src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_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_ElecEwSw_VdwLJSw_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  75     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  84     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  85     __m256i          ewitab;
  86     __m128i          ewitab_lo,ewitab_hi;
  87     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  88     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  89     real             *ewtab;
  90     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  91     real             rswitch_scalar,d_scalar;
  92     __m256           dummy_mask,cutoff_mask;
  93     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  94     __m256           one     = _mm256_set1_ps(1.0);
  95     __m256           two     = _mm256_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm256_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 114     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 115     beta2            = _mm256_mul_ps(beta,beta);
 116     beta3            = _mm256_mul_ps(beta,beta2);
 117
 118     ewtab            = fr->ic->tabq_coul_FDV0;
 119     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 120     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 126
 127     rswitch_scalar   = fr->rcoulomb_switch;
 128     rswitch          = _mm256_set1_ps(rswitch_scalar);
 129     /* Setup switch parameters */
 130     d_scalar         = rcutoff_scalar-rswitch_scalar;
 131     d                = _mm256_set1_ps(d_scalar);
 132     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 133     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 134     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 135     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 136     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 137     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145     j_coord_offsetE = 0;
 146     j_coord_offsetF = 0;
 147     j_coord_offsetG = 0;
 148     j_coord_offsetH = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 174
 175         fix0             = _mm256_setzero_ps();
 176         fiy0             = _mm256_setzero_ps();
 177         fiz0             = _mm256_setzero_ps();
 178
 179         /* Load parameters for i particles */
 180         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 181         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm256_setzero_ps();
 185         vvdwsum          = _mm256_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             jnrE             = jjnr[jidx+4];
 197             jnrF             = jjnr[jidx+5];
 198             jnrG             = jjnr[jidx+6];
 199             jnrH             = jjnr[jidx+7];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204             j_coord_offsetE  = DIM*jnrE;
 205             j_coord_offsetF  = DIM*jnrF;
 206             j_coord_offsetG  = DIM*jnrG;
 207             j_coord_offsetH  = DIM*jnrH;
 208
 209             /* load j atom coordinates */
 210             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 212                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 213                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 214                                                  &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm256_sub_ps(ix0,jx0);
 218             dy00             = _mm256_sub_ps(iy0,jy0);
 219             dz00             = _mm256_sub_ps(iz0,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 223
 224             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 225
 226             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 230                                                                  charge+jnrC+0,charge+jnrD+0,
 231                                                                  charge+jnrE+0,charge+jnrF+0,
 232                                                                  charge+jnrG+0,charge+jnrH+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235             vdwjidx0C        = 2*vdwtype[jnrC+0];
 236             vdwjidx0D        = 2*vdwtype[jnrD+0];
 237             vdwjidx0E        = 2*vdwtype[jnrE+0];
 238             vdwjidx0F        = 2*vdwtype[jnrF+0];
 239             vdwjidx0G        = 2*vdwtype[jnrG+0];
 240             vdwjidx0H        = 2*vdwtype[jnrH+0];
 241
 242             /**************************
 243              * CALCULATE INTERACTIONS *
 244              **************************/
 245
 246             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 247             {
 248
 249             r00              = _mm256_mul_ps(rsq00,rinv00);
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm256_mul_ps(iq0,jq0);
 253             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 254                                             vdwioffsetptr0+vdwjidx0B,
 255                                             vdwioffsetptr0+vdwjidx0C,
 256                                             vdwioffsetptr0+vdwjidx0D,
 257                                             vdwioffsetptr0+vdwjidx0E,
 258                                             vdwioffsetptr0+vdwjidx0F,
 259                                             vdwioffsetptr0+vdwjidx0G,
 260                                             vdwioffsetptr0+vdwjidx0H,
 261                                             &c6_00,&c12_00);
 262
 263             /* EWALD ELECTROSTATICS */
 264
 265             /* Analytical PME correction */
 266             zeta2            = _mm256_mul_ps(beta2,rsq00);
 267             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 268             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 269             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 270             felec            = _mm256_mul_ps(qq00,felec);
 271             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 272             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 273             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 274             velec            = _mm256_mul_ps(qq00,velec);
 275
 276             /* LENNARD-JONES DISPERSION/REPULSION */
 277
