_ps(fiz2,tz);
 439
 440             fjx0             = _mm256_add_ps(fjx0,tx);
 441             fjy0             = _mm256_add_ps(fjy0,ty);
 442             fjz0             = _mm256_add_ps(fjz0,tz);
 443
 444             fjptrA             = f+j_coord_offsetA;
 445             fjptrB             = f+j_coord_offsetB;
 446             fjptrC             = f+j_coord_offsetC;
 447             fjptrD             = f+j_coord_offsetD;
 448             fjptrE             = f+j_coord_offsetE;
 449             fjptrF             = f+j_coord_offsetF;
 450             fjptrG             = f+j_coord_offsetG;
 451             fjptrH             = f+j_coord_offsetH;
 452
 453             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 454
 455             /* Inner loop uses 289 flops */
 456         }
 457
 458         if(jidx<j_index_end)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrlistA         = jjnr[jidx];
 463             jnrlistB         = jjnr[jidx+1];
 464             jnrlistC         = jjnr[jidx+2];
 465             jnrlistD         = jjnr[jidx+3];
 466             jnrlistE         = jjnr[jidx+4];
 467             jnrlistF         = jjnr[jidx+5];
 468             jnrlistG         = jjnr[jidx+6];
 469             jnrlistH         = jjnr[jidx+7];
 470             /* Sign of each element will be negative for non-real atoms.
 471              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 472              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 473              */
 474             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 475                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 476
 477             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 478             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 479             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 480             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 481             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 482             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 483             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 484             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 485             j_coord_offsetA  = DIM*jnrA;
 486             j_coord_offsetB  = DIM*jnrB;
 487             j_coord_offsetC  = DIM*jnrC;
 488             j_coord_offsetD  = DIM*jnrD;
 489             j_coord_offsetE  = DIM*jnrE;
 490             j_coord_offsetF  = DIM*jnrF;
 491             j_coord_offsetG  = DIM*jnrG;
 492             j_coord_offsetH  = DIM*jnrH;
 493
 494             /* load j atom coordinates */
 495             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 496                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 497                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 498                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 499                                                  &jx0,&jy0,&jz0);
 500
 501             /* Calculate displacement vector */
 502             dx00             = _mm256_sub_ps(ix0,jx0);
 503             dy00             = _mm256_sub_ps(iy0,jy0);
 504             dz00             = _mm256_sub_ps(iz0,jz0);
 505             dx10             = _mm256_sub_ps(ix1,jx0);
 506             dy10             = _mm256_sub_ps(iy1,jy0);
 507             dz10             = _mm256_sub_ps(iz1,jz0);
 508             dx20             = _mm256_sub_ps(ix2,jx0);
 509             dy20             = _mm256_sub_ps(iy2,jy0);
 510             dz20             = _mm256_sub_ps(iz2,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 514             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 515             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 516
 517             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 518             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 519             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 520
 521             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 522             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 523             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 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,3);
 570             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 571
 572             /* EWALD ELECTROSTATICS */
 573
 574             /* Analytical PME correction */
 575             zeta2            = _mm256_mul_ps(beta2,rsq00);
 576             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 577             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 578             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 579             felec            = _mm256_mul_ps(qq00,felec);
 580             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 581             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 582             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 583             velec            = _mm256_mul_ps(qq00,velec);
 584
 585             /* CUBIC SPLINE TABLE DISPERSION */
 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             vvdw6            = _mm256_mul_ps(c6_00,VV);
 599             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 600             fvdw6            = _mm256_mul_ps(c6_00,FF);
 601
 602             /* CUBIC SPLINE TABLE REPULSION */
 603             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 604             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 605             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 606                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 607             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 608                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 609             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 610                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 611             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 612                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 613             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 614             Heps             = _mm256_mul_ps(vfeps,H);
 615             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 616             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 617             vvdw12           = _mm256_mul_ps(c12_00,VV);
 618             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 619             fvdw12           = _mm256_mul_ps(c12_00,FF);
 620             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 621             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             velec            = _mm256_andnot_ps(dummy_mask,velec);
 625             velecsum         = _mm256_add_ps(velecsum,velec);
