src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwCSTab_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_ElecEw_VdwCSTab_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_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     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 101     real             *vftab;
 102     __m128i          ewitab;
 103     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 128
 129     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vvdwsum          = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_ps(ix0,jx0);
 199             dy00             = _mm_sub_ps(iy0,jy0);
 200             dz00             = _mm_sub_ps(iz0,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 204
 205             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 206
 207             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 211                                                               charge+jnrC+0,charge+jnrD+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214             vdwjidx0C        = 2*vdwtype[jnrC+0];
 215             vdwjidx0D        = 2*vdwtype[jnrD+0];
 216
 217             /**************************
 218              * CALCULATE INTERACTIONS *
 219              **************************/
 220
 221             r00              = _mm_mul_ps(rsq00,rinv00);
 222
 223             /* Compute parameters for interactions between i and j atoms */
 224             qq00             = _mm_mul_ps(iq0,jq0);
 225             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 226                                          vdwparam+vdwioffset0+vdwjidx0B,
 227                                          vdwparam+vdwioffset0+vdwjidx0C,
 228                                          vdwparam+vdwioffset0+vdwjidx0D,
 229                                          &c6_00,&c12_00);
 230
 231             /* Calculate table index by multiplying r with table scale and truncate to integer */
 232             rt               = _mm_mul_ps(r00,vftabscale);
 233             vfitab           = _mm_cvttps_epi32(rt);
 234             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 235             vfitab           = _mm_slli_epi32(vfitab,3);
 236
 237             /* EWALD ELECTROSTATICS */
 238
 239             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 240             ewrt             = _mm_mul_ps(r00,ewtabscale);
 241             ewitab           = _mm_cvttps_epi32(ewrt);
 242             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 243             ewitab           = _mm_slli_epi32(ewitab,2);
 244             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 245             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 246             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 247             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 248             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 249             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 250             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 251             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 252             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 253
 254             /* CUBIC SPLINE TABLE DISPERSION */
 255             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 256             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 257             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 258             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 259             _MM_TRANSPOSE4_PS(Y,F,G,H);
 260             Heps             = _mm_mul_ps(vfeps,H);
 261             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 262             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 263             vvdw6            = _mm_mul_ps(c6_00,VV);
 264             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 265             fvdw6            = _mm_mul_ps(c6_00,FF);
 266
 267             /* CUBIC SPLINE TABLE REPULSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Heps             = _mm_mul_ps(vfeps,H);
 275             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 276             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 277             vvdw12           = _mm_mul_ps(c12_00,VV);
 278             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 279             fvdw12           = _mm_mul_ps(c12_00,FF);
 280             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 281             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm_add_ps(velecsum,velec);
 285             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 286
 287             fscal            = _mm_add_ps(felec,fvdw);
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm_mul_ps(fscal,dx00);
 291             ty               = _mm_mul_ps(fscal,dy00);
 292             tz               = _mm_mul_ps(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm_add_ps(fix0,tx);
 296             fiy0             = _mm_add_ps(fiy0,ty);
 297             fiz0             = _mm_add_ps(fiz0,tz);
 298
 299             fjptrA             = f+j_coord_offsetA;
 300             fjptrB             = f+j_coord_offsetB;
 301             fjptrC             = f+j_coord_offsetC;
 302             fjptrD             = f+j_coord_offsetD;
 303             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 304
 305             /* Inner loop uses 75 flops */
 306         }
 307
 308         if(jidx<j_index_end)
 309         {
 310
 311             /* Get j neighbor index, and coordinate index */
 312             jnrlistA         = jjnr[jidx];
 313             jnrlistB         = jjnr[jidx+1];
 314             jnrlistC         = jjnr[jidx+2];
 315             jnrlistD         = jjnr[jidx+3];
 316             /* Sign of each element will be negative for non-real atoms.
