src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecGB_VdwCSTab_GeomP1P1_sse2_double.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_double 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_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     __m128i          gbitab;
  87     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  88     __m128d          minushalf = _mm_set1_pd(-0.5);
  89     real             *invsqrta,*dvda,*gbtab;
  90     int              nvdwtype;
  91     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  95     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->ic->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     vftab            = kernel_data->table_vdw->data;
 122     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 123
 124     invsqrta         = fr->invsqrta;
 125     dvda             = fr->dvda;
 126     gbtabscale       = _mm_set1_pd(fr->gbtab->scale);
 127     gbtab            = fr->gbtab->data;
 128     gbinvepsdiff     = _mm_set1_pd((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 161         isai0            = _mm_load1_pd(invsqrta+inr+0);
 162         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166         vgbsum           = _mm_setzero_pd();
 167         vvdwsum          = _mm_setzero_pd();
 168         dvdasum          = _mm_setzero_pd();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179
 180             /* load j atom coordinates */
 181             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm_sub_pd(ix0,jx0);
 186             dy00             = _mm_sub_pd(iy0,jy0);
 187             dz00             = _mm_sub_pd(iz0,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 191
 192             rinv00           = sse2_invsqrt_d(rsq00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 196             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 197             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             r00              = _mm_mul_pd(rsq00,rinv00);
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm_mul_pd(iq0,jq0);
 208             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 210
 211             /* Calculate table index by multiplying r with table scale and truncate to integer */
 212             rt               = _mm_mul_pd(r00,vftabscale);
 213             vfitab           = _mm_cvttpd_epi32(rt);
 214             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 215             vfitab           = _mm_slli_epi32(vfitab,3);
 216
 217             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 218             isaprod          = _mm_mul_pd(isai0,isaj0);
 219             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 220             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 221
 222             /* Calculate generalized born table index - this is a separate table from the normal one,
 223              * but we use the same procedure by multiplying r with scale and truncating to integer.
 224              */
 225             rt               = _mm_mul_pd(r00,gbscale);
 226             gbitab           = _mm_cvttpd_epi32(rt);
 227             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 228             gbitab           = _mm_slli_epi32(gbitab,2);
 229
 230             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 231             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 232             GMX_MM_TRANSPOSE2_PD(Y,F);
 233             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 234             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 235             GMX_MM_TRANSPOSE2_PD(G,H);
 236             Heps             = _mm_mul_pd(gbeps,H);
 237             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 238             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 239             vgb              = _mm_mul_pd(gbqqfactor,VV);
 240
 241             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 242             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 243             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 244             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 245             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 246             velec            = _mm_mul_pd(qq00,rinv00);
 247             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 248
 249             /* CUBIC SPLINE TABLE DISPERSION */
 250             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 251             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 252             GMX_MM_TRANSPOSE2_PD(Y,F);
 253             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 254             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 255             GMX_MM_TRANSPOSE2_PD(G,H);
 256             Heps             = _mm_mul_pd(vfeps,H);
 257             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 258             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 259             vvdw6            = _mm_mul_pd(c6_00,VV);
 260             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 261             fvdw6            = _mm_mul_pd(c6_00,FF);
 262
 263             /* CUBIC SPLINE TABLE REPULSION */
 264             vfitab           = _mm_add_epi32(vfitab,ifour);
 265             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 266             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 267             GMX_MM_TRANSPOSE2_PD(Y,F);
 268             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 269             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 270             GMX_MM_TRANSPOSE2_PD(G,H);
 271             Heps             = _mm_mul_pd(vfeps,H);
 272             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 273             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 274             vvdw12           = _mm_mul_pd(c12_00,VV);
 275             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 276             fvdw12           = _mm_mul_pd(c12_00,FF);
 277             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 278             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm_add_pd(velecsum,velec);
 282             vgbsum           = _mm_add_pd(vgbsum,vgb);
 283             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 284
 285             fscal            = _mm_add_pd(felec,fvdw);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_pd(fscal,dx00);
 289             ty               = _mm_mul_pd(fscal,dy00);
 290             tz               = _mm_mul_pd(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_pd(fix0,tx);
 294             fiy0             = _mm_add_pd(fiy0,ty);
 295             fiz0             = _mm_add_pd(fiz0,tz);
 296
 297             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 298
 299             /* Inner loop uses 92 flops */
 300         }
 301
 302         if(jidx<j_index_end)
 303         {
 304
 305             jnrA             = jjnr[jidx];
 306             j_coord_offsetA  = DIM*jnrA;
 307
 308             /* load j atom coordinates */
 309             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 310                                               &jx0,&jy0,&jz0);
 311
 312             /* Calculate displacement vector */
 313             dx00             = _mm_sub_pd(ix0,jx0);
 314             dy00             = _mm_sub_pd(iy0,jy0);
 315             dz00             = _mm_sub_pd(iz0,jz0);
 316
 317             /* Calculate squared distance and things based on it */
 318             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 319
