src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: None
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     real *           vdwgridioffsetptr0;
  90     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     int              nvdwtype;
  95     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  96     int              *vdwtype;
  97     real             *vdwparam;
  98     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  99     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 100     __m256           c6grid_00;
 101     real             *vdwgridparam;
 102     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 103     __m256           one_half  = _mm256_set1_ps(0.5);
 104     __m256           minus_one = _mm256_set1_ps(-1.0);
 105     __m256           dummy_mask,cutoff_mask;
 106     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 107     __m256           one     = _mm256_set1_ps(1.0);
 108     __m256           two     = _mm256_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123     vdwgridparam     = fr->ljpme_c6grid;
 124     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 125     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 126     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 127
 128     rcutoff_scalar   = fr->rvdw;
 129     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 131
 132     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 133     rvdw             = _mm256_set1_ps(fr->rvdw);
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141     j_coord_offsetE = 0;
 142     j_coord_offsetF = 0;
 143     j_coord_offsetG = 0;
 144     j_coord_offsetH = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 170
 171         fix0             = _mm256_setzero_ps();
 172         fiy0             = _mm256_setzero_ps();
 173         fiz0             = _mm256_setzero_ps();
 174
 175         /* Load parameters for i particles */
 176         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 177         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 178
 179         /* Reset potential sums */
 180         vvdwsum          = _mm256_setzero_ps();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             jnrC             = jjnr[jidx+2];
 190             jnrD             = jjnr[jidx+3];
 191             jnrE             = jjnr[jidx+4];
 192             jnrF             = jjnr[jidx+5];
 193             jnrG             = jjnr[jidx+6];
 194             jnrH             = jjnr[jidx+7];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199             j_coord_offsetE  = DIM*jnrE;
 200             j_coord_offsetF  = DIM*jnrF;
 201             j_coord_offsetG  = DIM*jnrG;
 202             j_coord_offsetH  = DIM*jnrH;
 203
 204             /* load j atom coordinates */
 205             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 207                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 208                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 209                                                  &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm256_sub_ps(ix0,jx0);
 213             dy00             = _mm256_sub_ps(iy0,jy0);
 214             dz00             = _mm256_sub_ps(iz0,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 218
 219             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 220
 221             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 222
 223             /* Load parameters for j particles */
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228             vdwjidx0E        = 2*vdwtype[jnrE+0];
 229             vdwjidx0F        = 2*vdwtype[jnrF+0];
 230             vdwjidx0G        = 2*vdwtype[jnrG+0];
 231             vdwjidx0H        = 2*vdwtype[jnrH+0];
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 238             {
 239
 240             r00              = _mm256_mul_ps(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 244                                             vdwioffsetptr0+vdwjidx0B,
 245                                             vdwioffsetptr0+vdwjidx0C,
 246                                             vdwioffsetptr0+vdwjidx0D,
 247                                             vdwioffsetptr0+vdwjidx0E,
 248                                             vdwioffsetptr0+vdwjidx0F,
 249                                             vdwioffsetptr0+vdwjidx0G,
 250                                             vdwioffsetptr0+vdwjidx0H,
 251                                             &c6_00,&c12_00);
 252
 253             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 254                                                                   vdwgridioffsetptr0+vdwjidx0B,
 255                                                                   vdwgridioffsetptr0+vdwjidx0C,
 256                                                                   vdwgridioffsetptr0+vdwjidx0D,
 257                                                                   vdwgridioffsetptr0+vdwjidx0E,
 258                                                                   vdwgridioffsetptr0+vdwjidx0F,
 259                                                                   vdwgridioffsetptr0+vdwjidx0G,
 260                                                                   vdwgridioffsetptr0+vdwjidx0H);
 261
 262             /* Analytical LJ-PME */
 263             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 264             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 265             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 266             exponent         = gmx_simd_exp_r(ewcljrsq);
