src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecGB_VdwLJ_GeomP1P1_sse4_1_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 sse4_1_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_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_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_sse4_1_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           = sse41_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_round_ps(rt, _MM_FROUND_FLOOR));
 243             gbitab           = _mm_slli_epi32(gbitab,2);
 244             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 245             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 246             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 247             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 248             _MM_TRANSPOSE4_PS(Y,F,G,H);
 249             Heps             = _mm_mul_ps(gbeps,H);
 250             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 251             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 252             vgb              = _mm_mul_ps(gbqqfactor,VV);
 253
 254             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 255             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 256             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 257             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 258             fjptrA           = dvda+jnrA;
 259             fjptrB           = dvda+jnrB;
 260             fjptrC           = dvda+jnrC;
 261             fjptrD           = dvda+jnrD;
 262             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 263             velec            = _mm_mul_ps(qq00,rinv00);
 264             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 265
 266             /* LENNARD-JONES DISPERSION/REPULSION */
 267
 268             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 269             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 270             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 271             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 272             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velecsum         = _mm_add_ps(velecsum,velec);
 276             vgbsum           = _mm_add_ps(vgbsum,vgb);
 277             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 278
 279             fscal            = _mm_add_ps(felec,fvdw);
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm_mul_ps(fscal,dx00);
 283             ty               = _mm_mul_ps(fscal,dy00);
 284             tz               = _mm_mul_ps(fscal,dz00);
 285
 286             /* Update vectorial force */
 287             fix0             = _mm_add_ps(fix0,tx);
 288             fiy0             = _mm_add_ps(fiy0,ty);
 289             fiz0             = _mm_add_ps(fiz0,tz);
 290
 291             fjptrA             = f+j_coord_offsetA;
 292             fjptrB             = f+j_coord_offsetB;
 293             fjptrC             = f+j_coord_offsetC;
 294             fjptrD             = f+j_coord_offsetD;
 295             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 296
 297             /* Inner loop uses 71 flops */
 298         }
 299
 300         if(jidx<j_index_end)
 301         {
 302
 303             /* Get j neighbor index, and coordinate index */
 304             jnrlistA         = jjnr[jidx];
 305             jnrlistB         = jjnr[jidx+1];
 306             jnrlistC         = jjnr[jidx+2];
 307             jnrlistD         = jjnr[jidx+3];
 308             /* Sign of each element will be negative for non-real atoms.
 309              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 310              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 311              */
 312             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 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             j_coord_offsetA  = DIM*jnrA;
 318             j_coord_offsetB  = DIM*jnrB;
 319             j_coord_offsetC  = DIM*jnrC;
 320             j_coord_offsetD  = DIM*jnrD;
 321
 322             /* load j atom coordinates */
 323             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 324                                               x+j_coord_offsetC,x+j_coord_offsetD,
 325                                               &jx0,&jy0,&jz0);
 326
 327             /* Calculate displacement vector */
 328             dx00             = _mm_sub_ps(ix0,jx0);
 329             dy00             = _mm_sub_ps(iy0,jy0);
 330             dz00             = _mm_sub_ps(iz0,jz0);
 331
 332             /* Calculate squared distance and things based on it */
 333             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 334
 335             rinv00           = sse41_invsqrt_f(rsq00);
 336
 337             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 338
 339             /* Load parameters for j particles */
 340             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 341                                                               charge+jnrC+0,charge+jnrD+0);
 342             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 343                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 344             vdwjidx0A        = 2*vdwtype[jnrA+0];
 345             vdwjidx0B        = 2*vdwtype[jnrB+0];
 346             vdwjidx0C        = 2*vdwtype[jnrC+0];
 347             vdwjidx0D        = 2*vdwtype[jnrD+0];
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             r00              = _mm_mul_ps(rsq00,rinv00);
 354             r00              = _mm_andnot_ps(dummy_mask,r00);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq00             = _mm_mul_ps(iq0,jq0);
 358             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 359                                          vdwparam+vdwioffset0+vdwjidx0B,
 360                                          vdwparam+vdwioffset0+vdwjidx0C,
 361                                          vdwparam+vdwioffset0+vdwjidx0D,
 362                                          &c6_00,&c12_00);
 363
 364             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 365             isaprod          = _mm_mul_ps(isai0,isaj0);
 366             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 367             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 368
 369             /* Calculate generalized born table index - this is a separate table from the normal one,
 370              * but we use the same procedure by multiplying r with scale and truncating to integer.
