src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecGB_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse2_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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          gbitab;
  90     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  91     __m128           minushalf = _mm_set1_ps(-0.5);
  92     real             *invsqrta,*dvda,*gbtab;
  93     int              nvdwtype;
  94     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  98     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m128           dummy_mask,cutoff_mask;
 104     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 105     __m128           one     = _mm_set1_ps(1.0);
 106     __m128           two     = _mm_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_ps(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     invsqrta         = fr->invsqrta;
 125     dvda             = fr->dvda;
 126     gbtabscale       = _mm_set1_ps(fr->gbtab->scale);
 127     gbtab            = fr->gbtab->data;
 128     gbinvepsdiff     = _mm_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165
 166         /* Load parameters for i particles */
 167         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 168         isai0            = _mm_load1_ps(invsqrta+inr+0);
 169         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_ps();
 173         vgbsum           = _mm_setzero_ps();
 174         vvdwsum          = _mm_setzero_ps();
 175         dvdasum          = _mm_setzero_ps();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               x+j_coord_offsetC,x+j_coord_offsetD,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_ps(ix0,jx0);
 198             dy00             = _mm_sub_ps(iy0,jy0);
 199             dz00             = _mm_sub_ps(iz0,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 203
 204             rinv00           = sse2_invsqrt_f(rsq00);
 205
 206             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 210                                                               charge+jnrC+0,charge+jnrD+0);
 211             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 212                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 213             vdwjidx0A        = 2*vdwtype[jnrA+0];
 214             vdwjidx0B        = 2*vdwtype[jnrB+0];
 215             vdwjidx0C        = 2*vdwtype[jnrC+0];
 216             vdwjidx0D        = 2*vdwtype[jnrD+0];
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             r00              = _mm_mul_ps(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm_mul_ps(iq0,jq0);
 226             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 227                                          vdwparam+vdwioffset0+vdwjidx0B,
 228                                          vdwparam+vdwioffset0+vdwjidx0C,
 229                                          vdwparam+vdwioffset0+vdwjidx0D,
 230                                          &c6_00,&c12_00);
 231
 232             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 233             isaprod          = _mm_mul_ps(isai0,isaj0);
 234             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 235             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 236
 237             /* Calculate generalized born table index - this is a separate table from the normal one,
 238              * but we use the same procedure by multiplying r with scale and truncating to integer.
 239              */
 240             rt               = _mm_mul_ps(r00,gbscale);
 241             gbitab           = _mm_cvttps_epi32(rt);
 242             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 243             gbitab           = _mm_slli_epi32(gbitab,2);
 244
 245             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 246             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 247             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 248             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 249             _MM_TRANSPOSE4_PS(Y,F,G,H);
 250             Heps             = _mm_mul_ps(gbeps,H);
 251             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 252             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 253             vgb              = _mm_mul_ps(gbqqfactor,VV);
 254
 255             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 256             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 257             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 258             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 259             fjptrA           = dvda+jnrA;
 260             fjptrB           = dvda+jnrB;
 261             fjptrC           = dvda+jnrC;
 262             fjptrD           = dvda+jnrD;
 263             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 264             velec            = _mm_mul_ps(qq00,rinv00);
 265             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 266
 267             /* LENNARD-JONES DISPERSION/REPULSION */
 268
 269             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 270             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 271             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 272             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 273             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velecsum         = _mm_add_ps(velecsum,velec);
 277             vgbsum           = _mm_add_ps(vgbsum,vgb);
 278             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 279
 280             fscal            = _mm_add_ps(felec,fvdw);
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_ps(fscal,dx00);
 284             ty               = _mm_mul_ps(fscal,dy00);
 285             tz               = _mm_mul_ps(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_ps(fix0,tx);
 289             fiy0             = _mm_add_ps(fiy0,ty);
 290             fiz0             = _mm_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             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 297
 298             /* Inner loop uses 71 flops */
 299         }
 300
 301         if(jidx<j_index_end)
 302         {
 303
 304             /* Get j neighbor index, and coordinate index */
 305             jnrlistA         = jjnr[jidx];
 306             jnrlistB         = jjnr[jidx+1];
 307             jnrlistC         = jjnr[jidx+2];
 308             jnrlistD         = jjnr[jidx+3];
 309             /* Sign of each element will be negative for non-real atoms.