 320             rinv00           = sse2_invsqrt_d(rsq00);
 321
 322             /* Load parameters for j particles */
 323             jq0              = _mm_load_sd(charge+jnrA+0);
 324             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 325             vdwjidx0A        = 2*vdwtype[jnrA+0];
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             r00              = _mm_mul_pd(rsq00,rinv00);
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq00             = _mm_mul_pd(iq0,jq0);
 335             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 336
 337             /* Calculate table index by multiplying r with table scale and truncate to integer */
 338             rt               = _mm_mul_pd(r00,vftabscale);
 339             vfitab           = _mm_cvttpd_epi32(rt);
 340             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 341             vfitab           = _mm_slli_epi32(vfitab,3);
 342
 343             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 344             isaprod          = _mm_mul_pd(isai0,isaj0);
 345             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 346             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 347
 348             /* Calculate generalized born table index - this is a separate table from the normal one,
 349              * but we use the same procedure by multiplying r with scale and truncating to integer.
 350              */
 351             rt               = _mm_mul_pd(r00,gbscale);
 352             gbitab           = _mm_cvttpd_epi32(rt);
 353             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 354             gbitab           = _mm_slli_epi32(gbitab,2);
 355
 356             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 357             F                = _mm_setzero_pd();
 358             GMX_MM_TRANSPOSE2_PD(Y,F);
 359             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 360             H                = _mm_setzero_pd();
 361             GMX_MM_TRANSPOSE2_PD(G,H);
 362             Heps             = _mm_mul_pd(gbeps,H);
 363             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 364             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 365             vgb              = _mm_mul_pd(gbqqfactor,VV);
 366
 367             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 368             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 369             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 370             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 371             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 372             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 373             velec            = _mm_mul_pd(qq00,rinv00);
 374             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 375
 376             /* CUBIC SPLINE TABLE DISPERSION */
 377             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 378             F                = _mm_setzero_pd();
 379             GMX_MM_TRANSPOSE2_PD(Y,F);
 380             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 381             H                = _mm_setzero_pd();
 382             GMX_MM_TRANSPOSE2_PD(G,H);
 383             Heps             = _mm_mul_pd(vfeps,H);
 384             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 385             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 386             vvdw6            = _mm_mul_pd(c6_00,VV);
 387             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 388             fvdw6            = _mm_mul_pd(c6_00,FF);
 389
 390             /* CUBIC SPLINE TABLE REPULSION */
 391             vfitab           = _mm_add_epi32(vfitab,ifour);
 392             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 393             F                = _mm_setzero_pd();
 394             GMX_MM_TRANSPOSE2_PD(Y,F);
 395             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 396             H                = _mm_setzero_pd();
 397             GMX_MM_TRANSPOSE2_PD(G,H);
 398             Heps             = _mm_mul_pd(vfeps,H);
 399             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 400             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 401             vvdw12           = _mm_mul_pd(c12_00,VV);
 402             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 403             fvdw12           = _mm_mul_pd(c12_00,FF);
 404             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 405             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 409             velecsum         = _mm_add_pd(velecsum,velec);
 410             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 411             vgbsum           = _mm_add_pd(vgbsum,vgb);
 412             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 413             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 414
 415             fscal            = _mm_add_pd(felec,fvdw);
 416
 417             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm_mul_pd(fscal,dx00);
 421             ty               = _mm_mul_pd(fscal,dy00);
 422             tz               = _mm_mul_pd(fscal,dz00);
 423
 424             /* Update vectorial force */
 425             fix0             = _mm_add_pd(fix0,tx);
 426             fiy0             = _mm_add_pd(fiy0,ty);
 427             fiz0             = _mm_add_pd(fiz0,tz);
 428
 429             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 430
 431             /* Inner loop uses 92 flops */
 432         }
 433
 434         /* End of innermost loop */
 435
 436         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 437                                               f+i_coord_offset,fshift+i_shift_offset);
 438
 439         ggid                        = gid[iidx];
 440         /* Update potential energies */
 441         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 442         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 443         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 444         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 445         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 446
 447         /* Increment number of inner iterations */
 448         inneriter                  += j_index_end - j_index_start;
 449
 450         /* Outer loop uses 10 flops */
 451     }
 452
 453     /* Increment number of outer iterations */
 454     outeriter        += nri;
 455
 456     /* Update outer/inner flops */
 457
 458     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*92);
 459 }
 460 /*
 461  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_double
 462  * Electrostatics interaction: GeneralizedBorn
 463  * VdW interaction:            CubicSplineTable
 464  * Geometry:                   Particle-Particle
 465  * Calculate force/pot:        Force
 466  */
 467 void
 468 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_double
 469                     (t_nblist                    * gmx_restrict       nlist,
 470                      rvec                        * gmx_restrict          xx,
 471                      rvec                        * gmx_restrict          ff,
 472                      struct t_forcerec           * gmx_restrict          fr,
 473                      t_mdatoms                   * gmx_restrict     mdatoms,
 474                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 475                      t_nrnb                      * gmx_restrict        nrnb)
 476 {
 477     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 478      * just 0 for non-waters.
