src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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_ElecEw_VdwLJEw_GeomW3P1_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128           c6grid_00;
 103     __m128           c6grid_10;
 104     __m128           c6grid_20;
 105     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 106     real             *vdwgridparam;
 107     __m128           one_half  = _mm_set1_ps(0.5);
 108     __m128           minus_one = _mm_set1_ps(-1.0);
 109     __m128i          ewitab;
 110     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     real             *ewtab;
 112     __m128           dummy_mask,cutoff_mask;
 113     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 114     __m128           one     = _mm_set1_ps(1.0);
 115     __m128           two     = _mm_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132     vdwgridparam     = fr->ljpme_c6grid;
 133     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 134     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 135     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 136
 137     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 145     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 146     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = jnrC = jnrD = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153     j_coord_offsetC = 0;
 154     j_coord_offsetD = 0;
 155
 156     outeriter        = 0;
 157     inneriter        = 0;
 158
 159     for(iidx=0;iidx<4*DIM;iidx++)
 160     {
 161         scratch[iidx] = 0.0;
 162     }
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 181
 182         fix0             = _mm_setzero_ps();
 183         fiy0             = _mm_setzero_ps();
 184         fiz0             = _mm_setzero_ps();
 185         fix1             = _mm_setzero_ps();
 186         fiy1             = _mm_setzero_ps();
 187         fiz1             = _mm_setzero_ps();
 188         fix2             = _mm_setzero_ps();
 189         fiy2             = _mm_setzero_ps();
 190         fiz2             = _mm_setzero_ps();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_ps();
 194         vvdwsum          = _mm_setzero_ps();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209
 210             /* load j atom coordinates */
 211             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 212                                               x+j_coord_offsetC,x+j_coord_offsetD,
 213                                               &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm_sub_ps(ix0,jx0);
 217             dy00             = _mm_sub_ps(iy0,jy0);
 218             dz00             = _mm_sub_ps(iz0,jz0);
 219             dx10             = _mm_sub_ps(ix1,jx0);
 220             dy10             = _mm_sub_ps(iy1,jy0);
 221             dz10             = _mm_sub_ps(iz1,jz0);
 222             dx20             = _mm_sub_ps(ix2,jx0);
 223             dy20             = _mm_sub_ps(iy2,jy0);
 224             dz20             = _mm_sub_ps(iz2,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 228             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 229             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 230
 231             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 232             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 234
 235             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 236             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 237             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 241                                                               charge+jnrC+0,charge+jnrD+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246
 247             fjx0             = _mm_setzero_ps();
 248             fjy0             = _mm_setzero_ps();
 249             fjz0             = _mm_setzero_ps();
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             r00              = _mm_mul_ps(rsq00,rinv00);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq00             = _mm_mul_ps(iq0,jq0);
 259             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 260                                          vdwparam+vdwioffset0+vdwjidx0B,
 261                                          vdwparam+vdwioffset0+vdwjidx0C,
 262                                          vdwparam+vdwioffset0+vdwjidx0D,
 263                                          &c6_00,&c12_00);
 264
 265             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 266                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 267                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 268                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 269
 270             /* EWALD ELECTROSTATICS */
 271
 272             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 273             ewrt             = _mm_mul_ps(r00,ewtabscale);
 274             ewitab           = _mm_cvttps_epi32(ewrt);
 275             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 276             ewitab           = _mm_slli_epi32(ewitab,2);
 277             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 278             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 279             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 280             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 281             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 282             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 283             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 284             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 285             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 286
 287             /* Analytical LJ-PME */