 317              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 318              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 319              */
 320             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 321             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 322             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 323             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 324             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 325             j_coord_offsetA  = DIM*jnrA;
 326             j_coord_offsetB  = DIM*jnrB;
 327             j_coord_offsetC  = DIM*jnrC;
 328             j_coord_offsetD  = DIM*jnrD;
 329
 330             /* load j atom coordinates */
 331             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 332                                               x+j_coord_offsetC,x+j_coord_offsetD,
 333                                               &jx0,&jy0,&jz0);
 334
 335             /* Calculate displacement vector */
 336             dx00             = _mm_sub_ps(ix0,jx0);
 337             dy00             = _mm_sub_ps(iy0,jy0);
 338             dz00             = _mm_sub_ps(iz0,jz0);
 339
 340             /* Calculate squared distance and things based on it */
 341             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 342
 343             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 344
 345             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 346
 347             /* Load parameters for j particles */
 348             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 349                                                               charge+jnrC+0,charge+jnrD+0);
 350             vdwjidx0A        = 2*vdwtype[jnrA+0];
 351             vdwjidx0B        = 2*vdwtype[jnrB+0];
 352             vdwjidx0C        = 2*vdwtype[jnrC+0];
 353             vdwjidx0D        = 2*vdwtype[jnrD+0];
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             r00              = _mm_mul_ps(rsq00,rinv00);
 360             r00              = _mm_andnot_ps(dummy_mask,r00);
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq00             = _mm_mul_ps(iq0,jq0);
 364             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 365                                          vdwparam+vdwioffset0+vdwjidx0B,
 366                                          vdwparam+vdwioffset0+vdwjidx0C,
 367                                          vdwparam+vdwioffset0+vdwjidx0D,
 368                                          &c6_00,&c12_00);
 369
 370             /* Calculate table index by multiplying r with table scale and truncate to integer */
 371             rt               = _mm_mul_ps(r00,vftabscale);
 372             vfitab           = _mm_cvttps_epi32(rt);
 373             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 374             vfitab           = _mm_slli_epi32(vfitab,3);
 375
 376             /* EWALD ELECTROSTATICS */
 377
 378             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 379             ewrt             = _mm_mul_ps(r00,ewtabscale);
 380             ewitab           = _mm_cvttps_epi32(ewrt);
 381             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 382             ewitab           = _mm_slli_epi32(ewitab,2);
 383             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 384             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 385             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 386             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 387             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 388             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 389             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 390             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 391             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 392
 393             /* CUBIC SPLINE TABLE DISPERSION */
 394             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 395             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 396             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 397             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 398             _MM_TRANSPOSE4_PS(Y,F,G,H);
 399             Heps             = _mm_mul_ps(vfeps,H);
 400             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 401             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 402             vvdw6            = _mm_mul_ps(c6_00,VV);
 403             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 404             fvdw6            = _mm_mul_ps(c6_00,FF);
 405
 406             /* CUBIC SPLINE TABLE REPULSION */
 407             vfitab           = _mm_add_epi32(vfitab,ifour);
 408             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 409             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 410             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 411             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 412             _MM_TRANSPOSE4_PS(Y,F,G,H);
 413             Heps             = _mm_mul_ps(vfeps,H);
 414             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 415             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 416             vvdw12           = _mm_mul_ps(c12_00,VV);
 417             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 418             fvdw12           = _mm_mul_ps(c12_00,FF);
 419             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 420             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velec            = _mm_andnot_ps(dummy_mask,velec);
 424             velecsum         = _mm_add_ps(velecsum,velec);
 425             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 426             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 427
 428             fscal            = _mm_add_ps(felec,fvdw);
 429
 430             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm_mul_ps(fscal,dx00);
 434             ty               = _mm_mul_ps(fscal,dy00);
 435             tz               = _mm_mul_ps(fscal,dz00);
 436
 437             /* Update vectorial force */
 438             fix0             = _mm_add_ps(fix0,tx);
 439             fiy0             = _mm_add_ps(fiy0,ty);
 440             fiz0             = _mm_add_ps(fiz0,tz);
 441
 442             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 443             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 444             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 445             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 446             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 447
 448             /* Inner loop uses 76 flops */
 449         }
 450
 451         /* End of innermost loop */
 452
 453         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 454                                               f+i_coord_offset,fshift+i_shift_offset);
 455
 456         ggid                        = gid[iidx];
 457         /* Update potential energies */
 458         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 459         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 460
 461         /* Increment number of inner iterations */
 462         inneriter                  += j_index_end - j_index_start;
 463
 464         /* Outer loop uses 9 flops */
 465     }
 466
 467     /* Increment number of outer iterations */
 468     outeriter        += nri;
 469
 470     /* Update outer/inner flops */
 471
 472     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*76);
 473 }
 474 /*
 475  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse4_1_single
 476  * Electrostatics interaction: Ewald
 477  * VdW interaction:            CubicSplineTable
 478  * Geometry:                   Particle-Particle
 479  * Calculate force/pot:        Force
 480  */
 481 void
 482 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse4_1_single
 483                     (t_nblist                    * gmx_restrict       nlist,
 484                      rvec                        * gmx_restrict          xx,
 485                      rvec                        * gmx_restrict          ff,
 486                      t_forcerec                  * gmx_restrict          fr,
 487                      t_mdatoms                   * gmx_restrict     mdatoms,
 488                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 489                      t_nrnb                      * gmx_restrict        nrnb)
 490 {
 491     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 492      * just 0 for non-waters.
