src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecNone_VdwLJSw_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 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_VdwLJSw_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: None
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwLJSw_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     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     int              nvdwtype;
  94     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  98     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  99     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 100     real             rswitch_scalar,d_scalar;
 101     __m256           dummy_mask,cutoff_mask;
 102     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 103     __m256           one     = _mm256_set1_ps(1.0);
 104     __m256           two     = _mm256_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     rcutoff_scalar   = fr->rvdw;
 121     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 122     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 123
 124     rswitch_scalar   = fr->rvdw_switch;
 125     rswitch          = _mm256_set1_ps(rswitch_scalar);
 126     /* Setup switch parameters */
 127     d_scalar         = rcutoff_scalar-rswitch_scalar;
 128     d                = _mm256_set1_ps(d_scalar);
 129     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 130     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 131     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 132     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 133     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 134     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142     j_coord_offsetE = 0;
 143     j_coord_offsetF = 0;
 144     j_coord_offsetG = 0;
 145     j_coord_offsetH = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 171
 172         fix0             = _mm256_setzero_ps();
 173         fiy0             = _mm256_setzero_ps();
 174         fiz0             = _mm256_setzero_ps();
 175
 176         /* Load parameters for i particles */
 177         vdwioffsetptr0   = vdwparam+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             /* LENNARD-JONES DISPERSION/REPULSION */
 254
 255             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 256             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 257             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 258             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 259             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 260
 261             d                = _mm256_sub_ps(r00,rswitch);
 262             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 263             d2               = _mm256_mul_ps(d,d);
 264             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 265
 266             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 267
 268             /* Evaluate switch function */
 269             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 270             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 271             vvdw             = _mm256_mul_ps(vvdw,sw);
 272             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 276             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 277
 278             fscal            = fvdw;
 279
 280             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm256_mul_ps(fscal,dx00);
 284             ty               = _mm256_mul_ps(fscal,dy00);
 285             tz               = _mm256_mul_ps(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm256_add_ps(fix0,tx);
 289             fiy0             = _mm256_add_ps(fiy0,ty);
 290             fiz0             = _mm256_add_ps(fiz0,tz);
 291
 292             fjptrA             = f+j_coord_offsetA;
 293             fjptrB             = f+j_coord_offsetB;
 294             fjptrC             = f+j_coord_offsetC;
 295             fjptrD             = f+j_coord_offsetD;
 296             fjptrE             = f+j_coord_offsetE;
 297             fjptrF             = f+j_coord_offsetF;
 298             fjptrG             = f+j_coord_offsetG;
 299             fjptrH             = f+j_coord_offsetH;
 300             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 301
 302             }
 303
 304             /* Inner loop uses 59 flops */
 305         }
 306
 307         if(jidx<j_index_end)
 308         {
 309
 310             /* Get j neighbor index, and coordinate index */
 311             jnrlistA         = jjnr[jidx];
 312             jnrlistB         = jjnr[jidx+1];
 313             jnrlistC         = jjnr[jidx+2];
 314             jnrlistD         = jjnr[jidx+3];
 315             jnrlistE         = jjnr[jidx+4];
 316             jnrlistF         = jjnr[jidx+5];
 317             jnrlistG         = jjnr[jidx+6];
 318             jnrlistH         = jjnr[jidx+7];
 319             /* Sign of each element will be negative for non-real atoms.