 626             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 627             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 628
 629             fscal            = _mm256_add_ps(felec,fvdw);
 630
 631             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm256_mul_ps(fscal,dx00);
 635             ty               = _mm256_mul_ps(fscal,dy00);
 636             tz               = _mm256_mul_ps(fscal,dz00);
 637
 638             /* Update vectorial force */
 639             fix0             = _mm256_add_ps(fix0,tx);
 640             fiy0             = _mm256_add_ps(fiy0,ty);
 641             fiz0             = _mm256_add_ps(fiz0,tz);
 642
 643             fjx0             = _mm256_add_ps(fjx0,tx);
 644             fjy0             = _mm256_add_ps(fjy0,ty);
 645             fjz0             = _mm256_add_ps(fjz0,tz);
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             r10              = _mm256_mul_ps(rsq10,rinv10);
 652             r10              = _mm256_andnot_ps(dummy_mask,r10);
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq10             = _mm256_mul_ps(iq1,jq0);
 656
 657             /* EWALD ELECTROSTATICS */
 658
 659             /* Analytical PME correction */
 660             zeta2            = _mm256_mul_ps(beta2,rsq10);
 661             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 662             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 663             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 664             felec            = _mm256_mul_ps(qq10,felec);
 665             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 666             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 667             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 668             velec            = _mm256_mul_ps(qq10,velec);
 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,dx10);
 680             ty               = _mm256_mul_ps(fscal,dy10);
 681             tz               = _mm256_mul_ps(fscal,dz10);
 682
 683             /* Update vectorial force */
 684             fix1             = _mm256_add_ps(fix1,tx);
 685             fiy1             = _mm256_add_ps(fiy1,ty);
 686             fiz1             = _mm256_add_ps(fiz1,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             r20              = _mm256_mul_ps(rsq20,rinv20);
 697             r20              = _mm256_andnot_ps(dummy_mask,r20);
 698
 699             /* Compute parameters for interactions between i and j atoms */
 700             qq20             = _mm256_mul_ps(iq2,jq0);
 701
 702             /* EWALD ELECTROSTATICS */
 703
 704             /* Analytical PME correction */
 705             zeta2            = _mm256_mul_ps(beta2,rsq20);
 706             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 707             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 708             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 709             felec            = _mm256_mul_ps(qq20,felec);
 710             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 711             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 712             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 713             velec            = _mm256_mul_ps(qq20,velec);
 714
 715             /* Update potential sum for this i atom from the interaction with this j atom. */
 716             velec            = _mm256_andnot_ps(dummy_mask,velec);
 717             velecsum         = _mm256_add_ps(velecsum,velec);
 718
 719             fscal            = felec;
 720
 721             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 722
 723             /* Calculate temporary vectorial force */
 724             tx               = _mm256_mul_ps(fscal,dx20);
 725             ty               = _mm256_mul_ps(fscal,dy20);
 726             tz               = _mm256_mul_ps(fscal,dz20);
 727
 728             /* Update vectorial force */
 729             fix2             = _mm256_add_ps(fix2,tx);
 730             fiy2             = _mm256_add_ps(fiy2,ty);
 731             fiz2             = _mm256_add_ps(fiz2,tz);
 732
 733             fjx0             = _mm256_add_ps(fjx0,tx);
 734             fjy0             = _mm256_add_ps(fjy0,ty);
 735             fjz0             = _mm256_add_ps(fjz0,tz);
 736
 737             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 738             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 739             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 740             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 741             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 742             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 743             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 744             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 745
 746             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 747
 748             /* Inner loop uses 292 flops */
 749         }
 750
 751         /* End of innermost loop */
 752
 753         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 754                                                  f+i_coord_offset,fshift+i_shift_offset);
 755
 756         ggid                        = gid[iidx];
 757         /* Update potential energies */
 758         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 759         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 760
 761         /* Increment number of inner iterations */
 762         inneriter                  += j_index_end - j_index_start;
 763
 764         /* Outer loop uses 20 flops */
 765     }
 766
 767     /* Increment number of outer iterations */
 768     outeriter        += nri;
 769
 770     /* Update outer/inner flops */
 771
 772     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*292);
 773 }
 774 /*
 775  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_single
 776  * Electrostatics interaction: Ewald
 777  * VdW interaction:            CubicSplineTable
 778  * Geometry:                   Water3-Particle
 779  * Calculate force/pot:        Force
 780  */
 781 void
 782 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_single
 783                     (t_nblist * gmx_restrict                nlist,
 784                      rvec * gmx_restrict                    xx,
 785                      rvec * gmx_restrict                    ff,
 786                      t_forcerec * gmx_restrict              fr,
 787                      t_mdatoms * gmx_restrict               mdatoms,
 788                      nb_kernel_data_t * gmx_restrict        kernel_data,
 789                      t_nrnb * gmx_restrict                  nrnb)
 790 {
 791     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 792      * just 0 for non-waters.
