src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSw_VdwNone_GeomW4P1_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_VdwNone_GeomW4P1_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_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              vdwioffset1;
  84     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  85     int              vdwioffset2;
  86     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  87     int              vdwioffset3;
  88     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  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
 119     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 127     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 128     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 129
 130     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 131     rcutoff_scalar   = fr->rcoulomb;
 132     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 133     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 134
 135     rswitch_scalar   = fr->rcoulomb_switch;
 136     rswitch          = _mm_set1_ps(rswitch_scalar);
 137     /* Setup switch parameters */
 138     d_scalar         = rcutoff_scalar-rswitch_scalar;
 139     d                = _mm_set1_ps(d_scalar);
 140     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 141     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 142     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 143     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 144     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 145     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 178                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 179
 180         fix1             = _mm_setzero_ps();
 181         fiy1             = _mm_setzero_ps();
 182         fiz1             = _mm_setzero_ps();
 183         fix2             = _mm_setzero_ps();
 184         fiy2             = _mm_setzero_ps();
 185         fiz2             = _mm_setzero_ps();
 186         fix3             = _mm_setzero_ps();
 187         fiy3             = _mm_setzero_ps();
 188         fiz3             = _mm_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_ps();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             jnrC             = jjnr[jidx+2];
 201             jnrD             = jjnr[jidx+3];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206
 207             /* load j atom coordinates */
 208             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 209                                               x+j_coord_offsetC,x+j_coord_offsetD,
 210                                               &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219             dx30             = _mm_sub_ps(ix3,jx0);
 220             dy30             = _mm_sub_ps(iy3,jy0);
 221             dz30             = _mm_sub_ps(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 227
 228             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 229             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 230             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 231
 232             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 234             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 238                                                               charge+jnrC+0,charge+jnrD+0);
 239
 240             fjx0             = _mm_setzero_ps();
 241             fjy0             = _mm_setzero_ps();
 242             fjz0             = _mm_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             if (gmx_mm_any_lt(rsq10,rcutoff2))
 249             {
 250
 251             r10              = _mm_mul_ps(rsq10,rinv10);
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             qq10             = _mm_mul_ps(iq1,jq0);
 255
 256             /* EWALD ELECTROSTATICS */
 257
 258             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 259             ewrt             = _mm_mul_ps(r10,ewtabscale);
 260             ewitab           = _mm_cvttps_epi32(ewrt);
 261             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 262             ewitab           = _mm_slli_epi32(ewitab,2);
 263             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 264             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 265             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 266             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 267             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 268             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 269             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 270             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 271             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 272
 273             d                = _mm_sub_ps(r10,rswitch);
 274             d                = _mm_max_ps(d,_mm_setzero_ps());
 275             d2               = _mm_mul_ps(d,d);
 276             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)))))));
 277
 278             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 279
 280             /* Evaluate switch function */
 281             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 282             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 283             velec            = _mm_mul_ps(velec,sw);
 284             cutoff_mask      = _mm_cmplt_ps(rsq10,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
 290             fscal            = felec;
 291
 292             fscal            = _mm_and_ps(fscal,cutoff_mask);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm_mul_ps(fscal,dx10);
 296             ty               = _mm_mul_ps(fscal,dy10);
 297             tz               = _mm_mul_ps(fscal,dz10);
 298
 299             /* Update vectorial force */
 300             fix1             = _mm_add_ps(fix1,tx);
 301             fiy1             = _mm_add_ps(fiy1,ty);
 302             fiz1             = _mm_add_ps(fiz1,tz);
 303
 304             fjx0             = _mm_add_ps(fjx0,tx);
 305             fjy0             = _mm_add_ps(fjy0,ty);
 306             fjz0             = _mm_add_ps(fjz0,tz);
 307
 308             }
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             if (gmx_mm_any_lt(rsq20,rcutoff2))
 315             {
 316
 317             r20              = _mm_mul_ps(rsq20,rinv20);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq20             = _mm_mul_ps(iq2,jq0);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 325             ewrt             = _mm_mul_ps(r20,ewtabscale);