 479      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 480      * jnr indices corresponding to data put in the four positions in the SIMD register.
 481      */
 482     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 483     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 484     int              jnrA,jnrB;
 485     int              j_coord_offsetA,j_coord_offsetB;
 486     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 487     real             rcutoff_scalar;
 488     real             *shiftvec,*fshift,*x,*f;
 489     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 490     int              vdwioffset0;
 491     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 492     int              vdwjidx0A,vdwjidx0B;
 493     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 494     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 495     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 496     real             *charge;
 497     __m128i          gbitab;
 498     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 499     __m128d          minushalf = _mm_set1_pd(-0.5);
 500     real             *invsqrta,*dvda,*gbtab;
 501     int              nvdwtype;
 502     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 503     int              *vdwtype;
 504     real             *vdwparam;
 505     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 506     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 507     __m128i          vfitab;
 508     __m128i          ifour       = _mm_set1_epi32(4);
 509     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 510     real             *vftab;
 511     __m128d          dummy_mask,cutoff_mask;
 512     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 513     __m128d          one     = _mm_set1_pd(1.0);
 514     __m128d          two     = _mm_set1_pd(2.0);
 515     x                = xx[0];
 516     f                = ff[0];
 517
 518     nri              = nlist->nri;
 519     iinr             = nlist->iinr;
 520     jindex           = nlist->jindex;
 521     jjnr             = nlist->jjnr;
 522     shiftidx         = nlist->shift;
 523     gid              = nlist->gid;
 524     shiftvec         = fr->shift_vec[0];
 525     fshift           = fr->fshift[0];
 526     facel            = _mm_set1_pd(fr->ic->epsfac);
 527     charge           = mdatoms->chargeA;
 528     nvdwtype         = fr->ntype;
 529     vdwparam         = fr->nbfp;
 530     vdwtype          = mdatoms->typeA;
 531
 532     vftab            = kernel_data->table_vdw->data;
 533     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 534
 535     invsqrta         = fr->invsqrta;
 536     dvda             = fr->dvda;
 537     gbtabscale       = _mm_set1_pd(fr->gbtab->scale);
 538     gbtab            = fr->gbtab->data;
 539     gbinvepsdiff     = _mm_set1_pd((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 540
 541     /* Avoid stupid compiler warnings */
 542     jnrA = jnrB = 0;
 543     j_coord_offsetA = 0;
 544     j_coord_offsetB = 0;
 545
 546     outeriter        = 0;
 547     inneriter        = 0;
 548
 549     /* Start outer loop over neighborlists */
 550     for(iidx=0; iidx<nri; iidx++)
 551     {
 552         /* Load shift vector for this list */
 553         i_shift_offset   = DIM*shiftidx[iidx];
 554
 555         /* Load limits for loop over neighbors */
 556         j_index_start    = jindex[iidx];
 557         j_index_end      = jindex[iidx+1];
 558
 559         /* Get outer coordinate index */
 560         inr              = iinr[iidx];
 561         i_coord_offset   = DIM*inr;
 562
 563         /* Load i particle coords and add shift vector */
 564         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 565
 566         fix0             = _mm_setzero_pd();
 567         fiy0             = _mm_setzero_pd();
 568         fiz0             = _mm_setzero_pd();
 569
 570         /* Load parameters for i particles */
 571         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 572         isai0            = _mm_load1_pd(invsqrta+inr+0);
 573         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 574
 575         dvdasum          = _mm_setzero_pd();
 576
 577         /* Start inner kernel loop */
 578         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 579         {
 580
 581             /* Get j neighbor index, and coordinate index */
 582             jnrA             = jjnr[jidx];
 583             jnrB             = jjnr[jidx+1];
 584             j_coord_offsetA  = DIM*jnrA;
 585             j_coord_offsetB  = DIM*jnrB;
 586
 587             /* load j atom coordinates */
 588             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 589                                               &jx0,&jy0,&jz0);
 590
 591             /* Calculate displacement vector */
 592             dx00             = _mm_sub_pd(ix0,jx0);
 593             dy00             = _mm_sub_pd(iy0,jy0);
 594             dz00             = _mm_sub_pd(iz0,jz0);
 595
 596             /* Calculate squared distance and things based on it */
 597             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 598
 599             rinv00           = sse2_invsqrt_d(rsq00);
 600
 601             /* Load parameters for j particles */
 602             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 603             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 604             vdwjidx0A        = 2*vdwtype[jnrA+0];
 605             vdwjidx0B        = 2*vdwtype[jnrB+0];
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             r00              = _mm_mul_pd(rsq00,rinv00);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq00             = _mm_mul_pd(iq0,jq0);