 278             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 279             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 280             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 281             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 282             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 283
 284             d                = _mm256_sub_ps(r00,rswitch);
 285             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 286             d2               = _mm256_mul_ps(d,d);
 287             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 288
 289             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 290
 291             /* Evaluate switch function */
 292             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 293             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 294             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 295             velec            = _mm256_mul_ps(velec,sw);
 296             vvdw             = _mm256_mul_ps(vvdw,sw);
 297             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm256_and_ps(velec,cutoff_mask);
 301             velecsum         = _mm256_add_ps(velecsum,velec);
 302             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 303             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 304
 305             fscal            = _mm256_add_ps(felec,fvdw);
 306
 307             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm256_mul_ps(fscal,dx00);
 311             ty               = _mm256_mul_ps(fscal,dy00);
 312             tz               = _mm256_mul_ps(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm256_add_ps(fix0,tx);
 316             fiy0             = _mm256_add_ps(fiy0,ty);
 317             fiz0             = _mm256_add_ps(fiz0,tz);
 318
 319             fjptrA             = f+j_coord_offsetA;
 320             fjptrB             = f+j_coord_offsetB;
 321             fjptrC             = f+j_coord_offsetC;
 322             fjptrD             = f+j_coord_offsetD;
 323             fjptrE             = f+j_coord_offsetE;
 324             fjptrF             = f+j_coord_offsetF;
 325             fjptrG             = f+j_coord_offsetG;
 326             fjptrH             = f+j_coord_offsetH;
 327             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 328
 329             }
 330
 331             /* Inner loop uses 126 flops */
 332         }
 333
 334         if(jidx<j_index_end)
 335         {
 336
 337             /* Get j neighbor index, and coordinate index */
 338             jnrlistA         = jjnr[jidx];
 339             jnrlistB         = jjnr[jidx+1];
 340             jnrlistC         = jjnr[jidx+2];
 341             jnrlistD         = jjnr[jidx+3];
 342             jnrlistE         = jjnr[jidx+4];
 343             jnrlistF         = jjnr[jidx+5];
 344             jnrlistG         = jjnr[jidx+6];
 345             jnrlistH         = jjnr[jidx+7];
 346             /* Sign of each element will be negative for non-real atoms.
 347              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 348              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 349              */
 350             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 351                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 352
 353             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 354             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 355             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 356             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 357             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 358             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 359             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 360             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 361             j_coord_offsetA  = DIM*jnrA;
 362             j_coord_offsetB  = DIM*jnrB;
 363             j_coord_offsetC  = DIM*jnrC;
 364             j_coord_offsetD  = DIM*jnrD;
 365             j_coord_offsetE  = DIM*jnrE;
 366             j_coord_offsetF  = DIM*jnrF;
 367             j_coord_offsetG  = DIM*jnrG;
 368             j_coord_offsetH  = DIM*jnrH;
 369
 370             /* load j atom coordinates */
 371             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 372                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 373                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 374                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 375                                                  &jx0,&jy0,&jz0);
 376
 377             /* Calculate displacement vector */
 378             dx00             = _mm256_sub_ps(ix0,jx0);
 379             dy00             = _mm256_sub_ps(iy0,jy0);
 380             dz00             = _mm256_sub_ps(iz0,jz0);
 381
 382             /* Calculate squared distance and things based on it */
 383             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 384
 385             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 386
 387             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 388
 389             /* Load parameters for j particles */
 390             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 391                                                                  charge+jnrC+0,charge+jnrD+0,
 392                                                                  charge+jnrE+0,charge+jnrF+0,
 393                                                                  charge+jnrG+0,charge+jnrH+0);
 394             vdwjidx0A        = 2*vdwtype[jnrA+0];
 395             vdwjidx0B        = 2*vdwtype[jnrB+0];
 396             vdwjidx0C        = 2*vdwtype[jnrC+0];
 397             vdwjidx0D        = 2*vdwtype[jnrD+0];
 398             vdwjidx0E        = 2*vdwtype[jnrE+0];
 399             vdwjidx0F        = 2*vdwtype[jnrF+0];
 400             vdwjidx0G        = 2*vdwtype[jnrG+0];
 401             vdwjidx0H        = 2*vdwtype[jnrH+0];
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 408             {
 409
 410             r00              = _mm256_mul_ps(rsq00,rinv00);
 411             r00              = _mm256_andnot_ps(dummy_mask,r00);
 412
 413             /* Compute parameters for interactions between i and j atoms */