 793      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 794      * jnr indices corresponding to data put in the four positions in the SIMD register.
 795      */
 796     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 797     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 798     int              jnrA,jnrB,jnrC,jnrD;
 799     int              jnrE,jnrF,jnrG,jnrH;
 800     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 801     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 802     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 803     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 804     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 805     real             rcutoff_scalar;
 806     real             *shiftvec,*fshift,*x,*f;
 807     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 808     real             scratch[4*DIM];
 809     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 810     real *           vdwioffsetptr0;
 811     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 812     real *           vdwioffsetptr1;
 813     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 814     real *           vdwioffsetptr2;
 815     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 816     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 817     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 818     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 819     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 820     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 821     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 822     real             *charge;
 823     int              nvdwtype;
 824     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 825     int              *vdwtype;
 826     real             *vdwparam;
 827     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 828     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 829     __m256i          vfitab;
 830     __m128i          vfitab_lo,vfitab_hi;
 831     __m128i          ifour       = _mm_set1_epi32(4);
 832     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 833     real             *vftab;
 834     __m256i          ewitab;
 835     __m128i          ewitab_lo,ewitab_hi;
 836     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 837     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 838     real             *ewtab;
 839     __m256           dummy_mask,cutoff_mask;
 840     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 841     __m256           one     = _mm256_set1_ps(1.0);
 842     __m256           two     = _mm256_set1_ps(2.0);
 843     x                = xx[0];
 844     f                = ff[0];
 845
 846     nri              = nlist->nri;
 847     iinr             = nlist->iinr;
 848     jindex           = nlist->jindex;
 849     jjnr             = nlist->jjnr;
 850     shiftidx         = nlist->shift;
 851     gid              = nlist->gid;
 852     shiftvec         = fr->shift_vec[0];
 853     fshift           = fr->fshift[0];
 854     facel            = _mm256_set1_ps(fr->epsfac);
 855     charge           = mdatoms->chargeA;
 856     nvdwtype         = fr->ntype;
 857     vdwparam         = fr->nbfp;
 858     vdwtype          = mdatoms->typeA;
 859
 860     vftab            = kernel_data->table_vdw->data;
 861     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 862
 863     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 864     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 865     beta2            = _mm256_mul_ps(beta,beta);
 866     beta3            = _mm256_mul_ps(beta,beta2);
 867
 868     ewtab            = fr->ic->tabq_coul_F;
 869     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 870     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 871
 872     /* Setup water-specific parameters */
 873     inr              = nlist->iinr[0];
 874     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 875     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 876     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 877     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 878
 879     /* Avoid stupid compiler warnings */
 880     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 881     j_coord_offsetA = 0;
 882     j_coord_offsetB = 0;
 883     j_coord_offsetC = 0;
 884     j_coord_offsetD = 0;
 885     j_coord_offsetE = 0;
 886     j_coord_offsetF = 0;
 887     j_coord_offsetG = 0;
 888     j_coord_offsetH = 0;
 889
 890     outeriter        = 0;
 891     inneriter        = 0;
 892
 893     for(iidx=0;iidx<4*DIM;iidx++)
