src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwCSTab_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  94     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128i          ewitab;
 100     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     real             *ewtab;
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->ic->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 125
 126     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 162
 163         fix0             = _mm_setzero_ps();
 164         fiy0             = _mm_setzero_ps();
 165         fiz0             = _mm_setzero_ps();
 166
 167         /* Load parameters for i particles */
 168         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 169         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_ps();
 173         vvdwsum          = _mm_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               x+j_coord_offsetC,x+j_coord_offsetD,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_ps(ix0,jx0);
 196             dy00             = _mm_sub_ps(iy0,jy0);
 197             dz00             = _mm_sub_ps(iz0,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 201
 202             rinv00           = sse2_invsqrt_f(rsq00);
 203
 204             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 208                                                               charge+jnrC+0,charge+jnrD+0);
 209             vdwjidx0A        = 2*vdwtype[jnrA+0];
 210             vdwjidx0B        = 2*vdwtype[jnrB+0];
 211             vdwjidx0C        = 2*vdwtype[jnrC+0];
 212             vdwjidx0D        = 2*vdwtype[jnrD+0];
 213
 214             /**************************
 215              * CALCULATE INTERACTIONS *
 216              **************************/
 217
 218             r00              = _mm_mul_ps(rsq00,rinv00);
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             qq00             = _mm_mul_ps(iq0,jq0);
 222             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 223                                          vdwparam+vdwioffset0+vdwjidx0B,
 224                                          vdwparam+vdwioffset0+vdwjidx0C,
 225                                          vdwparam+vdwioffset0+vdwjidx0D,
 226                                          &c6_00,&c12_00);
 227
 228             /* Calculate table index by multiplying r with table scale and truncate to integer */
 229             rt               = _mm_mul_ps(r00,vftabscale);
 230             vfitab           = _mm_cvttps_epi32(rt);
 231             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 232             vfitab           = _mm_slli_epi32(vfitab,3);
 233
 234             /* EWALD ELECTROSTATICS */
 235
 236             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 237             ewrt             = _mm_mul_ps(r00,ewtabscale);
 238             ewitab           = _mm_cvttps_epi32(ewrt);
 239             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 240             ewitab           = _mm_slli_epi32(ewitab,2);
 241             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 242             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 243             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 244             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 245             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 246             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 247             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 248             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 249             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 250
 251             /* CUBIC SPLINE TABLE DISPERSION */
 252             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 253             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 254             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 255             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 256             _MM_TRANSPOSE4_PS(Y,F,G,H);
 257             Heps             = _mm_mul_ps(vfeps,H);
 258             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 259             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 260             vvdw6            = _mm_mul_ps(c6_00,VV);
 261             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 262             fvdw6            = _mm_mul_ps(c6_00,FF);
 263
 264             /* CUBIC SPLINE TABLE REPULSION */
 265             vfitab           = _mm_add_epi32(vfitab,ifour);
 266             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 267             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 268             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 269             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 270             _MM_TRANSPOSE4_PS(Y,F,G,H);
 271             Heps             = _mm_mul_ps(vfeps,H);
 272             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 273             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 274             vvdw12           = _mm_mul_ps(c12_00,VV);
 275             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 276             fvdw12           = _mm_mul_ps(c12_00,FF);
 277             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 278             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm_add_ps(velecsum,velec);
 282             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_ps(felec,fvdw);
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_ps(fscal,dx00);
 288             ty               = _mm_mul_ps(fscal,dy00);
 289             tz               = _mm_mul_ps(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_add_ps(fix0,tx);
 293             fiy0             = _mm_add_ps(fiy0,ty);
 294             fiz0             = _mm_add_ps(fiz0,tz);
 295
 296             fjptrA             = f+j_coord_offsetA;
 297             fjptrB             = f+j_coord_offsetB;
 298             fjptrC             = f+j_coord_offsetC;
 299             fjptrD             = f+j_coord_offsetD;
 300             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 301
 302             /* Inner loop uses 75 flops */
 303         }
 304
 305         if(jidx<j_index_end)
 306         {
 307
 308             /* Get j neighbor index, and coordinate index */
 309             jnrlistA         = jjnr[jidx];
 310             jnrlistB         = jjnr[jidx+1];
 311             jnrlistC         = jjnr[jidx+2];
 312             jnrlistD         = jjnr[jidx+3];
 313             /* Sign of each element will be negative for non-real atoms.
