src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecNone_VdwLJEw_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_VdwLJEw_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_VdwLJEw_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     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134     j_coord_offsetE = 0;
 135     j_coord_offsetF = 0;
 136     j_coord_offsetG = 0;
 137     j_coord_offsetH = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 163
 164         fix0             = _mm256_setzero_ps();
 165         fiy0             = _mm256_setzero_ps();
 166         fiz0             = _mm256_setzero_ps();
 167
 168         /* Load parameters for i particles */
 169         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 170         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 171
 172         /* Reset potential sums */
 173         vvdwsum          = _mm256_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             jnrE             = jjnr[jidx+4];
 185             jnrF             = jjnr[jidx+5];
 186             jnrG             = jjnr[jidx+6];
 187             jnrH             = jjnr[jidx+7];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192             j_coord_offsetE  = DIM*jnrE;
 193             j_coord_offsetF  = DIM*jnrF;
 194             j_coord_offsetG  = DIM*jnrG;
 195             j_coord_offsetH  = DIM*jnrH;
 196
 197             /* load j atom coordinates */
 198             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 200                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 201                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 202                                                  &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm256_sub_ps(ix0,jx0);
 206             dy00             = _mm256_sub_ps(iy0,jy0);
 207             dz00             = _mm256_sub_ps(iz0,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 211
 212             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 213
 214             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 215
 216             /* Load parameters for j particles */
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219             vdwjidx0C        = 2*vdwtype[jnrC+0];
 220             vdwjidx0D        = 2*vdwtype[jnrD+0];
 221             vdwjidx0E        = 2*vdwtype[jnrE+0];
 222             vdwjidx0F        = 2*vdwtype[jnrF+0];
 223             vdwjidx0G        = 2*vdwtype[jnrG+0];
 224             vdwjidx0H        = 2*vdwtype[jnrH+0];
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             r00              = _mm256_mul_ps(rsq00,rinv00);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 234                                             vdwioffsetptr0+vdwjidx0B,
 235                                             vdwioffsetptr0+vdwjidx0C,
 236                                             vdwioffsetptr0+vdwjidx0D,
 237                                             vdwioffsetptr0+vdwjidx0E,
 238                                             vdwioffsetptr0+vdwjidx0F,
 239                                             vdwioffsetptr0+vdwjidx0G,
 240                                             vdwioffsetptr0+vdwjidx0H,
 241                                             &c6_00,&c12_00);
 242
 243             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 244                                                                   vdwgridioffsetptr0+vdwjidx0B,
 245                                                                   vdwgridioffsetptr0+vdwjidx0C,
 246                                                                   vdwgridioffsetptr0+vdwjidx0D,
 247                                                                   vdwgridioffsetptr0+vdwjidx0E,
 248                                                                   vdwgridioffsetptr0+vdwjidx0F,
 249                                                                   vdwgridioffsetptr0+vdwjidx0G,
 250                                                                   vdwgridioffsetptr0+vdwjidx0H);
 251
 252             /* Analytical LJ-PME */
 253             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 254             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 255             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 256             exponent         = gmx_simd_exp_r(ewcljrsq);
 257             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 258             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 259             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 260             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 261             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 262             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 263             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 264             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);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = fvdw;
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm256_mul_ps(fscal,dx00);
 273             ty               = _mm256_mul_ps(fscal,dy00);
 274             tz               = _mm256_mul_ps(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm256_add_ps(fix0,tx);
 278             fiy0             = _mm256_add_ps(fiy0,ty);
 279             fiz0             = _mm256_add_ps(fiz0,tz);
 280
 281             fjptrA             = f+j_coord_offsetA;
 282             fjptrB             = f+j_coord_offsetB;
 283             fjptrC             = f+j_coord_offsetC;
 284             fjptrD             = f+j_coord_offsetD;
 285             fjptrE             = f+j_coord_offsetE;
 286             fjptrF             = f+j_coord_offsetF;
 287             fjptrG             = f+j_coord_offsetG;
 288             fjptrH             = f+j_coord_offsetH;
 289             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 290
 291             /* Inner loop uses 51 flops */
 292         }
 293
 294         if(jidx<j_index_end)
 295         {
 296
 297             /* Get j neighbor index, and coordinate index */
 298             jnrlistA         = jjnr[jidx];
 299             jnrlistB         = jjnr[jidx+1];
 300             jnrlistC         = jjnr[jidx+2];
 301             jnrlistD         = jjnr[jidx+3];
 302             jnrlistE         = jjnr[jidx+4];
 303             jnrlistF         = jjnr[jidx+5];
 304             jnrlistG         = jjnr[jidx+6];
 305             jnrlistH         = jjnr[jidx+7];
 306             /* Sign of each element will be negative for non-real atoms.