 320              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 321              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 322              */
 323             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 324                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 325
 326             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 327             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 328             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 329             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 330             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 331             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 332             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 333             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 334             j_coord_offsetA  = DIM*jnrA;
 335             j_coord_offsetB  = DIM*jnrB;
 336             j_coord_offsetC  = DIM*jnrC;
 337             j_coord_offsetD  = DIM*jnrD;
 338             j_coord_offsetE  = DIM*jnrE;
 339             j_coord_offsetF  = DIM*jnrF;
 340             j_coord_offsetG  = DIM*jnrG;
 341             j_coord_offsetH  = DIM*jnrH;
 342
 343             /* load j atom coordinates */
 344             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 345                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 346                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 347                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 348                                                  &jx0,&jy0,&jz0);
 349
 350             /* Calculate displacement vector */
 351             dx00             = _mm256_sub_ps(ix0,jx0);
 352             dy00             = _mm256_sub_ps(iy0,jy0);
 353             dz00             = _mm256_sub_ps(iz0,jz0);
 354
 355             /* Calculate squared distance and things based on it */
 356             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 357
 358             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 359
 360             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 361
 362             /* Load parameters for j particles */
 363             vdwjidx0A        = 2*vdwtype[jnrA+0];
 364             vdwjidx0B        = 2*vdwtype[jnrB+0];
 365             vdwjidx0C        = 2*vdwtype[jnrC+0];
 366             vdwjidx0D        = 2*vdwtype[jnrD+0];
 367             vdwjidx0E        = 2*vdwtype[jnrE+0];
 368             vdwjidx0F        = 2*vdwtype[jnrF+0];
 369             vdwjidx0G        = 2*vdwtype[jnrG+0];
 370             vdwjidx0H        = 2*vdwtype[jnrH+0];
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 377             {
 378
 379             r00              = _mm256_mul_ps(rsq00,rinv00);
 380             r00              = _mm256_andnot_ps(dummy_mask,r00);
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 384                                             vdwioffsetptr0+vdwjidx0B,
 385                                             vdwioffsetptr0+vdwjidx0C,
 386                                             vdwioffsetptr0+vdwjidx0D,
 387                                             vdwioffsetptr0+vdwjidx0E,
 388                                             vdwioffsetptr0+vdwjidx0F,
 389                                             vdwioffsetptr0+vdwjidx0G,
 390                                             vdwioffsetptr0+vdwjidx0H,
 391                                             &c6_00,&c12_00);
 392
 393             /* LENNARD-JONES DISPERSION/REPULSION */
 394
 395             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 396             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 397             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 398             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 399             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 400
 401             d                = _mm256_sub_ps(r00,rswitch);
 402             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 403             d2               = _mm256_mul_ps(d,d);
 404             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 405
 406             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 407
 408             /* Evaluate switch function */
 409             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 410             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 411             vvdw             = _mm256_mul_ps(vvdw,sw);
 412             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 416             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 417             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 418
 419             fscal            = fvdw;
 420
 421             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 422
 423             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 424
 425             /* Calculate temporary vectorial force */
 426             tx               = _mm256_mul_ps(fscal,dx00);
 427             ty               = _mm256_mul_ps(fscal,dy00);
 428             tz               = _mm256_mul_ps(fscal,dz00);
 429
 430             /* Update vectorial force */
 431             fix0             = _mm256_add_ps(fix0,tx);
 432             fiy0             = _mm256_add_ps(fiy0,ty);
 433             fiz0             = _mm256_add_ps(fiz0,tz);
 434
 435             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 436             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 437             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 438             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 439             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 440             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 441             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 442             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 443             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 444
 445             }
 446
 447             /* Inner loop uses 60 flops */
 448         }
 449
 450         /* End of innermost loop */
 451
 452         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 453                                                  f+i_coord_offset,fshift+i_shift_offset);
 454
 455         ggid                        = gid[iidx];
 456         /* Update potential energies */
 457         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 458
 459         /* Increment number of inner iterations */
 460         inneriter                  += j_index_end - j_index_start;
 461
 462         /* Outer loop uses 7 flops */
 463     }
 464
 465     /* Increment number of outer iterations */
 466     outeriter        += nri;
 467
 468     /* Update outer/inner flops */
 469
 470     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*60);
 471 }
 472 /*
 473  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_256_single
 474  * Electrostatics interaction: None
 475  * VdW interaction:            LennardJones
 476  * Geometry:                   Particle-Particle
 477  * Calculate force/pot:        Force
 478  */
 479 void
 480 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_256_single
 481                     (t_nblist                    * gmx_restrict       nlist,
 482                      rvec                        * gmx_restrict          xx,
 483                      rvec                        * gmx_restrict          ff,
 484                      t_forcerec                  * gmx_restrict          fr,
 485                      t_mdatoms                   * gmx_restrict     mdatoms,
 486                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 487                      t_nrnb                      * gmx_restrict        nrnb)
 488 {
 489     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 490      * just 0 for non-waters.
 491      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 492      * jnr indices corresponding to data put in the four positions in the SIMD register.