 371              */
 372             rt               = _mm_mul_ps(r00,gbscale);
 373             gbitab           = _mm_cvttps_epi32(rt);
 374             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 375             gbitab           = _mm_slli_epi32(gbitab,2);
 376             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 377             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 378             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 379             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 380             _MM_TRANSPOSE4_PS(Y,F,G,H);
 381             Heps             = _mm_mul_ps(gbeps,H);
 382             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 383             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 384             vgb              = _mm_mul_ps(gbqqfactor,VV);
 385
 386             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 387             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 388             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 389             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 390             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 391             /* 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. */
 392             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 393             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 394             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 395             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 396             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 397             velec            = _mm_mul_ps(qq00,rinv00);
 398             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 399
 400             /* LENNARD-JONES DISPERSION/REPULSION */
 401
 402             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 403             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 404             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 405             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 406             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velec            = _mm_andnot_ps(dummy_mask,velec);
 410             velecsum         = _mm_add_ps(velecsum,velec);
 411             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 412             vgbsum           = _mm_add_ps(vgbsum,vgb);
 413             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 414             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 415
 416             fscal            = _mm_add_ps(felec,fvdw);
 417
 418             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 419
 420             /* Calculate temporary vectorial force */
 421             tx               = _mm_mul_ps(fscal,dx00);
 422             ty               = _mm_mul_ps(fscal,dy00);
 423             tz               = _mm_mul_ps(fscal,dz00);
 424
 425             /* Update vectorial force */
 426             fix0             = _mm_add_ps(fix0,tx);
 427             fiy0             = _mm_add_ps(fiy0,ty);
 428             fiz0             = _mm_add_ps(fiz0,tz);
 429
 430             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 431             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 432             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 433             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 434             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 435
 436             /* Inner loop uses 72 flops */
 437         }
 438
 439         /* End of innermost loop */
 440
 441         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 442                                               f+i_coord_offset,fshift+i_shift_offset);
 443
 444         ggid                        = gid[iidx];
 445         /* Update potential energies */
 446         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 447         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 448         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 449         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 450         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 451
 452         /* Increment number of inner iterations */
 453         inneriter                  += j_index_end - j_index_start;
 454
 455         /* Outer loop uses 10 flops */
 456     }
 457
 458     /* Increment number of outer iterations */
 459     outeriter        += nri;
 460
 461     /* Update outer/inner flops */
 462
 463     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*72);
 464 }
 465 /*
 466  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_single
 467  * Electrostatics interaction: GeneralizedBorn
 468  * VdW interaction:            LennardJones
 469  * Geometry:                   Particle-Particle
 470  * Calculate force/pot:        Force
 471  */
 472 void
 473 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_single
 474                     (t_nblist                    * gmx_restrict       nlist,
 475                      rvec                        * gmx_restrict          xx,
 476                      rvec                        * gmx_restrict          ff,
 477                      struct t_forcerec           * gmx_restrict          fr,
 478                      t_mdatoms                   * gmx_restrict     mdatoms,
 479                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 480                      t_nrnb                      * gmx_restrict        nrnb)
 481 {
 482     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 483      * just 0 for non-waters.
 484      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 485      * jnr indices corresponding to data put in the four positions in the SIMD register.