 894     {
 895         scratch[iidx] = 0.0;
 896     }
 897
 898     /* Start outer loop over neighborlists */
 899     for(iidx=0; iidx<nri; iidx++)
 900     {
 901         /* Load shift vector for this list */
 902         i_shift_offset   = DIM*shiftidx[iidx];
 903
 904         /* Load limits for loop over neighbors */
 905         j_index_start    = jindex[iidx];
 906         j_index_end      = jindex[iidx+1];
 907
 908         /* Get outer coordinate index */
 909         inr              = iinr[iidx];
 910         i_coord_offset   = DIM*inr;
 911
 912         /* Load i particle coords and add shift vector */
 913         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 914                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 915
 916         fix0             = _mm256_setzero_ps();
 917         fiy0             = _mm256_setzero_ps();
 918         fiz0             = _mm256_setzero_ps();
 919         fix1             = _mm256_setzero_ps();
 920         fiy1             = _mm256_setzero_ps();
 921         fiz1             = _mm256_setzero_ps();
 922         fix2             = _mm256_setzero_ps();
 923         fiy2             = _mm256_setzero_ps();
 924         fiz2             = _mm256_setzero_ps();
 925
 926         /* Start inner kernel loop */
 927         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 928         {
 929
 930             /* Get j neighbor index, and coordinate index */
 931             jnrA             = jjnr[jidx];
 932             jnrB             = jjnr[jidx+1];
 933             jnrC             = jjnr[jidx+2];
 934             jnrD             = jjnr[jidx+3];
 935             jnrE             = jjnr[jidx+4];
 936             jnrF             = jjnr[jidx+5];
 937             jnrG             = jjnr[jidx+6];
 938             jnrH             = jjnr[jidx+7];
 939             j_coord_offsetA  = DIM*jnrA;
 940             j_coord_offsetB  = DIM*jnrB;
 941             j_coord_offsetC  = DIM*jnrC;
 942             j_coord_offsetD  = DIM*jnrD;
 943             j_coord_offsetE  = DIM*jnrE;
 944             j_coord_offsetF  = DIM*jnrF;
 945             j_coord_offsetG  = DIM*jnrG;
 946             j_coord_offsetH  = DIM*jnrH;
 947
 948             /* load j atom coordinates */
 949             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 950                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 951                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 952                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 953                                                  &jx0,&jy0,&jz0);
 954
 955             /* Calculate displacement vector */
 956             dx00             = _mm256_sub_ps(ix0,jx0);
 957             dy00             = _mm256_sub_ps(iy0,jy0);
 958             dz00             = _mm256_sub_ps(iz0,jz0);
 959             dx10             = _mm256_sub_ps(ix1,jx0);
 960             dy10             = _mm256_sub_ps(iy1,jy0);
 961             dz10             = _mm256_sub_ps(iz1,jz0);
 962             dx20             = _mm256_sub_ps(ix2,jx0);
 963             dy20             = _mm256_sub_ps(iy2,jy0);
 964             dz20             = _mm256_sub_ps(iz2,jz0);
 965
 966             /* Calculate squared distance and things based on it */
 967             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 968             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 969             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 970
 971             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 972             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 973             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 974
 975             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 976             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 977             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 978
 979             /* Load parameters for j particles */
 980             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 981                                                                  charge+jnrC+0,charge+jnrD+0,
 982                                                                  charge+jnrE+0,charge+jnrF+0,
 983                                                                  charge+jnrG+0,charge+jnrH+0);
 984             vdwjidx0A        = 2*vdwtype[jnrA+0];
 985             vdwjidx0B        = 2*vdwtype[jnrB+0];
 986             vdwjidx0C        = 2*vdwtype[jnrC+0];
 987             vdwjidx0D        = 2*vdwtype[jnrD+0];
 988             vdwjidx0E        = 2*vdwtype[jnrE+0];