 267             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 268             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 269             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 270             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 271             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 272             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 273                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 274             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 275             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
 276
 277             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 281             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 282
 283             fscal            = fvdw;
 284
 285             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm256_mul_ps(fscal,dx00);
 289             ty               = _mm256_mul_ps(fscal,dy00);
 290             tz               = _mm256_mul_ps(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm256_add_ps(fix0,tx);
 294             fiy0             = _mm256_add_ps(fiy0,ty);
 295             fiz0             = _mm256_add_ps(fiz0,tz);
 296
 297             fjptrA             = f+j_coord_offsetA;
 298             fjptrB             = f+j_coord_offsetB;
 299             fjptrC             = f+j_coord_offsetC;
 300             fjptrD             = f+j_coord_offsetD;
 301             fjptrE             = f+j_coord_offsetE;
 302             fjptrF             = f+j_coord_offsetF;
 303             fjptrG             = f+j_coord_offsetG;
 304             fjptrH             = f+j_coord_offsetH;
 305             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 306
 307             }
 308
 309             /* Inner loop uses 62 flops */
 310         }
 311
 312         if(jidx<j_index_end)
 313         {
 314
 315             /* Get j neighbor index, and coordinate index */
 316             jnrlistA         = jjnr[jidx];
 317             jnrlistB         = jjnr[jidx+1];
 318             jnrlistC         = jjnr[jidx+2];
 319             jnrlistD         = jjnr[jidx+3];
 320             jnrlistE         = jjnr[jidx+4];
 321             jnrlistF         = jjnr[jidx+5];
 322             jnrlistG         = jjnr[jidx+6];
 323             jnrlistH         = jjnr[jidx+7];
 324             /* Sign of each element will be negative for non-real atoms.
 325              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 326              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 327              */
 328             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 329                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 330
 331             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 332             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 333             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 334             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 335             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 336             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 337             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 338             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 339             j_coord_offsetA  = DIM*jnrA;
 340             j_coord_offsetB  = DIM*jnrB;
 341             j_coord_offsetC  = DIM*jnrC;
 342             j_coord_offsetD  = DIM*jnrD;
 343             j_coord_offsetE  = DIM*jnrE;
 344             j_coord_offsetF  = DIM*jnrF;
 345             j_coord_offsetG  = DIM*jnrG;
 346             j_coord_offsetH  = DIM*jnrH;
 347
 348             /* load j atom coordinates */
 349             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 350                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 351                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 352                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 353                                                  &jx0,&jy0,&jz0);
 354
 355             /* Calculate displacement vector */
 356             dx00             = _mm256_sub_ps(ix0,jx0);
 357             dy00             = _mm256_sub_ps(iy0,jy0);
 358             dz00             = _mm256_sub_ps(iz0,jz0);
 359
 360             /* Calculate squared distance and things based on it */
 361             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 362
 363             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 364
 365             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 366
 367             /* Load parameters for j particles */
 368             vdwjidx0A        = 2*vdwtype[jnrA+0];
 369             vdwjidx0B        = 2*vdwtype[jnrB+0];
 370             vdwjidx0C        = 2*vdwtype[jnrC+0];
 371             vdwjidx0D        = 2*vdwtype[jnrD+0];
 372             vdwjidx0E        = 2*vdwtype[jnrE+0];
 373             vdwjidx0F        = 2*vdwtype[jnrF+0];
 374             vdwjidx0G        = 2*vdwtype[jnrG+0];
 375             vdwjidx0H        = 2*vdwtype[jnrH+0];
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 382             {
 383
 384             r00              = _mm256_mul_ps(rsq00,rinv00);
 385             r00              = _mm256_andnot_ps(dummy_mask,r00);
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 389                                             vdwioffsetptr0+vdwjidx0B,
 390                                             vdwioffsetptr0+vdwjidx0C,
 391                                             vdwioffsetptr0+vdwjidx0D,
 392                                             vdwioffsetptr0+vdwjidx0E,
