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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_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 refer to j loop unrolling done with SSE, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128i          gbitab;
  93     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  94     __m128           minushalf = _mm_set1_ps(-0.5);
  95     real             *invsqrta,*dvda,*gbtab;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     invsqrta         = fr->invsqrta;
 128     dvda             = fr->dvda;
 129     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 130     gbtab            = fr->gbtab.data;
 131     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168
 169         /* Load parameters for i particles */
 170         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 171         isai0            = _mm_load1_ps(invsqrta+inr+0);
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vgbsum           = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178         dvdasum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               x+j_coord_offsetC,x+j_coord_offsetD,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_ps(ix0,jx0);
 201             dy00             = _mm_sub_ps(iy0,jy0);
 202             dz00             = _mm_sub_ps(iz0,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206
 207             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 208
 209             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 215                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_ps(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_ps(iq0,jq0);
 229             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,
 231                                          vdwparam+vdwioffset0+vdwjidx0C,
 232                                          vdwparam+vdwioffset0+vdwjidx0D,
 233                                          &c6_00,&c12_00);
 234
 235             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 236             isaprod          = _mm_mul_ps(isai0,isaj0);
 237             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 238             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 239
 240             /* Calculate generalized born table index - this is a separate table from the normal one,
 241              * but we use the same procedure by multiplying r with scale and truncating to integer.
 242              */
 243             rt               = _mm_mul_ps(r00,gbscale);
 244             gbitab           = _mm_cvttps_epi32(rt);
 245             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 246             gbitab           = _mm_slli_epi32(gbitab,2);
 247             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 248             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 249             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 250             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 251             _MM_TRANSPOSE4_PS(Y,F,G,H);
 252             Heps             = _mm_mul_ps(gbeps,H);
 253             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 254             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 255             vgb              = _mm_mul_ps(gbqqfactor,VV);
 256
 257             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 258             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 259             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 260             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 261             fjptrA           = dvda+jnrA;
 262             fjptrB           = dvda+jnrB;
 263             fjptrC           = dvda+jnrC;
 264             fjptrD           = dvda+jnrD;
 265             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 266             velec            = _mm_mul_ps(qq00,rinv00);
 267             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 268
 269             /* LENNARD-JONES DISPERSION/REPULSION */
 270
 271             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 272             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 273             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 274             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 275             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velecsum         = _mm_add_ps(velecsum,velec);
 279             vgbsum           = _mm_add_ps(vgbsum,vgb);
 280             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 281
 282             fscal            = _mm_add_ps(felec,fvdw);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_ps(fscal,dx00);
 286             ty               = _mm_mul_ps(fscal,dy00);
 287             tz               = _mm_mul_ps(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_add_ps(fix0,tx);
 291             fiy0             = _mm_add_ps(fiy0,ty);
 292             fiz0             = _mm_add_ps(fiz0,tz);
 293
 294             fjptrA             = f+j_coord_offsetA;
 295             fjptrB             = f+j_coord_offsetB;
 296             fjptrC             = f+j_coord_offsetC;
 297             fjptrD             = f+j_coord_offsetD;
 298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 299
 300             /* Inner loop uses 71 flops */
 301         }
 302
 303         if(jidx<j_index_end)
 304         {
 305
 306             /* Get j neighbor index, and coordinate index */
 307             jnrlistA         = jjnr[jidx];
 308             jnrlistB         = jjnr[jidx+1];
 309             jnrlistC         = jjnr[jidx+2];
 310             jnrlistD         = jjnr[jidx+3];
 311             /* Sign of each element will be negative for non-real atoms.