 493      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 494      * jnr indices corresponding to data put in the four positions in the SIMD register.
 495      */
 496     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 497     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 498     int              jnrA,jnrB,jnrC,jnrD;
 499     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 500     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 501     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 502     real             rcutoff_scalar;
 503     real             *shiftvec,*fshift,*x,*f;
 504     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 505     real             scratch[4*DIM];
 506     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 507     int              vdwioffset0;
 508     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 509     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 510     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 511     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 512     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 513     real             *charge;
 514     int              nvdwtype;
 515     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 516     int              *vdwtype;
 517     real             *vdwparam;
 518     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 519     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 520     __m128i          vfitab;
 521     __m128i          ifour       = _mm_set1_epi32(4);
 522     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 523     real             *vftab;
 524     __m128i          ewitab;
 525     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 526     real             *ewtab;
 527     __m128           dummy_mask,cutoff_mask;
 528     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 529     __m128           one     = _mm_set1_ps(1.0);
 530     __m128           two     = _mm_set1_ps(2.0);
 531     x                = xx[0];
 532     f                = ff[0];
 533
 534     nri              = nlist->nri;
 535     iinr             = nlist->iinr;
 536     jindex           = nlist->jindex;
 537     jjnr             = nlist->jjnr;
 538     shiftidx         = nlist->shift;
 539     gid              = nlist->gid;
 540     shiftvec         = fr->shift_vec[0];
 541     fshift           = fr->fshift[0];
 542     facel            = _mm_set1_ps(fr->epsfac);
 543     charge           = mdatoms->chargeA;
 544     nvdwtype         = fr->ntype;
 545     vdwparam         = fr->nbfp;
 546     vdwtype          = mdatoms->typeA;
 547
 548     vftab            = kernel_data->table_vdw->data;
 549     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 550
 551     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 552     ewtab            = fr->ic->tabq_coul_F;
 553     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 554     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 555
 556     /* Avoid stupid compiler warnings */
 557     jnrA = jnrB = jnrC = jnrD = 0;
 558     j_coord_offsetA = 0;
 559     j_coord_offsetB = 0;
 560     j_coord_offsetC = 0;
 561     j_coord_offsetD = 0;
 562
 563     outeriter        = 0;
 564     inneriter        = 0;
 565
 566     for(iidx=0;iidx<4*DIM;iidx++)
 567     {
 568         scratch[iidx] = 0.0;
 569     }
 570
 571     /* Start outer loop over neighborlists */
 572     for(iidx=0; iidx<nri; iidx++)
 573     {
 574         /* Load shift vector for this list */
 575         i_shift_offset   = DIM*shiftidx[iidx];
 576
 577         /* Load limits for loop over neighbors */
 578         j_index_start    = jindex[iidx];
 579         j_index_end      = jindex[iidx+1];
 580
 581         /* Get outer coordinate index */
 582         inr              = iinr[iidx];
 583         i_coord_offset   = DIM*inr;
 584
 585         /* Load i particle coords and add shift vector */
 586         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 587
 588         fix0             = _mm_setzero_ps();
 589         fiy0             = _mm_setzero_ps();
 590         fiz0             = _mm_setzero_ps();
 591
 592         /* Load parameters for i particles */
 593         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 594         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 595
 596         /* Start inner kernel loop */
 597         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 598         {
 599
 600             /* Get j neighbor index, and coordinate index */
 601             jnrA             = jjnr[jidx];
 602             jnrB             = jjnr[jidx+1];
 603             jnrC             = jjnr[jidx+2];
 604             jnrD             = jjnr[jidx+3];
 605             j_coord_offsetA  = DIM*jnrA;
 606             j_coord_offsetB  = DIM*jnrB;
 607             j_coord_offsetC  = DIM*jnrC;
 608             j_coord_offsetD  = DIM*jnrD;
 609
 610             /* load j atom coordinates */
 611             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 612                                               x+j_coord_offsetC,x+j_coord_offsetD,
 613                                               &jx0,&jy0,&jz0);
 614
 615             /* Calculate displacement vector */