 989             vdwjidx0F        = 2*vdwtype[jnrF+0];
 990             vdwjidx0G        = 2*vdwtype[jnrG+0];
 991             vdwjidx0H        = 2*vdwtype[jnrH+0];
 992
 993             fjx0             = _mm256_setzero_ps();
 994             fjy0             = _mm256_setzero_ps();
 995             fjz0             = _mm256_setzero_ps();
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             r00              = _mm256_mul_ps(rsq00,rinv00);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq00             = _mm256_mul_ps(iq0,jq0);
1005             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1006                                             vdwioffsetptr0+vdwjidx0B,
1007                                             vdwioffsetptr0+vdwjidx0C,
1008                                             vdwioffsetptr0+vdwjidx0D,
1009                                             vdwioffsetptr0+vdwjidx0E,
1010                                             vdwioffsetptr0+vdwjidx0F,
1011                                             vdwioffsetptr0+vdwjidx0G,
1012                                             vdwioffsetptr0+vdwjidx0H,
1013                                             &c6_00,&c12_00);
1014
1015             /* Calculate table index by multiplying r with table scale and truncate to integer */
1016             rt               = _mm256_mul_ps(r00,vftabscale);
1017             vfitab           = _mm256_cvttps_epi32(rt);
1018             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1019             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1020             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1021             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1022             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1023             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1024
1025             /* EWALD ELECTROSTATICS */
1026
1027             /* Analytical PME correction */
1028             zeta2            = _mm256_mul_ps(beta2,rsq00);
1029             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1030             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1031             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1032             felec            = _mm256_mul_ps(qq00,felec);
1033
1034             /* CUBIC SPLINE TABLE DISPERSION */
1035             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1036                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1037             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1038                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1039             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1040                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1041             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1042                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1043             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1044             Heps             = _mm256_mul_ps(vfeps,H);
1045             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1046             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1047             fvdw6            = _mm256_mul_ps(c6_00,FF);
1048
1049             /* CUBIC SPLINE TABLE REPULSION */
1050             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1051             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1052             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1053                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1054             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1055                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1056             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1057                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1058             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1059                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1060             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1061             Heps             = _mm256_mul_ps(vfeps,H);
1062             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1063             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1064             fvdw12           = _mm256_mul_ps(c12_00,FF);
1065             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1066
1067             fscal            = _mm256_add_ps(felec,fvdw);
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm256_mul_ps(fscal,dx00);
1071             ty               = _mm256_mul_ps(fscal,dy00);