 414             qq00             = _mm256_mul_ps(iq0,jq0);
 415             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 416                                             vdwioffsetptr0+vdwjidx0B,
 417                                             vdwioffsetptr0+vdwjidx0C,
 418                                             vdwioffsetptr0+vdwjidx0D,
 419                                             vdwioffsetptr0+vdwjidx0E,
 420                                             vdwioffsetptr0+vdwjidx0F,
 421                                             vdwioffsetptr0+vdwjidx0G,
 422                                             vdwioffsetptr0+vdwjidx0H,
 423                                             &c6_00,&c12_00);
 424
 425             /* EWALD ELECTROSTATICS */
 426
 427             /* Analytical PME correction */
 428             zeta2            = _mm256_mul_ps(beta2,rsq00);
 429             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 430             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 431             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 432             felec            = _mm256_mul_ps(qq00,felec);
 433             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 434             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 435             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 436             velec            = _mm256_mul_ps(qq00,velec);
 437
 438             /* LENNARD-JONES DISPERSION/REPULSION */
 439
 440             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 441             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 442             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 443             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 444             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 445
 446             d                = _mm256_sub_ps(r00,rswitch);
 447             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 448             d2               = _mm256_mul_ps(d,d);
 449             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 450
 451             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 452
 453             /* Evaluate switch function */
 454             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 455             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 456             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 457             velec            = _mm256_mul_ps(velec,sw);
 458             vvdw             = _mm256_mul_ps(vvdw,sw);
 459             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velec            = _mm256_and_ps(velec,cutoff_mask);
 463             velec            = _mm256_andnot_ps(dummy_mask,velec);
 464             velecsum         = _mm256_add_ps(velecsum,velec);
 465             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 466             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 467             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 468
 469             fscal            = _mm256_add_ps(felec,fvdw);
 470
 471             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 472
 473             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 474
 475             /* Calculate temporary vectorial force */
 476             tx               = _mm256_mul_ps(fscal,dx00);
 477             ty               = _mm256_mul_ps(fscal,dy00);
 478             tz               = _mm256_mul_ps(fscal,dz00);
 479
 480             /* Update vectorial force */
 481             fix0             = _mm256_add_ps(fix0,tx);
 482             fiy0             = _mm256_add_ps(fiy0,ty);
 483             fiz0             = _mm256_add_ps(fiz0,tz);
 484
 485             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 486             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 487             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 488             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 489             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 490             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 491             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 492             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 493             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 494
 495             }
 496
 497             /* Inner loop uses 127 flops */
 498         }
 499
 500         /* End of innermost loop */
 501
 502         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 503                                                  f+i_coord_offset,fshift+i_shift_offset);
 504
 505         ggid                        = gid[iidx];
 506         /* Update potential energies */
 507         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 508         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 509
 510         /* Increment number of inner iterations */
 511         inneriter                  += j_index_end - j_index_start;
 512
 513         /* Outer loop uses 9 flops */
 514     }
 515
 516     /* Increment number of outer iterations */
 517     outeriter        += nri;
 518
 519     /* Update outer/inner flops */
 520
 521     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*127);
 522 }
 523 /*
 524  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_avx_256_single
 525  * Electrostatics interaction: Ewald
 526  * VdW interaction:            LennardJones
 527  * Geometry:                   Particle-Particle
 528  * Calculate force/pot:        Force
 529  */
 530 void
 531 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_avx_256_single
 532                     (t_nblist * gmx_restrict                nlist,
 533                      rvec * gmx_restrict                    xx,
 534                      rvec * gmx_restrict                    ff,
 535                      t_forcerec * gmx_restrict              fr,
 536                      t_mdatoms * gmx_restrict               mdatoms,
 537                      nb_kernel_data_t * gmx_restrict        kernel_data,
 538                      t_nrnb * gmx_restrict                  nrnb)
 539 {
 540     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 541      * just 0 for non-waters.
