src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_sse2_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 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse2_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          ewitab;
  99     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 102     real             rswitch_scalar,d_scalar;
 103     __m128           dummy_mask,cutoff_mask;
 104     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 105     __m128           one     = _mm_set1_ps(1.0);
 106     __m128           two     = _mm_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_ps(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 128
 129     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->rcoulomb;
 131     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 132     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 133
 134     rswitch_scalar   = fr->rcoulomb_switch;
 135     rswitch          = _mm_set1_ps(rswitch_scalar);
 136     /* Setup switch parameters */
 137     d_scalar         = rcutoff_scalar-rswitch_scalar;
 138     d                = _mm_set1_ps(d_scalar);
 139     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 140     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 143     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 144     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     for(iidx=0;iidx<4*DIM;iidx++)
 157     {
 158         scratch[iidx] = 0.0;
 159     }
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181
 182         /* Load parameters for i particles */
 183         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 184         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 216
 217             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 218
 219             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             if (gmx_mm_any_lt(rsq00,rcutoff2))
 234             {
 235
 236             r00              = _mm_mul_ps(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq00             = _mm_mul_ps(iq0,jq0);
 240             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,
 242                                          vdwparam+vdwioffset0+vdwjidx0C,
 243                                          vdwparam+vdwioffset0+vdwjidx0D,
 244                                          &c6_00,&c12_00);
 245
 246             /* EWALD ELECTROSTATICS */
 247
 248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 249             ewrt             = _mm_mul_ps(r00,ewtabscale);
 250             ewitab           = _mm_cvttps_epi32(ewrt);
 251             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 252             ewitab           = _mm_slli_epi32(ewitab,2);
 253             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 254             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 255             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 256             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 257             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 258             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 259             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 260             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 261             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 262
 263             /* LENNARD-JONES DISPERSION/REPULSION */
 264
 265             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 266             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 267             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 268             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 269             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 270
 271             d                = _mm_sub_ps(r00,rswitch);
 272             d                = _mm_max_ps(d,_mm_setzero_ps());
 273             d2               = _mm_mul_ps(d,d);
 274             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 275
 276             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 277
 278             /* Evaluate switch function */
 279             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 280             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 281             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 282             velec            = _mm_mul_ps(velec,sw);
 283             vvdw             = _mm_mul_ps(vvdw,sw);
 284             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velec            = _mm_and_ps(velec,cutoff_mask);
 288             velecsum         = _mm_add_ps(velecsum,velec);
 289             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 290             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 291
 292             fscal            = _mm_add_ps(felec,fvdw);
 293
 294             fscal            = _mm_and_ps(fscal,cutoff_mask);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm_mul_ps(fscal,dx00);
 298             ty               = _mm_mul_ps(fscal,dy00);
 299             tz               = _mm_mul_ps(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm_add_ps(fix0,tx);
 303             fiy0             = _mm_add_ps(fiy0,ty);
 304             fiz0             = _mm_add_ps(fiz0,tz);
 305
 306             fjptrA             = f+j_coord_offsetA;
 307             fjptrB             = f+j_coord_offsetB;
 308             fjptrC             = f+j_coord_offsetC;
 309             fjptrD             = f+j_coord_offsetD;
 310             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 311
 312             }
 313
 314             /* Inner loop uses 83 flops */
 315         }
 316
 317         if(jidx<j_index_end)
 318         {
 319
 320             /* Get j neighbor index, and coordinate index */
 321             jnrlistA         = jjnr[jidx];
 322             jnrlistB         = jjnr[jidx+1];
 323             jnrlistC         = jjnr[jidx+2];
 324             jnrlistD         = jjnr[jidx+3];
 325             /* Sign of each element will be negative for non-real atoms.