1072             tz               = _mm256_mul_ps(fscal,dz00);
1073
1074             /* Update vectorial force */
1075             fix0             = _mm256_add_ps(fix0,tx);
1076             fiy0             = _mm256_add_ps(fiy0,ty);
1077             fiz0             = _mm256_add_ps(fiz0,tz);
1078
1079             fjx0             = _mm256_add_ps(fjx0,tx);
1080             fjy0             = _mm256_add_ps(fjy0,ty);
1081             fjz0             = _mm256_add_ps(fjz0,tz);
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             r10              = _mm256_mul_ps(rsq10,rinv10);
1088
1089             /* Compute parameters for interactions between i and j atoms */
1090             qq10             = _mm256_mul_ps(iq1,jq0);
1091
1092             /* EWALD ELECTROSTATICS */
1093
1094             /* Analytical PME correction */
1095             zeta2            = _mm256_mul_ps(beta2,rsq10);
1096             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1097             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1098             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1099             felec            = _mm256_mul_ps(qq10,felec);
1100
1101             fscal            = felec;
1102
1103             /* Calculate temporary vectorial force */
1104             tx               = _mm256_mul_ps(fscal,dx10);
1105             ty               = _mm256_mul_ps(fscal,dy10);
1106             tz               = _mm256_mul_ps(fscal,dz10);
1107
1108             /* Update vectorial force */
1109             fix1             = _mm256_add_ps(fix1,tx);
1110             fiy1             = _mm256_add_ps(fiy1,ty);
1111             fiz1             = _mm256_add_ps(fiz1,tz);
1112
1113             fjx0             = _mm256_add_ps(fjx0,tx);
1114             fjy0             = _mm256_add_ps(fjy0,ty);
1115             fjz0             = _mm256_add_ps(fjz0,tz);
1116
1117             /**************************
1118              * CALCULATE INTERACTIONS *
1119              **************************/
1120
1121             r20              = _mm256_mul_ps(rsq20,rinv20);
1122
1123             /* Compute parameters for interactions between i and j atoms */
1124             qq20             = _mm256_mul_ps(iq2,jq0);
1125
1126             /* EWALD ELECTROSTATICS */
1127
1128             /* Analytical PME correction */
1129             zeta2            = _mm256_mul_ps(beta2,rsq20);
1130             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1131             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1132             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1133             felec            = _mm256_mul_ps(qq20,felec);
1134
1135             fscal            = felec;
1136
1137             /* Calculate temporary vectorial force */
1138             tx               = _mm256_mul_ps(fscal,dx20);
1139             ty               = _mm256_mul_ps(fscal,dy20);
1140             tz               = _mm256_mul_ps(fscal,dz20);
1141
1142             /* Update vectorial force */
1143             fix2             = _mm256_add_ps(fix2,tx);
1144             fiy2             = _mm256_add_ps(fiy2,ty);
1145             fiz2             = _mm256_add_ps(fiz2,tz);
1146
1147             fjx0             = _mm256_add_ps(fjx0,tx);
1148             fjy0             = _mm256_add_ps(fjy0,ty);
1149             fjz0             = _mm256_add_ps(fjz0,tz);
1150
1151             fjptrA             = f+j_coord_offsetA;
1152             fjptrB             = f+j_coord_offsetB;
1153             fjptrC             = f+j_coord_offsetC;
1154             fjptrD             = f+j_coord_offsetD;
1155             fjptrE             = f+j_coord_offsetE;
1156             fjptrF             = f+j_coord_offsetF;
1157             fjptrG             = f+j_coord_offsetG;
1158             fjptrH             = f+j_coord_offsetH;
1159
1160             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1161
1162             /* Inner loop uses 197 flops */
1163         }
1164
1165         if(jidx<j_index_end)
1166         {
1167
1168             /* Get j neighbor index, and coordinate index */
1169             jnrlistA         = jjnr[jidx];
1170             jnrlistB         = jjnr[jidx+1];
1171             jnrlistC         = jjnr[jidx+2];
1172             jnrlistD         = jjnr[jidx+3];
1173             jnrlistE         = jjnr[jidx+4];
1174             jnrlistF         = jjnr[jidx+5];
1175             jnrlistG         = jjnr[jidx+6];
1176             jnrlistH         = jjnr[jidx+7];
1177             /* Sign of each element will be negative for non-real atoms.