 288             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 289             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 290             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 291             exponent         = gmx_simd_exp_r(ewcljrsq);
 292             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 293             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 294             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 295             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 296             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 297             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 298             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 299             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velecsum         = _mm_add_ps(velecsum,velec);
 303             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 304
 305             fscal            = _mm_add_ps(felec,fvdw);
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm_mul_ps(fscal,dx00);
 309             ty               = _mm_mul_ps(fscal,dy00);
 310             tz               = _mm_mul_ps(fscal,dz00);
 311
 312             /* Update vectorial force */
 313             fix0             = _mm_add_ps(fix0,tx);
 314             fiy0             = _mm_add_ps(fiy0,ty);
 315             fiz0             = _mm_add_ps(fiz0,tz);
 316
 317             fjx0             = _mm_add_ps(fjx0,tx);
 318             fjy0             = _mm_add_ps(fjy0,ty);
 319             fjz0             = _mm_add_ps(fjz0,tz);
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             r10              = _mm_mul_ps(rsq10,rinv10);
 326
 327             /* Compute parameters for interactions between i and j atoms */
 328             qq10             = _mm_mul_ps(iq1,jq0);
 329
 330             /* EWALD ELECTROSTATICS */
 331
 332             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 333             ewrt             = _mm_mul_ps(r10,ewtabscale);
 334             ewitab           = _mm_cvttps_epi32(ewrt);
 335             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 336             ewitab           = _mm_slli_epi32(ewitab,2);
 337             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 338             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 339             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 340             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 341             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 342             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 343             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 344             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 345             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm_add_ps(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm_mul_ps(fscal,dx10);
 354             ty               = _mm_mul_ps(fscal,dy10);
 355             tz               = _mm_mul_ps(fscal,dz10);
 356
 357             /* Update vectorial force */
 358             fix1             = _mm_add_ps(fix1,tx);
 359             fiy1             = _mm_add_ps(fiy1,ty);
 360             fiz1             = _mm_add_ps(fiz1,tz);
 361
 362             fjx0             = _mm_add_ps(fjx0,tx);
 363             fjy0             = _mm_add_ps(fjy0,ty);
 364             fjz0             = _mm_add_ps(fjz0,tz);
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             r20              = _mm_mul_ps(rsq20,rinv20);
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq20             = _mm_mul_ps(iq2,jq0);
 374
 375             /* EWALD ELECTROSTATICS */
 376
 377             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 378             ewrt             = _mm_mul_ps(r20,ewtabscale);
 379             ewitab           = _mm_cvttps_epi32(ewrt);
 380             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 381             ewitab           = _mm_slli_epi32(ewitab,2);
 382             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 383             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 384             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 385             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 386             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 387             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 388             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 389             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 390             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm_mul_ps(fscal,dx20);
 399             ty               = _mm_mul_ps(fscal,dy20);
 400             tz               = _mm_mul_ps(fscal,dz20);
 401
 402             /* Update vectorial force */
 403             fix2             = _mm_add_ps(fix2,tx);
 404             fiy2             = _mm_add_ps(fiy2,ty);
 405             fiz2             = _mm_add_ps(fiz2,tz);
 406
 407             fjx0             = _mm_add_ps(fjx0,tx);
 408             fjy0             = _mm_add_ps(fjy0,ty);
 409             fjz0             = _mm_add_ps(fjz0,tz);
 410
 411             fjptrA             = f+j_coord_offsetA;
 412             fjptrB             = f+j_coord_offsetB;
 413             fjptrC             = f+j_coord_offsetC;
 414             fjptrD             = f+j_coord_offsetD;
 415
 416             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 417
 418             /* Inner loop uses 151 flops */
 419         }
 420
 421         if(jidx<j_index_end)
 422         {
 423
 424             /* Get j neighbor index, and coordinate index */
 425             jnrlistA         = jjnr[jidx];
 426             jnrlistB         = jjnr[jidx+1];
 427             jnrlistC         = jjnr[jidx+2];
 428             jnrlistD         = jjnr[jidx+3];
 429             /* Sign of each element will be negative for non-real atoms.