 393                                             vdwioffsetptr0+vdwjidx0F,
 394                                             vdwioffsetptr0+vdwjidx0G,
 395                                             vdwioffsetptr0+vdwjidx0H,
 396                                             &c6_00,&c12_00);
 397
 398             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 399                                                                   vdwgridioffsetptr0+vdwjidx0B,
 400                                                                   vdwgridioffsetptr0+vdwjidx0C,
 401                                                                   vdwgridioffsetptr0+vdwjidx0D,
 402                                                                   vdwgridioffsetptr0+vdwjidx0E,
 403                                                                   vdwgridioffsetptr0+vdwjidx0F,
 404                                                                   vdwgridioffsetptr0+vdwjidx0G,
 405                                                                   vdwgridioffsetptr0+vdwjidx0H);
 406
 407             /* Analytical LJ-PME */
 408             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 409             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 410             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 411             exponent         = gmx_simd_exp_r(ewcljrsq);
 412             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 413             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 414             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 415             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 416             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 417             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 418                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 419             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 420             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
 421
 422             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 423
 424             /* Update potential sum for this i atom from the interaction with this j atom. */
 425             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 426             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 427             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 428
 429             fscal            = fvdw;
 430
 431             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 432
 433             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 434
 435             /* Calculate temporary vectorial force */
 436             tx               = _mm256_mul_ps(fscal,dx00);
 437             ty               = _mm256_mul_ps(fscal,dy00);
 438             tz               = _mm256_mul_ps(fscal,dz00);
 439
 440             /* Update vectorial force */
 441             fix0             = _mm256_add_ps(fix0,tx);
 442             fiy0             = _mm256_add_ps(fiy0,ty);
 443             fiz0             = _mm256_add_ps(fiz0,tz);
 444
 445             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 446             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 447             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 448             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 449             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 450             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 451             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 452             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 453             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 454
 455             }
 456
 457             /* Inner loop uses 63 flops */
 458         }
 459
 460         /* End of innermost loop */
 461
 462         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 463                                                  f+i_coord_offset,fshift+i_shift_offset);
 464
 465         ggid                        = gid[iidx];
 466         /* Update potential energies */
 467         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 468
 469         /* Increment number of inner iterations */
 470         inneriter                  += j_index_end - j_index_start;
 471
 472         /* Outer loop uses 7 flops */
 473     }
 474
 475     /* Increment number of outer iterations */
 476     outeriter        += nri;
 477
 478     /* Update outer/inner flops */
 479
 480     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*63);
 481 }
 482 /*
 483  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_avx_256_single
 484  * Electrostatics interaction: None
 485  * VdW interaction:            LJEwald
 486  * Geometry:                   Particle-Particle
 487  * Calculate force/pot:        Force
 488  */
 489 void
 490 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_avx_256_single
 491                     (t_nblist                    * gmx_restrict       nlist,
 492                      rvec                        * gmx_restrict          xx,
 493                      rvec                        * gmx_restrict          ff,
 494                      t_forcerec                  * gmx_restrict          fr,
 495                      t_mdatoms                   * gmx_restrict     mdatoms,
 496                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 497                      t_nrnb                      * gmx_restrict        nrnb)
 498 {
 499     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 500      * just 0 for non-waters.