 326              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 327              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 328              */
 329             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 330             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 331             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 332             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 333             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 334             j_coord_offsetA  = DIM*jnrA;
 335             j_coord_offsetB  = DIM*jnrB;
 336             j_coord_offsetC  = DIM*jnrC;
 337             j_coord_offsetD  = DIM*jnrD;
 338
 339             /* load j atom coordinates */
 340             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 341                                               x+j_coord_offsetC,x+j_coord_offsetD,
 342                                               &jx0,&jy0,&jz0);
 343
 344             /* Calculate displacement vector */
 345             dx00             = _mm_sub_ps(ix0,jx0);
 346             dy00             = _mm_sub_ps(iy0,jy0);
 347             dz00             = _mm_sub_ps(iz0,jz0);
 348
 349             /* Calculate squared distance and things based on it */
 350             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 351
 352             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 353
 354             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 355
 356             /* Load parameters for j particles */
 357             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 358                                                               charge+jnrC+0,charge+jnrD+0);
 359             vdwjidx0A        = 2*vdwtype[jnrA+0];
 360             vdwjidx0B        = 2*vdwtype[jnrB+0];
 361             vdwjidx0C        = 2*vdwtype[jnrC+0];
 362             vdwjidx0D        = 2*vdwtype[jnrD+0];
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             if (gmx_mm_any_lt(rsq00,rcutoff2))
 369             {
 370
 371             r00              = _mm_mul_ps(rsq00,rinv00);
 372             r00              = _mm_andnot_ps(dummy_mask,r00);
 373
 374             /* Compute parameters for interactions between i and j atoms */
 375             qq00             = _mm_mul_ps(iq0,jq0);
 376             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 377                                          vdwparam+vdwioffset0+vdwjidx0B,
 378                                          vdwparam+vdwioffset0+vdwjidx0C,
 379                                          vdwparam+vdwioffset0+vdwjidx0D,
 380                                          &c6_00,&c12_00);
 381
 382             /* EWALD ELECTROSTATICS */
 383
 384             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 385             ewrt             = _mm_mul_ps(r00,ewtabscale);
 386             ewitab           = _mm_cvttps_epi32(ewrt);
 387             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 388             ewitab           = _mm_slli_epi32(ewitab,2);
 389             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 390             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 391             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 392             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 393             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 394             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 395             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 396             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 397             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 398
 399             /* LENNARD-JONES DISPERSION/REPULSION */
 400
 401             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 402             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 403             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 404             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 405             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 406
 407             d                = _mm_sub_ps(r00,rswitch);
 408             d                = _mm_max_ps(d,_mm_setzero_ps());
 409             d2               = _mm_mul_ps(d,d);
 410             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 411
 412             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 413
 414             /* Evaluate switch function */
 415             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 416             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 417             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 418             velec            = _mm_mul_ps(velec,sw);
 419             vvdw             = _mm_mul_ps(vvdw,sw);
 420             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velec            = _mm_and_ps(velec,cutoff_mask);
 424             velec            = _mm_andnot_ps(dummy_mask,velec);
 425             velecsum         = _mm_add_ps(velecsum,velec);
 426             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 427             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 428             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 429
 430             fscal            = _mm_add_ps(felec,fvdw);
 431
 432             fscal            = _mm_and_ps(fscal,cutoff_mask);
 433
 434             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm_mul_ps(fscal,dx00);
 438             ty               = _mm_mul_ps(fscal,dy00);
 439             tz               = _mm_mul_ps(fscal,dz00);
 440
 441             /* Update vectorial force */
 442             fix0             = _mm_add_ps(fix0,tx);
 443             fiy0             = _mm_add_ps(fiy0,ty);
 444             fiz0             = _mm_add_ps(fiz0,tz);
 445
 446             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 447             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 448             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 449             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 450             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 451
 452             }
 453
 454             /* Inner loop uses 84 flops */
 455         }
 456
 457         /* End of innermost loop */
 458
 459         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 460                                               f+i_coord_offset,fshift+i_shift_offset);
 461
 462         ggid                        = gid[iidx];
 463         /* Update potential energies */
 464         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 465         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 466
 467         /* Increment number of inner iterations */
 468         inneriter                  += j_index_end - j_index_start;
 469
 470         /* Outer loop uses 9 flops */
 471     }
 472
 473     /* Increment number of outer iterations */
 474     outeriter        += nri;
 475
 476     /* Update outer/inner flops */
 477
 478     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*84);
 479 }
 480 /*
 481  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse2_single
 482  * Electrostatics interaction: Ewald
 483  * VdW interaction:            LennardJones
 484  * Geometry:                   Particle-Particle
 485  * Calculate force/pot:        Force
 486  */
 487 void
 488 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse2_single
 489                     (t_nblist                    * gmx_restrict       nlist,
 490                      rvec                        * gmx_restrict          xx,
 491                      rvec                        * gmx_restrict          ff,
 492                      t_forcerec                  * gmx_restrict          fr,
 493                      t_mdatoms                   * gmx_restrict     mdatoms,
 494                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 495                      t_nrnb                      * gmx_restrict        nrnb)
 496 {
 497     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 498      * just 0 for non-waters.