 430              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 431              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 432              */
 433             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 434             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 435             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 436             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 437             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 438             j_coord_offsetA  = DIM*jnrA;
 439             j_coord_offsetB  = DIM*jnrB;
 440             j_coord_offsetC  = DIM*jnrC;
 441             j_coord_offsetD  = DIM*jnrD;
 442
 443             /* load j atom coordinates */
 444             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 445                                               x+j_coord_offsetC,x+j_coord_offsetD,
 446                                               &jx0,&jy0,&jz0);
 447
 448             /* Calculate displacement vector */
 449             dx00             = _mm_sub_ps(ix0,jx0);
 450             dy00             = _mm_sub_ps(iy0,jy0);
 451             dz00             = _mm_sub_ps(iz0,jz0);
 452             dx10             = _mm_sub_ps(ix1,jx0);
 453             dy10             = _mm_sub_ps(iy1,jy0);
 454             dz10             = _mm_sub_ps(iz1,jz0);
 455             dx20             = _mm_sub_ps(ix2,jx0);
 456             dy20             = _mm_sub_ps(iy2,jy0);
 457             dz20             = _mm_sub_ps(iz2,jz0);
 458
 459             /* Calculate squared distance and things based on it */
 460             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 461             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 462             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 463
 464             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 465             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 466             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 467
 468             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 469             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 470             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 471
 472             /* Load parameters for j particles */
 473             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 474                                                               charge+jnrC+0,charge+jnrD+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479
 480             fjx0             = _mm_setzero_ps();
 481             fjy0             = _mm_setzero_ps();
 482             fjz0             = _mm_setzero_ps();
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             r00              = _mm_mul_ps(rsq00,rinv00);
 489             r00              = _mm_andnot_ps(dummy_mask,r00);
 490
 491             /* Compute parameters for interactions between i and j atoms */
 492             qq00             = _mm_mul_ps(iq0,jq0);
 493             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 494                                          vdwparam+vdwioffset0+vdwjidx0B,
 495                                          vdwparam+vdwioffset0+vdwjidx0C,
 496                                          vdwparam+vdwioffset0+vdwjidx0D,
 497                                          &c6_00,&c12_00);
 498
 499             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 500                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 501                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 502                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 503
 504             /* EWALD ELECTROSTATICS */
 505
 506             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 507             ewrt             = _mm_mul_ps(r00,ewtabscale);
 508             ewitab           = _mm_cvttps_epi32(ewrt);
 509             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 510             ewitab           = _mm_slli_epi32(ewitab,2);
 511             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 512             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 513             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 514             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 515             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 516             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 517             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 518             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 519             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 520
 521             /* Analytical LJ-PME */
 522             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 523             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 524             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 525             exponent         = gmx_simd_exp_r(ewcljrsq);
 526             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 527             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 528             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 529             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 530             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 531             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 532             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 533             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             velec            = _mm_andnot_ps(dummy_mask,velec);
 537             velecsum         = _mm_add_ps(velecsum,velec);
 538             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 539             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 540
 541             fscal            = _mm_add_ps(felec,fvdw);
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm_mul_ps(fscal,dx00);
 547             ty               = _mm_mul_ps(fscal,dy00);
 548             tz               = _mm_mul_ps(fscal,dz00);
 549
 550             /* Update vectorial force */
 551             fix0             = _mm_add_ps(fix0,tx);
 552             fiy0             = _mm_add_ps(fiy0,ty);
 553             fiz0             = _mm_add_ps(fiz0,tz);
 554