 310              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 311              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 312              */
 313             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 314             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 315             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 316             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 317             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 318             j_coord_offsetA  = DIM*jnrA;
 319             j_coord_offsetB  = DIM*jnrB;
 320             j_coord_offsetC  = DIM*jnrC;
 321             j_coord_offsetD  = DIM*jnrD;
 322
 323             /* load j atom coordinates */
 324             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 325                                               x+j_coord_offsetC,x+j_coord_offsetD,
 326                                               &jx0,&jy0,&jz0);
 327
 328             /* Calculate displacement vector */
 329             dx00             = _mm_sub_ps(ix0,jx0);
 330             dy00             = _mm_sub_ps(iy0,jy0);
 331             dz00             = _mm_sub_ps(iz0,jz0);
 332
 333             /* Calculate squared distance and things based on it */
 334             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 335
 336             rinv00           = sse2_invsqrt_f(rsq00);
 337
 338             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 339
 340             /* Load parameters for j particles */
 341             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 342                                                               charge+jnrC+0,charge+jnrD+0);
 343             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 344                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 345             vdwjidx0A        = 2*vdwtype[jnrA+0];
 346             vdwjidx0B        = 2*vdwtype[jnrB+0];
 347             vdwjidx0C        = 2*vdwtype[jnrC+0];
 348             vdwjidx0D        = 2*vdwtype[jnrD+0];
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             r00              = _mm_mul_ps(rsq00,rinv00);
 355             r00              = _mm_andnot_ps(dummy_mask,r00);
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq00             = _mm_mul_ps(iq0,jq0);
 359             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 360                                          vdwparam+vdwioffset0+vdwjidx0B,
 361                                          vdwparam+vdwioffset0+vdwjidx0C,
 362                                          vdwparam+vdwioffset0+vdwjidx0D,
 363                                          &c6_00,&c12_00);
 364
 365             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 366             isaprod          = _mm_mul_ps(isai0,isaj0);
 367             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 368             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 369
 370             /* Calculate generalized born table index - this is a separate table from the normal one,
 371              * but we use the same procedure by multiplying r with scale and truncating to integer.
 372              */
 373             rt               = _mm_mul_ps(r00,gbscale);
 374             gbitab           = _mm_cvttps_epi32(rt);
 375             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 376             gbitab           = _mm_slli_epi32(gbitab,2);
 377
 378             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 379             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 380             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 381             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 382             _MM_TRANSPOSE4_PS(Y,F,G,H);
 383             Heps             = _mm_mul_ps(gbeps,H);
 384             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 385             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 386             vgb              = _mm_mul_ps(gbqqfactor,VV);
 387
 388             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 389             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 390             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 391             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 392             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 393             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 394             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 395             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 396             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 397             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 398             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 399             velec            = _mm_mul_ps(qq00,rinv00);
 400             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 401
 402             /* LENNARD-JONES DISPERSION/REPULSION */
 403
 404             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 405             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 406             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 407             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 408             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velec            = _mm_andnot_ps(dummy_mask,velec);
 412             velecsum         = _mm_add_ps(velecsum,velec);
 413             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 414             vgbsum           = _mm_add_ps(vgbsum,vgb);
 415             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 416             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 417
 418             fscal            = _mm_add_ps(felec,fvdw);
 419
 420             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 421
 422             /* Calculate temporary vectorial force */
 423             tx               = _mm_mul_ps(fscal,dx00);
 424             ty               = _mm_mul_ps(fscal,dy00);
 425             tz               = _mm_mul_ps(fscal,dz00);
 426
 427             /* Update vectorial force */
 428             fix0             = _mm_add_ps(fix0,tx);
 429             fiy0             = _mm_add_ps(fiy0,ty);
 430             fiz0             = _mm_add_ps(fiz0,tz);
 431
 432             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 433             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 434             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 435             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 436             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 437
 438             /* Inner loop uses 72 flops */
 439         }
 440
 441         /* End of innermost loop */
 442
 443         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 444                                               f+i_coord_offset,fshift+i_shift_offset);
 445
 446         ggid                        = gid[iidx];
 447         /* Update potential energies */
 448         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 449         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 450         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 451         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 452         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 453
 454         /* Increment number of inner iterations */
 455         inneriter                  += j_index_end - j_index_start;
 456
 457         /* Outer loop uses 10 flops */
 458     }
 459
 460     /* Increment number of outer iterations */
 461     outeriter        += nri;
 462
 463     /* Update outer/inner flops */
 464
 465     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*72);
 466 }
 467 /*
 468  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse2_single
 469  * Electrostatics interaction: GeneralizedBorn
 470  * VdW interaction:            LennardJones
 471  * Geometry:                   Particle-Particle
 472  * Calculate force/pot:        Force
 473  */
 474 void
 475 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse2_single
 476                     (t_nblist                    * gmx_restrict       nlist,
 477                      rvec                        * gmx_restrict          xx,
 478                      rvec                        * gmx_restrict          ff,
 479                      struct t_forcerec           * gmx_restrict          fr,
 480                      t_mdatoms                   * gmx_restrict     mdatoms,
 481                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 482                      t_nrnb                      * gmx_restrict        nrnb)
 483 {
 484     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 485      * just 0 for non-waters.