1178              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1179              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1180              */
1181             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1182                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1183
1184             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1185             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1186             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1187             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1188             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1189             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1190             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1191             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1192             j_coord_offsetA  = DIM*jnrA;
1193             j_coord_offsetB  = DIM*jnrB;
1194             j_coord_offsetC  = DIM*jnrC;
1195             j_coord_offsetD  = DIM*jnrD;
1196             j_coord_offsetE  = DIM*jnrE;
1197             j_coord_offsetF  = DIM*jnrF;
1198             j_coord_offsetG  = DIM*jnrG;
1199             j_coord_offsetH  = DIM*jnrH;
1200
1201             /* load j atom coordinates */
1202             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1203                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1204                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1205                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1206                                                  &jx0,&jy0,&jz0);
1207
1208             /* Calculate displacement vector */
1209             dx00             = _mm256_sub_ps(ix0,jx0);
1210             dy00             = _mm256_sub_ps(iy0,jy0);
1211             dz00             = _mm256_sub_ps(iz0,jz0);
1212             dx10             = _mm256_sub_ps(ix1,jx0);
1213             dy10             = _mm256_sub_ps(iy1,jy0);
1214             dz10             = _mm256_sub_ps(iz1,jz0);
1215             dx20             = _mm256_sub_ps(ix2,jx0);
1216             dy20             = _mm256_sub_ps(iy2,jy0);
1217             dz20             = _mm256_sub_ps(iz2,jz0);
1218
1219             /* Calculate squared distance and things based on it */
1220             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1221             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1222             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1223
1224             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1225             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1226             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1227
1228             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1229             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1230             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1231
1232             /* Load parameters for j particles */
1233             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1234                                                                  charge+jnrC+0,charge+jnrD+0,
1235                                                                  charge+jnrE+0,charge+jnrF+0,
1236                                                                  charge+jnrG+0,charge+jnrH+0);
1237             vdwjidx0A        = 2*vdwtype[jnrA+0];
1238             vdwjidx0B        = 2*vdwtype[jnrB+0];
1239             vdwjidx0C        = 2*vdwtype[jnrC+0];
1240             vdwjidx0D        = 2*vdwtype[jnrD+0];
1241             vdwjidx0E        = 2*vdwtype[jnrE+0];
1242             vdwjidx0F        = 2*vdwtype[jnrF+0];
1243             vdwjidx0G        = 2*vdwtype[jnrG+0];
1244             vdwjidx0H        = 2*vdwtype[jnrH+0];
1245
1246             fjx0             = _mm256_setzero_ps();
1247             fjy0             = _mm256_setzero_ps();
1248             fjz0             = _mm256_setzero_ps();
1249
1250             /**************************
1251              * CALCULATE INTERACTIONS *
1252              **************************/
1253
1254             r00              = _mm256_mul_ps(rsq00,rinv00);
1255             r00              = _mm256_andnot_ps(dummy_mask,r00);
1256
1257             /* Compute parameters for interactions between i and j atoms */
1258             qq00             = _mm256_mul_ps(iq0,jq0);
1259             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1260                                             vdwioffsetptr0+vdwjidx0B,
1261                                             vdwioffsetptr0+vdwjidx0C,
1262                                             vdwioffsetptr0+vdwjidx0D,
1263                                             vdwioffsetptr0+vdwjidx0E,
1264                                             vdwioffsetptr0+vdwjidx0F,
1265                                             vdwioffsetptr0+vdwjidx0G,
1266                                             vdwioffsetptr0+vdwjidx0H,
1267                                             &c6_00,&c12_00);
1268
1269             /* Calculate table index by multiplying r with table scale and truncate to integer */
1270             rt               = _mm256_mul_ps(r00,vftabscale);
1271             vfitab           = _mm256_cvttps_epi32(rt);
1272             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1273             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1274             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1275             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1276             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1277             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1278
1279             /* EWALD ELECTROSTATICS */
1280
1281             /* Analytical PME correction */
1282             zeta2            = _mm256_mul_ps(beta2,rsq00);
1283             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1284             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1285             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1286             felec            = _mm256_mul_ps(qq00,felec);
1287
1288             /* CUBIC SPLINE TABLE DISPERSION */
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             fvdw6            = _mm256_mul_ps(c6_00,FF);