 312              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 313              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 314              */
 315             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 316             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 317             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 318             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 319             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 320             j_coord_offsetA  = DIM*jnrA;
 321             j_coord_offsetB  = DIM*jnrB;
 322             j_coord_offsetC  = DIM*jnrC;
 323             j_coord_offsetD  = DIM*jnrD;
 324
 325             /* load j atom coordinates */
 326             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 327                                               x+j_coord_offsetC,x+j_coord_offsetD,
 328                                               &jx0,&jy0,&jz0);
 329
 330             /* Calculate displacement vector */
 331             dx00             = _mm_sub_ps(ix0,jx0);
 332             dy00             = _mm_sub_ps(iy0,jy0);
 333             dz00             = _mm_sub_ps(iz0,jz0);
 334
 335             /* Calculate squared distance and things based on it */
 336             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 337
 338             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 339
 340             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 341
 342             /* Load parameters for j particles */
 343             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 344                                                               charge+jnrC+0,charge+jnrD+0);
 345             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 346                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 347             vdwjidx0A        = 2*vdwtype[jnrA+0];
 348             vdwjidx0B        = 2*vdwtype[jnrB+0];
 349             vdwjidx0C        = 2*vdwtype[jnrC+0];
 350             vdwjidx0D        = 2*vdwtype[jnrD+0];
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             r00              = _mm_mul_ps(rsq00,rinv00);
 357             r00              = _mm_andnot_ps(dummy_mask,r00);
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq00             = _mm_mul_ps(iq0,jq0);
 361             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 362                                          vdwparam+vdwioffset0+vdwjidx0B,
 363                                          vdwparam+vdwioffset0+vdwjidx0C,
 364                                          vdwparam+vdwioffset0+vdwjidx0D,
 365                                          &c6_00,&c12_00);
 366
 367             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 368             isaprod          = _mm_mul_ps(isai0,isaj0);
 369             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 370             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 371
 372             /* Calculate generalized born table index - this is a separate table from the normal one,
 373              * but we use the same procedure by multiplying r with scale and truncating to integer.
 374              */
 375             rt               = _mm_mul_ps(r00,gbscale);
 376             gbitab           = _mm_cvttps_epi32(rt);
 377             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 378             gbitab           = _mm_slli_epi32(gbitab,2);
 379             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 380             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 381             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 382             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 383             _MM_TRANSPOSE4_PS(Y,F,G,H);
 384             Heps             = _mm_mul_ps(gbeps,H);
 385             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 386             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 387             vgb              = _mm_mul_ps(gbqqfactor,VV);
 388
 389             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 390             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 391             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 392             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 393             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 394             /* 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. */
 395             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 396             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 397             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 398             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 399             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 400             velec            = _mm_mul_ps(qq00,rinv00);
 401             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 402
 403             /* LENNARD-JONES DISPERSION/REPULSION */
 404
 405             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 406             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 407             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 408             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 409             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm_andnot_ps(dummy_mask,velec);
 413             velecsum         = _mm_add_ps(velecsum,velec);
 414             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 415             vgbsum           = _mm_add_ps(vgbsum,vgb);
 416             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 417             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 418
 419             fscal            = _mm_add_ps(felec,fvdw);
 420
 421             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 422
 423             /* Calculate temporary vectorial force */
 424             tx               = _mm_mul_ps(fscal,dx00);
 425             ty               = _mm_mul_ps(fscal,dy00);
 426             tz               = _mm_mul_ps(fscal,dz00);
 427
 428             /* Update vectorial force */
 429             fix0             = _mm_add_ps(fix0,tx);
 430             fiy0             = _mm_add_ps(fiy0,ty);
 431             fiz0             = _mm_add_ps(fiz0,tz);
 432
 433             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 434             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 435             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 436             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 437             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 438
 439             /* Inner loop uses 72 flops */
 440         }
 441
 442         /* End of innermost loop */
 443
 444         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 445                                               f+i_coord_offset,fshift+i_shift_offset);
 446
 447         ggid                        = gid[iidx];
 448         /* Update potential energies */
 449         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 450         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 451         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 452         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 453         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 454
 455         /* Increment number of inner iterations */
 456         inneriter                  += j_index_end - j_index_start;
 457
 458         /* Outer loop uses 10 flops */
 459     }
 460
 461     /* Increment number of outer iterations */
 462     outeriter        += nri;
 463
 464     /* Update outer/inner flops */
 465
 466     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*72);
 467 }
 468 /*
 469  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_single
 470  * Electrostatics interaction: GeneralizedBorn
 471  * VdW interaction:            LennardJones
 472  * Geometry:                   Particle-Particle
 473  * Calculate force/pot:        Force
 474  */
 475 void
 476 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_single
 477                     (t_nblist                    * gmx_restrict       nlist,
 478                      rvec                        * gmx_restrict          xx,
 479                      rvec                        * gmx_restrict          ff,
 480                      t_forcerec                  * gmx_restrict          fr,
 481                      t_mdatoms                   * gmx_restrict     mdatoms,
 482                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 483                      t_nrnb                      * gmx_restrict        nrnb)
 484 {
 485     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 486      * just 0 for non-waters.