 542      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 543      * jnr indices corresponding to data put in the four positions in the SIMD register.
 544      */
 545     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 546     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 547     int              jnrA,jnrB,jnrC,jnrD;
 548     int              jnrE,jnrF,jnrG,jnrH;
 549     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 550     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 551     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 552     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 553     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 554     real             rcutoff_scalar;
 555     real             *shiftvec,*fshift,*x,*f;
 556     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 557     real             scratch[4*DIM];
 558     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 559     real *           vdwioffsetptr0;
 560     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 561     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 562     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 563     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 564     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 565     real             *charge;
 566     int              nvdwtype;
 567     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 568     int              *vdwtype;
 569     real             *vdwparam;
 570     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 571     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 572     __m256i          ewitab;
 573     __m128i          ewitab_lo,ewitab_hi;
 574     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 575     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 576     real             *ewtab;
 577     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 578     real             rswitch_scalar,d_scalar;
 579     __m256           dummy_mask,cutoff_mask;
 580     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 581     __m256           one     = _mm256_set1_ps(1.0);
 582     __m256           two     = _mm256_set1_ps(2.0);
 583     x                = xx[0];
 584     f                = ff[0];
 585
 586     nri              = nlist->nri;
 587     iinr             = nlist->iinr;
 588     jindex           = nlist->jindex;
 589     jjnr             = nlist->jjnr;
 590     shiftidx         = nlist->shift;
 591     gid              = nlist->gid;
 592     shiftvec         = fr->shift_vec[0];
 593     fshift           = fr->fshift[0];
 594     facel            = _mm256_set1_ps(fr->epsfac);
 595     charge           = mdatoms->chargeA;
 596     nvdwtype         = fr->ntype;
 597     vdwparam         = fr->nbfp;
 598     vdwtype          = mdatoms->typeA;
 599
 600     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 601     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 602     beta2            = _mm256_mul_ps(beta,beta);
 603     beta3            = _mm256_mul_ps(beta,beta2);
 604
 605     ewtab            = fr->ic->tabq_coul_FDV0;
 606     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 607     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 608
 609     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 610     rcutoff_scalar   = fr->rcoulomb;
 611     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 612     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 613
 614     rswitch_scalar   = fr->rcoulomb_switch;
 615     rswitch          = _mm256_set1_ps(rswitch_scalar);
 616     /* Setup switch parameters */
 617     d_scalar         = rcutoff_scalar-rswitch_scalar;
 618     d                = _mm256_set1_ps(d_scalar);
 619     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 620     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 621     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 622     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 623     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 624     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 625
 626     /* Avoid stupid compiler warnings */
 627     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 628     j_coord_offsetA = 0;
 629     j_coord_offsetB = 0;
 630     j_coord_offsetC = 0;
 631     j_coord_offsetD = 0;
 632     j_coord_offsetE = 0;