 314              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 315              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 316              */
 317             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 318             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 319             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 320             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 321             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 322             j_coord_offsetA  = DIM*jnrA;
 323             j_coord_offsetB  = DIM*jnrB;
 324             j_coord_offsetC  = DIM*jnrC;
 325             j_coord_offsetD  = DIM*jnrD;
 326
 327             /* load j atom coordinates */
 328             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 329                                               x+j_coord_offsetC,x+j_coord_offsetD,
 330                                               &jx0,&jy0,&jz0);
 331
 332             /* Calculate displacement vector */
 333             dx00             = _mm_sub_ps(ix0,jx0);
 334             dy00             = _mm_sub_ps(iy0,jy0);
 335             dz00             = _mm_sub_ps(iz0,jz0);
 336
 337             /* Calculate squared distance and things based on it */
 338             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 339
 340             rinv00           = sse2_invsqrt_f(rsq00);
 341
 342             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 343
 344             /* Load parameters for j particles */
 345             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 346                                                               charge+jnrC+0,charge+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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 368             rt               = _mm_mul_ps(r00,vftabscale);
 369             vfitab           = _mm_cvttps_epi32(rt);
 370             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 371             vfitab           = _mm_slli_epi32(vfitab,3);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 376             ewrt             = _mm_mul_ps(r00,ewtabscale);
 377             ewitab           = _mm_cvttps_epi32(ewrt);
 378             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 379             ewitab           = _mm_slli_epi32(ewitab,2);
 380             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 381             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 382             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 383             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 384             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 385             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 386             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 387             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 388             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 389
 390             /* CUBIC SPLINE TABLE DISPERSION */
 391             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 392             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 393             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 394             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 395             _MM_TRANSPOSE4_PS(Y,F,G,H);
 396             Heps             = _mm_mul_ps(vfeps,H);
 397             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 398             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 399             vvdw6            = _mm_mul_ps(c6_00,VV);
 400             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 401             fvdw6            = _mm_mul_ps(c6_00,FF);
 402
 403             /* CUBIC SPLINE TABLE REPULSION */
 404             vfitab           = _mm_add_epi32(vfitab,ifour);
 405             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 406             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 407             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 408             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 409             _MM_TRANSPOSE4_PS(Y,F,G,H);
 410             Heps             = _mm_mul_ps(vfeps,H);
 411             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 412             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 413             vvdw12           = _mm_mul_ps(c12_00,VV);
 414             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 415             fvdw12           = _mm_mul_ps(c12_00,FF);
 416             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 417             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm_andnot_ps(dummy_mask,velec);
 421             velecsum         = _mm_add_ps(velecsum,velec);
 422             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 423             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 424
 425             fscal            = _mm_add_ps(felec,fvdw);
 426
 427             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm_mul_ps(fscal,dx00);
 431             ty               = _mm_mul_ps(fscal,dy00);
 432             tz               = _mm_mul_ps(fscal,dz00);
 433
 434             /* Update vectorial force */
 435             fix0             = _mm_add_ps(fix0,tx);
 436             fiy0             = _mm_add_ps(fiy0,ty);
 437             fiz0             = _mm_add_ps(fiz0,tz);
 438
 439             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 440             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 441             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 442             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 443             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 444
 445             /* Inner loop uses 76 flops */
 446         }
 447
 448         /* End of innermost loop */
 449
 450         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 451                                               f+i_coord_offset,fshift+i_shift_offset);
 452
 453         ggid                        = gid[iidx];
 454         /* Update potential energies */
 455         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 456         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 457
 458         /* Increment number of inner iterations */
 459         inneriter                  += j_index_end - j_index_start;
 460
 461         /* Outer loop uses 9 flops */
 462     }
 463
 464     /* Increment number of outer iterations */
 465     outeriter        += nri;
 466
 467     /* Update outer/inner flops */
 468
 469     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*76);
 470 }
 471 /*
 472  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse2_single
 473  * Electrostatics interaction: Ewald
 474  * VdW interaction:            CubicSplineTable
 475  * Geometry:                   Particle-Particle
 476  * Calculate force/pot:        Force
 477  */
 478 void
 479 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse2_single
 480                     (t_nblist                    * gmx_restrict       nlist,
 481                      rvec                        * gmx_restrict          xx,
 482                      rvec                        * gmx_restrict          ff,
 483                      struct t_forcerec           * gmx_restrict          fr,
 484                      t_mdatoms                   * gmx_restrict     mdatoms,
 485                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 486                      t_nrnb                      * gmx_restrict        nrnb)
 487 {
 488     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 489      * just 0 for non-waters.