 487      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 488      * jnr indices corresponding to data put in the four positions in the SIMD register.
 489      */
 490     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 491     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 492     int              jnrA,jnrB,jnrC,jnrD;
 493     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 494     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 495     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 496     real             rcutoff_scalar;
 497     real             *shiftvec,*fshift,*x,*f;
 498     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 499     real             scratch[4*DIM];
 500     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 501     int              vdwioffset0;
 502     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 503     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 504     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 505     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 506     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 507     real             *charge;
 508     __m128i          gbitab;
 509     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 510     __m128           minushalf = _mm_set1_ps(-0.5);
 511     real             *invsqrta,*dvda,*gbtab;
 512     int              nvdwtype;
 513     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 514     int              *vdwtype;
 515     real             *vdwparam;
 516     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 517     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 518     __m128i          vfitab;
 519     __m128i          ifour       = _mm_set1_epi32(4);
 520     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 521     real             *vftab;
 522     __m128           dummy_mask,cutoff_mask;
 523     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 524     __m128           one     = _mm_set1_ps(1.0);
 525     __m128           two     = _mm_set1_ps(2.0);
 526     x                = xx[0];
 527     f                = ff[0];
 528
 529     nri              = nlist->nri;
 530     iinr             = nlist->iinr;
 531     jindex           = nlist->jindex;
 532     jjnr             = nlist->jjnr;
 533     shiftidx         = nlist->shift;
 534     gid              = nlist->gid;
 535     shiftvec         = fr->shift_vec[0];
 536     fshift           = fr->fshift[0];
 537     facel            = _mm_set1_ps(fr->epsfac);
 538     charge           = mdatoms->chargeA;
 539     nvdwtype         = fr->ntype;
 540     vdwparam         = fr->nbfp;
 541     vdwtype          = mdatoms->typeA;
 542
 543     invsqrta         = fr->invsqrta;
 544     dvda             = fr->dvda;
 545     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 546     gbtab            = fr->gbtab.data;
 547     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 548
 549     /* Avoid stupid compiler warnings */
 550     jnrA = jnrB = jnrC = jnrD = 0;
 551     j_coord_offsetA = 0;
 552     j_coord_offsetB = 0;
 553     j_coord_offsetC = 0;
 554     j_coord_offsetD = 0;
 555
 556     outeriter        = 0;
 557     inneriter        = 0;
 558
 559     for(iidx=0;iidx<4*DIM;iidx++)
 560     {
 561         scratch[iidx] = 0.0;
 562     }
 563
 564     /* Start outer loop over neighborlists */
 565     for(iidx=0; iidx<nri; iidx++)
 566     {
 567         /* Load shift vector for this list */
 568         i_shift_offset   = DIM*shiftidx[iidx];
 569
 570         /* Load limits for loop over neighbors */
 571         j_index_start    = jindex[iidx];
 572         j_index_end      = jindex[iidx+1];
 573
 574         /* Get outer coordinate index */
 575         inr              = iinr[iidx];
 576         i_coord_offset   = DIM*inr;
 577
 578         /* Load i particle coords and add shift vector */
 579         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 580
 581         fix0             = _mm_setzero_ps();
 582         fiy0             = _mm_setzero_ps();
 583         fiz0             = _mm_setzero_ps();
 584
 585         /* Load parameters for i particles */
 586         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 587         isai0            = _mm_load1_ps(invsqrta+inr+0);
 588         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 589
 590         dvdasum          = _mm_setzero_ps();
 591
 592         /* Start inner kernel loop */
 593         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 594         {
 595
 596             /* Get j neighbor index, and coordinate index */
 597             jnrA             = jjnr[jidx];
 598             jnrB             = jjnr[jidx+1];
 599             jnrC             = jjnr[jidx+2];
 600             jnrD             = jjnr[jidx+3];
 601             j_coord_offsetA  = DIM*jnrA;
 602             j_coord_offsetB  = DIM*jnrB;
 603             j_coord_offsetC  = DIM*jnrC;
 604             j_coord_offsetD  = DIM*jnrD;
 605
 606             /* load j atom coordinates */
 607             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 608                                               x+j_coord_offsetC,x+j_coord_offsetD,
 609                                               &jx0,&jy0,&jz0);
 610
 611             /* Calculate displacement vector */
 612             dx00             = _mm_sub_ps(ix0,jx0);
 613             dy00             = _mm_sub_ps(iy0,jy0);
 614             dz00             = _mm_sub_ps(iz0,jz0);
 615
 616             /* Calculate squared distance and things based on it */
 617             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 618
 619             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 620
 621             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 622