 501      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 502      * jnr indices corresponding to data put in the four positions in the SIMD register.
 503      */
 504     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 505     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 506     int              jnrA,jnrB,jnrC,jnrD;
 507     int              jnrE,jnrF,jnrG,jnrH;
 508     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 509     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 510     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 511     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 512     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 513     real             rcutoff_scalar;
 514     real             *shiftvec,*fshift,*x,*f;
 515     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 516     real             scratch[4*DIM];
 517     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 518     real *           vdwioffsetptr0;
 519     real *           vdwgridioffsetptr0;
 520     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 521     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 522     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 523     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 524     int              nvdwtype;
 525     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 526     int              *vdwtype;
 527     real             *vdwparam;
 528     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 529     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 530     __m256           c6grid_00;
 531     real             *vdwgridparam;
 532     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 533     __m256           one_half  = _mm256_set1_ps(0.5);
 534     __m256           minus_one = _mm256_set1_ps(-1.0);
 535     __m256           dummy_mask,cutoff_mask;
 536     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 537     __m256           one     = _mm256_set1_ps(1.0);
 538     __m256           two     = _mm256_set1_ps(2.0);
 539     x                = xx[0];
 540     f                = ff[0];
 541
 542     nri              = nlist->nri;
 543     iinr             = nlist->iinr;
 544     jindex           = nlist->jindex;
 545     jjnr             = nlist->jjnr;
 546     shiftidx         = nlist->shift;
 547     gid              = nlist->gid;
 548     shiftvec         = fr->shift_vec[0];
 549     fshift           = fr->fshift[0];
 550     nvdwtype         = fr->ntype;
 551     vdwparam         = fr->nbfp;
 552     vdwtype          = mdatoms->typeA;
 553     vdwgridparam     = fr->ljpme_c6grid;
 554     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 555     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 556     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 557
 558     rcutoff_scalar   = fr->rvdw;
 559     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 560     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 561
 562     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 563     rvdw             = _mm256_set1_ps(fr->rvdw);
 564
 565     /* Avoid stupid compiler warnings */
 566     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 567     j_coord_offsetA = 0;
 568     j_coord_offsetB = 0;
 569     j_coord_offsetC = 0;
 570     j_coord_offsetD = 0;
 571     j_coord_offsetE = 0;
 572     j_coord_offsetF = 0;
 573     j_coord_offsetG = 0;
 574     j_coord_offsetH = 0;
 575
 576     outeriter        = 0;
 577     inneriter        = 0;
 578
 579     for(iidx=0;iidx<4*DIM;iidx++)
 580     {
 581         scratch[iidx] = 0.0;
 582     }
 583
 584     /* Start outer loop over neighborlists */
 585     for(iidx=0; iidx<nri; iidx++)
 586     {
 587         /* Load shift vector for this list */
 588         i_shift_offset   = DIM*shiftidx[iidx];
 589
 590         /* Load limits for loop over neighbors */
 591         j_index_start    = jindex[iidx];
 592         j_index_end      = jindex[iidx+1];
 593
 594         /* Get outer coordinate index */
 595         inr              = iinr[iidx];
 596         i_coord_offset   = DIM*inr;
 597
 598         /* Load i particle coords and add shift vector */
 599         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 600
 601         fix0             = _mm256_setzero_ps();
 602         fiy0             = _mm256_setzero_ps();
 603         fiz0             = _mm256_setzero_ps();
 604
 605         /* Load parameters for i particles */
 606         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 607         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 608
 609         /* Start inner kernel loop */
 610         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 611         {
 612
 613             /* Get j neighbor index, and coordinate index */
 614             jnrA             = jjnr[jidx];
 615             jnrB             = jjnr[jidx+1];
 616             jnrC             = jjnr[jidx+2];
 617             jnrD             = jjnr[jidx+3];
 618             jnrE             = jjnr[jidx+4];
 619             jnrF             = jjnr[jidx+5];
 620             jnrG             = jjnr[jidx+6];
 621             jnrH             = jjnr[jidx+7];
 622             j_coord_offsetA  = DIM*jnrA;
 623             j_coord_offsetB  = DIM*jnrB;
 624             j_coord_offsetC  = DIM*jnrC;
 625             j_coord_offsetD  = DIM*jnrD;
 626             j_coord_offsetE  = DIM*jnrE;
 627             j_coord_offsetF  = DIM*jnrF;
 628             j_coord_offsetG  = DIM*jnrG;
 629             j_coord_offsetH  = DIM*jnrH;
 630
 631             /* load j atom coordinates */