 326             ewitab           = _mm_cvttps_epi32(ewrt);
 327             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 328             ewitab           = _mm_slli_epi32(ewitab,2);
 329             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 330             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 331             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 332             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 333             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 334             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 335             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 336             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 337             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 338
 339             d                = _mm_sub_ps(r20,rswitch);
 340             d                = _mm_max_ps(d,_mm_setzero_ps());
 341             d2               = _mm_mul_ps(d,d);
 342             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)))))));
 343
 344             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 345
 346             /* Evaluate switch function */
 347             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 348             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 349             velec            = _mm_mul_ps(velec,sw);
 350             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm_and_ps(velec,cutoff_mask);
 354             velecsum         = _mm_add_ps(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             fscal            = _mm_and_ps(fscal,cutoff_mask);
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm_mul_ps(fscal,dx20);
 362             ty               = _mm_mul_ps(fscal,dy20);
 363             tz               = _mm_mul_ps(fscal,dz20);
 364
 365             /* Update vectorial force */
 366             fix2             = _mm_add_ps(fix2,tx);
 367             fiy2             = _mm_add_ps(fiy2,ty);
 368             fiz2             = _mm_add_ps(fiz2,tz);
 369
 370             fjx0             = _mm_add_ps(fjx0,tx);
 371             fjy0             = _mm_add_ps(fjy0,ty);
 372             fjz0             = _mm_add_ps(fjz0,tz);
 373
 374             }
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm_any_lt(rsq30,rcutoff2))
 381             {
 382
 383             r30              = _mm_mul_ps(rsq30,rinv30);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq30             = _mm_mul_ps(iq3,jq0);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 391             ewrt             = _mm_mul_ps(r30,ewtabscale);
 392             ewitab           = _mm_cvttps_epi32(ewrt);
 393             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 394             ewitab           = _mm_slli_epi32(ewitab,2);
 395             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 396             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 397             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 398             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 399             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 400             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 401             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 402             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 403             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 404
 405             d                = _mm_sub_ps(r30,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(rinv30,_mm_mul_ps(velec,dsw)) );
 415             velec            = _mm_mul_ps(velec,sw);
 416             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velec            = _mm_and_ps(velec,cutoff_mask);
 420             velecsum         = _mm_add_ps(velecsum,velec);
 421
 422             fscal            = felec;
 423
 424             fscal            = _mm_and_ps(fscal,cutoff_mask);
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm_mul_ps(fscal,dx30);
 428             ty               = _mm_mul_ps(fscal,dy30);
 429             tz               = _mm_mul_ps(fscal,dz30);
 430
 431             /* Update vectorial force */
 432             fix3             = _mm_add_ps(fix3,tx);
 433             fiy3             = _mm_add_ps(fiy3,ty);
 434             fiz3             = _mm_add_ps(fiz3,tz);
 435
 436             fjx0             = _mm_add_ps(fjx0,tx);
 437             fjy0             = _mm_add_ps(fjy0,ty);
 438             fjz0             = _mm_add_ps(fjz0,tz);
 439
 440             }
 441
 442             fjptrA             = f+j_coord_offsetA;
 443             fjptrB             = f+j_coord_offsetB;
 444             fjptrC             = f+j_coord_offsetC;
 445             fjptrD             = f+j_coord_offsetD;
 446
 447             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 448
 449             /* Inner loop uses 195 flops */
 450         }
 451
 452         if(jidx<j_index_end)
 453         {
 454
 455             /* Get j neighbor index, and coordinate index */
 456             jnrlistA         = jjnr[jidx];
 457             jnrlistB         = jjnr[jidx+1];
 458             jnrlistC         = jjnr[jidx+2];
 459             jnrlistD         = jjnr[jidx+3];
 460             /* Sign of each element will be negative for non-real atoms.
 461              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 462              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 463              */
 464             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 465             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 466             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 467             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 468             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 469             j_coord_offsetA  = DIM*jnrA;
 470             j_coord_offsetB  = DIM*jnrB;
 471             j_coord_offsetC  = DIM*jnrC;
 472             j_coord_offsetD  = DIM*jnrD;
 473
 474             /* load j atom coordinates */
 475             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 476                                               x+j_coord_offsetC,x+j_coord_offsetD,
 477                                               &jx0,&jy0,&jz0);
 478
 479             /* Calculate displacement vector */