1302
1303             /* CUBIC SPLINE TABLE REPULSION */
1304             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1305             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1306             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1308             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1310             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1312             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1314             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1315             Heps             = _mm256_mul_ps(vfeps,H);
1316             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1317             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1318             fvdw12           = _mm256_mul_ps(c12_00,FF);
1319             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1320
1321             fscal            = _mm256_add_ps(felec,fvdw);
1322
1323             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1324
1325             /* Calculate temporary vectorial force */
1326             tx               = _mm256_mul_ps(fscal,dx00);
1327             ty               = _mm256_mul_ps(fscal,dy00);
1328             tz               = _mm256_mul_ps(fscal,dz00);
1329
1330             /* Update vectorial force */
1331             fix0             = _mm256_add_ps(fix0,tx);
1332             fiy0             = _mm256_add_ps(fiy0,ty);
1333             fiz0             = _mm256_add_ps(fiz0,tz);
1334
1335             fjx0             = _mm256_add_ps(fjx0,tx);
1336             fjy0             = _mm256_add_ps(fjy0,ty);
1337             fjz0             = _mm256_add_ps(fjz0,tz);
1338
1339             /**************************
1340              * CALCULATE INTERACTIONS *
1341              **************************/
1342
1343             r10              = _mm256_mul_ps(rsq10,rinv10);
1344             r10              = _mm256_andnot_ps(dummy_mask,r10);
1345
1346             /* Compute parameters for interactions between i and j atoms */
1347             qq10             = _mm256_mul_ps(iq1,jq0);
1348
1349             /* EWALD ELECTROSTATICS */
1350
1351             /* Analytical PME correction */
1352             zeta2            = _mm256_mul_ps(beta2,rsq10);
1353             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1354             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1355             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1356             felec            = _mm256_mul_ps(qq10,felec);
1357
1358             fscal            = felec;
1359
1360             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1361
1362             /* Calculate temporary vectorial force */
1363             tx               = _mm256_mul_ps(fscal,dx10);
1364             ty               = _mm256_mul_ps(fscal,dy10);
1365             tz               = _mm256_mul_ps(fscal,dz10);
1366
1367             /* Update vectorial force */
1368             fix1             = _mm256_add_ps(fix1,tx);
1369             fiy1             = _mm256_add_ps(fiy1,ty);
1370             fiz1             = _mm256_add_ps(fiz1,tz);
1371
1372             fjx0             = _mm256_add_ps(fjx0,tx);
1373             fjy0             = _mm256_add_ps(fjy0,ty);
1374             fjz0             = _mm256_add_ps(fjz0,tz);
1375
1376             /**************************
1377              * CALCULATE INTERACTIONS *
1378              **************************/
1379
1380             r20              = _mm256_mul_ps(rsq20,rinv20);
1381             r20              = _mm256_andnot_ps(dummy_mask,r20);
1382
1383             /* Compute parameters for interactions between i and j atoms */
1384             qq20             = _mm256_mul_ps(iq2,jq0);
1385
1386             /* EWALD ELECTROSTATICS */
1387
1388             /* Analytical PME correction */
1389             zeta2            = _mm256_mul_ps(beta2,rsq20);
1390             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1391             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1392             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1393             felec            = _mm256_mul_ps(qq20,felec);
1394
1395             fscal            = felec;
1396
1397             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1398
1399             /* Calculate temporary vectorial force */
1400             tx               = _mm256_mul_ps(fscal,dx20);
1401             ty               = _mm256_mul_ps(fscal,dy20);
1402             tz               = _mm256_mul_ps(fscal,dz20);
1403
1404             /* Update vectorial force */
1405             fix2             = _mm256_add_ps(fix2,tx);
1406             fiy2             = _mm256_add_ps(fiy2,ty);
1407             fiz2             = _mm256_add_ps(fiz2,tz);
1408
1409             fjx0             = _mm256_add_ps(fjx0,tx);
1410             fjy0             = _mm256_add_ps(fjy0,ty);
1411             fjz0             = _mm256_add_ps(fjz0,tz);
1412
1413             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1414             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1415             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1416             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1417             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1418             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1419             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1420             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1421
1422             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1423
1424             /* Inner loop uses 200 flops */
1425         }
1426
1427         /* End of innermost loop */
1428
1429         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1430                                                  f+i_coord_offset,fshift+i_shift_offset);
1431
1432         /* Increment number of inner iterations */
1433         inneriter                  += j_index_end - j_index_start;
1434
1435         /* Outer loop uses 18 flops */
1436     }
1437
1438     /* Increment number of outer iterations */
1439     outeriter        += nri;
1440
1441     /* Update outer/inner flops */
1442
1443     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*200);
1444 }