 633     j_coord_offsetF = 0;
 634     j_coord_offsetG = 0;
 635     j_coord_offsetH = 0;
 636
 637     outeriter        = 0;
 638     inneriter        = 0;
 639
 640     for(iidx=0;iidx<4*DIM;iidx++)
 641     {
 642         scratch[iidx] = 0.0;
 643     }
 644
 645     /* Start outer loop over neighborlists */
 646     for(iidx=0; iidx<nri; iidx++)
 647     {
 648         /* Load shift vector for this list */
 649         i_shift_offset   = DIM*shiftidx[iidx];
 650
 651         /* Load limits for loop over neighbors */
 652         j_index_start    = jindex[iidx];
 653         j_index_end      = jindex[iidx+1];
 654
 655         /* Get outer coordinate index */
 656         inr              = iinr[iidx];
 657         i_coord_offset   = DIM*inr;
 658
 659         /* Load i particle coords and add shift vector */
 660         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 661
 662         fix0             = _mm256_setzero_ps();
 663         fiy0             = _mm256_setzero_ps();
 664         fiz0             = _mm256_setzero_ps();
 665
 666         /* Load parameters for i particles */
 667         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 668         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 669
 670         /* Start inner kernel loop */
 671         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 672         {
 673
 674             /* Get j neighbor index, and coordinate index */
 675             jnrA             = jjnr[jidx];
 676             jnrB             = jjnr[jidx+1];
 677             jnrC             = jjnr[jidx+2];
 678             jnrD             = jjnr[jidx+3];
 679             jnrE             = jjnr[jidx+4];
 680             jnrF             = jjnr[jidx+5];
 681             jnrG             = jjnr[jidx+6];
 682             jnrH             = jjnr[jidx+7];
 683             j_coord_offsetA  = DIM*jnrA;
 684             j_coord_offsetB  = DIM*jnrB;
 685             j_coord_offsetC  = DIM*jnrC;
 686             j_coord_offsetD  = DIM*jnrD;
 687             j_coord_offsetE  = DIM*jnrE;
 688             j_coord_offsetF  = DIM*jnrF;
 689             j_coord_offsetG  = DIM*jnrG;
 690             j_coord_offsetH  = DIM*jnrH;
 691
 692             /* load j atom coordinates */
 693             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 694                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 695                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 696                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 697                                                  &jx0,&jy0,&jz0);
 698
 699             /* Calculate displacement vector */
 700             dx00             = _mm256_sub_ps(ix0,jx0);
 701             dy00             = _mm256_sub_ps(iy0,jy0);
 702             dz00             = _mm256_sub_ps(iz0,jz0);
 703
 704             /* Calculate squared distance and things based on it */
 705             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 706
 707             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 708
 709             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 710
 711             /* Load parameters for j particles */
 712             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 713                                                                  charge+jnrC+0,charge+jnrD+0,
 714                                                                  charge+jnrE+0,charge+jnrF+0,
 715                                                                  charge+jnrG+0,charge+jnrH+0);
 716             vdwjidx0A        = 2*vdwtype[jnrA+0];
 717             vdwjidx0B        = 2*vdwtype[jnrB+0];
 718             vdwjidx0C        = 2*vdwtype[jnrC+0];
 719             vdwjidx0D        = 2*vdwtype[jnrD+0];
 720             vdwjidx0E        = 2*vdwtype[jnrE+0];
 721             vdwjidx0F        = 2*vdwtype[jnrF+0];
 722             vdwjidx0G        = 2*vdwtype[jnrG+0];
 723             vdwjidx0H        = 2*vdwtype[jnrH+0];
 724
 725             /**************************
 726              * CALCULATE INTERACTIONS *
 727              **************************/
 728
 729             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 730             {
 731
 732             r00              = _mm256_mul_ps(rsq00,rinv00);
 733
 734             /* Compute parameters for interactions between i and j atoms */