 480             dx10             = _mm_sub_ps(ix1,jx0);
 481             dy10             = _mm_sub_ps(iy1,jy0);
 482             dz10             = _mm_sub_ps(iz1,jz0);
 483             dx20             = _mm_sub_ps(ix2,jx0);
 484             dy20             = _mm_sub_ps(iy2,jy0);
 485             dz20             = _mm_sub_ps(iz2,jz0);
 486             dx30             = _mm_sub_ps(ix3,jx0);
 487             dy30             = _mm_sub_ps(iy3,jy0);
 488             dz30             = _mm_sub_ps(iz3,jz0);
 489
 490             /* Calculate squared distance and things based on it */
 491             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 492             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 493             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 494
 495             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 496             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 497             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 498
 499             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 500             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 501             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 502
 503             /* Load parameters for j particles */
 504             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 505                                                               charge+jnrC+0,charge+jnrD+0);
 506
 507             fjx0             = _mm_setzero_ps();
 508             fjy0             = _mm_setzero_ps();
 509             fjz0             = _mm_setzero_ps();
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             if (gmx_mm_any_lt(rsq10,rcutoff2))
 516             {
 517
 518             r10              = _mm_mul_ps(rsq10,rinv10);
 519             r10              = _mm_andnot_ps(dummy_mask,r10);
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq10             = _mm_mul_ps(iq1,jq0);
 523
 524             /* EWALD ELECTROSTATICS */
 525
 526             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 527             ewrt             = _mm_mul_ps(r10,ewtabscale);
 528             ewitab           = _mm_cvttps_epi32(ewrt);
 529             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 530             ewitab           = _mm_slli_epi32(ewitab,2);
 531             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 532             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 533             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 534             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 535             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 536             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 537             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 538             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 539             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 540
 541             d                = _mm_sub_ps(r10,rswitch);
 542             d                = _mm_max_ps(d,_mm_setzero_ps());
 543             d2               = _mm_mul_ps(d,d);
 544             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)))))));
 545
 546             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 547
 548             /* Evaluate switch function */
 549             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 550             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 551             velec            = _mm_mul_ps(velec,sw);
 552             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_and_ps(velec,cutoff_mask);
 556             velec            = _mm_andnot_ps(dummy_mask,velec);
 557             velecsum         = _mm_add_ps(velecsum,velec);
 558
 559             fscal            = felec;
 560
 561             fscal            = _mm_and_ps(fscal,cutoff_mask);
 562
 563             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_ps(fscal,dx10);
 567             ty               = _mm_mul_ps(fscal,dy10);
 568             tz               = _mm_mul_ps(fscal,dz10);
 569
 570             /* Update vectorial force */
 571             fix1             = _mm_add_ps(fix1,tx);
 572             fiy1             = _mm_add_ps(fiy1,ty);
 573             fiz1             = _mm_add_ps(fiz1,tz);
 574
 575             fjx0             = _mm_add_ps(fjx0,tx);
 576             fjy0             = _mm_add_ps(fjy0,ty);
 577             fjz0             = _mm_add_ps(fjz0,tz);
 578
 579             }
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             if (gmx_mm_any_lt(rsq20,rcutoff2))
 586             {
 587
 588             r20              = _mm_mul_ps(rsq20,rinv20);
 589             r20              = _mm_andnot_ps(dummy_mask,r20);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq20             = _mm_mul_ps(iq2,jq0);
 593
 594             /* EWALD ELECTROSTATICS */
 595
 596             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 597             ewrt             = _mm_mul_ps(r20,ewtabscale);
 598             ewitab           = _mm_cvttps_epi32(ewrt);
 599             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 600             ewitab           = _mm_slli_epi32(ewitab,2);
 601             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 602             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 603             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 604             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 605             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 606             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 607             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 608             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 609             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 610
 611             d                = _mm_sub_ps(r20,rswitch);
 612             d                = _mm_max_ps(d,_mm_setzero_ps());
 613             d2               = _mm_mul_ps(d,d);
 614             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)))))));
 615
 616             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 617
 618             /* Evaluate switch function */
 619             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 620             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 621             velec            = _mm_mul_ps(velec,sw);
 622             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 623