 493      */
 494     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 495     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 496     int              jnrA,jnrB,jnrC,jnrD;
 497     int              jnrE,jnrF,jnrG,jnrH;
 498     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 499     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 500     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 501     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 502     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 503     real             rcutoff_scalar;
 504     real             *shiftvec,*fshift,*x,*f;
 505     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 506     real             scratch[4*DIM];
 507     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 508     real *           vdwioffsetptr0;
 509     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 510     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 511     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 512     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 513     int              nvdwtype;
 514     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 515     int              *vdwtype;
 516     real             *vdwparam;
 517     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 518     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 519     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 520     real             rswitch_scalar,d_scalar;
 521     __m256           dummy_mask,cutoff_mask;
 522     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 523     __m256           one     = _mm256_set1_ps(1.0);
 524     __m256           two     = _mm256_set1_ps(2.0);
 525     x                = xx[0];
 526     f                = ff[0];
 527
 528     nri              = nlist->nri;
 529     iinr             = nlist->iinr;
 530     jindex           = nlist->jindex;
 531     jjnr             = nlist->jjnr;
 532     shiftidx         = nlist->shift;
 533     gid              = nlist->gid;
 534     shiftvec         = fr->shift_vec[0];
 535     fshift           = fr->fshift[0];
 536     nvdwtype         = fr->ntype;
 537     vdwparam         = fr->nbfp;
 538     vdwtype          = mdatoms->typeA;
 539
 540     rcutoff_scalar   = fr->rvdw;
 541     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 542     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 543
 544     rswitch_scalar   = fr->rvdw_switch;
 545     rswitch          = _mm256_set1_ps(rswitch_scalar);
 546     /* Setup switch parameters */
 547     d_scalar         = rcutoff_scalar-rswitch_scalar;
 548     d                = _mm256_set1_ps(d_scalar);
 549     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 550     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 551     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 552     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 553     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 554     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 555
 556     /* Avoid stupid compiler warnings */
 557     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 558     j_coord_offsetA = 0;
 559     j_coord_offsetB = 0;
 560     j_coord_offsetC = 0;
 561     j_coord_offsetD = 0;
 562     j_coord_offsetE = 0;
 563     j_coord_offsetF = 0;
 564     j_coord_offsetG = 0;
 565     j_coord_offsetH = 0;
 566
 567     outeriter        = 0;
 568     inneriter        = 0;
 569
 570     for(iidx=0;iidx<4*DIM;iidx++)
 571     {
 572         scratch[iidx] = 0.0;
 573     }
 574
 575     /* Start outer loop over neighborlists */
 576     for(iidx=0; iidx<nri; iidx++)
 577     {
 578         /* Load shift vector for this list */
 579         i_shift_offset   = DIM*shiftidx[iidx];
 580
 581         /* Load limits for loop over neighbors */
 582         j_index_start    = jindex[iidx];
 583         j_index_end      = jindex[iidx+1];
 584
 585         /* Get outer coordinate index */
 586         inr              = iinr[iidx];
 587         i_coord_offset   = DIM*inr;
 588
 589         /* Load i particle coords and add shift vector */
 590         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 591
 592         fix0             = _mm256_setzero_ps();
 593         fiy0             = _mm256_setzero_ps();
 594         fiz0             = _mm256_setzero_ps();
 595
 596         /* Load parameters for i particles */
 597         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 598
 599         /* Start inner kernel loop */
 600         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 601         {
 602
 603             /* Get j neighbor index, and coordinate index */
 604             jnrA             = jjnr[jidx];
 605             jnrB             = jjnr[jidx+1];
 606             jnrC             = jjnr[jidx+2];
 607             jnrD             = jjnr[jidx+3];
 608             jnrE             = jjnr[jidx+4];
 609             jnrF             = jjnr[jidx+5];
 610             jnrG             = jjnr[jidx+6];
 611             jnrH             = jjnr[jidx+7];
 612             j_coord_offsetA  = DIM*jnrA;
 613             j_coord_offsetB  = DIM*jnrB;
 614             j_coord_offsetC  = DIM*jnrC;
 615             j_coord_offsetD  = DIM*jnrD;