 735             qq00             = _mm256_mul_ps(iq0,jq0);
 736             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 737                                             vdwioffsetptr0+vdwjidx0B,
 738                                             vdwioffsetptr0+vdwjidx0C,
 739                                             vdwioffsetptr0+vdwjidx0D,
 740                                             vdwioffsetptr0+vdwjidx0E,
 741                                             vdwioffsetptr0+vdwjidx0F,
 742                                             vdwioffsetptr0+vdwjidx0G,
 743                                             vdwioffsetptr0+vdwjidx0H,
 744                                             &c6_00,&c12_00);
 745
 746             /* EWALD ELECTROSTATICS */
 747
 748             /* Analytical PME correction */
 749             zeta2            = _mm256_mul_ps(beta2,rsq00);
 750             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 751             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 752             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 753             felec            = _mm256_mul_ps(qq00,felec);
 754             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 755             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 756             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 757             velec            = _mm256_mul_ps(qq00,velec);
 758
 759             /* LENNARD-JONES DISPERSION/REPULSION */
 760
 761             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 762             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 763             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 764             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 765             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 766
 767             d                = _mm256_sub_ps(r00,rswitch);
 768             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 769             d2               = _mm256_mul_ps(d,d);
 770             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 771
 772             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 773
 774             /* Evaluate switch function */
 775             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 776             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 777             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 778             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 779
 780             fscal            = _mm256_add_ps(felec,fvdw);
 781
 782             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 783
 784             /* Calculate temporary vectorial force */
 785             tx               = _mm256_mul_ps(fscal,dx00);
 786             ty               = _mm256_mul_ps(fscal,dy00);
 787             tz               = _mm256_mul_ps(fscal,dz00);
 788
 789             /* Update vectorial force */
 790             fix0             = _mm256_add_ps(fix0,tx);
 791             fiy0             = _mm256_add_ps(fiy0,ty);
 792             fiz0             = _mm256_add_ps(fiz0,tz);
 793
 794             fjptrA             = f+j_coord_offsetA;
 795             fjptrB             = f+j_coord_offsetB;
 796             fjptrC             = f+j_coord_offsetC;
 797             fjptrD             = f+j_coord_offsetD;
 798             fjptrE             = f+j_coord_offsetE;
 799             fjptrF             = f+j_coord_offsetF;
 800             fjptrG             = f+j_coord_offsetG;
 801             fjptrH             = f+j_coord_offsetH;
 802             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 803
 804             }
 805
 806             /* Inner loop uses 120 flops */
 807         }
 808
 809         if(jidx<j_index_end)
 810         {
 811
 812             /* Get j neighbor index, and coordinate index */
 813             jnrlistA         = jjnr[jidx];
 814             jnrlistB         = jjnr[jidx+1];
 815             jnrlistC         = jjnr[jidx+2];
 816             jnrlistD         = jjnr[jidx+3];
 817             jnrlistE         = jjnr[jidx+4];
 818             jnrlistF         = jjnr[jidx+5];
 819             jnrlistG         = jjnr[jidx+6];
 820             jnrlistH         = jjnr[jidx+7];
 821             /* Sign of each element will be negative for non-real atoms.