 486      */
 487     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 488     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 489     int              jnrA,jnrB,jnrC,jnrD;
 490     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 491     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 492     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 493     real             rcutoff_scalar;
 494     real             *shiftvec,*fshift,*x,*f;
 495     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 496     real             scratch[4*DIM];
 497     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 498     int              vdwioffset0;
 499     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 500     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 501     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 502     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 503     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 504     real             *charge;
 505     __m128i          gbitab;
 506     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 507     __m128           minushalf = _mm_set1_ps(-0.5);
 508     real             *invsqrta,*dvda,*gbtab;
 509     int              nvdwtype;
 510     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 511     int              *vdwtype;
 512     real             *vdwparam;
 513     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 514     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 515     __m128i          vfitab;
 516     __m128i          ifour       = _mm_set1_epi32(4);
 517     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 518     real             *vftab;
 519     __m128           dummy_mask,cutoff_mask;
 520     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 521     __m128           one     = _mm_set1_ps(1.0);
 522     __m128           two     = _mm_set1_ps(2.0);
 523     x                = xx[0];
 524     f                = ff[0];
 525
 526     nri              = nlist->nri;
 527     iinr             = nlist->iinr;
 528     jindex           = nlist->jindex;
 529     jjnr             = nlist->jjnr;
 530     shiftidx         = nlist->shift;
 531     gid              = nlist->gid;
 532     shiftvec         = fr->shift_vec[0];
 533     fshift           = fr->fshift[0];
 534     facel            = _mm_set1_ps(fr->ic->epsfac);
 535     charge           = mdatoms->chargeA;
 536     nvdwtype         = fr->ntype;
 537     vdwparam         = fr->nbfp;
 538     vdwtype          = mdatoms->typeA;
 539
 540     invsqrta         = fr->invsqrta;
 541     dvda             = fr->dvda;
 542     gbtabscale       = _mm_set1_ps(fr->gbtab->scale);
 543     gbtab            = fr->gbtab->data;
 544     gbinvepsdiff     = _mm_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 545
 546     /* Avoid stupid compiler warnings */
 547     jnrA = jnrB = jnrC = jnrD = 0;
 548     j_coord_offsetA = 0;
 549     j_coord_offsetB = 0;
 550     j_coord_offsetC = 0;
 551     j_coord_offsetD = 0;
 552
 553     outeriter        = 0;
 554     inneriter        = 0;
 555
 556     for(iidx=0;iidx<4*DIM;iidx++)
 557     {
 558         scratch[iidx] = 0.0;
 559     }
 560
 561     /* Start outer loop over neighborlists */
 562     for(iidx=0; iidx<nri; iidx++)
 563     {
 564         /* Load shift vector for this list */
 565         i_shift_offset   = DIM*shiftidx[iidx];
 566
 567         /* Load limits for loop over neighbors */
 568         j_index_start    = jindex[iidx];
 569         j_index_end      = jindex[iidx+1];
 570
 571         /* Get outer coordinate index */
 572         inr              = iinr[iidx];
 573         i_coord_offset   = DIM*inr;
 574
 575         /* Load i particle coords and add shift vector */
 576         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 577
 578         fix0             = _mm_setzero_ps();
 579         fiy0             = _mm_setzero_ps();
 580         fiz0             = _mm_setzero_ps();
 581
 582         /* Load parameters for i particles */
 583         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 584         isai0            = _mm_load1_ps(invsqrta+inr+0);
 585         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 586
 587         dvdasum          = _mm_setzero_ps();
 588
 589         /* Start inner kernel loop */
 590         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 591         {
 592
 593             /* Get j neighbor index, and coordinate index */
 594             jnrA             = jjnr[jidx];
 595             jnrB             = jjnr[jidx+1];
 596             jnrC             = jjnr[jidx+2];
 597             jnrD             = jjnr[jidx+3];
 598             j_coord_offsetA  = DIM*jnrA;
 599             j_coord_offsetB  = DIM*jnrB;
 600             j_coord_offsetC  = DIM*jnrC;
 601             j_coord_offsetD  = DIM*jnrD;
 602
 603             /* load j atom coordinates */
 604             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 605                                               x+j_coord_offsetC,x+j_coord_offsetD,
 606                                               &jx0,&jy0,&jz0);
 607