 499      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 500      * jnr indices corresponding to data put in the four positions in the SIMD register.
 501      */
 502     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 503     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 504     int              jnrA,jnrB,jnrC,jnrD;
 505     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 506     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 507     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 508     real             rcutoff_scalar;
 509     real             *shiftvec,*fshift,*x,*f;
 510     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 511     real             scratch[4*DIM];
 512     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 513     int              vdwioffset0;
 514     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 515     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 516     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 517     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 518     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 519     real             *charge;
 520     int              nvdwtype;
 521     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 522     int              *vdwtype;
 523     real             *vdwparam;
 524     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 525     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 526     __m128i          ewitab;
 527     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 528     real             *ewtab;
 529     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 530     real             rswitch_scalar,d_scalar;
 531     __m128           dummy_mask,cutoff_mask;
 532     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 533     __m128           one     = _mm_set1_ps(1.0);
 534     __m128           two     = _mm_set1_ps(2.0);
 535     x                = xx[0];
 536     f                = ff[0];
 537
 538     nri              = nlist->nri;
 539     iinr             = nlist->iinr;
 540     jindex           = nlist->jindex;
 541     jjnr             = nlist->jjnr;
 542     shiftidx         = nlist->shift;
 543     gid              = nlist->gid;
 544     shiftvec         = fr->shift_vec[0];
 545     fshift           = fr->fshift[0];
 546     facel            = _mm_set1_ps(fr->epsfac);
 547     charge           = mdatoms->chargeA;
 548     nvdwtype         = fr->ntype;
 549     vdwparam         = fr->nbfp;
 550     vdwtype          = mdatoms->typeA;
 551
 552     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 553     ewtab            = fr->ic->tabq_coul_FDV0;
 554     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 555     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 556
 557     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 558     rcutoff_scalar   = fr->rcoulomb;
 559     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 560     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 561
 562     rswitch_scalar   = fr->rcoulomb_switch;
 563     rswitch          = _mm_set1_ps(rswitch_scalar);
 564     /* Setup switch parameters */
 565     d_scalar         = rcutoff_scalar-rswitch_scalar;
 566     d                = _mm_set1_ps(d_scalar);
 567     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 568     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 569     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 570     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 571     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 572     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 573
 574     /* Avoid stupid compiler warnings */
 575     jnrA = jnrB = jnrC = jnrD = 0;
 576     j_coord_offsetA = 0;
 577     j_coord_offsetB = 0;
 578     j_coord_offsetC = 0;
 579     j_coord_offsetD = 0;
 580
 581     outeriter        = 0;
 582     inneriter        = 0;
 583
 584     for(iidx=0;iidx<4*DIM;iidx++)
 585     {
 586         scratch[iidx] = 0.0;
 587     }
 588
 589     /* Start outer loop over neighborlists */
 590     for(iidx=0; iidx<nri; iidx++)
 591     {
 592         /* Load shift vector for this list */
 593         i_shift_offset   = DIM*shiftidx[iidx];
 594
 595         /* Load limits for loop over neighbors */
 596         j_index_start    = jindex[iidx];
 597         j_index_end      = jindex[iidx+1];
 598
 599         /* Get outer coordinate index */
 600         inr              = iinr[iidx];
 601         i_coord_offset   = DIM*inr;
 602
 603         /* Load i particle coords and add shift vector */
 604         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 605
 606         fix0             = _mm_setzero_ps();
 607         fiy0             = _mm_setzero_ps();
 608         fiz0             = _mm_setzero_ps();
 609
 610         /* Load parameters for i particles */
 611         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 612         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 613
 614         /* Start inner kernel loop */
 615         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 616         {
 617
 618             /* Get j neighbor index, and coordinate index */
 619             jnrA             = jjnr[jidx];
 620             jnrB             = jjnr[jidx+1];
 621             jnrC             = jjnr[jidx+2];
 622             jnrD             = jjnr[jidx+3];
 623             j_coord_offsetA  = DIM*jnrA;
 624             j_coord_offsetB  = DIM*jnrB;