 822              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 823              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 824              */
 825             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 826                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 827
 828             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 829             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 830             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 831             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 832             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 833             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 834             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 835             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 836             j_coord_offsetA  = DIM*jnrA;
 837             j_coord_offsetB  = DIM*jnrB;
 838             j_coord_offsetC  = DIM*jnrC;
 839             j_coord_offsetD  = DIM*jnrD;
 840             j_coord_offsetE  = DIM*jnrE;
 841             j_coord_offsetF  = DIM*jnrF;
 842             j_coord_offsetG  = DIM*jnrG;
 843             j_coord_offsetH  = DIM*jnrH;
 844
 845             /* load j atom coordinates */
 846             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 847                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 848                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 849                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 850                                                  &jx0,&jy0,&jz0);
 851
 852             /* Calculate displacement vector */
 853             dx00             = _mm256_sub_ps(ix0,jx0);
 854             dy00             = _mm256_sub_ps(iy0,jy0);
 855             dz00             = _mm256_sub_ps(iz0,jz0);
 856
 857             /* Calculate squared distance and things based on it */
 858             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 859
 860             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 861
 862             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 863
 864             /* Load parameters for j particles */
 865             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 866                                                                  charge+jnrC+0,charge+jnrD+0,
 867                                                                  charge+jnrE+0,charge+jnrF+0,
 868                                                                  charge+jnrG+0,charge+jnrH+0);
 869             vdwjidx0A        = 2*vdwtype[jnrA+0];
 870             vdwjidx0B        = 2*vdwtype[jnrB+0];
 871             vdwjidx0C        = 2*vdwtype[jnrC+0];
 872             vdwjidx0D        = 2*vdwtype[jnrD+0];
 873             vdwjidx0E        = 2*vdwtype[jnrE+0];
 874             vdwjidx0F        = 2*vdwtype[jnrF+0];
 875             vdwjidx0G        = 2*vdwtype[jnrG+0];
 876             vdwjidx0H        = 2*vdwtype[jnrH+0];
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 883             {
 884
 885             r00              = _mm256_mul_ps(rsq00,rinv00);
 886             r00              = _mm256_andnot_ps(dummy_mask,r00);
 887
 888             /* Compute parameters for interactions between i and j atoms */
 889             qq00             = _mm256_mul_ps(iq0,jq0);
 890             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 891                                             vdwioffsetptr0+vdwjidx0B,
 892                                             vdwioffsetptr0+vdwjidx0C,
 893                                             vdwioffsetptr0+vdwjidx0D,
 894                                             vdwioffsetptr0+vdwjidx0E,
 895                                             vdwioffsetptr0+vdwjidx0F,
 896                                             vdwioffsetptr0+vdwjidx0G,
 897                                             vdwioffsetptr0+vdwjidx0H,
 898                                             &c6_00,&c12_00);
 899
 900             /* EWALD ELECTROSTATICS */
 901
 902             /* Analytical PME correction */
 903             zeta2            = _mm256_mul_ps(beta2,rsq00);
 904             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 905             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 906             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 907             felec            = _mm256_mul_ps(qq00,felec);
 908             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 909             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 910             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 911             velec            = _mm256_mul_ps(qq00,velec);
 912
 913             /* LENNARD-JONES DISPERSION/REPULSION */
 914
 915             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 916             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 917             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 918             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 919             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 920
 921             d                = _mm256_sub_ps(r00,rswitch);
 922             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 923             d2               = _mm256_mul_ps(d,d);
 924             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 925
 926             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 927
 928             /* Evaluate switch function */
 929             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 930             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 931             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 932             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 933
 934             fscal            = _mm256_add_ps(felec,fvdw);
 935
 936             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 937
 938             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm256_mul_ps(fscal,dx00);
 942             ty               = _mm256_mul_ps(fscal,dy00);
 943             tz               = _mm256_mul_ps(fscal,dz00);
 944
 945             /* Update vectorial force */
 946             fix0             = _mm256_add_ps(fix0,tx);
 947             fiy0             = _mm256_add_ps(fiy0,ty);
 948             fiz0             = _mm256_add_ps(fiz0,tz);
 949
 950             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 951             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 952             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 953             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 954             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 955             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 956             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 957             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 958             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 959
 960             }
 961
 962             /* Inner loop uses 121 flops */
 963         }
 964
 965         /* End of innermost loop */
 966
 967         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 968                                                  f+i_coord_offset,fshift+i_shift_offset);
 969
 970         /* Increment number of inner iterations */
 971         inneriter                  += j_index_end - j_index_start;
 972
 973         /* Outer loop uses 7 flops */
 974     }
 975
 976     /* Increment number of outer iterations */
 977     outeriter        += nri;
 978
 979     /* Update outer/inner flops */
 980
 981     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*121);
 982 }