 490      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 491      * jnr indices corresponding to data put in the four positions in the SIMD register.
 492      */
 493     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 494     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 495     int              jnrA,jnrB,jnrC,jnrD;
 496     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 497     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 498     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 499     real             rcutoff_scalar;
 500     real             *shiftvec,*fshift,*x,*f;
 501     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 502     real             scratch[4*DIM];
 503     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 504     int              vdwioffset0;
 505     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 506     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 507     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 508     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 509     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 510     real             *charge;
 511     int              nvdwtype;
 512     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 513     int              *vdwtype;
 514     real             *vdwparam;
 515     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 516     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 517     __m128i          vfitab;
 518     __m128i          ifour       = _mm_set1_epi32(4);
 519     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 520     real             *vftab;
 521     __m128i          ewitab;
 522     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 523     real             *ewtab;
 524     __m128           dummy_mask,cutoff_mask;
 525     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 526     __m128           one     = _mm_set1_ps(1.0);
 527     __m128           two     = _mm_set1_ps(2.0);
 528     x                = xx[0];
 529     f                = ff[0];
 530
 531     nri              = nlist->nri;
 532     iinr             = nlist->iinr;
 533     jindex           = nlist->jindex;
 534     jjnr             = nlist->jjnr;
 535     shiftidx         = nlist->shift;
 536     gid              = nlist->gid;
 537     shiftvec         = fr->shift_vec[0];
 538     fshift           = fr->fshift[0];
 539     facel            = _mm_set1_ps(fr->ic->epsfac);
 540     charge           = mdatoms->chargeA;
 541     nvdwtype         = fr->ntype;
 542     vdwparam         = fr->nbfp;
 543     vdwtype          = mdatoms->typeA;
 544
 545     vftab            = kernel_data->table_vdw->data;
 546     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 547
 548     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 549     ewtab            = fr->ic->tabq_coul_F;
 550     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 551     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 552
 553     /* Avoid stupid compiler warnings */
 554     jnrA = jnrB = jnrC = jnrD = 0;
 555     j_coord_offsetA = 0;
 556     j_coord_offsetB = 0;
 557     j_coord_offsetC = 0;
 558     j_coord_offsetD = 0;
 559
 560     outeriter        = 0;
 561     inneriter        = 0;
 562
 563     for(iidx=0;iidx<4*DIM;iidx++)
 564     {
 565         scratch[iidx] = 0.0;
 566     }
 567
 568     /* Start outer loop over neighborlists */
 569     for(iidx=0; iidx<nri; iidx++)
 570     {
 571         /* Load shift vector for this list */
 572         i_shift_offset   = DIM*shiftidx[iidx];
 573
 574         /* Load limits for loop over neighbors */
 575         j_index_start    = jindex[iidx];
 576         j_index_end      = jindex[iidx+1];
 577
 578         /* Get outer coordinate index */
 579         inr              = iinr[iidx];
 580         i_coord_offset   = DIM*inr;
 581
 582         /* Load i particle coords and add shift vector */
 583         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 584
 585         fix0             = _mm_setzero_ps();
 586         fiy0             = _mm_setzero_ps();
 587         fiz0             = _mm_setzero_ps();
 588
 589         /* Load parameters for i particles */
 590         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 591         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 592
 593         /* Start inner kernel loop */
 594         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 595         {
 596
 597             /* Get j neighbor index, and coordinate index */
 598             jnrA             = jjnr[jidx];
 599             jnrB             = jjnr[jidx+1];
 600             jnrC             = jjnr[jidx+2];
 601             jnrD             = jjnr[jidx+3];
 602             j_coord_offsetA  = DIM*jnrA;
 603             j_coord_offsetB  = DIM*jnrB;
 604             j_coord_offsetC  = DIM*jnrC;
 605             j_coord_offsetD  = DIM*jnrD;
 606
 607             /* load j atom coordinates */
 608             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 609                                               x+j_coord_offsetC,x+j_coord_offsetD,
 610                                               &jx0,&jy0,&jz0);
 611
 612             /* Calculate displacement vector */