 616             dx00             = _mm_sub_ps(ix0,jx0);
 617             dy00             = _mm_sub_ps(iy0,jy0);
 618             dz00             = _mm_sub_ps(iz0,jz0);
 619
 620             /* Calculate squared distance and things based on it */
 621             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 622
 623             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 624
 625             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 626
 627             /* Load parameters for j particles */
 628             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 629                                                               charge+jnrC+0,charge+jnrD+0);
 630             vdwjidx0A        = 2*vdwtype[jnrA+0];
 631             vdwjidx0B        = 2*vdwtype[jnrB+0];
 632             vdwjidx0C        = 2*vdwtype[jnrC+0];
 633             vdwjidx0D        = 2*vdwtype[jnrD+0];
 634
 635             /**************************
 636              * CALCULATE INTERACTIONS *
 637              **************************/
 638
 639             r00              = _mm_mul_ps(rsq00,rinv00);
 640
 641             /* Compute parameters for interactions between i and j atoms */
 642             qq00             = _mm_mul_ps(iq0,jq0);
 643             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 644                                          vdwparam+vdwioffset0+vdwjidx0B,
 645                                          vdwparam+vdwioffset0+vdwjidx0C,
 646                                          vdwparam+vdwioffset0+vdwjidx0D,
 647                                          &c6_00,&c12_00);
 648
 649             /* Calculate table index by multiplying r with table scale and truncate to integer */
 650             rt               = _mm_mul_ps(r00,vftabscale);
 651             vfitab           = _mm_cvttps_epi32(rt);
 652             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 653             vfitab           = _mm_slli_epi32(vfitab,3);
 654
 655             /* EWALD ELECTROSTATICS */
 656
 657             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 658             ewrt             = _mm_mul_ps(r00,ewtabscale);
 659             ewitab           = _mm_cvttps_epi32(ewrt);
 660             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 661             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 662                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 663                                          &ewtabF,&ewtabFn);
 664             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 665             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 666
 667             /* CUBIC SPLINE TABLE DISPERSION */
 668             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 669             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 670             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 671             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 672             _MM_TRANSPOSE4_PS(Y,F,G,H);
 673             Heps             = _mm_mul_ps(vfeps,H);
 674             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 675             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 676             fvdw6            = _mm_mul_ps(c6_00,FF);
 677
 678             /* CUBIC SPLINE TABLE REPULSION */
 679             vfitab           = _mm_add_epi32(vfitab,ifour);
 680             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 681             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 682             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 683             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 684             _MM_TRANSPOSE4_PS(Y,F,G,H);
 685             Heps             = _mm_mul_ps(vfeps,H);
 686             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 687             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 688             fvdw12           = _mm_mul_ps(c12_00,FF);
 689             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 690
 691             fscal            = _mm_add_ps(felec,fvdw);
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm_mul_ps(fscal,dx00);
 695             ty               = _mm_mul_ps(fscal,dy00);
 696             tz               = _mm_mul_ps(fscal,dz00);
 697
 698             /* Update vectorial force */
 699             fix0             = _mm_add_ps(fix0,tx);
 700             fiy0             = _mm_add_ps(fiy0,ty);
 701             fiz0             = _mm_add_ps(fiz0,tz);
 702
 703             fjptrA             = f+j_coord_offsetA;
 704             fjptrB             = f+j_coord_offsetB;
 705             fjptrC             = f+j_coord_offsetC;
 706             fjptrD             = f+j_coord_offsetD;
 707             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 708
 709             /* Inner loop uses 62 flops */
 710         }
 711
 712         if(jidx<j_index_end)
 713         {
 714
 715             /* Get j neighbor index, and coordinate index */
 716             jnrlistA         = jjnr[jidx];
 717             jnrlistB         = jjnr[jidx+1];
 718             jnrlistC         = jjnr[jidx+2];
 719             jnrlistD         = jjnr[jidx+3];
 720             /* Sign of each element will be negative for non-real atoms.