 624             /* Update potential sum for this i atom from the interaction with this j atom. */
 625             velec            = _mm_and_ps(velec,cutoff_mask);
 626             velec            = _mm_andnot_ps(dummy_mask,velec);
 627             velecsum         = _mm_add_ps(velecsum,velec);
 628
 629             fscal            = felec;
 630
 631             fscal            = _mm_and_ps(fscal,cutoff_mask);
 632
 633             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm_mul_ps(fscal,dx20);
 637             ty               = _mm_mul_ps(fscal,dy20);
 638             tz               = _mm_mul_ps(fscal,dz20);
 639
 640             /* Update vectorial force */
 641             fix2             = _mm_add_ps(fix2,tx);
 642             fiy2             = _mm_add_ps(fiy2,ty);
 643             fiz2             = _mm_add_ps(fiz2,tz);
 644
 645             fjx0             = _mm_add_ps(fjx0,tx);
 646             fjy0             = _mm_add_ps(fjy0,ty);
 647             fjz0             = _mm_add_ps(fjz0,tz);
 648
 649             }
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             if (gmx_mm_any_lt(rsq30,rcutoff2))
 656             {
 657
 658             r30              = _mm_mul_ps(rsq30,rinv30);
 659             r30              = _mm_andnot_ps(dummy_mask,r30);
 660
 661             /* Compute parameters for interactions between i and j atoms */
 662             qq30             = _mm_mul_ps(iq3,jq0);
 663
 664             /* EWALD ELECTROSTATICS */
 665
 666             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 667             ewrt             = _mm_mul_ps(r30,ewtabscale);
 668             ewitab           = _mm_cvttps_epi32(ewrt);
 669             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 670             ewitab           = _mm_slli_epi32(ewitab,2);
 671             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 672             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 673             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 674             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 675             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 676             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 677             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 678             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 679             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 680
 681             d                = _mm_sub_ps(r30,rswitch);
 682             d                = _mm_max_ps(d,_mm_setzero_ps());
 683             d2               = _mm_mul_ps(d,d);
 684             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)))))));
 685
 686             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 687
 688             /* Evaluate switch function */
 689             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 690             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 691             velec            = _mm_mul_ps(velec,sw);
 692             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 693
 694             /* Update potential sum for this i atom from the interaction with this j atom. */
 695             velec            = _mm_and_ps(velec,cutoff_mask);
 696             velec            = _mm_andnot_ps(dummy_mask,velec);
 697             velecsum         = _mm_add_ps(velecsum,velec);
 698
 699             fscal            = felec;
 700
 701             fscal            = _mm_and_ps(fscal,cutoff_mask);
 702
 703             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm_mul_ps(fscal,dx30);
 707             ty               = _mm_mul_ps(fscal,dy30);
 708             tz               = _mm_mul_ps(fscal,dz30);
 709
 710             /* Update vectorial force */
 711             fix3             = _mm_add_ps(fix3,tx);
 712             fiy3             = _mm_add_ps(fiy3,ty);
 713             fiz3             = _mm_add_ps(fiz3,tz);
 714
 715             fjx0             = _mm_add_ps(fjx0,tx);
 716             fjy0             = _mm_add_ps(fjy0,ty);
 717             fjz0             = _mm_add_ps(fjz0,tz);
 718
 719             }
 720
 721             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 722             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 723             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 724             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 725
 726             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 727
 728             /* Inner loop uses 198 flops */
 729         }
 730
 731         /* End of innermost loop */
 732
 733         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 734                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 735
 736         ggid                        = gid[iidx];
 737         /* Update potential energies */
 738         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 739
 740         /* Increment number of inner iterations */
 741         inneriter                  += j_index_end - j_index_start;
 742
 743         /* Outer loop uses 19 flops */
 744     }
 745
 746     /* Increment number of outer iterations */
 747     outeriter        += nri;
 748
 749     /* Update outer/inner flops */
 750
 751     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*198);
 752 }
 753 /*
 754  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse4_1_single
 755  * Electrostatics interaction: Ewald
 756  * VdW interaction:            None
 757  * Geometry:                   Water4-Particle
 758  * Calculate force/pot:        Force
 759  */
 760 void
 761 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse4_1_single
 762                     (t_nblist                    * gmx_restrict       nlist,
 763                      rvec                        * gmx_restrict          xx,
 764                      rvec                        * gmx_restrict          ff,
 765                      t_forcerec                  * gmx_restrict          fr,
 766                      t_mdatoms                   * gmx_restrict     mdatoms,
 767                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 768                      t_nrnb                      * gmx_restrict        nrnb)
 769 {
 770     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 771      * just 0 for non-waters.
 772      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 773      * jnr indices corresponding to data put in the four positions in the SIMD register.