 632             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 633                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 634                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 635                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 636                                                  &jx0,&jy0,&jz0);
 637
 638             /* Calculate displacement vector */
 639             dx00             = _mm256_sub_ps(ix0,jx0);
 640             dy00             = _mm256_sub_ps(iy0,jy0);
 641             dz00             = _mm256_sub_ps(iz0,jz0);
 642
 643             /* Calculate squared distance and things based on it */
 644             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 645
 646             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 647
 648             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 649
 650             /* Load parameters for j particles */
 651             vdwjidx0A        = 2*vdwtype[jnrA+0];
 652             vdwjidx0B        = 2*vdwtype[jnrB+0];
 653             vdwjidx0C        = 2*vdwtype[jnrC+0];
 654             vdwjidx0D        = 2*vdwtype[jnrD+0];
 655             vdwjidx0E        = 2*vdwtype[jnrE+0];
 656             vdwjidx0F        = 2*vdwtype[jnrF+0];
 657             vdwjidx0G        = 2*vdwtype[jnrG+0];
 658             vdwjidx0H        = 2*vdwtype[jnrH+0];
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 665             {
 666
 667             r00              = _mm256_mul_ps(rsq00,rinv00);
 668
 669             /* Compute parameters for interactions between i and j atoms */
 670             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 671                                             vdwioffsetptr0+vdwjidx0B,
 672                                             vdwioffsetptr0+vdwjidx0C,
 673                                             vdwioffsetptr0+vdwjidx0D,
 674                                             vdwioffsetptr0+vdwjidx0E,
 675                                             vdwioffsetptr0+vdwjidx0F,
 676                                             vdwioffsetptr0+vdwjidx0G,
 677                                             vdwioffsetptr0+vdwjidx0H,
 678                                             &c6_00,&c12_00);
 679
 680             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 681                                                                   vdwgridioffsetptr0+vdwjidx0B,
 682                                                                   vdwgridioffsetptr0+vdwjidx0C,
 683                                                                   vdwgridioffsetptr0+vdwjidx0D,
 684                                                                   vdwgridioffsetptr0+vdwjidx0E,
 685                                                                   vdwgridioffsetptr0+vdwjidx0F,
 686                                                                   vdwgridioffsetptr0+vdwjidx0G,
 687                                                                   vdwgridioffsetptr0+vdwjidx0H);
 688
 689             /* Analytical LJ-PME */
 690             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 691             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 692             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 693             exponent         = gmx_simd_exp_r(ewcljrsq);
 694             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 695             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 696             /* f6A = 6 * C6grid * (1 - poly) */
 697             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
 698             /* f6B = C6grid * exponent * beta^6 */
 699             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
 700             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 701             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 702
 703             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 704
 705             fscal            = fvdw;
 706
 707             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 708
 709             /* Calculate temporary vectorial force */
 710             tx               = _mm256_mul_ps(fscal,dx00);
 711             ty               = _mm256_mul_ps(fscal,dy00);
 712             tz               = _mm256_mul_ps(fscal,dz00);
 713
 714             /* Update vectorial force */
 715             fix0             = _mm256_add_ps(fix0,tx);
 716             fiy0             = _mm256_add_ps(fiy0,ty);
 717             fiz0             = _mm256_add_ps(fiz0,tz);
 718
 719             fjptrA             = f+j_coord_offsetA;
 720             fjptrB             = f+j_coord_offsetB;
 721             fjptrC             = f+j_coord_offsetC;
 722             fjptrD             = f+j_coord_offsetD;
 723             fjptrE             = f+j_coord_offsetE;
 724             fjptrF             = f+j_coord_offsetF;
 725             fjptrG             = f+j_coord_offsetG;
 726             fjptrH             = f+j_coord_offsetH;
 727             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 728
 729             }
 730
 731             /* Inner loop uses 49 flops */
 732         }
 733
 734         if(jidx<j_index_end)
 735         {
 736
 737             /* Get j neighbor index, and coordinate index */
 738             jnrlistA         = jjnr[jidx];
 739             jnrlistB         = jjnr[jidx+1];
 740             jnrlistC         = jjnr[jidx+2];
 741             jnrlistD         = jjnr[jidx+3];
 742             jnrlistE         = jjnr[jidx+4];
 743             jnrlistF         = jjnr[jidx+5];
 744             jnrlistG         = jjnr[jidx+6];
 745             jnrlistH         = jjnr[jidx+7];
 746             /* Sign of each element will be negative for non-real atoms.