 721              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 722              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 723              */
 724             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 725             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 726             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 727             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 728             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 729             j_coord_offsetA  = DIM*jnrA;
 730             j_coord_offsetB  = DIM*jnrB;
 731             j_coord_offsetC  = DIM*jnrC;
 732             j_coord_offsetD  = DIM*jnrD;
 733
 734             /* load j atom coordinates */
 735             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 736                                               x+j_coord_offsetC,x+j_coord_offsetD,
 737                                               &jx0,&jy0,&jz0);
 738
 739             /* Calculate displacement vector */
 740             dx00             = _mm_sub_ps(ix0,jx0);
 741             dy00             = _mm_sub_ps(iy0,jy0);
 742             dz00             = _mm_sub_ps(iz0,jz0);
 743
 744             /* Calculate squared distance and things based on it */
 745             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 746
 747             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 748
 749             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 750
 751             /* Load parameters for j particles */
 752             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 753                                                               charge+jnrC+0,charge+jnrD+0);
 754             vdwjidx0A        = 2*vdwtype[jnrA+0];
 755             vdwjidx0B        = 2*vdwtype[jnrB+0];
 756             vdwjidx0C        = 2*vdwtype[jnrC+0];
 757             vdwjidx0D        = 2*vdwtype[jnrD+0];
 758
 759             /**************************
 760              * CALCULATE INTERACTIONS *
 761              **************************/
 762
 763             r00              = _mm_mul_ps(rsq00,rinv00);
 764             r00              = _mm_andnot_ps(dummy_mask,r00);
 765
 766             /* Compute parameters for interactions between i and j atoms */
 767             qq00             = _mm_mul_ps(iq0,jq0);
 768             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 769                                          vdwparam+vdwioffset0+vdwjidx0B,
 770                                          vdwparam+vdwioffset0+vdwjidx0C,
 771                                          vdwparam+vdwioffset0+vdwjidx0D,
 772                                          &c6_00,&c12_00);
 773
 774             /* Calculate table index by multiplying r with table scale and truncate to integer */
 775             rt               = _mm_mul_ps(r00,vftabscale);
 776             vfitab           = _mm_cvttps_epi32(rt);
 777             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 778             vfitab           = _mm_slli_epi32(vfitab,3);
 779
 780             /* EWALD ELECTROSTATICS */
 781
 782             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 783             ewrt             = _mm_mul_ps(r00,ewtabscale);
 784             ewitab           = _mm_cvttps_epi32(ewrt);
 785             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 786             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 787                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 788                                          &ewtabF,&ewtabFn);
 789             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 790             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 791
 792             /* CUBIC SPLINE TABLE DISPERSION */
 793             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 794             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 795             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 796             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 797             _MM_TRANSPOSE4_PS(Y,F,G,H);
 798             Heps             = _mm_mul_ps(vfeps,H);
 799             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 800             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 801             fvdw6            = _mm_mul_ps(c6_00,FF);
 802
 803             /* CUBIC SPLINE TABLE REPULSION */
 804             vfitab           = _mm_add_epi32(vfitab,ifour);
 805             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 806             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 807             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 808             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 809             _MM_TRANSPOSE4_PS(Y,F,G,H);
 810             Heps             = _mm_mul_ps(vfeps,H);
 811             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 812             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 813             fvdw12           = _mm_mul_ps(c12_00,FF);
 814             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 815
 816             fscal            = _mm_add_ps(felec,fvdw);
 817
 818             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 819
 820             /* Calculate temporary vectorial force */
 821             tx               = _mm_mul_ps(fscal,dx00);
 822             ty               = _mm_mul_ps(fscal,dy00);
 823             tz               = _mm_mul_ps(fscal,dz00);
 824
 825             /* Update vectorial force */
 826             fix0             = _mm_add_ps(fix0,tx);
 827             fiy0             = _mm_add_ps(fiy0,ty);
 828             fiz0             = _mm_add_ps(fiz0,tz);
 829
 830             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 831             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 832             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 833             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 834             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 835
 836             /* Inner loop uses 63 flops */
 837         }
 838
 839         /* End of innermost loop */
 840
 841         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 842                                               f+i_coord_offset,fshift+i_shift_offset);
 843
 844         /* Increment number of inner iterations */
 845         inneriter                  += j_index_end - j_index_start;
 846
 847         /* Outer loop uses 7 flops */
 848     }
 849
 850     /* Increment number of outer iterations */
 851     outeriter        += nri;
 852
 853     /* Update outer/inner flops */
 854
 855     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);
 856 }