 613             dx00             = _mm_sub_ps(ix0,jx0);
 614             dy00             = _mm_sub_ps(iy0,jy0);
 615             dz00             = _mm_sub_ps(iz0,jz0);
 616
 617             /* Calculate squared distance and things based on it */
 618             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 619
 620             rinv00           = sse2_invsqrt_f(rsq00);
 621
 622             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 623
 624             /* Load parameters for j particles */
 625             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 626                                                               charge+jnrC+0,charge+jnrD+0);
 627             vdwjidx0A        = 2*vdwtype[jnrA+0];
 628             vdwjidx0B        = 2*vdwtype[jnrB+0];
 629             vdwjidx0C        = 2*vdwtype[jnrC+0];
 630             vdwjidx0D        = 2*vdwtype[jnrD+0];
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             r00              = _mm_mul_ps(rsq00,rinv00);
 637
 638             /* Compute parameters for interactions between i and j atoms */
 639             qq00             = _mm_mul_ps(iq0,jq0);
 640             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 641                                          vdwparam+vdwioffset0+vdwjidx0B,
 642                                          vdwparam+vdwioffset0+vdwjidx0C,
 643                                          vdwparam+vdwioffset0+vdwjidx0D,
 644                                          &c6_00,&c12_00);
 645
 646             /* Calculate table index by multiplying r with table scale and truncate to integer */
 647             rt               = _mm_mul_ps(r00,vftabscale);
 648             vfitab           = _mm_cvttps_epi32(rt);
 649             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 650             vfitab           = _mm_slli_epi32(vfitab,3);
 651
 652             /* EWALD ELECTROSTATICS */
 653
 654             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 655             ewrt             = _mm_mul_ps(r00,ewtabscale);
 656             ewitab           = _mm_cvttps_epi32(ewrt);
 657             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 658             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 659                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 660                                          &ewtabF,&ewtabFn);
 661             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 662             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 663
 664             /* CUBIC SPLINE TABLE DISPERSION */
 665             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 666             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 667             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 668             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 669             _MM_TRANSPOSE4_PS(Y,F,G,H);
 670             Heps             = _mm_mul_ps(vfeps,H);
 671             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 672             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 673             fvdw6            = _mm_mul_ps(c6_00,FF);
 674
 675             /* CUBIC SPLINE TABLE REPULSION */
 676             vfitab           = _mm_add_epi32(vfitab,ifour);
 677             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 678             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 679             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 680             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 681             _MM_TRANSPOSE4_PS(Y,F,G,H);
 682             Heps             = _mm_mul_ps(vfeps,H);
 683             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 684             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 685             fvdw12           = _mm_mul_ps(c12_00,FF);
 686             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 687
 688             fscal            = _mm_add_ps(felec,fvdw);
 689
 690             /* Calculate temporary vectorial force */
 691             tx               = _mm_mul_ps(fscal,dx00);
 692             ty               = _mm_mul_ps(fscal,dy00);
 693             tz               = _mm_mul_ps(fscal,dz00);
 694
 695             /* Update vectorial force */
 696             fix0             = _mm_add_ps(fix0,tx);
 697             fiy0             = _mm_add_ps(fiy0,ty);
 698             fiz0             = _mm_add_ps(fiz0,tz);
 699
 700             fjptrA             = f+j_coord_offsetA;
 701             fjptrB             = f+j_coord_offsetB;
 702             fjptrC             = f+j_coord_offsetC;
 703             fjptrD             = f+j_coord_offsetD;
 704             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 705
 706             /* Inner loop uses 62 flops */
 707         }
 708
 709         if(jidx<j_index_end)
 710         {
 711
 712             /* Get j neighbor index, and coordinate index */
 713             jnrlistA         = jjnr[jidx];
 714             jnrlistB         = jjnr[jidx+1];
 715             jnrlistC         = jjnr[jidx+2];
 716             jnrlistD         = jjnr[jidx+3];
 717             /* Sign of each element will be negative for non-real atoms.