 486      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 487      * jnr indices corresponding to data put in the four positions in the SIMD register.
 488      */
 489     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 490     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 491     int              jnrA,jnrB,jnrC,jnrD;
 492     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 493     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 494     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 495     real             rcutoff_scalar;
 496     real             *shiftvec,*fshift,*x,*f;
 497     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 498     real             scratch[4*DIM];
 499     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 500     int              vdwioffset0;
 501     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 502     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 503     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 504     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 505     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 506     real             *charge;
 507     __m128i          gbitab;
 508     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 509     __m128           minushalf = _mm_set1_ps(-0.5);
 510     real             *invsqrta,*dvda,*gbtab;
 511     int              nvdwtype;
 512     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 513     int              *vdwtype;
 514     real             *vdwparam;
 515     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 516     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 517     __m128i          vfitab;
 518     __m128i          ifour       = _mm_set1_epi32(4);
 519     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 520     real             *vftab;
 521     __m128           dummy_mask,cutoff_mask;
 522     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 523     __m128           one     = _mm_set1_ps(1.0);
 524     __m128           two     = _mm_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     facel            = _mm_set1_ps(fr->ic->epsfac);
 537     charge           = mdatoms->chargeA;
 538     nvdwtype         = fr->ntype;
 539     vdwparam         = fr->nbfp;
 540     vdwtype          = mdatoms->typeA;
 541
 542     invsqrta         = fr->invsqrta;
 543     dvda             = fr->dvda;
 544     gbtabscale       = _mm_set1_ps(fr->gbtab->scale);
 545     gbtab            = fr->gbtab->data;
 546     gbinvepsdiff     = _mm_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 547
 548     /* Avoid stupid compiler warnings */
 549     jnrA = jnrB = jnrC = jnrD = 0;
 550     j_coord_offsetA = 0;
 551     j_coord_offsetB = 0;
 552     j_coord_offsetC = 0;
 553     j_coord_offsetD = 0;
 554
 555     outeriter        = 0;
 556     inneriter        = 0;
 557
 558     for(iidx=0;iidx<4*DIM;iidx++)
 559     {
 560         scratch[iidx] = 0.0;
 561     }
 562
 563     /* Start outer loop over neighborlists */
 564     for(iidx=0; iidx<nri; iidx++)
 565     {
 566         /* Load shift vector for this list */
 567         i_shift_offset   = DIM*shiftidx[iidx];
 568
 569         /* Load limits for loop over neighbors */
 570         j_index_start    = jindex[iidx];
 571         j_index_end      = jindex[iidx+1];
 572
 573         /* Get outer coordinate index */
 574         inr              = iinr[iidx];
 575         i_coord_offset   = DIM*inr;
 576
 577         /* Load i particle coords and add shift vector */
 578         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 579
 580         fix0             = _mm_setzero_ps();
 581         fiy0             = _mm_setzero_ps();
 582         fiz0             = _mm_setzero_ps();
 583
 584         /* Load parameters for i particles */
 585         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 586         isai0            = _mm_load1_ps(invsqrta+inr+0);
 587         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 588
 589         dvdasum          = _mm_setzero_ps();
 590
 591         /* Start inner kernel loop */
 592         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 593         {
 594
 595             /* Get j neighbor index, and coordinate index */
 596             jnrA             = jjnr[jidx];
 597             jnrB             = jjnr[jidx+1];
 598             jnrC             = jjnr[jidx+2];
 599             jnrD             = jjnr[jidx+3];
 600             j_coord_offsetA  = DIM*jnrA;
 601             j_coord_offsetB  = DIM*jnrB;
 602             j_coord_offsetC  = DIM*jnrC;
 603             j_coord_offsetD  = DIM*jnrD;
 604
 605             /* load j atom coordinates */
 606             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 607                                               x+j_coord_offsetC,x+j_coord_offsetD,
 608                                               &jx0,&jy0,&jz0);
 609
 610             /* Calculate displacement vector */
 611             dx00             = _mm_sub_ps(ix0,jx0);
 612             dy00             = _mm_sub_ps(iy0,jy0);
