src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSw_VdwLJSw_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128i          ewitab;
  97     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     real             *ewtab;
  99     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 100     real             rswitch_scalar,d_scalar;
 101     __m128           dummy_mask,cutoff_mask;
 102     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 103     __m128           one     = _mm_set1_ps(1.0);
 104     __m128           two     = _mm_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 126
 127     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 131
 132     rswitch_scalar   = fr->rcoulomb_switch;
 133     rswitch          = _mm_set1_ps(rswitch_scalar);
 134     /* Setup switch parameters */
 135     d_scalar         = rcutoff_scalar-rswitch_scalar;
 136     d                = _mm_set1_ps(d_scalar);
 137     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 138     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 141     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 142     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179
 180         /* Load parameters for i particles */
 181         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 182         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_ps();
 186         vvdwsum          = _mm_setzero_ps();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 214
 215             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 216
 217             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                               charge+jnrC+0,charge+jnrD+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224             vdwjidx0C        = 2*vdwtype[jnrC+0];
 225             vdwjidx0D        = 2*vdwtype[jnrD+0];
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             if (gmx_mm_any_lt(rsq00,rcutoff2))
 232             {
 233
 234             r00              = _mm_mul_ps(rsq00,rinv00);
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq00             = _mm_mul_ps(iq0,jq0);
 238             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,
 240                                          vdwparam+vdwioffset0+vdwjidx0C,
 241                                          vdwparam+vdwioffset0+vdwjidx0D,
 242                                          &c6_00,&c12_00);
 243
 244             /* EWALD ELECTROSTATICS */
 245
 246             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 247             ewrt             = _mm_mul_ps(r00,ewtabscale);
 248             ewitab           = _mm_cvttps_epi32(ewrt);
 249             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 250             ewitab           = _mm_slli_epi32(ewitab,2);
 251             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 252             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 253             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 254             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 255             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 256             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 257             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 258             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 259             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 265             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 266             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 267             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 268
 269             d                = _mm_sub_ps(r00,rswitch);
 270             d                = _mm_max_ps(d,_mm_setzero_ps());
 271             d2               = _mm_mul_ps(d,d);
 272             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)))))));
 273
 274             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 275
 276             /* Evaluate switch function */
 277             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 278             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 279             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 280             velec            = _mm_mul_ps(velec,sw);
 281             vvdw             = _mm_mul_ps(vvdw,sw);
 282             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velec            = _mm_and_ps(velec,cutoff_mask);
 286             velecsum         = _mm_add_ps(velecsum,velec);
 287             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 288             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 289
 290             fscal            = _mm_add_ps(felec,fvdw);
 291
 292             fscal            = _mm_and_ps(fscal,cutoff_mask);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm_mul_ps(fscal,dx00);
 296             ty               = _mm_mul_ps(fscal,dy00);
 297             tz               = _mm_mul_ps(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm_add_ps(fix0,tx);
 301             fiy0             = _mm_add_ps(fiy0,ty);
 302             fiz0             = _mm_add_ps(fiz0,tz);
 303
 304             fjptrA             = f+j_coord_offsetA;
 305             fjptrB             = f+j_coord_offsetB;
 306             fjptrC             = f+j_coord_offsetC;
 307             fjptrD             = f+j_coord_offsetD;
 308             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 309
 310             }
 311
 312             /* Inner loop uses 83 flops */
 313         }
 314
 315         if(jidx<j_index_end)
 316         {
 317
 318             /* Get j neighbor index, and coordinate index */
 319             jnrlistA         = jjnr[jidx];
 320             jnrlistB         = jjnr[jidx+1];
 321             jnrlistC         = jjnr[jidx+2];
 322             jnrlistD         = jjnr[jidx+3];
 323             /* Sign of each element will be negative for non-real atoms.