 615             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 616                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 617
 618             /* Calculate table index by multiplying r with table scale and truncate to integer */
 619             rt               = _mm_mul_pd(r00,vftabscale);
 620             vfitab           = _mm_cvttpd_epi32(rt);
 621             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 622             vfitab           = _mm_slli_epi32(vfitab,3);
 623
 624             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 625             isaprod          = _mm_mul_pd(isai0,isaj0);
 626             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 627             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 628
 629             /* Calculate generalized born table index - this is a separate table from the normal one,
 630              * but we use the same procedure by multiplying r with scale and truncating to integer.
 631              */
 632             rt               = _mm_mul_pd(r00,gbscale);
 633             gbitab           = _mm_cvttpd_epi32(rt);
 634             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 635             gbitab           = _mm_slli_epi32(gbitab,2);
 636
 637             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 638             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 639             GMX_MM_TRANSPOSE2_PD(Y,F);
 640             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 641             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 642             GMX_MM_TRANSPOSE2_PD(G,H);
 643             Heps             = _mm_mul_pd(gbeps,H);
 644             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 645             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 646             vgb              = _mm_mul_pd(gbqqfactor,VV);
 647
 648             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 649             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 650             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 651             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 652             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 653             velec            = _mm_mul_pd(qq00,rinv00);
 654             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 655
 656             /* CUBIC SPLINE TABLE DISPERSION */
 657             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 658             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 659             GMX_MM_TRANSPOSE2_PD(Y,F);
 660             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 661             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 662             GMX_MM_TRANSPOSE2_PD(G,H);
 663             Heps             = _mm_mul_pd(vfeps,H);
 664             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 665             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 666             fvdw6            = _mm_mul_pd(c6_00,FF);
 667
 668             /* CUBIC SPLINE TABLE REPULSION */
 669             vfitab           = _mm_add_epi32(vfitab,ifour);
 670             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 671             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 672             GMX_MM_TRANSPOSE2_PD(Y,F);
 673             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 674             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 675             GMX_MM_TRANSPOSE2_PD(G,H);
 676             Heps             = _mm_mul_pd(vfeps,H);
 677             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 678             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 679             fvdw12           = _mm_mul_pd(c12_00,FF);
 680             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 681
 682             fscal            = _mm_add_pd(felec,fvdw);
 683
 684             /* Calculate temporary vectorial force */
 685             tx               = _mm_mul_pd(fscal,dx00);
 686             ty               = _mm_mul_pd(fscal,dy00);
 687             tz               = _mm_mul_pd(fscal,dz00);
 688
 689             /* Update vectorial force */
 690             fix0             = _mm_add_pd(fix0,tx);
 691             fiy0             = _mm_add_pd(fiy0,ty);
 692             fiz0             = _mm_add_pd(fiz0,tz);
 693
 694             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 695
 696             /* Inner loop uses 82 flops */
 697         }
 698
 699         if(jidx<j_index_end)
 700         {
 701
 702             jnrA             = jjnr[jidx];
 703             j_coord_offsetA  = DIM*jnrA;
 704
 705             /* load j atom coordinates */
 706             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 707                                               &jx0,&jy0,&jz0);
 708
 709             /* Calculate displacement vector */
 710             dx00             = _mm_sub_pd(ix0,jx0);
 711             dy00             = _mm_sub_pd(iy0,jy0);
 712             dz00             = _mm_sub_pd(iz0,jz0);
 713
 714             /* Calculate squared distance and things based on it */
 715             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 716
 717             rinv00           = sse2_invsqrt_d(rsq00);
 718
 719             /* Load parameters for j particles */
 720             jq0              = _mm_load_sd(charge+jnrA+0);
 721             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 722             vdwjidx0A        = 2*vdwtype[jnrA+0];