 613             dz00             = _mm_sub_ps(iz0,jz0);
 614
 615             /* Calculate squared distance and things based on it */
 616             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 617
 618             rinv00           = sse2_invsqrt_f(rsq00);
 619
 620             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 621
 622             /* Load parameters for j particles */
 623             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 624                                                               charge+jnrC+0,charge+jnrD+0);
 625             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 626                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 627             vdwjidx0A        = 2*vdwtype[jnrA+0];
 628             vdwjidx0B        = 2*vdwtype[jnrB+0];
 629             vdwjidx0C        = 2*vdwtype[jnrC+0];
 630             vdwjidx0D        = 2*vdwtype[jnrD+0];
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             r00              = _mm_mul_ps(rsq00,rinv00);
 637
 638             /* Compute parameters for interactions between i and j atoms */
 639             qq00             = _mm_mul_ps(iq0,jq0);
 640             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 641                                          vdwparam+vdwioffset0+vdwjidx0B,
 642                                          vdwparam+vdwioffset0+vdwjidx0C,
 643                                          vdwparam+vdwioffset0+vdwjidx0D,
 644                                          &c6_00,&c12_00);
 645
 646             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 647             isaprod          = _mm_mul_ps(isai0,isaj0);
 648             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 649             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 650
 651             /* Calculate generalized born table index - this is a separate table from the normal one,
 652              * but we use the same procedure by multiplying r with scale and truncating to integer.
 653              */
 654             rt               = _mm_mul_ps(r00,gbscale);
 655             gbitab           = _mm_cvttps_epi32(rt);
 656             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 657             gbitab           = _mm_slli_epi32(gbitab,2);
 658
 659             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 660             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 661             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 662             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 663             _MM_TRANSPOSE4_PS(Y,F,G,H);
 664             Heps             = _mm_mul_ps(gbeps,H);
 665             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 666             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 667             vgb              = _mm_mul_ps(gbqqfactor,VV);
 668
 669             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 670             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 671             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 672             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 673             fjptrA           = dvda+jnrA;
 674             fjptrB           = dvda+jnrB;
 675             fjptrC           = dvda+jnrC;
 676             fjptrD           = dvda+jnrD;
 677             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 678             velec            = _mm_mul_ps(qq00,rinv00);
 679             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 680
 681             /* LENNARD-JONES DISPERSION/REPULSION */
 682
 683             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 684             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 685
 686             fscal            = _mm_add_ps(felec,fvdw);
 687
 688             /* Calculate temporary vectorial force */
 689             tx               = _mm_mul_ps(fscal,dx00);
 690             ty               = _mm_mul_ps(fscal,dy00);
 691             tz               = _mm_mul_ps(fscal,dz00);
 692
 693             /* Update vectorial force */
 694             fix0             = _mm_add_ps(fix0,tx);
 695             fiy0             = _mm_add_ps(fiy0,ty);
 696             fiz0             = _mm_add_ps(fiz0,tz);
 697
 698             fjptrA             = f+j_coord_offsetA;
 699             fjptrB             = f+j_coord_offsetB;
 700             fjptrC             = f+j_coord_offsetC;
 701             fjptrD             = f+j_coord_offsetD;
 702             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 703
 704             /* Inner loop uses 64 flops */
 705         }
 706
 707         if(jidx<j_index_end)
 708         {
 709
 710             /* Get j neighbor index, and coordinate index */
 711             jnrlistA         = jjnr[jidx];
 712             jnrlistB         = jjnr[jidx+1];
 713             jnrlistC         = jjnr[jidx+2];
 714             jnrlistD         = jjnr[jidx+3];
 715             /* Sign of each element will be negative for non-real atoms.