 324              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 325              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 326              */
 327             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 328             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 329             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 330             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 331             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 332             j_coord_offsetA  = DIM*jnrA;
 333             j_coord_offsetB  = DIM*jnrB;
 334             j_coord_offsetC  = DIM*jnrC;
 335             j_coord_offsetD  = DIM*jnrD;
 336
 337             /* load j atom coordinates */
 338             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 339                                               x+j_coord_offsetC,x+j_coord_offsetD,
 340                                               &jx0,&jy0,&jz0);
 341
 342             /* Calculate displacement vector */
 343             dx00             = _mm_sub_ps(ix0,jx0);
 344             dy00             = _mm_sub_ps(iy0,jy0);
 345             dz00             = _mm_sub_ps(iz0,jz0);
 346
 347             /* Calculate squared distance and things based on it */
 348             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 349
 350             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 351
 352             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 353
 354             /* Load parameters for j particles */
 355             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 356                                                               charge+jnrC+0,charge+jnrD+0);
 357             vdwjidx0A        = 2*vdwtype[jnrA+0];
 358             vdwjidx0B        = 2*vdwtype[jnrB+0];
 359             vdwjidx0C        = 2*vdwtype[jnrC+0];
 360             vdwjidx0D        = 2*vdwtype[jnrD+0];
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm_any_lt(rsq00,rcutoff2))
 367             {
 368
 369             r00              = _mm_mul_ps(rsq00,rinv00);
 370             r00              = _mm_andnot_ps(dummy_mask,r00);
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq00             = _mm_mul_ps(iq0,jq0);
 374             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 375                                          vdwparam+vdwioffset0+vdwjidx0B,
 376                                          vdwparam+vdwioffset0+vdwjidx0C,
 377                                          vdwparam+vdwioffset0+vdwjidx0D,
 378                                          &c6_00,&c12_00);
 379
 380             /* EWALD ELECTROSTATICS */
 381
 382             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 383             ewrt             = _mm_mul_ps(r00,ewtabscale);
 384             ewitab           = _mm_cvttps_epi32(ewrt);
 385             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 386             ewitab           = _mm_slli_epi32(ewitab,2);
 387             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 388             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 389             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 390             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 391             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 392             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 393             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 394             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 395             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 396
 397             /* LENNARD-JONES DISPERSION/REPULSION */
 398
 399             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 400             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 401             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 402             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 403             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 404
 405             d                = _mm_sub_ps(r00,rswitch);
 406             d                = _mm_max_ps(d,_mm_setzero_ps());
 407             d2               = _mm_mul_ps(d,d);
 408             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)))))));
 409
 410             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 411
 412             /* Evaluate switch function */
 413             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 414             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 415             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 416             velec            = _mm_mul_ps(velec,sw);
 417             vvdw             = _mm_mul_ps(vvdw,sw);
 418             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velec            = _mm_and_ps(velec,cutoff_mask);
 422             velec            = _mm_andnot_ps(dummy_mask,velec);
 423             velecsum         = _mm_add_ps(velecsum,velec);
 424             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 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_and_ps(fscal,cutoff_mask);
 431
 432             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm_mul_ps(fscal,dx00);
 436             ty               = _mm_mul_ps(fscal,dy00);
 437             tz               = _mm_mul_ps(fscal,dz00);
 438
 439             /* Update vectorial force */
 440             fix0             = _mm_add_ps(fix0,tx);
 441             fiy0             = _mm_add_ps(fiy0,ty);
 442             fiz0             = _mm_add_ps(fiz0,tz);
 443
 444             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 445             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 446             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 447             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 448             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 449
 450             }
 451
 452             /* Inner loop uses 84 flops */
 453         }
 454
 455         /* End of innermost loop */
 456
 457         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 458                                               f+i_coord_offset,fshift+i_shift_offset);
 459
 460         ggid                        = gid[iidx];
 461         /* Update potential energies */
 462         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 463         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 464
 465         /* Increment number of inner iterations */
 466         inneriter                  += j_index_end - j_index_start;
 467
 468         /* Outer loop uses 9 flops */
 469     }
 470
 471     /* Increment number of outer iterations */
 472     outeriter        += nri;
 473
 474     /* Update outer/inner flops */
 475
 476     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*84);
 477 }
 478 /*
 479  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse4_1_single
 480  * Electrostatics interaction: Ewald
 481  * VdW interaction:            LennardJones
 482  * Geometry:                   Particle-Particle
 483  * Calculate force/pot:        Force
 484  */
 485 void
 486 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse4_1_single
 487                     (t_nblist                    * gmx_restrict       nlist,
 488                      rvec                        * gmx_restrict          xx,
 489                      rvec                        * gmx_restrict          ff,
 490                      t_forcerec                  * gmx_restrict          fr,
 491                      t_mdatoms                   * gmx_restrict     mdatoms,
 492                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 493                      t_nrnb                      * gmx_restrict        nrnb)
 494 {
 495     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 496      * just 0 for non-waters.