 307              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 308              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 309              */
 310             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 311                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 312
 313             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 314             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 315             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 316             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 317             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 318             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 319             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 320             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 321             j_coord_offsetA  = DIM*jnrA;
 322             j_coord_offsetB  = DIM*jnrB;
 323             j_coord_offsetC  = DIM*jnrC;
 324             j_coord_offsetD  = DIM*jnrD;
 325             j_coord_offsetE  = DIM*jnrE;
 326             j_coord_offsetF  = DIM*jnrF;
 327             j_coord_offsetG  = DIM*jnrG;
 328             j_coord_offsetH  = DIM*jnrH;
 329
 330             /* load j atom coordinates */
 331             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 332                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 333                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 334                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 335                                                  &jx0,&jy0,&jz0);
 336
 337             /* Calculate displacement vector */
 338             dx00             = _mm256_sub_ps(ix0,jx0);
 339             dy00             = _mm256_sub_ps(iy0,jy0);
 340             dz00             = _mm256_sub_ps(iz0,jz0);
 341
 342             /* Calculate squared distance and things based on it */
 343             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 344
 345             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 346
 347             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 348
 349             /* Load parameters for j particles */
 350             vdwjidx0A        = 2*vdwtype[jnrA+0];
 351             vdwjidx0B        = 2*vdwtype[jnrB+0];
 352             vdwjidx0C        = 2*vdwtype[jnrC+0];
 353             vdwjidx0D        = 2*vdwtype[jnrD+0];
 354             vdwjidx0E        = 2*vdwtype[jnrE+0];
 355             vdwjidx0F        = 2*vdwtype[jnrF+0];
 356             vdwjidx0G        = 2*vdwtype[jnrG+0];
 357             vdwjidx0H        = 2*vdwtype[jnrH+0];
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             r00              = _mm256_mul_ps(rsq00,rinv00);
 364             r00              = _mm256_andnot_ps(dummy_mask,r00);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 368                                             vdwioffsetptr0+vdwjidx0B,
 369                                             vdwioffsetptr0+vdwjidx0C,
 370                                             vdwioffsetptr0+vdwjidx0D,
 371                                             vdwioffsetptr0+vdwjidx0E,
 372                                             vdwioffsetptr0+vdwjidx0F,
 373                                             vdwioffsetptr0+vdwjidx0G,
 374                                             vdwioffsetptr0+vdwjidx0H,
 375                                             &c6_00,&c12_00);
 376
 377             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 378                                                                   vdwgridioffsetptr0+vdwjidx0B,
 379                                                                   vdwgridioffsetptr0+vdwjidx0C,
 380                                                                   vdwgridioffsetptr0+vdwjidx0D,
 381                                                                   vdwgridioffsetptr0+vdwjidx0E,
 382                                                                   vdwgridioffsetptr0+vdwjidx0F,
 383                                                                   vdwgridioffsetptr0+vdwjidx0G,
 384                                                                   vdwgridioffsetptr0+vdwjidx0H);
 385
 386             /* Analytical LJ-PME */
 387             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 388             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 389             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 390             exponent         = gmx_simd_exp_r(ewcljrsq);