 774      */
 775     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 776     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 777     int              jnrA,jnrB,jnrC,jnrD;
 778     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 779     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 780     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 781     real             rcutoff_scalar;
 782     real             *shiftvec,*fshift,*x,*f;
 783     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 784     real             scratch[4*DIM];
 785     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 786     int              vdwioffset1;
 787     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 788     int              vdwioffset2;
 789     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 790     int              vdwioffset3;
 791     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 792     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 793     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 794     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 795     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 796     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 797     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 798     real             *charge;
 799     __m128i          ewitab;
 800     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 801     real             *ewtab;
 802     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 803     real             rswitch_scalar,d_scalar;
 804     __m128           dummy_mask,cutoff_mask;
 805     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 806     __m128           one     = _mm_set1_ps(1.0);
 807     __m128           two     = _mm_set1_ps(2.0);
 808     x                = xx[0];
 809     f                = ff[0];
 810
 811     nri              = nlist->nri;
 812     iinr             = nlist->iinr;
 813     jindex           = nlist->jindex;
 814     jjnr             = nlist->jjnr;
 815     shiftidx         = nlist->shift;
 816     gid              = nlist->gid;
 817     shiftvec         = fr->shift_vec[0];
 818     fshift           = fr->fshift[0];
 819     facel            = _mm_set1_ps(fr->epsfac);
 820     charge           = mdatoms->chargeA;
 821
 822     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 823     ewtab            = fr->ic->tabq_coul_FDV0;
 824     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 825     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 826
 827     /* Setup water-specific parameters */
 828     inr              = nlist->iinr[0];
 829     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 830     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 831     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 832
 833     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 834     rcutoff_scalar   = fr->rcoulomb;
 835     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 836     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 837
 838     rswitch_scalar   = fr->rcoulomb_switch;
 839     rswitch          = _mm_set1_ps(rswitch_scalar);
 840     /* Setup switch parameters */
 841     d_scalar         = rcutoff_scalar-rswitch_scalar;
 842     d                = _mm_set1_ps(d_scalar);
 843     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 844     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 845     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 846     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 847     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 848     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 849
 850     /* Avoid stupid compiler warnings */
 851     jnrA = jnrB = jnrC = jnrD = 0;
 852     j_coord_offsetA = 0;
 853     j_coord_offsetB = 0;
 854     j_coord_offsetC = 0;
 855     j_coord_offsetD = 0;
 856
 857     outeriter        = 0;
 858     inneriter        = 0;
 859
 860     for(iidx=0;iidx<4*DIM;iidx++)
 861     {
 862         scratch[iidx] = 0.0;
 863     }
 864
 865     /* Start outer loop over neighborlists */
 866     for(iidx=0; iidx<nri; iidx++)
 867     {
 868         /* Load shift vector for this list */
 869         i_shift_offset   = DIM*shiftidx[iidx];
 870
 871         /* Load limits for loop over neighbors */
 872         j_index_start    = jindex[iidx];
 873         j_index_end      = jindex[iidx+1];
 874
 875         /* Get outer coordinate index */
 876         inr              = iinr[iidx];
 877         i_coord_offset   = DIM*inr;
 878
 879         /* Load i particle coords and add shift vector */
 880         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 881                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 882
 883         fix1             = _mm_setzero_ps();
 884         fiy1             = _mm_setzero_ps();
 885         fiz1             = _mm_setzero_ps();
 886         fix2             = _mm_setzero_ps();
 887         fiy2             = _mm_setzero_ps();
 888         fiz2             = _mm_setzero_ps();
 889         fix3             = _mm_setzero_ps();
 890         fiy3             = _mm_setzero_ps();
 891         fiz3             = _mm_setzero_ps();
 892
 893         /* Start inner kernel loop */
 894         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 895         {
 896
 897             /* Get j neighbor index, and coordinate index */
 898             jnrA             = jjnr[jidx];
 899             jnrB             = jjnr[jidx+1];
 900             jnrC             = jjnr[jidx+2];
 901             jnrD             = jjnr[jidx+3];
 902             j_coord_offsetA  = DIM*jnrA;
 903             j_coord_offsetB  = DIM*jnrB;
 904             j_coord_offsetC  = DIM*jnrC;
 905             j_coord_offsetD  = DIM*jnrD;
 906
 907             /* load j atom coordinates */
 908             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 909                                               x+j_coord_offsetC,x+j_coord_offsetD,
 910                                               &jx0,&jy0,&jz0);
 911
 912             /* Calculate displacement vector */
 913             dx10             = _mm_sub_ps(ix1,jx0);
 914             dy10             = _mm_sub_ps(iy1,jy0);
 915             dz10             = _mm_sub_ps(iz1,jz0);
 916             dx20             = _mm_sub_ps(ix2,jx0);
 917             dy20             = _mm_sub_ps(iy2,jy0);
 918             dz20             = _mm_sub_ps(iz2,jz0);
 919             dx30             = _mm_sub_ps(ix3,jx0);
 920             dy30             = _mm_sub_ps(iy3,jy0);
 921             dz30             = _mm_sub_ps(iz3,jz0);
 922
 923             /* Calculate squared distance and things based on it */
 924             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 925             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 926             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 927
 928             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 929             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 930             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 931
 932             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 933             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 934             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 935
 936             /* Load parameters for j particles */
 937             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 938                                                               charge+jnrC+0,charge+jnrD+0);
 939
 940             fjx0             = _mm_setzero_ps();
 941             fjy0             = _mm_setzero_ps();
 942             fjz0             = _mm_setzero_ps();
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             if (gmx_mm_any_lt(rsq10,rcutoff2))
 949             {
 950
 951             r10              = _mm_mul_ps(rsq10,rinv10);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq10             = _mm_mul_ps(iq1,jq0);
 955
 956             /* EWALD ELECTROSTATICS */
 957
 958             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 959             ewrt             = _mm_mul_ps(r10,ewtabscale);