 497      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 498      * jnr indices corresponding to data put in the four positions in the SIMD register.
 499      */
 500     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 501     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 502     int              jnrA,jnrB,jnrC,jnrD;
 503     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 504     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 505     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 506     real             rcutoff_scalar;
 507     real             *shiftvec,*fshift,*x,*f;
 508     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 509     real             scratch[4*DIM];
 510     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 511     int              vdwioffset0;
 512     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 513     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 514     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 515     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 516     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 517     real             *charge;
 518     int              nvdwtype;
 519     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 520     int              *vdwtype;
 521     real             *vdwparam;
 522     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 523     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 524     __m128i          ewitab;
 525     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 526     real             *ewtab;
 527     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 528     real             rswitch_scalar,d_scalar;
 529     __m128           dummy_mask,cutoff_mask;
 530     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 531     __m128           one     = _mm_set1_ps(1.0);
 532     __m128           two     = _mm_set1_ps(2.0);
 533     x                = xx[0];
 534     f                = ff[0];
 535
 536     nri              = nlist->nri;
 537     iinr             = nlist->iinr;
 538     jindex           = nlist->jindex;
 539     jjnr             = nlist->jjnr;
 540     shiftidx         = nlist->shift;
 541     gid              = nlist->gid;
 542     shiftvec         = fr->shift_vec[0];
 543     fshift           = fr->fshift[0];
 544     facel            = _mm_set1_ps(fr->epsfac);
 545     charge           = mdatoms->chargeA;
 546     nvdwtype         = fr->ntype;
 547     vdwparam         = fr->nbfp;
 548     vdwtype          = mdatoms->typeA;
 549
 550     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 551     ewtab            = fr->ic->tabq_coul_FDV0;
 552     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 553     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 554
 555     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 556     rcutoff_scalar   = fr->rcoulomb;
 557     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 558     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 559
 560     rswitch_scalar   = fr->rcoulomb_switch;
 561     rswitch          = _mm_set1_ps(rswitch_scalar);
 562     /* Setup switch parameters */
 563     d_scalar         = rcutoff_scalar-rswitch_scalar;
 564     d                = _mm_set1_ps(d_scalar);
 565     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 566     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 567     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 568     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 569     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 570     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 571
 572     /* Avoid stupid compiler warnings */
 573     jnrA = jnrB = jnrC = jnrD = 0;
 574     j_coord_offsetA = 0;
 575     j_coord_offsetB = 0;
 576     j_coord_offsetC = 0;
 577     j_coord_offsetD = 0;
 578
 579     outeriter        = 0;
 580     inneriter        = 0;
 581
 582     for(iidx=0;iidx<4*DIM;iidx++)
 583     {
 584         scratch[iidx] = 0.0;
 585     }
 586
 587     /* Start outer loop over neighborlists */
 588     for(iidx=0; iidx<nri; iidx++)
 589     {
 590         /* Load shift vector for this list */
 591         i_shift_offset   = DIM*shiftidx[iidx];
 592
 593         /* Load limits for loop over neighbors */
 594         j_index_start    = jindex[iidx];
 595         j_index_end      = jindex[iidx+1];
 596
 597         /* Get outer coordinate index */
 598         inr              = iinr[iidx];
 599         i_coord_offset   = DIM*inr;
 600
 601         /* Load i particle coords and add shift vector */
 602         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 603
 604         fix0             = _mm_setzero_ps();
 605         fiy0             = _mm_setzero_ps();
 606         fiz0             = _mm_setzero_ps();
 607
 608         /* Load parameters for i particles */
 609         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 610         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 611
 612         /* Start inner kernel loop */
 613         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 614         {
 615
 616             /* Get j neighbor index, and coordinate index */
 617             jnrA             = jjnr[jidx];
 618             jnrB             = jjnr[jidx+1];
 619             jnrC             = jjnr[jidx+2];
 620             jnrD             = jjnr[jidx+3];
 621             j_coord_offsetA  = DIM*jnrA;
 622             j_coord_offsetB  = DIM*jnrB;