 391             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 392             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 393             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 394             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 395             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 396             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 397             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 398             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);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 402             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 403
 404             fscal            = fvdw;
 405
 406             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_ps(fscal,dx00);
 410             ty               = _mm256_mul_ps(fscal,dy00);
 411             tz               = _mm256_mul_ps(fscal,dz00);
 412
 413             /* Update vectorial force */
 414             fix0             = _mm256_add_ps(fix0,tx);
 415             fiy0             = _mm256_add_ps(fiy0,ty);
 416             fiz0             = _mm256_add_ps(fiz0,tz);
 417
 418             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 419             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 420             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 421             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 422             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 423             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 424             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 425             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 426             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 427
 428             /* Inner loop uses 52 flops */
 429         }
 430
 431         /* End of innermost loop */
 432
 433         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 434                                                  f+i_coord_offset,fshift+i_shift_offset);
 435
 436         ggid                        = gid[iidx];
 437         /* Update potential energies */
 438         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 439
 440         /* Increment number of inner iterations */
 441         inneriter                  += j_index_end - j_index_start;
 442
 443         /* Outer loop uses 7 flops */
 444     }
 445
 446     /* Increment number of outer iterations */
 447     outeriter        += nri;
 448
 449     /* Update outer/inner flops */
 450
 451     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*52);
 452 }
 453 /*
 454  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_avx_256_single
 455  * Electrostatics interaction: None
 456  * VdW interaction:            LJEwald
 457  * Geometry:                   Particle-Particle
 458  * Calculate force/pot:        Force
 459  */
 460 void
 461 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_avx_256_single
 462                     (t_nblist                    * gmx_restrict       nlist,
 463                      rvec                        * gmx_restrict          xx,
 464                      rvec                        * gmx_restrict          ff,
 465                      t_forcerec                  * gmx_restrict          fr,
 466                      t_mdatoms                   * gmx_restrict     mdatoms,
 467                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 468                      t_nrnb                      * gmx_restrict        nrnb)
 469 {
 470     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 471      * just 0 for non-waters.
 472      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 473      * jnr indices corresponding to data put in the four positions in the SIMD register.
 474      */
 475     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 476     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 477     int              jnrA,jnrB,jnrC,jnrD;
 478     int              jnrE,jnrF,jnrG,jnrH;
 479     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 480     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 481     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 482     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 483     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 484     real             rcutoff_scalar;
 485     real             *shiftvec,*fshift,*x,*f;
 486     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 487     real             scratch[4*DIM];
 488     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 489     real *           vdwioffsetptr0;
 490     real *           vdwgridioffsetptr0;
 491     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 492     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 493     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 494     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 495     int              nvdwtype;