 616             j_coord_offsetE  = DIM*jnrE;
 617             j_coord_offsetF  = DIM*jnrF;
 618             j_coord_offsetG  = DIM*jnrG;
 619             j_coord_offsetH  = DIM*jnrH;
 620
 621             /* load j atom coordinates */
 622             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 623                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 624                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 625                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 626                                                  &jx0,&jy0,&jz0);
 627
 628             /* Calculate displacement vector */
 629             dx00             = _mm256_sub_ps(ix0,jx0);
 630             dy00             = _mm256_sub_ps(iy0,jy0);
 631             dz00             = _mm256_sub_ps(iz0,jz0);
 632
 633             /* Calculate squared distance and things based on it */
 634             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 635
 636             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 637
 638             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 639
 640             /* Load parameters for j particles */
 641             vdwjidx0A        = 2*vdwtype[jnrA+0];
 642             vdwjidx0B        = 2*vdwtype[jnrB+0];
 643             vdwjidx0C        = 2*vdwtype[jnrC+0];
 644             vdwjidx0D        = 2*vdwtype[jnrD+0];
 645             vdwjidx0E        = 2*vdwtype[jnrE+0];
 646             vdwjidx0F        = 2*vdwtype[jnrF+0];
 647             vdwjidx0G        = 2*vdwtype[jnrG+0];
 648             vdwjidx0H        = 2*vdwtype[jnrH+0];
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 655             {
 656
 657             r00              = _mm256_mul_ps(rsq00,rinv00);
 658
 659             /* Compute parameters for interactions between i and j atoms */
 660             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 661                                             vdwioffsetptr0+vdwjidx0B,
 662                                             vdwioffsetptr0+vdwjidx0C,
 663                                             vdwioffsetptr0+vdwjidx0D,
 664                                             vdwioffsetptr0+vdwjidx0E,
 665                                             vdwioffsetptr0+vdwjidx0F,
 666                                             vdwioffsetptr0+vdwjidx0G,
 667                                             vdwioffsetptr0+vdwjidx0H,
 668                                             &c6_00,&c12_00);
 669
 670             /* LENNARD-JONES DISPERSION/REPULSION */
 671
 672             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 673             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 674             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 675             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 676             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 677
 678             d                = _mm256_sub_ps(r00,rswitch);
 679             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 680             d2               = _mm256_mul_ps(d,d);
 681             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 682
 683             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 684
 685             /* Evaluate switch function */
 686             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 687             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 688             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 689
 690             fscal            = fvdw;
 691
 692             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm256_mul_ps(fscal,dx00);
 696             ty               = _mm256_mul_ps(fscal,dy00);
 697             tz               = _mm256_mul_ps(fscal,dz00);
 698
 699             /* Update vectorial force */
 700             fix0             = _mm256_add_ps(fix0,tx);
 701             fiy0             = _mm256_add_ps(fiy0,ty);
 702             fiz0             = _mm256_add_ps(fiz0,tz);
 703
 704             fjptrA             = f+j_coord_offsetA;
 705             fjptrB             = f+j_coord_offsetB;
 706             fjptrC             = f+j_coord_offsetC;
 707             fjptrD             = f+j_coord_offsetD;
 708             fjptrE             = f+j_coord_offsetE;
 709             fjptrF             = f+j_coord_offsetF;
 710             fjptrG             = f+j_coord_offsetG;
 711             fjptrH             = f+j_coord_offsetH;
 712             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 713
 714             }
 715
 716             /* Inner loop uses 56 flops */
 717         }
 718
 719         if(jidx<j_index_end)
 720         {
 721
 722             /* Get j neighbor index, and coordinate index */
 723             jnrlistA         = jjnr[jidx];
 724             jnrlistB         = jjnr[jidx+1];
 725             jnrlistC         = jjnr[jidx+2];
 726             jnrlistD         = jjnr[jidx+3];
 727             jnrlistE         = jjnr[jidx+4];
 728             jnrlistF         = jjnr[jidx+5];
 729             jnrlistG         = jjnr[jidx+6];
 730             jnrlistH         = jjnr[jidx+7];
 731             /* Sign of each element will be negative for non-real atoms.