 608             /* Calculate displacement vector */
 609             dx00             = _mm_sub_ps(ix0,jx0);
 610             dy00             = _mm_sub_ps(iy0,jy0);
 611             dz00             = _mm_sub_ps(iz0,jz0);
 612
 613             /* Calculate squared distance and things based on it */
 614             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 615
 616             rinv00           = sse41_invsqrt_f(rsq00);
 617
 618             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 619
 620             /* Load parameters for j particles */
 621             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 622                                                               charge+jnrC+0,charge+jnrD+0);
 623             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 624                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 625             vdwjidx0A        = 2*vdwtype[jnrA+0];
 626             vdwjidx0B        = 2*vdwtype[jnrB+0];
 627             vdwjidx0C        = 2*vdwtype[jnrC+0];
 628             vdwjidx0D        = 2*vdwtype[jnrD+0];
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             r00              = _mm_mul_ps(rsq00,rinv00);
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq00             = _mm_mul_ps(iq0,jq0);
 638             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 639                                          vdwparam+vdwioffset0+vdwjidx0B,
 640                                          vdwparam+vdwioffset0+vdwjidx0C,
 641                                          vdwparam+vdwioffset0+vdwjidx0D,
 642                                          &c6_00,&c12_00);
 643
 644             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 645             isaprod          = _mm_mul_ps(isai0,isaj0);
 646             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 647             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 648
 649             /* Calculate generalized born table index - this is a separate table from the normal one,
 650              * but we use the same procedure by multiplying r with scale and truncating to integer.
 651              */
 652             rt               = _mm_mul_ps(r00,gbscale);
 653             gbitab           = _mm_cvttps_epi32(rt);
 654             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 655             gbitab           = _mm_slli_epi32(gbitab,2);
 656             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 657             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 658             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 659             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 660             _MM_TRANSPOSE4_PS(Y,F,G,H);
 661             Heps             = _mm_mul_ps(gbeps,H);
 662             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 663             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 664             vgb              = _mm_mul_ps(gbqqfactor,VV);
 665
 666             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 667             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 668             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 669             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 670             fjptrA           = dvda+jnrA;
 671             fjptrB           = dvda+jnrB;
 672             fjptrC           = dvda+jnrC;
 673             fjptrD           = dvda+jnrD;
 674             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 675             velec            = _mm_mul_ps(qq00,rinv00);
 676             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 677
 678             /* LENNARD-JONES DISPERSION/REPULSION */
 679
 680             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 681             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 682
 683             fscal            = _mm_add_ps(felec,fvdw);
 684
 685             /* Calculate temporary vectorial force */
 686             tx               = _mm_mul_ps(fscal,dx00);
 687             ty               = _mm_mul_ps(fscal,dy00);
 688             tz               = _mm_mul_ps(fscal,dz00);
 689
 690             /* Update vectorial force */
 691             fix0             = _mm_add_ps(fix0,tx);
 692             fiy0             = _mm_add_ps(fiy0,ty);
 693             fiz0             = _mm_add_ps(fiz0,tz);
 694
 695             fjptrA             = f+j_coord_offsetA;
 696             fjptrB             = f+j_coord_offsetB;
 697             fjptrC             = f+j_coord_offsetC;
 698             fjptrD             = f+j_coord_offsetD;
 699             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 700
 701             /* Inner loop uses 64 flops */
 702         }
 703
 704         if(jidx<j_index_end)
 705         {
 706
 707             /* Get j neighbor index, and coordinate index */
 708             jnrlistA         = jjnr[jidx];
 709             jnrlistB         = jjnr[jidx+1];
 710             jnrlistC         = jjnr[jidx+2];
 711             jnrlistD         = jjnr[jidx+3];
 712             /* Sign of each element will be negative for non-real atoms.