 496     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 497     int              *vdwtype;
 498     real             *vdwparam;
 499     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 500     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 501     __m256           c6grid_00;
 502     real             *vdwgridparam;
 503     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 504     __m256           one_half  = _mm256_set1_ps(0.5);
 505     __m256           minus_one = _mm256_set1_ps(-1.0);
 506     __m256           dummy_mask,cutoff_mask;
 507     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 508     __m256           one     = _mm256_set1_ps(1.0);
 509     __m256           two     = _mm256_set1_ps(2.0);
 510     x                = xx[0];
 511     f                = ff[0];
 512
 513     nri              = nlist->nri;
 514     iinr             = nlist->iinr;
 515     jindex           = nlist->jindex;
 516     jjnr             = nlist->jjnr;
 517     shiftidx         = nlist->shift;
 518     gid              = nlist->gid;
 519     shiftvec         = fr->shift_vec[0];
 520     fshift           = fr->fshift[0];
 521     nvdwtype         = fr->ntype;
 522     vdwparam         = fr->nbfp;
 523     vdwtype          = mdatoms->typeA;
 524     vdwgridparam     = fr->ljpme_c6grid;
 525     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 526     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 527     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 528
 529     /* Avoid stupid compiler warnings */
 530     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 531     j_coord_offsetA = 0;
 532     j_coord_offsetB = 0;
 533     j_coord_offsetC = 0;
 534     j_coord_offsetD = 0;
 535     j_coord_offsetE = 0;
 536     j_coord_offsetF = 0;
 537     j_coord_offsetG = 0;
 538     j_coord_offsetH = 0;
 539
 540     outeriter        = 0;
 541     inneriter        = 0;
 542
 543     for(iidx=0;iidx<4*DIM;iidx++)
 544     {
 545         scratch[iidx] = 0.0;
 546     }
 547
 548     /* Start outer loop over neighborlists */
 549     for(iidx=0; iidx<nri; iidx++)
 550     {
 551         /* Load shift vector for this list */
 552         i_shift_offset   = DIM*shiftidx[iidx];
 553
 554         /* Load limits for loop over neighbors */
 555         j_index_start    = jindex[iidx];
 556         j_index_end      = jindex[iidx+1];
 557
 558         /* Get outer coordinate index */
 559         inr              = iinr[iidx];
 560         i_coord_offset   = DIM*inr;
 561
 562         /* Load i particle coords and add shift vector */
 563         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 564
 565         fix0             = _mm256_setzero_ps();
 566         fiy0             = _mm256_setzero_ps();
 567         fiz0             = _mm256_setzero_ps();
 568
 569         /* Load parameters for i particles */
 570         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 571         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 572
 573         /* Start inner kernel loop */
 574         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 575         {
 576
 577             /* Get j neighbor index, and coordinate index */
 578             jnrA             = jjnr[jidx];
 579             jnrB             = jjnr[jidx+1];
 580             jnrC             = jjnr[jidx+2];
 581             jnrD             = jjnr[jidx+3];
 582             jnrE             = jjnr[jidx+4];
 583             jnrF             = jjnr[jidx+5];
 584             jnrG             = jjnr[jidx+6];
 585             jnrH             = jjnr[jidx+7];
 586             j_coord_offsetA  = DIM*jnrA;
 587             j_coord_offsetB  = DIM*jnrB;
 588             j_coord_offsetC  = DIM*jnrC;
 589             j_coord_offsetD  = DIM*jnrD;
 590             j_coord_offsetE  = DIM*jnrE;
 591             j_coord_offsetF  = DIM*jnrF;
 592             j_coord_offsetG  = DIM*jnrG;
 593             j_coord_offsetH  = DIM*jnrH;
 594
 595             /* load j atom coordinates */
 596             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 597                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 598                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 599                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 600                                                  &jx0,&jy0,&jz0);
 601
 602             /* Calculate displacement vector */
 603             dx00             = _mm256_sub_ps(ix0,jx0);
 604             dy00             = _mm256_sub_ps(iy0,jy0);
 605             dz00             = _mm256_sub_ps(iz0,jz0);
 606
 607             /* Calculate squared distance and things based on it */