 960             ewitab           = _mm_cvttps_epi32(ewrt);
 961             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 962             ewitab           = _mm_slli_epi32(ewitab,2);
 963             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 964             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 965             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 966             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 967             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 968             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 969             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 970             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 971             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 972
 973             d                = _mm_sub_ps(r10,rswitch);
 974             d                = _mm_max_ps(d,_mm_setzero_ps());
 975             d2               = _mm_mul_ps(d,d);
 976             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)))))));
 977
 978             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 979
 980             /* Evaluate switch function */
 981             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 982             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 983             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 984
 985             fscal            = felec;
 986
 987             fscal            = _mm_and_ps(fscal,cutoff_mask);
 988
 989             /* Calculate temporary vectorial force */
 990             tx               = _mm_mul_ps(fscal,dx10);
 991             ty               = _mm_mul_ps(fscal,dy10);
 992             tz               = _mm_mul_ps(fscal,dz10);
 993
 994             /* Update vectorial force */
 995             fix1             = _mm_add_ps(fix1,tx);
 996             fiy1             = _mm_add_ps(fiy1,ty);
 997             fiz1             = _mm_add_ps(fiz1,tz);
 998
 999             fjx0             = _mm_add_ps(fjx0,tx);
1000             fjy0             = _mm_add_ps(fjy0,ty);
1001             fjz0             = _mm_add_ps(fjz0,tz);
1002
1003             }
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             if (gmx_mm_any_lt(rsq20,rcutoff2))
1010             {
1011
1012             r20              = _mm_mul_ps(rsq20,rinv20);
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq20             = _mm_mul_ps(iq2,jq0);
1016
1017             /* EWALD ELECTROSTATICS */
1018
1019             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1020             ewrt             = _mm_mul_ps(r20,ewtabscale);
1021             ewitab           = _mm_cvttps_epi32(ewrt);
1022             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1023             ewitab           = _mm_slli_epi32(ewitab,2);
1024             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1025             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1026             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1027             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1028             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1029             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1030             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1031             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1032             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1033
1034             d                = _mm_sub_ps(r20,rswitch);
1035             d                = _mm_max_ps(d,_mm_setzero_ps());
1036             d2               = _mm_mul_ps(d,d);
1037             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)))))));
1038
1039             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1040
1041             /* Evaluate switch function */
1042             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1043             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1044             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm_and_ps(fscal,cutoff_mask);
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm_mul_ps(fscal,dx20);
1052             ty               = _mm_mul_ps(fscal,dy20);
1053             tz               = _mm_mul_ps(fscal,dz20);
1054
1055             /* Update vectorial force */
1056             fix2             = _mm_add_ps(fix2,tx);
1057             fiy2             = _mm_add_ps(fiy2,ty);
1058             fiz2             = _mm_add_ps(fiz2,tz);
1059
1060             fjx0             = _mm_add_ps(fjx0,tx);
1061             fjy0             = _mm_add_ps(fjy0,ty);
1062             fjz0             = _mm_add_ps(fjz0,tz);
1063
1064             }
1065
1066             /**************************
1067              * CALCULATE INTERACTIONS *
1068              **************************/
1069
1070             if (gmx_mm_any_lt(rsq30,rcutoff2))
1071             {
1072
1073             r30              = _mm_mul_ps(rsq30,rinv30);
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq30             = _mm_mul_ps(iq3,jq0);
1077
1078             /* EWALD ELECTROSTATICS */
1079
1080             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1081             ewrt             = _mm_mul_ps(r30,ewtabscale);
1082             ewitab           = _mm_cvttps_epi32(ewrt);
1083             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1084             ewitab           = _mm_slli_epi32(ewitab,2);
1085             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1086             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1087             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1088             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1089             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1090             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1091             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1092             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
1093             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1094
1095             d                = _mm_sub_ps(r30,rswitch);
1096             d                = _mm_max_ps(d,_mm_setzero_ps());
1097             d2               = _mm_mul_ps(d,d);
1098             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)))))));
1099
1100             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1101
1102             /* Evaluate switch function */
1103             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1104             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1105             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_and_ps(fscal,cutoff_mask);
1110
1111             /* Calculate temporary vectorial force */
1112             tx               = _mm_mul_ps(fscal,dx30);
1113             ty               = _mm_mul_ps(fscal,dy30);
1114             tz               = _mm_mul_ps(fscal,dz30);
1115
1116             /* Update vectorial force */
1117             fix3             = _mm_add_ps(fix3,tx);
1118             fiy3             = _mm_add_ps(fiy3,ty);
1119             fiz3             = _mm_add_ps(fiz3,tz);
1120
1121             fjx0             = _mm_add_ps(fjx0,tx);
1122             fjy0             = _mm_add_ps(fjy0,ty);
1123             fjz0             = _mm_add_ps(fjz0,tz);
1124
1125             }
1126
1127             fjptrA             = f+j_coord_offsetA;
1128             fjptrB             = f+j_coord_offsetB;
1129             fjptrC             = f+j_coord_offsetC;
1130             fjptrD             = f+j_coord_offsetD;
1131
1132             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1133
1134             /* Inner loop uses 186 flops */
1135         }
1136
1137         if(jidx<j_index_end)
1138         {
1139
1140             /* Get j neighbor index, and coordinate index */
1141             jnrlistA         = jjnr[jidx];
1142             jnrlistB         = jjnr[jidx+1];
1143             jnrlistC         = jjnr[jidx+2];
1144             jnrlistD         = jjnr[jidx+3];
1145             /* Sign of each element will be negative for non-real atoms.