 555             fjx0             = _mm_add_ps(fjx0,tx);
 556             fjy0             = _mm_add_ps(fjy0,ty);
 557             fjz0             = _mm_add_ps(fjz0,tz);
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r10              = _mm_mul_ps(rsq10,rinv10);
 564             r10              = _mm_andnot_ps(dummy_mask,r10);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm_mul_ps(iq1,jq0);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 572             ewrt             = _mm_mul_ps(r10,ewtabscale);
 573             ewitab           = _mm_cvttps_epi32(ewrt);
 574             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 575             ewitab           = _mm_slli_epi32(ewitab,2);
 576             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 577             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 578             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 579             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 580             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 581             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 582             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 583             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 584             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_andnot_ps(dummy_mask,velec);
 588             velecsum         = _mm_add_ps(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm_mul_ps(fscal,dx10);
 596             ty               = _mm_mul_ps(fscal,dy10);
 597             tz               = _mm_mul_ps(fscal,dz10);
 598
 599             /* Update vectorial force */
 600             fix1             = _mm_add_ps(fix1,tx);
 601             fiy1             = _mm_add_ps(fiy1,ty);
 602             fiz1             = _mm_add_ps(fiz1,tz);
 603
 604             fjx0             = _mm_add_ps(fjx0,tx);
 605             fjy0             = _mm_add_ps(fjy0,ty);
 606             fjz0             = _mm_add_ps(fjz0,tz);
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             r20              = _mm_mul_ps(rsq20,rinv20);
 613             r20              = _mm_andnot_ps(dummy_mask,r20);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq20             = _mm_mul_ps(iq2,jq0);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 621             ewrt             = _mm_mul_ps(r20,ewtabscale);
 622             ewitab           = _mm_cvttps_epi32(ewrt);
 623             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 624             ewitab           = _mm_slli_epi32(ewitab,2);
 625             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 626             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 627             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 628             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 629             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 630             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 631             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 632             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 633             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velec            = _mm_andnot_ps(dummy_mask,velec);
 637             velecsum         = _mm_add_ps(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm_mul_ps(fscal,dx20);
 645             ty               = _mm_mul_ps(fscal,dy20);
 646             tz               = _mm_mul_ps(fscal,dz20);
 647
 648             /* Update vectorial force */
 649             fix2             = _mm_add_ps(fix2,tx);
 650             fiy2             = _mm_add_ps(fiy2,ty);
 651             fiz2             = _mm_add_ps(fiz2,tz);
 652
 653             fjx0             = _mm_add_ps(fjx0,tx);
 654             fjy0             = _mm_add_ps(fjy0,ty);
 655             fjz0             = _mm_add_ps(fjz0,tz);
 656
 657             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 658             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 659             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 660             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 661
 662             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 663
 664             /* Inner loop uses 154 flops */
 665         }
 666
 667         /* End of innermost loop */
 668
 669         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 670                                               f+i_coord_offset,fshift+i_shift_offset);
 671
 672         ggid                        = gid[iidx];
 673         /* Update potential energies */
 674         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 675         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 676
 677         /* Increment number of inner iterations */
 678         inneriter                  += j_index_end - j_index_start;
 679
 680         /* Outer loop uses 20 flops */
 681     }
 682
 683     /* Increment number of outer iterations */
 684     outeriter        += nri;
 685
 686     /* Update outer/inner flops */
 687
 688     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 689 }
 690 /*
 691  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_single
 692  * Electrostatics interaction: Ewald
 693  * VdW interaction:            LJEwald
 694  * Geometry:                   Water3-Particle
 695  * Calculate force/pot:        Force
 696  */
 697 void
 698 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_single
 699                     (t_nblist                    * gmx_restrict       nlist,
 700                      rvec                        * gmx_restrict          xx,
 701                      rvec                        * gmx_restrict          ff,
 702                      t_forcerec                  * gmx_restrict          fr,
 703                      t_mdatoms                   * gmx_restrict     mdatoms,
 704                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 705                      t_nrnb                      * gmx_restrict        nrnb)
 706 {
 707     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 708      * just 0 for non-waters.
 709      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 710      * jnr indices corresponding to data put in the four positions in the SIMD register.