 608             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 609
 610             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 611
 612             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 613
 614             /* Load parameters for j particles */
 615             vdwjidx0A        = 2*vdwtype[jnrA+0];
 616             vdwjidx0B        = 2*vdwtype[jnrB+0];
 617             vdwjidx0C        = 2*vdwtype[jnrC+0];
 618             vdwjidx0D        = 2*vdwtype[jnrD+0];
 619             vdwjidx0E        = 2*vdwtype[jnrE+0];
 620             vdwjidx0F        = 2*vdwtype[jnrF+0];
 621             vdwjidx0G        = 2*vdwtype[jnrG+0];
 622             vdwjidx0H        = 2*vdwtype[jnrH+0];
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             r00              = _mm256_mul_ps(rsq00,rinv00);
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 632                                             vdwioffsetptr0+vdwjidx0B,
 633                                             vdwioffsetptr0+vdwjidx0C,
 634                                             vdwioffsetptr0+vdwjidx0D,
 635                                             vdwioffsetptr0+vdwjidx0E,
 636                                             vdwioffsetptr0+vdwjidx0F,
 637                                             vdwioffsetptr0+vdwjidx0G,
 638                                             vdwioffsetptr0+vdwjidx0H,
 639                                             &c6_00,&c12_00);
 640
 641             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 642                                                                   vdwgridioffsetptr0+vdwjidx0B,
 643                                                                   vdwgridioffsetptr0+vdwjidx0C,
 644                                                                   vdwgridioffsetptr0+vdwjidx0D,
 645                                                                   vdwgridioffsetptr0+vdwjidx0E,
 646                                                                   vdwgridioffsetptr0+vdwjidx0F,
 647                                                                   vdwgridioffsetptr0+vdwjidx0G,
 648                                                                   vdwgridioffsetptr0+vdwjidx0H);
 649
 650             /* Analytical LJ-PME */
 651             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 652             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 653             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 654             exponent         = gmx_simd_exp_r(ewcljrsq);
 655             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 656             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 657             /* f6A = 6 * C6grid * (1 - poly) */
 658             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
 659             /* f6B = C6grid * exponent * beta^6 */
 660             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
 661             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 662             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);
 663
 664             fscal            = fvdw;
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm256_mul_ps(fscal,dx00);
 668             ty               = _mm256_mul_ps(fscal,dy00);
 669             tz               = _mm256_mul_ps(fscal,dz00);
 670
 671             /* Update vectorial force */
 672             fix0             = _mm256_add_ps(fix0,tx);
 673             fiy0             = _mm256_add_ps(fiy0,ty);
 674             fiz0             = _mm256_add_ps(fiz0,tz);
 675
 676             fjptrA             = f+j_coord_offsetA;
 677             fjptrB             = f+j_coord_offsetB;
 678             fjptrC             = f+j_coord_offsetC;
 679             fjptrD             = f+j_coord_offsetD;
 680             fjptrE             = f+j_coord_offsetE;
 681             fjptrF             = f+j_coord_offsetF;
 682             fjptrG             = f+j_coord_offsetG;
 683             fjptrH             = f+j_coord_offsetH;
 684             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 685
 686             /* Inner loop uses 46 flops */
 687         }
 688
 689         if(jidx<j_index_end)
 690         {
 691
 692             /* Get j neighbor index, and coordinate index */
 693             jnrlistA         = jjnr[jidx];
 694             jnrlistB         = jjnr[jidx+1];
 695             jnrlistC         = jjnr[jidx+2];
 696             jnrlistD         = jjnr[jidx+3];
 697             jnrlistE         = jjnr[jidx+4];
 698             jnrlistF         = jjnr[jidx+5];
 699             jnrlistG         = jjnr[jidx+6];
 700             jnrlistH         = jjnr[jidx+7];
 701             /* Sign of each element will be negative for non-real atoms.