1146              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1147              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1148              */
1149             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1150             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1151             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1152             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1153             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1154             j_coord_offsetA  = DIM*jnrA;
1155             j_coord_offsetB  = DIM*jnrB;
1156             j_coord_offsetC  = DIM*jnrC;
1157             j_coord_offsetD  = DIM*jnrD;
1158
1159             /* load j atom coordinates */
1160             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1161                                               x+j_coord_offsetC,x+j_coord_offsetD,
1162                                               &jx0,&jy0,&jz0);
1163
1164             /* Calculate displacement vector */
1165             dx10             = _mm_sub_ps(ix1,jx0);
1166             dy10             = _mm_sub_ps(iy1,jy0);
1167             dz10             = _mm_sub_ps(iz1,jz0);
1168             dx20             = _mm_sub_ps(ix2,jx0);
1169             dy20             = _mm_sub_ps(iy2,jy0);
1170             dz20             = _mm_sub_ps(iz2,jz0);
1171             dx30             = _mm_sub_ps(ix3,jx0);
1172             dy30             = _mm_sub_ps(iy3,jy0);
1173             dz30             = _mm_sub_ps(iz3,jz0);
1174
1175             /* Calculate squared distance and things based on it */
1176             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1177             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1178             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1179
1180             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1181             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1182             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1183
1184             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1185             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1186             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1187
1188             /* Load parameters for j particles */
1189             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1190                                                               charge+jnrC+0,charge+jnrD+0);
1191
1192             fjx0             = _mm_setzero_ps();
1193             fjy0             = _mm_setzero_ps();
1194             fjz0             = _mm_setzero_ps();
1195
1196             /**************************
1197              * CALCULATE INTERACTIONS *
1198              **************************/
1199
1200             if (gmx_mm_any_lt(rsq10,rcutoff2))
1201             {
1202
1203             r10              = _mm_mul_ps(rsq10,rinv10);
1204             r10              = _mm_andnot_ps(dummy_mask,r10);
1205
1206             /* Compute parameters for interactions between i and j atoms */
1207             qq10             = _mm_mul_ps(iq1,jq0);
1208
1209             /* EWALD ELECTROSTATICS */
1210
1211             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1212             ewrt             = _mm_mul_ps(r10,ewtabscale);
1213             ewitab           = _mm_cvttps_epi32(ewrt);
1214             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1215             ewitab           = _mm_slli_epi32(ewitab,2);
1216             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1217             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1218             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1219             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1220             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1221             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1222             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1223             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1224             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1225
1226             d                = _mm_sub_ps(r10,rswitch);
1227             d                = _mm_max_ps(d,_mm_setzero_ps());
1228             d2               = _mm_mul_ps(d,d);
1229             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)))))));
1230
1231             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1232
1233             /* Evaluate switch function */
1234             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1235             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1236             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1237
1238             fscal            = felec;
1239
1240             fscal            = _mm_and_ps(fscal,cutoff_mask);
1241
1242             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1243
1244             /* Calculate temporary vectorial force */
1245             tx               = _mm_mul_ps(fscal,dx10);
1246             ty               = _mm_mul_ps(fscal,dy10);
1247             tz               = _mm_mul_ps(fscal,dz10);
1248
1249             /* Update vectorial force */
1250             fix1             = _mm_add_ps(fix1,tx);
1251             fiy1             = _mm_add_ps(fiy1,ty);
1252             fiz1             = _mm_add_ps(fiz1,tz);
1253
1254             fjx0             = _mm_add_ps(fjx0,tx);
1255             fjy0             = _mm_add_ps(fjy0,ty);
1256             fjz0             = _mm_add_ps(fjz0,tz);
1257
1258             }
1259
1260             /**************************
1261              * CALCULATE INTERACTIONS *
1262              **************************/
1263
1264             if (gmx_mm_any_lt(rsq20,rcutoff2))
1265             {
1266
1267             r20              = _mm_mul_ps(rsq20,rinv20);
1268             r20              = _mm_andnot_ps(dummy_mask,r20);
1269
1270             /* Compute parameters for interactions between i and j atoms */
1271             qq20             = _mm_mul_ps(iq2,jq0);
1272
1273             /* EWALD ELECTROSTATICS */
1274
1275             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1276             ewrt             = _mm_mul_ps(r20,ewtabscale);