 732              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 733              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 734              */
 735             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 736                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 737
 738             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 739             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 740             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 741             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 742             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 743             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 744             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 745             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 746             j_coord_offsetA  = DIM*jnrA;
 747             j_coord_offsetB  = DIM*jnrB;
 748             j_coord_offsetC  = DIM*jnrC;
 749             j_coord_offsetD  = DIM*jnrD;
 750             j_coord_offsetE  = DIM*jnrE;
 751             j_coord_offsetF  = DIM*jnrF;
 752             j_coord_offsetG  = DIM*jnrG;
 753             j_coord_offsetH  = DIM*jnrH;
 754
 755             /* load j atom coordinates */
 756             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 757                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 758                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 759                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 760                                                  &jx0,&jy0,&jz0);
 761
 762             /* Calculate displacement vector */
 763             dx00             = _mm256_sub_ps(ix0,jx0);
 764             dy00             = _mm256_sub_ps(iy0,jy0);
 765             dz00             = _mm256_sub_ps(iz0,jz0);
 766
 767             /* Calculate squared distance and things based on it */
 768             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 769
 770             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 771
 772             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 773
 774             /* Load parameters for j particles */
 775             vdwjidx0A        = 2*vdwtype[jnrA+0];
 776             vdwjidx0B        = 2*vdwtype[jnrB+0];
 777             vdwjidx0C        = 2*vdwtype[jnrC+0];
 778             vdwjidx0D        = 2*vdwtype[jnrD+0];
 779             vdwjidx0E        = 2*vdwtype[jnrE+0];
 780             vdwjidx0F        = 2*vdwtype[jnrF+0];
 781             vdwjidx0G        = 2*vdwtype[jnrG+0];
 782             vdwjidx0H        = 2*vdwtype[jnrH+0];
 783
 784             /**************************
 785              * CALCULATE INTERACTIONS *
 786              **************************/
 787
 788             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 789             {
 790
 791             r00              = _mm256_mul_ps(rsq00,rinv00);
 792             r00              = _mm256_andnot_ps(dummy_mask,r00);
 793
 794             /* Compute parameters for interactions between i and j atoms */
 795             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 796                                             vdwioffsetptr0+vdwjidx0B,
 797                                             vdwioffsetptr0+vdwjidx0C,
 798                                             vdwioffsetptr0+vdwjidx0D,
 799                                             vdwioffsetptr0+vdwjidx0E,
 800                                             vdwioffsetptr0+vdwjidx0F,
 801                                             vdwioffsetptr0+vdwjidx0G,
 802                                             vdwioffsetptr0+vdwjidx0H,
 803                                             &c6_00,&c12_00);
 804
 805             /* LENNARD-JONES DISPERSION/REPULSION */
 806
 807             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 808             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 809             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 810             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 811             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 812
 813             d                = _mm256_sub_ps(r00,rswitch);
 814             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 815             d2               = _mm256_mul_ps(d,d);
 816             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 817
 818             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 819
 820             /* Evaluate switch function */
 821             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 822             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 823             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 824
 825             fscal            = fvdw;
 826
 827             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 828
 829             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 830
 831             /* Calculate temporary vectorial force */
 832             tx               = _mm256_mul_ps(fscal,dx00);
 833             ty               = _mm256_mul_ps(fscal,dy00);
 834             tz               = _mm256_mul_ps(fscal,dz00);
 835
 836             /* Update vectorial force */
 837             fix0             = _mm256_add_ps(fix0,tx);
 838             fiy0             = _mm256_add_ps(fiy0,ty);
 839             fiz0             = _mm256_add_ps(fiz0,tz);
 840
 841             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 842             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 843             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 844             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 845             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 846             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 847             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 848             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 849             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 850
 851             }
 852
 853             /* Inner loop uses 57 flops */
 854         }
 855
 856         /* End of innermost loop */
 857
 858         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 859                                                  f+i_coord_offset,fshift+i_shift_offset);
 860
 861         /* Increment number of inner iterations */
 862         inneriter                  += j_index_end - j_index_start;
 863
 864         /* Outer loop uses 6 flops */
 865     }
 866
 867     /* Increment number of outer iterations */
 868     outeriter        += nri;
 869
 870     /* Update outer/inner flops */
 871
 872     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*57);
 873 }