 623             /* Load parameters for j particles */
 624             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 625                                                               charge+jnrC+0,charge+jnrD+0);
 626             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 627                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 628             vdwjidx0A        = 2*vdwtype[jnrA+0];
 629             vdwjidx0B        = 2*vdwtype[jnrB+0];
 630             vdwjidx0C        = 2*vdwtype[jnrC+0];
 631             vdwjidx0D        = 2*vdwtype[jnrD+0];
 632
 633             /**************************
 634              * CALCULATE INTERACTIONS *
 635              **************************/
 636
 637             r00              = _mm_mul_ps(rsq00,rinv00);
 638
 639             /* Compute parameters for interactions between i and j atoms */
 640             qq00             = _mm_mul_ps(iq0,jq0);
 641             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 642                                          vdwparam+vdwioffset0+vdwjidx0B,
 643                                          vdwparam+vdwioffset0+vdwjidx0C,
 644                                          vdwparam+vdwioffset0+vdwjidx0D,
 645                                          &c6_00,&c12_00);
 646
 647             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 648             isaprod          = _mm_mul_ps(isai0,isaj0);
 649             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 650             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 651
 652             /* Calculate generalized born table index - this is a separate table from the normal one,
 653              * but we use the same procedure by multiplying r with scale and truncating to integer.
 654              */
 655             rt               = _mm_mul_ps(r00,gbscale);
 656             gbitab           = _mm_cvttps_epi32(rt);
 657             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 658             gbitab           = _mm_slli_epi32(gbitab,2);
 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           = gmx_mm_invsqrt_ps(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_round_ps(rt, _MM_FROUND_FLOOR));
 782             gbitab           = _mm_slli_epi32(gbitab,2);
 783             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 784             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 785             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 786             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 787             _MM_TRANSPOSE4_PS(Y,F,G,H);
 788             Heps             = _mm_mul_ps(gbeps,H);
 789             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 790             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 791             vgb              = _mm_mul_ps(gbqqfactor,VV);
 792
 793             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 794             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 795             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 796             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 797             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 798             /* 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. */
 799             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 800             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 801             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 802             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 803             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 804             velec            = _mm_mul_ps(qq00,rinv00);
 805             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 806
 807             /* LENNARD-JONES DISPERSION/REPULSION */
 808
 809             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 810             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 811
 812             fscal            = _mm_add_ps(felec,fvdw);
 813
 814             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 815
 816             /* Calculate temporary vectorial force */
 817             tx               = _mm_mul_ps(fscal,dx00);
 818             ty               = _mm_mul_ps(fscal,dy00);
 819             tz               = _mm_mul_ps(fscal,dz00);
 820
 821             /* Update vectorial force */
 822             fix0             = _mm_add_ps(fix0,tx);
 823             fiy0             = _mm_add_ps(fiy0,ty);
 824             fiz0             = _mm_add_ps(fiz0,tz);
 825
 826             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 827             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 828             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 829             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 830             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 831
 832             /* Inner loop uses 65 flops */
 833         }
 834
 835         /* End of innermost loop */
 836
 837         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 838                                               f+i_coord_offset,fshift+i_shift_offset);
 839
 840         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 841         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 842
 843         /* Increment number of inner iterations */
 844         inneriter                  += j_index_end - j_index_start;
 845
 846         /* Outer loop uses 7 flops */
 847     }
 848
 849     /* Increment number of outer iterations */
 850     outeriter        += nri;
 851
 852     /* Update outer/inner flops */
 853
 854     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);
 855 }