 711      */
 712     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 713     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 714     int              jnrA,jnrB,jnrC,jnrD;
 715     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 716     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 717     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 718     real             rcutoff_scalar;
 719     real             *shiftvec,*fshift,*x,*f;
 720     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 721     real             scratch[4*DIM];
 722     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 723     int              vdwioffset0;
 724     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 725     int              vdwioffset1;
 726     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 727     int              vdwioffset2;
 728     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 729     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 730     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 731     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 732     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 733     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 734     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 735     real             *charge;
 736     int              nvdwtype;
 737     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 738     int              *vdwtype;
 739     real             *vdwparam;
 740     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 741     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 742     __m128           c6grid_00;
 743     __m128           c6grid_10;
 744     __m128           c6grid_20;
 745     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 746     real             *vdwgridparam;
 747     __m128           one_half  = _mm_set1_ps(0.5);
 748     __m128           minus_one = _mm_set1_ps(-1.0);
 749     __m128i          ewitab;
 750     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 751     real             *ewtab;
 752     __m128           dummy_mask,cutoff_mask;
 753     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 754     __m128           one     = _mm_set1_ps(1.0);
 755     __m128           two     = _mm_set1_ps(2.0);
 756     x                = xx[0];
 757     f                = ff[0];
 758
 759     nri              = nlist->nri;
 760     iinr             = nlist->iinr;
 761     jindex           = nlist->jindex;
 762     jjnr             = nlist->jjnr;
 763     shiftidx         = nlist->shift;
 764     gid              = nlist->gid;
 765     shiftvec         = fr->shift_vec[0];
 766     fshift           = fr->fshift[0];
 767     facel            = _mm_set1_ps(fr->epsfac);
 768     charge           = mdatoms->chargeA;
 769     nvdwtype         = fr->ntype;
 770     vdwparam         = fr->nbfp;
 771     vdwtype          = mdatoms->typeA;
 772     vdwgridparam     = fr->ljpme_c6grid;
 773     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 774     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 775     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 776
 777     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 778     ewtab            = fr->ic->tabq_coul_F;
 779     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 780     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 781
 782     /* Setup water-specific parameters */
 783     inr              = nlist->iinr[0];
 784     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 785     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 786     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 787     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 788
 789     /* Avoid stupid compiler warnings */
 790     jnrA = jnrB = jnrC = jnrD = 0;
 791     j_coord_offsetA = 0;
 792     j_coord_offsetB = 0;
 793     j_coord_offsetC = 0;
 794     j_coord_offsetD = 0;
 795
 796     outeriter        = 0;
 797     inneriter        = 0;
 798
 799     for(iidx=0;iidx<4*DIM;iidx++)
 800     {
 801         scratch[iidx] = 0.0;
 802     }
 803
 804     /* Start outer loop over neighborlists */
 805     for(iidx=0; iidx<nri; iidx++)
 806     {
 807         /* Load shift vector for this list */
 808         i_shift_offset   = DIM*shiftidx[iidx];
 809
 810         /* Load limits for loop over neighbors */
 811         j_index_start    = jindex[iidx];
 812         j_index_end      = jindex[iidx+1];
 813
 814         /* Get outer coordinate index */
 815         inr              = iinr[iidx];
 816         i_coord_offset   = DIM*inr;
 817
 818         /* Load i particle coords and add shift vector */
 819         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 820                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 821
 822         fix0             = _mm_setzero_ps();
 823         fiy0             = _mm_setzero_ps();
 824         fiz0             = _mm_setzero_ps();
 825         fix1             = _mm_setzero_ps();
 826         fiy1             = _mm_setzero_ps();
 827         fiz1             = _mm_setzero_ps();
 828         fix2             = _mm_setzero_ps();
 829         fiy2             = _mm_setzero_ps();
 830         fiz2             = _mm_setzero_ps();
 831
 832         /* Start inner kernel loop */
 833         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrA             = jjnr[jidx];
 838             jnrB             = jjnr[jidx+1];
 839             jnrC             = jjnr[jidx+2];
 840             jnrD             = jjnr[jidx+3];
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843             j_coord_offsetC  = DIM*jnrC;
 844             j_coord_offsetD  = DIM*jnrD;
 845
 846             /* load j atom coordinates */
 847             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                               x+j_coord_offsetC,x+j_coord_offsetD,
 849                                               &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_ps(ix0,jx0);
 853             dy00             = _mm_sub_ps(iy0,jy0);
 854             dz00             = _mm_sub_ps(iz0,jz0);
 855             dx10             = _mm_sub_ps(ix1,jx0);
 856             dy10             = _mm_sub_ps(iy1,jy0);
 857             dz10             = _mm_sub_ps(iz1,jz0);
 858             dx20             = _mm_sub_ps(ix2,jx0);
 859             dy20             = _mm_sub_ps(iy2,jy0);
 860             dz20             = _mm_sub_ps(iz2,jz0);
 861
 862             /* Calculate squared distance and things based on it */
 863             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 864             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 866
 867             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 868             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 869             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 870
 871             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 872             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 873             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 874
 875             /* Load parameters for j particles */
 876             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 877                                                               charge+jnrC+0,charge+jnrD+0);