 747              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 748              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 749              */
 750             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 751                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 752
 753             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 754             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 755             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 756             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 757             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 758             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 759             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 760             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 761             j_coord_offsetA  = DIM*jnrA;
 762             j_coord_offsetB  = DIM*jnrB;
 763             j_coord_offsetC  = DIM*jnrC;
 764             j_coord_offsetD  = DIM*jnrD;
 765             j_coord_offsetE  = DIM*jnrE;
 766             j_coord_offsetF  = DIM*jnrF;
 767             j_coord_offsetG  = DIM*jnrG;
 768             j_coord_offsetH  = DIM*jnrH;
 769
 770             /* load j atom coordinates */
 771             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 772                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 773                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 774                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 775                                                  &jx0,&jy0,&jz0);
 776
 777             /* Calculate displacement vector */
 778             dx00             = _mm256_sub_ps(ix0,jx0);
 779             dy00             = _mm256_sub_ps(iy0,jy0);
 780             dz00             = _mm256_sub_ps(iz0,jz0);
 781
 782             /* Calculate squared distance and things based on it */
 783             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 784
 785             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 786
 787             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 788
 789             /* Load parameters for j particles */
 790             vdwjidx0A        = 2*vdwtype[jnrA+0];
 791             vdwjidx0B        = 2*vdwtype[jnrB+0];
 792             vdwjidx0C        = 2*vdwtype[jnrC+0];
 793             vdwjidx0D        = 2*vdwtype[jnrD+0];
 794             vdwjidx0E        = 2*vdwtype[jnrE+0];
 795             vdwjidx0F        = 2*vdwtype[jnrF+0];
 796             vdwjidx0G        = 2*vdwtype[jnrG+0];
 797             vdwjidx0H        = 2*vdwtype[jnrH+0];
 798
 799             /**************************
 800              * CALCULATE INTERACTIONS *
 801              **************************/
 802
 803             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 804             {
 805
 806             r00              = _mm256_mul_ps(rsq00,rinv00);
 807             r00              = _mm256_andnot_ps(dummy_mask,r00);
 808
 809             /* Compute parameters for interactions between i and j atoms */
 810             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 811                                             vdwioffsetptr0+vdwjidx0B,
 812                                             vdwioffsetptr0+vdwjidx0C,
 813                                             vdwioffsetptr0+vdwjidx0D,
 814                                             vdwioffsetptr0+vdwjidx0E,
 815                                             vdwioffsetptr0+vdwjidx0F,
 816                                             vdwioffsetptr0+vdwjidx0G,
 817                                             vdwioffsetptr0+vdwjidx0H,
 818                                             &c6_00,&c12_00);
 819
 820             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 821                                                                   vdwgridioffsetptr0+vdwjidx0B,
 822                                                                   vdwgridioffsetptr0+vdwjidx0C,
 823                                                                   vdwgridioffsetptr0+vdwjidx0D,
 824                                                                   vdwgridioffsetptr0+vdwjidx0E,
 825                                                                   vdwgridioffsetptr0+vdwjidx0F,
 826                                                                   vdwgridioffsetptr0+vdwjidx0G,
 827                                                                   vdwgridioffsetptr0+vdwjidx0H);
 828
 829             /* Analytical LJ-PME */
 830             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 831             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 832             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 833             exponent         = gmx_simd_exp_r(ewcljrsq);
 834             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 835             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 836             /* f6A = 6 * C6grid * (1 - poly) */
 837             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
 838             /* f6B = C6grid * exponent * beta^6 */
 839             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
 840             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 841             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 842
 843             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 844
 845             fscal            = fvdw;
 846
 847             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 848
 849             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm256_mul_ps(fscal,dx00);
 853             ty               = _mm256_mul_ps(fscal,dy00);
 854             tz               = _mm256_mul_ps(fscal,dz00);
 855
 856             /* Update vectorial force */
 857             fix0             = _mm256_add_ps(fix0,tx);
 858             fiy0             = _mm256_add_ps(fiy0,ty);
 859             fiz0             = _mm256_add_ps(fiz0,tz);
 860
 861             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 862             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 863             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 864             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 865             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 866             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 867             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 868             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 869             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 870
 871             }
 872
 873             /* Inner loop uses 50 flops */
 874         }
 875
 876         /* End of innermost loop */
 877
 878         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 879                                                  f+i_coord_offset,fshift+i_shift_offset);
 880
 881         /* Increment number of inner iterations */
 882         inneriter                  += j_index_end - j_index_start;
 883
 884         /* Outer loop uses 6 flops */
 885     }
 886
 887     /* Increment number of outer iterations */
 888     outeriter        += nri;
 889
 890     /* Update outer/inner flops */
 891
 892     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*50);
 893 }