 623             j_coord_offsetC  = DIM*jnrC;
 624             j_coord_offsetD  = DIM*jnrD;
 625
 626             /* load j atom coordinates */
 627             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 628                                               x+j_coord_offsetC,x+j_coord_offsetD,
 629                                               &jx0,&jy0,&jz0);
 630
 631             /* Calculate displacement vector */
 632             dx00             = _mm_sub_ps(ix0,jx0);
 633             dy00             = _mm_sub_ps(iy0,jy0);
 634             dz00             = _mm_sub_ps(iz0,jz0);
 635
 636             /* Calculate squared distance and things based on it */
 637             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 638
 639             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 640
 641             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 642
 643             /* Load parameters for j particles */
 644             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 645                                                               charge+jnrC+0,charge+jnrD+0);
 646             vdwjidx0A        = 2*vdwtype[jnrA+0];
 647             vdwjidx0B        = 2*vdwtype[jnrB+0];
 648             vdwjidx0C        = 2*vdwtype[jnrC+0];
 649             vdwjidx0D        = 2*vdwtype[jnrD+0];
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             if (gmx_mm_any_lt(rsq00,rcutoff2))
 656             {
 657
 658             r00              = _mm_mul_ps(rsq00,rinv00);
 659
 660             /* Compute parameters for interactions between i and j atoms */
 661             qq00             = _mm_mul_ps(iq0,jq0);
 662             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 663                                          vdwparam+vdwioffset0+vdwjidx0B,
 664                                          vdwparam+vdwioffset0+vdwjidx0C,
 665                                          vdwparam+vdwioffset0+vdwjidx0D,
 666                                          &c6_00,&c12_00);
 667
 668             /* EWALD ELECTROSTATICS */
 669
 670             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 671             ewrt             = _mm_mul_ps(r00,ewtabscale);
 672             ewitab           = _mm_cvttps_epi32(ewrt);
 673             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 674             ewitab           = _mm_slli_epi32(ewitab,2);
 675             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 676             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 677             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 678             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 679             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 680             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 681             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 682             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 683             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 684
 685             /* LENNARD-JONES DISPERSION/REPULSION */
 686
 687             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 688             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 689             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 690             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 691             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 692
 693             d                = _mm_sub_ps(r00,rswitch);
 694             d                = _mm_max_ps(d,_mm_setzero_ps());
 695             d2               = _mm_mul_ps(d,d);
 696             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)))))));
 697
 698             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 699
 700             /* Evaluate switch function */
 701             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 702             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 703             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 704             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 705
 706             fscal            = _mm_add_ps(felec,fvdw);
 707
 708             fscal            = _mm_and_ps(fscal,cutoff_mask);
 709
 710             /* Calculate temporary vectorial force */
 711             tx               = _mm_mul_ps(fscal,dx00);
 712             ty               = _mm_mul_ps(fscal,dy00);
 713             tz               = _mm_mul_ps(fscal,dz00);
 714
 715             /* Update vectorial force */
 716             fix0             = _mm_add_ps(fix0,tx);
 717             fiy0             = _mm_add_ps(fiy0,ty);
 718             fiz0             = _mm_add_ps(fiz0,tz);
 719
 720             fjptrA             = f+j_coord_offsetA;
 721             fjptrB             = f+j_coord_offsetB;
 722             fjptrC             = f+j_coord_offsetC;
 723             fjptrD             = f+j_coord_offsetD;
 724             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 725
 726             }
 727
 728             /* Inner loop uses 77 flops */
 729         }
 730
 731         if(jidx<j_index_end)
 732         {
 733
 734             /* Get j neighbor index, and coordinate index */
 735             jnrlistA         = jjnr[jidx];
 736             jnrlistB         = jjnr[jidx+1];
 737             jnrlistC         = jjnr[jidx+2];
 738             jnrlistD         = jjnr[jidx+3];
 739             /* Sign of each element will be negative for non-real atoms.