 878             vdwjidx0A        = 2*vdwtype[jnrA+0];
 879             vdwjidx0B        = 2*vdwtype[jnrB+0];
 880             vdwjidx0C        = 2*vdwtype[jnrC+0];
 881             vdwjidx0D        = 2*vdwtype[jnrD+0];
 882
 883             fjx0             = _mm_setzero_ps();
 884             fjy0             = _mm_setzero_ps();
 885             fjz0             = _mm_setzero_ps();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             r00              = _mm_mul_ps(rsq00,rinv00);
 892
 893             /* Compute parameters for interactions between i and j atoms */
 894             qq00             = _mm_mul_ps(iq0,jq0);
 895             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 896                                          vdwparam+vdwioffset0+vdwjidx0B,
 897                                          vdwparam+vdwioffset0+vdwjidx0C,
 898                                          vdwparam+vdwioffset0+vdwjidx0D,
 899                                          &c6_00,&c12_00);
 900
 901             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 902                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 903                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 904                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 905
 906             /* EWALD ELECTROSTATICS */
 907
 908             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 909             ewrt             = _mm_mul_ps(r00,ewtabscale);
 910             ewitab           = _mm_cvttps_epi32(ewrt);
 911             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 912             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 913                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 914                                          &ewtabF,&ewtabFn);
 915             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 916             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 917
 918             /* Analytical LJ-PME */
 919             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 920             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 921             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 922             exponent         = gmx_simd_exp_r(ewcljrsq);
 923             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 924             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 925             /* f6A = 6 * C6grid * (1 - poly) */
 926             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 927             /* f6B = C6grid * exponent * beta^6 */
 928             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 929             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 930             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 931
 932             fscal            = _mm_add_ps(felec,fvdw);
 933
 934             /* Calculate temporary vectorial force */
 935             tx               = _mm_mul_ps(fscal,dx00);
 936             ty               = _mm_mul_ps(fscal,dy00);
 937             tz               = _mm_mul_ps(fscal,dz00);
 938
 939             /* Update vectorial force */
 940             fix0             = _mm_add_ps(fix0,tx);
 941             fiy0             = _mm_add_ps(fiy0,ty);
 942             fiz0             = _mm_add_ps(fiz0,tz);
 943
 944             fjx0             = _mm_add_ps(fjx0,tx);
 945             fjy0             = _mm_add_ps(fjy0,ty);
 946             fjz0             = _mm_add_ps(fjz0,tz);
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             r10              = _mm_mul_ps(rsq10,rinv10);
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq10             = _mm_mul_ps(iq1,jq0);
 956
 957             /* EWALD ELECTROSTATICS */
 958
 959             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 960             ewrt             = _mm_mul_ps(r10,ewtabscale);
 961             ewitab           = _mm_cvttps_epi32(ewrt);
 962             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 963             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 964                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 965                                          &ewtabF,&ewtabFn);
 966             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 967             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 968
 969             fscal            = felec;
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm_mul_ps(fscal,dx10);
 973             ty               = _mm_mul_ps(fscal,dy10);
 974             tz               = _mm_mul_ps(fscal,dz10);
 975
 976             /* Update vectorial force */
 977             fix1             = _mm_add_ps(fix1,tx);
 978             fiy1             = _mm_add_ps(fiy1,ty);
 979             fiz1             = _mm_add_ps(fiz1,tz);
 980
 981             fjx0             = _mm_add_ps(fjx0,tx);
 982             fjy0             = _mm_add_ps(fjy0,ty);
 983             fjz0             = _mm_add_ps(fjz0,tz);
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             r20              = _mm_mul_ps(rsq20,rinv20);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq20             = _mm_mul_ps(iq2,jq0);
 993
 994             /* EWALD ELECTROSTATICS */
 995
 996             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 997             ewrt             = _mm_mul_ps(r20,ewtabscale);
 998             ewitab           = _mm_cvttps_epi32(ewrt);
 999             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1000             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1001                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1002                                          &ewtabF,&ewtabFn);
1003             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1004             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1005
1006             fscal            = felec;
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm_mul_ps(fscal,dx20);
1010             ty               = _mm_mul_ps(fscal,dy20);
1011             tz               = _mm_mul_ps(fscal,dz20);
1012
1013             /* Update vectorial force */
1014             fix2             = _mm_add_ps(fix2,tx);
1015             fiy2             = _mm_add_ps(fiy2,ty);
1016             fiz2             = _mm_add_ps(fiz2,tz);
1017
1018             fjx0             = _mm_add_ps(fjx0,tx);
1019             fjy0             = _mm_add_ps(fjy0,ty);
1020             fjz0             = _mm_add_ps(fjz0,tz);
1021
1022             fjptrA             = f+j_coord_offsetA;
1023             fjptrB             = f+j_coord_offsetB;
1024             fjptrC             = f+j_coord_offsetC;
1025             fjptrD             = f+j_coord_offsetD;
1026
1027             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1028
1029             /* Inner loop uses 131 flops */
1030         }
1031
1032         if(jidx<j_index_end)
1033         {
1034
1035             /* Get j neighbor index, and coordinate index */
1036             jnrlistA         = jjnr[jidx];
1037             jnrlistB         = jjnr[jidx+1];
1038             jnrlistC         = jjnr[jidx+2];
1039             jnrlistD         = jjnr[jidx+3];
1040             /* Sign of each element will be negative for non-real atoms.