 625             j_coord_offsetC  = DIM*jnrC;
 626             j_coord_offsetD  = DIM*jnrD;
 627
 628             /* load j atom coordinates */
 629             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 630                                               x+j_coord_offsetC,x+j_coord_offsetD,
 631                                               &jx0,&jy0,&jz0);
 632
 633             /* Calculate displacement vector */
 634             dx00             = _mm_sub_ps(ix0,jx0);
 635             dy00             = _mm_sub_ps(iy0,jy0);
 636             dz00             = _mm_sub_ps(iz0,jz0);
 637
 638             /* Calculate squared distance and things based on it */
 639             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 640
 641             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 642
 643             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 644
 645             /* Load parameters for j particles */
 646             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 647                                                               charge+jnrC+0,charge+jnrD+0);
 648             vdwjidx0A        = 2*vdwtype[jnrA+0];
 649             vdwjidx0B        = 2*vdwtype[jnrB+0];
 650             vdwjidx0C        = 2*vdwtype[jnrC+0];
 651             vdwjidx0D        = 2*vdwtype[jnrD+0];
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             if (gmx_mm_any_lt(rsq00,rcutoff2))
 658             {
 659
 660             r00              = _mm_mul_ps(rsq00,rinv00);
 661
 662             /* Compute parameters for interactions between i and j atoms */
 663             qq00             = _mm_mul_ps(iq0,jq0);
 664             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 665                                          vdwparam+vdwioffset0+vdwjidx0B,
 666                                          vdwparam+vdwioffset0+vdwjidx0C,
 667                                          vdwparam+vdwioffset0+vdwjidx0D,
 668                                          &c6_00,&c12_00);
 669
 670             /* EWALD ELECTROSTATICS */
 671
 672             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 673             ewrt             = _mm_mul_ps(r00,ewtabscale);
 674             ewitab           = _mm_cvttps_epi32(ewrt);
 675             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 676             ewitab           = _mm_slli_epi32(ewitab,2);
 677             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 678             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 679             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 680             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 681             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 682             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 683             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 684             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 685             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 686
 687             /* LENNARD-JONES DISPERSION/REPULSION */
 688
 689             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 690             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 691             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 692             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 693             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 694
 695             d                = _mm_sub_ps(r00,rswitch);
 696             d                = _mm_max_ps(d,_mm_setzero_ps());
 697             d2               = _mm_mul_ps(d,d);
 698             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 699
 700             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 701
 702             /* Evaluate switch function */
 703             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 704             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 705             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 706             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 707
 708             fscal            = _mm_add_ps(felec,fvdw);
 709
 710             fscal            = _mm_and_ps(fscal,cutoff_mask);
 711
 712             /* Calculate temporary vectorial force */
 713             tx               = _mm_mul_ps(fscal,dx00);
 714             ty               = _mm_mul_ps(fscal,dy00);
 715             tz               = _mm_mul_ps(fscal,dz00);
 716
 717             /* Update vectorial force */
 718             fix0             = _mm_add_ps(fix0,tx);
 719             fiy0             = _mm_add_ps(fiy0,ty);
 720             fiz0             = _mm_add_ps(fiz0,tz);
 721
 722             fjptrA             = f+j_coord_offsetA;
 723             fjptrB             = f+j_coord_offsetB;
 724             fjptrC             = f+j_coord_offsetC;
 725             fjptrD             = f+j_coord_offsetD;
 726             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 727
 728             }
 729
 730             /* Inner loop uses 77 flops */
 731         }
 732
 733         if(jidx<j_index_end)
 734         {
 735
 736             /* Get j neighbor index, and coordinate index */
 737             jnrlistA         = jjnr[jidx];
 738             jnrlistB         = jjnr[jidx+1];
 739             jnrlistC         = jjnr[jidx+2];
 740             jnrlistD         = jjnr[jidx+3];
 741             /* Sign of each element will be negative for non-real atoms.