 702              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 703              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 704              */
 705             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 706                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 707
 708             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 709             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 710             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 711             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 712             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 713             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 714             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 715             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 716             j_coord_offsetA  = DIM*jnrA;
 717             j_coord_offsetB  = DIM*jnrB;
 718             j_coord_offsetC  = DIM*jnrC;
 719             j_coord_offsetD  = DIM*jnrD;
 720             j_coord_offsetE  = DIM*jnrE;
 721             j_coord_offsetF  = DIM*jnrF;
 722             j_coord_offsetG  = DIM*jnrG;
 723             j_coord_offsetH  = DIM*jnrH;
 724
 725             /* load j atom coordinates */
 726             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 727                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 728                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 729                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 730                                                  &jx0,&jy0,&jz0);
 731
 732             /* Calculate displacement vector */
 733             dx00             = _mm256_sub_ps(ix0,jx0);
 734             dy00             = _mm256_sub_ps(iy0,jy0);
 735             dz00             = _mm256_sub_ps(iz0,jz0);
 736
 737             /* Calculate squared distance and things based on it */
 738             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 739
 740             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 741
 742             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 743
 744             /* Load parameters for j particles */
 745             vdwjidx0A        = 2*vdwtype[jnrA+0];
 746             vdwjidx0B        = 2*vdwtype[jnrB+0];
 747             vdwjidx0C        = 2*vdwtype[jnrC+0];
 748             vdwjidx0D        = 2*vdwtype[jnrD+0];
 749             vdwjidx0E        = 2*vdwtype[jnrE+0];
 750             vdwjidx0F        = 2*vdwtype[jnrF+0];
 751             vdwjidx0G        = 2*vdwtype[jnrG+0];
 752             vdwjidx0H        = 2*vdwtype[jnrH+0];
 753
 754             /**************************
 755              * CALCULATE INTERACTIONS *
 756              **************************/
 757
 758             r00              = _mm256_mul_ps(rsq00,rinv00);
 759             r00              = _mm256_andnot_ps(dummy_mask,r00);
 760
 761             /* Compute parameters for interactions between i and j atoms */
 762             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 763                                             vdwioffsetptr0+vdwjidx0B,
 764                                             vdwioffsetptr0+vdwjidx0C,
 765                                             vdwioffsetptr0+vdwjidx0D,
 766                                             vdwioffsetptr0+vdwjidx0E,
 767                                             vdwioffsetptr0+vdwjidx0F,
 768                                             vdwioffsetptr0+vdwjidx0G,
 769                                             vdwioffsetptr0+vdwjidx0H,
 770                                             &c6_00,&c12_00);
 771
 772             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 773                                                                   vdwgridioffsetptr0+vdwjidx0B,
 774                                                                   vdwgridioffsetptr0+vdwjidx0C,
 775                                                                   vdwgridioffsetptr0+vdwjidx0D,
 776                                                                   vdwgridioffsetptr0+vdwjidx0E,
 777                                                                   vdwgridioffsetptr0+vdwjidx0F,
 778                                                                   vdwgridioffsetptr0+vdwjidx0G,
 779                                                                   vdwgridioffsetptr0+vdwjidx0H);
 780
 781             /* Analytical LJ-PME */
 782             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 783             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 784             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 785             exponent         = gmx_simd_exp_r(ewcljrsq);
 786             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 787             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 788             /* f6A = 6 * C6grid * (1 - poly) */
 789             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
 790             /* f6B = C6grid * exponent * beta^6 */
 791             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
 792             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 793             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);
 794
 795             fscal            = fvdw;
 796
 797             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 798
 799             /* Calculate temporary vectorial force */
 800             tx               = _mm256_mul_ps(fscal,dx00);
 801             ty               = _mm256_mul_ps(fscal,dy00);
 802             tz               = _mm256_mul_ps(fscal,dz00);
 803
 804             /* Update vectorial force */
 805             fix0             = _mm256_add_ps(fix0,tx);
 806             fiy0             = _mm256_add_ps(fiy0,ty);
 807             fiz0             = _mm256_add_ps(fiz0,tz);
 808
 809             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 810             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 811             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 812             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 813             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 814             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 815             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 816             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 817             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 818
 819             /* Inner loop uses 47 flops */
 820         }
 821
 822         /* End of innermost loop */
 823
 824         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 825                                                  f+i_coord_offset,fshift+i_shift_offset);
 826
 827         /* Increment number of inner iterations */
 828         inneriter                  += j_index_end - j_index_start;
 829
 830         /* Outer loop uses 6 flops */
 831     }
 832
 833     /* Increment number of outer iterations */
 834     outeriter        += nri;
 835
 836     /* Update outer/inner flops */
 837
 838     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*47);
 839 }