1041              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1042              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1043              */
1044             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1045             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1046             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1047             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1048             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1049             j_coord_offsetA  = DIM*jnrA;
1050             j_coord_offsetB  = DIM*jnrB;
1051             j_coord_offsetC  = DIM*jnrC;
1052             j_coord_offsetD  = DIM*jnrD;
1053
1054             /* load j atom coordinates */
1055             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1056                                               x+j_coord_offsetC,x+j_coord_offsetD,
1057                                               &jx0,&jy0,&jz0);
1058
1059             /* Calculate displacement vector */
1060             dx00             = _mm_sub_ps(ix0,jx0);
1061             dy00             = _mm_sub_ps(iy0,jy0);
1062             dz00             = _mm_sub_ps(iz0,jz0);
1063             dx10             = _mm_sub_ps(ix1,jx0);
1064             dy10             = _mm_sub_ps(iy1,jy0);
1065             dz10             = _mm_sub_ps(iz1,jz0);
1066             dx20             = _mm_sub_ps(ix2,jx0);
1067             dy20             = _mm_sub_ps(iy2,jy0);
1068             dz20             = _mm_sub_ps(iz2,jz0);
1069
1070             /* Calculate squared distance and things based on it */
1071             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1072             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1073             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1074
1075             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1076             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1077             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1078
1079             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1080             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1081             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1082
1083             /* Load parameters for j particles */
1084             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1085                                                               charge+jnrC+0,charge+jnrD+0);
1086             vdwjidx0A        = 2*vdwtype[jnrA+0];
1087             vdwjidx0B        = 2*vdwtype[jnrB+0];
1088             vdwjidx0C        = 2*vdwtype[jnrC+0];
1089             vdwjidx0D        = 2*vdwtype[jnrD+0];
1090
1091             fjx0             = _mm_setzero_ps();
1092             fjy0             = _mm_setzero_ps();
1093             fjz0             = _mm_setzero_ps();
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r00              = _mm_mul_ps(rsq00,rinv00);
1100             r00              = _mm_andnot_ps(dummy_mask,r00);
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq00             = _mm_mul_ps(iq0,jq0);
1104             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1105                                          vdwparam+vdwioffset0+vdwjidx0B,
1106                                          vdwparam+vdwioffset0+vdwjidx0C,
1107                                          vdwparam+vdwioffset0+vdwjidx0D,
1108                                          &c6_00,&c12_00);
1109
1110             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1111                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1112                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1113                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1114
1115             /* EWALD ELECTROSTATICS */
1116
1117             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1118             ewrt             = _mm_mul_ps(r00,ewtabscale);
1119             ewitab           = _mm_cvttps_epi32(ewrt);
1120             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1121             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1122                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1123                                          &ewtabF,&ewtabFn);
1124             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1125             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1126
1127             /* Analytical LJ-PME */
1128             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1129             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1130             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1131             exponent         = gmx_simd_exp_r(ewcljrsq);
1132             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1133             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1134             /* f6A = 6 * C6grid * (1 - poly) */
1135             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1136             /* f6B = C6grid * exponent * beta^6 */
1137             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1138             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1139             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1140