1277             ewitab           = _mm_cvttps_epi32(ewrt);
1278             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1279             ewitab           = _mm_slli_epi32(ewitab,2);
1280             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1281             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1282             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1283             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1284             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1285             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1286             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1287             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1288             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1289
1290             d                = _mm_sub_ps(r20,rswitch);
1291             d                = _mm_max_ps(d,_mm_setzero_ps());
1292             d2               = _mm_mul_ps(d,d);
1293             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)))))));
1294
1295             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1296
1297             /* Evaluate switch function */
1298             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1299             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1300             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1301
1302             fscal            = felec;
1303
1304             fscal            = _mm_and_ps(fscal,cutoff_mask);
1305
1306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1307
1308             /* Calculate temporary vectorial force */
1309             tx               = _mm_mul_ps(fscal,dx20);
1310             ty               = _mm_mul_ps(fscal,dy20);
1311             tz               = _mm_mul_ps(fscal,dz20);
1312
1313             /* Update vectorial force */
1314             fix2             = _mm_add_ps(fix2,tx);
1315             fiy2             = _mm_add_ps(fiy2,ty);
1316             fiz2             = _mm_add_ps(fiz2,tz);
1317
1318             fjx0             = _mm_add_ps(fjx0,tx);
1319             fjy0             = _mm_add_ps(fjy0,ty);
1320             fjz0             = _mm_add_ps(fjz0,tz);
1321
1322             }
1323
1324             /**************************
1325              * CALCULATE INTERACTIONS *
1326              **************************/
1327
1328             if (gmx_mm_any_lt(rsq30,rcutoff2))
1329             {
1330
1331             r30              = _mm_mul_ps(rsq30,rinv30);
1332             r30              = _mm_andnot_ps(dummy_mask,r30);
1333
1334             /* Compute parameters for interactions between i and j atoms */
1335             qq30             = _mm_mul_ps(iq3,jq0);
1336
1337             /* EWALD ELECTROSTATICS */
1338
1339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1340             ewrt             = _mm_mul_ps(r30,ewtabscale);
1341             ewitab           = _mm_cvttps_epi32(ewrt);
1342             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1343             ewitab           = _mm_slli_epi32(ewitab,2);
1344             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1345             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1346             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1347             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1348             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1349             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1350             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1351             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
1352             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1353
1354             d                = _mm_sub_ps(r30,rswitch);
1355             d                = _mm_max_ps(d,_mm_setzero_ps());
1356             d2               = _mm_mul_ps(d,d);
1357             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)))))));
1358
1359             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1360
1361             /* Evaluate switch function */
1362             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1363             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1364             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1365
1366             fscal            = felec;
1367
1368             fscal            = _mm_and_ps(fscal,cutoff_mask);
1369
1370             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1371
1372             /* Calculate temporary vectorial force */
1373             tx               = _mm_mul_ps(fscal,dx30);
1374             ty               = _mm_mul_ps(fscal,dy30);
1375             tz               = _mm_mul_ps(fscal,dz30);
1376
1377             /* Update vectorial force */
1378             fix3             = _mm_add_ps(fix3,tx);
1379             fiy3             = _mm_add_ps(fiy3,ty);
1380             fiz3             = _mm_add_ps(fiz3,tz);
1381
1382             fjx0             = _mm_add_ps(fjx0,tx);
1383             fjy0             = _mm_add_ps(fjy0,ty);
1384             fjz0             = _mm_add_ps(fjz0,tz);
1385
1386             }
1387
1388             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1389             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1390             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1391             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1392
1393             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1394
1395             /* Inner loop uses 189 flops */
1396         }
1397
1398         /* End of innermost loop */
1399
1400         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1401                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1402
1403         /* Increment number of inner iterations */
1404         inneriter                  += j_index_end - j_index_start;
1405
1406         /* Outer loop uses 18 flops */
1407     }
1408
1409     /* Increment number of outer iterations */
1410     outeriter        += nri;
1411
1412     /* Update outer/inner flops */
1413
1414     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*189);
1415 }