 716              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 717              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 718              */
 719             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 720             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 721             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 722             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 723             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 724             j_coord_offsetA  = DIM*jnrA;
 725             j_coord_offsetB  = DIM*jnrB;
 726             j_coord_offsetC  = DIM*jnrC;
 727             j_coord_offsetD  = DIM*jnrD;
 728
 729             /* load j atom coordinates */
 730             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 731                                               x+j_coord_offsetC,x+j_coord_offsetD,
 732                                               &jx0,&jy0,&jz0);
 733
 734             /* Calculate displacement vector */
 735             dx00             = _mm_sub_ps(ix0,jx0);
 736             dy00             = _mm_sub_ps(iy0,jy0);
 737             dz00             = _mm_sub_ps(iz0,jz0);
 738
 739             /* Calculate squared distance and things based on it */
 740             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 741
 742             rinv00           = sse2_invsqrt_f(rsq00);
 743
 744             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 745
 746             /* Load parameters for j particles */
 747             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 748                                                               charge+jnrC+0,charge+jnrD+0);
 749             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 750                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 751             vdwjidx0A        = 2*vdwtype[jnrA+0];
 752             vdwjidx0B        = 2*vdwtype[jnrB+0];
 753             vdwjidx0C        = 2*vdwtype[jnrC+0];
 754             vdwjidx0D        = 2*vdwtype[jnrD+0];
 755
 756             /**************************
 757              * CALCULATE INTERACTIONS *
 758              **************************/
 759
 760             r00              = _mm_mul_ps(rsq00,rinv00);
 761             r00              = _mm_andnot_ps(dummy_mask,r00);
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq00             = _mm_mul_ps(iq0,jq0);
 765             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 766                                          vdwparam+vdwioffset0+vdwjidx0B,
 767                                          vdwparam+vdwioffset0+vdwjidx0C,
 768                                          vdwparam+vdwioffset0+vdwjidx0D,
 769                                          &c6_00,&c12_00);
 770
 771             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 772             isaprod          = _mm_mul_ps(isai0,isaj0);
 773             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 774             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 775
 776             /* Calculate generalized born table index - this is a separate table from the normal one,
 777              * but we use the same procedure by multiplying r with scale and truncating to integer.
 778              */
 779             rt               = _mm_mul_ps(r00,gbscale);
 780             gbitab           = _mm_cvttps_epi32(rt);
 781             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 782             gbitab           = _mm_slli_epi32(gbitab,2);
 783
 784             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 785             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 786             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 787             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 788             _MM_TRANSPOSE4_PS(Y,F,G,H);
 789             Heps             = _mm_mul_ps(gbeps,H);
 790             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 791             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 792             vgb              = _mm_mul_ps(gbqqfactor,VV);
 793
 794             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 795             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 796             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 797             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 798             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 799             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 800             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 801             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 802             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 803             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 804             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 805             velec            = _mm_mul_ps(qq00,rinv00);
 806             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 807
 808             /* LENNARD-JONES DISPERSION/REPULSION */
 809
 810             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 811             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 812
 813             fscal            = _mm_add_ps(felec,fvdw);
 814
 815             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 816
 817             /* Calculate temporary vectorial force */
 818             tx               = _mm_mul_ps(fscal,dx00);
 819             ty               = _mm_mul_ps(fscal,dy00);
 820             tz               = _mm_mul_ps(fscal,dz00);
 821
 822             /* Update vectorial force */
 823             fix0             = _mm_add_ps(fix0,tx);
 824             fiy0             = _mm_add_ps(fiy0,ty);
 825             fiz0             = _mm_add_ps(fiz0,tz);
 826
 827             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 828             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 829             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 830             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 831             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 832
 833             /* Inner loop uses 65 flops */
 834         }
 835
 836         /* End of innermost loop */
 837
 838         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 839                                               f+i_coord_offset,fshift+i_shift_offset);
 840
 841         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 842         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 843
 844         /* Increment number of inner iterations */
 845         inneriter                  += j_index_end - j_index_start;
 846
 847         /* Outer loop uses 7 flops */
 848     }
 849
 850     /* Increment number of outer iterations */
 851     outeriter        += nri;
 852
 853     /* Update outer/inner flops */
 854
 855     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);
 856 }