1141             fscal            = _mm_add_ps(felec,fvdw);
1142
1143             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1144
1145             /* Calculate temporary vectorial force */
1146             tx               = _mm_mul_ps(fscal,dx00);
1147             ty               = _mm_mul_ps(fscal,dy00);
1148             tz               = _mm_mul_ps(fscal,dz00);
1149
1150             /* Update vectorial force */
1151             fix0             = _mm_add_ps(fix0,tx);
1152             fiy0             = _mm_add_ps(fiy0,ty);
1153             fiz0             = _mm_add_ps(fiz0,tz);
1154
1155             fjx0             = _mm_add_ps(fjx0,tx);
1156             fjy0             = _mm_add_ps(fjy0,ty);
1157             fjz0             = _mm_add_ps(fjz0,tz);
1158
1159             /**************************
1160              * CALCULATE INTERACTIONS *
1161              **************************/
1162
1163             r10              = _mm_mul_ps(rsq10,rinv10);
1164             r10              = _mm_andnot_ps(dummy_mask,r10);
1165
1166             /* Compute parameters for interactions between i and j atoms */
1167             qq10             = _mm_mul_ps(iq1,jq0);
1168
1169             /* EWALD ELECTROSTATICS */
1170
1171             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1172             ewrt             = _mm_mul_ps(r10,ewtabscale);
1173             ewitab           = _mm_cvttps_epi32(ewrt);
1174             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1175             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1176                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1177                                          &ewtabF,&ewtabFn);
1178             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1179             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1180
1181             fscal            = felec;
1182
1183             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm_mul_ps(fscal,dx10);
1187             ty               = _mm_mul_ps(fscal,dy10);
1188             tz               = _mm_mul_ps(fscal,dz10);
1189
1190             /* Update vectorial force */
1191             fix1             = _mm_add_ps(fix1,tx);
1192             fiy1             = _mm_add_ps(fiy1,ty);
1193             fiz1             = _mm_add_ps(fiz1,tz);
1194
1195             fjx0             = _mm_add_ps(fjx0,tx);
1196             fjy0             = _mm_add_ps(fjy0,ty);
1197             fjz0             = _mm_add_ps(fjz0,tz);
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             r20              = _mm_mul_ps(rsq20,rinv20);
1204             r20              = _mm_andnot_ps(dummy_mask,r20);
1205
1206             /* Compute parameters for interactions between i and j atoms */
1207             qq20             = _mm_mul_ps(iq2,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(r20,ewtabscale);
1213             ewitab           = _mm_cvttps_epi32(ewrt);
1214             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1215             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1216                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1217                                          &ewtabF,&ewtabFn);
1218             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1219             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1220
1221             fscal            = felec;
1222
1223             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1224
1225             /* Calculate temporary vectorial force */
1226             tx               = _mm_mul_ps(fscal,dx20);
1227             ty               = _mm_mul_ps(fscal,dy20);
1228             tz               = _mm_mul_ps(fscal,dz20);
1229
1230             /* Update vectorial force */
1231             fix2             = _mm_add_ps(fix2,tx);
1232             fiy2             = _mm_add_ps(fiy2,ty);
1233             fiz2             = _mm_add_ps(fiz2,tz);
1234
1235             fjx0             = _mm_add_ps(fjx0,tx);
1236             fjy0             = _mm_add_ps(fjy0,ty);
1237             fjz0             = _mm_add_ps(fjz0,tz);
1238
1239             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1240             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1241             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1242             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1243
1244             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1245
1246             /* Inner loop uses 134 flops */
1247         }
1248
1249         /* End of innermost loop */
1250
1251         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1252                                               f+i_coord_offset,fshift+i_shift_offset);
1253
1254         /* Increment number of inner iterations */
1255         inneriter                  += j_index_end - j_index_start;
1256
1257         /* Outer loop uses 18 flops */
1258     }
1259
1260     /* Increment number of outer iterations */
1261     outeriter        += nri;
1262
1263     /* Update outer/inner flops */
1264
1265     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1266 }