 718              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 719              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 720              */
 721             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 722             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 723             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 724             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 725             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 726             j_coord_offsetA  = DIM*jnrA;
 727             j_coord_offsetB  = DIM*jnrB;
 728             j_coord_offsetC  = DIM*jnrC;
 729             j_coord_offsetD  = DIM*jnrD;
 730
 731             /* load j atom coordinates */
 732             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 733                                               x+j_coord_offsetC,x+j_coord_offsetD,
 734                                               &jx0,&jy0,&jz0);
 735
 736             /* Calculate displacement vector */
 737             dx00             = _mm_sub_ps(ix0,jx0);
 738             dy00             = _mm_sub_ps(iy0,jy0);
 739             dz00             = _mm_sub_ps(iz0,jz0);
 740
 741             /* Calculate squared distance and things based on it */
 742             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 743
 744             rinv00           = sse2_invsqrt_f(rsq00);
 745
 746             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 747
 748             /* Load parameters for j particles */
 749             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 750                                                               charge+jnrC+0,charge+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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 772             rt               = _mm_mul_ps(r00,vftabscale);
 773             vfitab           = _mm_cvttps_epi32(rt);
 774             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 775             vfitab           = _mm_slli_epi32(vfitab,3);
 776
 777             /* EWALD ELECTROSTATICS */
 778
 779             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 780             ewrt             = _mm_mul_ps(r00,ewtabscale);
 781             ewitab           = _mm_cvttps_epi32(ewrt);
 782             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 783             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 784                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 785                                          &ewtabF,&ewtabFn);
 786             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 787             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 788
 789             /* CUBIC SPLINE TABLE DISPERSION */
 790             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 791             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 792             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 793             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 794             _MM_TRANSPOSE4_PS(Y,F,G,H);
 795             Heps             = _mm_mul_ps(vfeps,H);
 796             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 797             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 798             fvdw6            = _mm_mul_ps(c6_00,FF);
 799
 800             /* CUBIC SPLINE TABLE REPULSION */
 801             vfitab           = _mm_add_epi32(vfitab,ifour);
 802             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 803             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 804             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 805             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 806             _MM_TRANSPOSE4_PS(Y,F,G,H);
 807             Heps             = _mm_mul_ps(vfeps,H);
 808             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 809             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 810             fvdw12           = _mm_mul_ps(c12_00,FF);
 811             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 812
 813             fscal            = _mm_add_ps(felec,fvdw);
 814
 815             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 816
 817             /* Calculate temporary vectorial force */
 818             tx               = _mm_mul_ps(fscal,dx00);
 819             ty               = _mm_mul_ps(fscal,dy00);
 820             tz               = _mm_mul_ps(fscal,dz00);
 821
 822             /* Update vectorial force */
 823             fix0             = _mm_add_ps(fix0,tx);
 824             fiy0             = _mm_add_ps(fiy0,ty);
 825             fiz0             = _mm_add_ps(fiz0,tz);
 826
 827             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 828             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 829             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 830             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 831             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 832
 833             /* Inner loop uses 63 flops */
 834         }
 835
 836         /* End of innermost loop */
 837
 838         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 839                                               f+i_coord_offset,fshift+i_shift_offset);
 840
 841         /* Increment number of inner iterations */
 842         inneriter                  += j_index_end - j_index_start;
 843
 844         /* Outer loop uses 7 flops */
 845     }
 846
 847     /* Increment number of outer iterations */
 848     outeriter        += nri;
 849
 850     /* Update outer/inner flops */
 851
 852     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);
 853 }