 723
 724             /**************************
 725              * CALCULATE INTERACTIONS *
 726              **************************/
 727
 728             r00              = _mm_mul_pd(rsq00,rinv00);
 729
 730             /* Compute parameters for interactions between i and j atoms */
 731             qq00             = _mm_mul_pd(iq0,jq0);
 732             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 733
 734             /* Calculate table index by multiplying r with table scale and truncate to integer */
 735             rt               = _mm_mul_pd(r00,vftabscale);
 736             vfitab           = _mm_cvttpd_epi32(rt);
 737             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 738             vfitab           = _mm_slli_epi32(vfitab,3);
 739
 740             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 741             isaprod          = _mm_mul_pd(isai0,isaj0);
 742             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 743             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 744
 745             /* Calculate generalized born table index - this is a separate table from the normal one,
 746              * but we use the same procedure by multiplying r with scale and truncating to integer.
 747              */
 748             rt               = _mm_mul_pd(r00,gbscale);
 749             gbitab           = _mm_cvttpd_epi32(rt);
 750             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 751             gbitab           = _mm_slli_epi32(gbitab,2);
 752
 753             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 754             F                = _mm_setzero_pd();
 755             GMX_MM_TRANSPOSE2_PD(Y,F);
 756             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 757             H                = _mm_setzero_pd();
 758             GMX_MM_TRANSPOSE2_PD(G,H);
 759             Heps             = _mm_mul_pd(gbeps,H);
 760             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 761             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 762             vgb              = _mm_mul_pd(gbqqfactor,VV);
 763
 764             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 765             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 766             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 767             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 768             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 769             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 770             velec            = _mm_mul_pd(qq00,rinv00);
 771             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 772
 773             /* CUBIC SPLINE TABLE DISPERSION */
 774             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 775             F                = _mm_setzero_pd();
 776             GMX_MM_TRANSPOSE2_PD(Y,F);
 777             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 778             H                = _mm_setzero_pd();
 779             GMX_MM_TRANSPOSE2_PD(G,H);
 780             Heps             = _mm_mul_pd(vfeps,H);
 781             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 782             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 783             fvdw6            = _mm_mul_pd(c6_00,FF);
 784
 785             /* CUBIC SPLINE TABLE REPULSION */
 786             vfitab           = _mm_add_epi32(vfitab,ifour);
 787             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 788             F                = _mm_setzero_pd();
 789             GMX_MM_TRANSPOSE2_PD(Y,F);
 790             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 791             H                = _mm_setzero_pd();
 792             GMX_MM_TRANSPOSE2_PD(G,H);
 793             Heps             = _mm_mul_pd(vfeps,H);
 794             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 795             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 796             fvdw12           = _mm_mul_pd(c12_00,FF);
 797             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 798
 799             fscal            = _mm_add_pd(felec,fvdw);
 800
 801             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 802
 803             /* Calculate temporary vectorial force */
 804             tx               = _mm_mul_pd(fscal,dx00);
 805             ty               = _mm_mul_pd(fscal,dy00);
 806             tz               = _mm_mul_pd(fscal,dz00);
 807
 808             /* Update vectorial force */
 809             fix0             = _mm_add_pd(fix0,tx);
 810             fiy0             = _mm_add_pd(fiy0,ty);
 811             fiz0             = _mm_add_pd(fiz0,tz);
 812
 813             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 814
 815             /* Inner loop uses 82 flops */
 816         }
 817
 818         /* End of innermost loop */
 819
 820         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 821                                               f+i_coord_offset,fshift+i_shift_offset);
 822
 823         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 824         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 825
 826         /* Increment number of inner iterations */
 827         inneriter                  += j_index_end - j_index_start;
 828
 829         /* Outer loop uses 7 flops */
 830     }
 831
 832     /* Increment number of outer iterations */
 833     outeriter        += nri;
 834
 835     /* Update outer/inner flops */
 836
 837     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*82);
 838 }