 740              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 741              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 742              */
 743             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 744             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 745             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 746             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 747             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 748             j_coord_offsetA  = DIM*jnrA;
 749             j_coord_offsetB  = DIM*jnrB;
 750             j_coord_offsetC  = DIM*jnrC;
 751             j_coord_offsetD  = DIM*jnrD;
 752
 753             /* load j atom coordinates */
 754             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 755                                               x+j_coord_offsetC,x+j_coord_offsetD,
 756                                               &jx0,&jy0,&jz0);
 757
 758             /* Calculate displacement vector */
 759             dx00             = _mm_sub_ps(ix0,jx0);
 760             dy00             = _mm_sub_ps(iy0,jy0);
 761             dz00             = _mm_sub_ps(iz0,jz0);
 762
 763             /* Calculate squared distance and things based on it */
 764             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 765
 766             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 767
 768             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 769
 770             /* Load parameters for j particles */
 771             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 772                                                               charge+jnrC+0,charge+jnrD+0);
 773             vdwjidx0A        = 2*vdwtype[jnrA+0];
 774             vdwjidx0B        = 2*vdwtype[jnrB+0];
 775             vdwjidx0C        = 2*vdwtype[jnrC+0];
 776             vdwjidx0D        = 2*vdwtype[jnrD+0];
 777
 778             /**************************
 779              * CALCULATE INTERACTIONS *
 780              **************************/
 781
 782             if (gmx_mm_any_lt(rsq00,rcutoff2))
 783             {
 784
 785             r00              = _mm_mul_ps(rsq00,rinv00);
 786             r00              = _mm_andnot_ps(dummy_mask,r00);
 787
 788             /* Compute parameters for interactions between i and j atoms */
 789             qq00             = _mm_mul_ps(iq0,jq0);
 790             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 791                                          vdwparam+vdwioffset0+vdwjidx0B,
 792                                          vdwparam+vdwioffset0+vdwjidx0C,
 793                                          vdwparam+vdwioffset0+vdwjidx0D,
 794                                          &c6_00,&c12_00);
 795
 796             /* EWALD ELECTROSTATICS */
 797
 798             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 799             ewrt             = _mm_mul_ps(r00,ewtabscale);
 800             ewitab           = _mm_cvttps_epi32(ewrt);
 801             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 802             ewitab           = _mm_slli_epi32(ewitab,2);
 803             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 804             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 805             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 806             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 807             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 808             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 809             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 810             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 811             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 812
 813             /* LENNARD-JONES DISPERSION/REPULSION */
 814
 815             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 816             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 817             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 818             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 819             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 820
 821             d                = _mm_sub_ps(r00,rswitch);
 822             d                = _mm_max_ps(d,_mm_setzero_ps());
 823             d2               = _mm_mul_ps(d,d);
 824             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)))))));
 825
 826             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 827
 828             /* Evaluate switch function */
 829             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 830             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 831             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 832             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 833
 834             fscal            = _mm_add_ps(felec,fvdw);
 835
 836             fscal            = _mm_and_ps(fscal,cutoff_mask);
 837
 838             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 839
 840             /* Calculate temporary vectorial force */
 841             tx               = _mm_mul_ps(fscal,dx00);
 842             ty               = _mm_mul_ps(fscal,dy00);
 843             tz               = _mm_mul_ps(fscal,dz00);
 844
 845             /* Update vectorial force */
 846             fix0             = _mm_add_ps(fix0,tx);
 847             fiy0             = _mm_add_ps(fiy0,ty);
 848             fiz0             = _mm_add_ps(fiz0,tz);
 849
 850             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 851             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 852             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 853             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 854             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 855
 856             }
 857
 858             /* Inner loop uses 78 flops */
 859         }
 860
 861         /* End of innermost loop */
 862
 863         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 864                                               f+i_coord_offset,fshift+i_shift_offset);
 865
 866         /* Increment number of inner iterations */
 867         inneriter                  += j_index_end - j_index_start;
 868
 869         /* Outer loop uses 7 flops */
 870     }
 871
 872     /* Increment number of outer iterations */
 873     outeriter        += nri;
 874
 875     /* Update outer/inner flops */
 876
 877     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*78);
 878 }