 742              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 743              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 744              */
 745             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 746             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 747             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 748             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 749             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 750             j_coord_offsetA  = DIM*jnrA;
 751             j_coord_offsetB  = DIM*jnrB;
 752             j_coord_offsetC  = DIM*jnrC;
 753             j_coord_offsetD  = DIM*jnrD;
 754
 755             /* load j atom coordinates */
 756             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 757                                               x+j_coord_offsetC,x+j_coord_offsetD,
 758                                               &jx0,&jy0,&jz0);
 759
 760             /* Calculate displacement vector */
 761             dx00             = _mm_sub_ps(ix0,jx0);
 762             dy00             = _mm_sub_ps(iy0,jy0);
 763             dz00             = _mm_sub_ps(iz0,jz0);
 764
 765             /* Calculate squared distance and things based on it */
 766             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 767
 768             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 769
 770             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 771
 772             /* Load parameters for j particles */
 773             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 774                                                               charge+jnrC+0,charge+jnrD+0);
 775             vdwjidx0A        = 2*vdwtype[jnrA+0];
 776             vdwjidx0B        = 2*vdwtype[jnrB+0];
 777             vdwjidx0C        = 2*vdwtype[jnrC+0];
 778             vdwjidx0D        = 2*vdwtype[jnrD+0];
 779
 780             /**************************
 781              * CALCULATE INTERACTIONS *
 782              **************************/
 783
 784             if (gmx_mm_any_lt(rsq00,rcutoff2))
 785             {
 786
 787             r00              = _mm_mul_ps(rsq00,rinv00);
 788             r00              = _mm_andnot_ps(dummy_mask,r00);
 789
 790             /* Compute parameters for interactions between i and j atoms */
 791             qq00             = _mm_mul_ps(iq0,jq0);
 792             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 793                                          vdwparam+vdwioffset0+vdwjidx0B,
 794                                          vdwparam+vdwioffset0+vdwjidx0C,
 795                                          vdwparam+vdwioffset0+vdwjidx0D,
 796                                          &c6_00,&c12_00);
 797
 798             /* EWALD ELECTROSTATICS */
 799
 800             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 801             ewrt             = _mm_mul_ps(r00,ewtabscale);
 802             ewitab           = _mm_cvttps_epi32(ewrt);
 803             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 804             ewitab           = _mm_slli_epi32(ewitab,2);
 805             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 806             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 807             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 808             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 809             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 810             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 811             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 812             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 813             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 814
 815             /* LENNARD-JONES DISPERSION/REPULSION */
 816
 817             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 818             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 819             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 820             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 821             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 822
 823             d                = _mm_sub_ps(r00,rswitch);
 824             d                = _mm_max_ps(d,_mm_setzero_ps());
 825             d2               = _mm_mul_ps(d,d);
 826             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 827
 828             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 829
 830             /* Evaluate switch function */
 831             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 832             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 833             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 834             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 835
 836             fscal            = _mm_add_ps(felec,fvdw);
 837
 838             fscal            = _mm_and_ps(fscal,cutoff_mask);
 839
 840             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 841
 842             /* Calculate temporary vectorial force */
 843             tx               = _mm_mul_ps(fscal,dx00);
 844             ty               = _mm_mul_ps(fscal,dy00);
 845             tz               = _mm_mul_ps(fscal,dz00);
 846
 847             /* Update vectorial force */
 848             fix0             = _mm_add_ps(fix0,tx);
 849             fiy0             = _mm_add_ps(fiy0,ty);
 850             fiz0             = _mm_add_ps(fiz0,tz);
 851
 852             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 853             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 854             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 855             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 856             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 857
 858             }
 859
 860             /* Inner loop uses 78 flops */
 861         }
 862
 863         /* End of innermost loop */
 864
 865         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 866                                               f+i_coord_offset,fshift+i_shift_offset);
 867
 868         /* Increment number of inner iterations */
 869         inneriter                  += j_index_end - j_index_start;
 870
 871         /* Outer loop uses 7 flops */
 872     }
 873
 874     /* Increment number of outer iterations */
 875     outeriter        += nri;
 876
 877     /* Update outer/inner flops */
 878
 879     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*78);
 880 }