 713              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 714              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 715              */
 716             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 717             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 718             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 719             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 720             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 721             j_coord_offsetA  = DIM*jnrA;
 722             j_coord_offsetB  = DIM*jnrB;
 723             j_coord_offsetC  = DIM*jnrC;
 724             j_coord_offsetD  = DIM*jnrD;
 725
 726             /* load j atom coordinates */
 727             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 728                                               x+j_coord_offsetC,x+j_coord_offsetD,
 729                                               &jx0,&jy0,&jz0);
 730
 731             /* Calculate displacement vector */
 732             dx00             = _mm_sub_ps(ix0,jx0);
 733             dy00             = _mm_sub_ps(iy0,jy0);
 734             dz00             = _mm_sub_ps(iz0,jz0);
 735
 736             /* Calculate squared distance and things based on it */
 737             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 738
 739             rinv00           = sse41_invsqrt_f(rsq00);
 740
 741             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 742
 743             /* Load parameters for j particles */
 744             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 745                                                               charge+jnrC+0,charge+jnrD+0);
 746             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 747                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 748             vdwjidx0A        = 2*vdwtype[jnrA+0];
 749             vdwjidx0B        = 2*vdwtype[jnrB+0];
 750             vdwjidx0C        = 2*vdwtype[jnrC+0];
 751             vdwjidx0D        = 2*vdwtype[jnrD+0];
 752
 753             /**************************
 754              * CALCULATE INTERACTIONS *
 755              **************************/
 756
 757             r00              = _mm_mul_ps(rsq00,rinv00);
 758             r00              = _mm_andnot_ps(dummy_mask,r00);
 759
 760             /* Compute parameters for interactions between i and j atoms */
 761             qq00             = _mm_mul_ps(iq0,jq0);
 762             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 763                                          vdwparam+vdwioffset0+vdwjidx0B,
 764                                          vdwparam+vdwioffset0+vdwjidx0C,
 765                                          vdwparam+vdwioffset0+vdwjidx0D,
 766                                          &c6_00,&c12_00);
 767
 768             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 769             isaprod          = _mm_mul_ps(isai0,isaj0);
 770             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 771             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 772
 773             /* Calculate generalized born table index - this is a separate table from the normal one,
 774              * but we use the same procedure by multiplying r with scale and truncating to integer.
 775              */
 776             rt               = _mm_mul_ps(r00,gbscale);
 777             gbitab           = _mm_cvttps_epi32(rt);
 778             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 779             gbitab           = _mm_slli_epi32(gbitab,2);
 780             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 781             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 782             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 783             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 784             _MM_TRANSPOSE4_PS(Y,F,G,H);
 785             Heps             = _mm_mul_ps(gbeps,H);
 786             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 787             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 788             vgb              = _mm_mul_ps(gbqqfactor,VV);
 789
 790             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 791             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 792             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 793             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 794             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 795             /* 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. */
 796             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 797             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 798             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 799             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 800             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 801             velec            = _mm_mul_ps(qq00,rinv00);
 802             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 803
 804             /* LENNARD-JONES DISPERSION/REPULSION */
 805
 806             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 807             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 808
 809             fscal            = _mm_add_ps(felec,fvdw);
 810
 811             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 812
 813             /* Calculate temporary vectorial force */
 814             tx               = _mm_mul_ps(fscal,dx00);
 815             ty               = _mm_mul_ps(fscal,dy00);
 816             tz               = _mm_mul_ps(fscal,dz00);
 817
 818             /* Update vectorial force */
 819             fix0             = _mm_add_ps(fix0,tx);
 820             fiy0             = _mm_add_ps(fiy0,ty);
 821             fiz0             = _mm_add_ps(fiz0,tz);
 822
 823             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 824             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 825             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 826             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 827             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 828
 829             /* Inner loop uses 65 flops */
 830         }
 831
 832         /* End of innermost loop */
 833
 834         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 835                                               f+i_coord_offset,fshift+i_shift_offset);
 836
 837         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 838         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 839
 840         /* Increment number of inner iterations */
 841         inneriter                  += j_index_end - j_index_start;
 842
 843         /* Outer loop uses 7 flops */
 844     }
 845
 846     /* Increment number of outer iterations */
 847     outeriter        += nri;
 848
 849     /* Update outer/inner flops */
 850
 851     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);
 852 }