src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  96     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  97     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  98     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  99     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 100     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 101     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 102     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 105     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 108     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 109     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 110     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 111     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 112     real             *charge;
 113     int              nvdwtype;
 114     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 115     int              *vdwtype;
 116     real             *vdwparam;
 117     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 118     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 119     __m128           c6grid_00;
 120     __m128           c6grid_11;
 121     __m128           c6grid_12;
 122     __m128           c6grid_13;
 123     __m128           c6grid_21;
 124     __m128           c6grid_22;
 125     __m128           c6grid_23;
 126     __m128           c6grid_31;
 127     __m128           c6grid_32;
 128     __m128           c6grid_33;
 129     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 130     real             *vdwgridparam;
 131     __m128           one_half  = _mm_set1_ps(0.5);
 132     __m128           minus_one = _mm_set1_ps(-1.0);
 133     __m128i          ewitab;
 134     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 135     real             *ewtab;
 136     __m128           dummy_mask,cutoff_mask;
 137     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 138     __m128           one     = _mm_set1_ps(1.0);
 139     __m128           two     = _mm_set1_ps(2.0);
 140     x                = xx[0];
 141     f                = ff[0];
 142
 143     nri              = nlist->nri;
 144     iinr             = nlist->iinr;
 145     jindex           = nlist->jindex;
 146     jjnr             = nlist->jjnr;
 147     shiftidx         = nlist->shift;
 148     gid              = nlist->gid;
 149     shiftvec         = fr->shift_vec[0];
 150     fshift           = fr->fshift[0];
 151     facel            = _mm_set1_ps(fr->epsfac);
 152     charge           = mdatoms->chargeA;
 153     nvdwtype         = fr->ntype;
 154     vdwparam         = fr->nbfp;
 155     vdwtype          = mdatoms->typeA;
 156     vdwgridparam     = fr->ljpme_c6grid;
 157     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 158     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 159     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 160
 161     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 162     ewtab            = fr->ic->tabq_coul_FDV0;
 163     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 164     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 165
 166     /* Setup water-specific parameters */
 167     inr              = nlist->iinr[0];
 168     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 169     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 170     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 171     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 172
 173     jq1              = _mm_set1_ps(charge[inr+1]);
 174     jq2              = _mm_set1_ps(charge[inr+2]);
 175     jq3              = _mm_set1_ps(charge[inr+3]);
 176     vdwjidx0A        = 2*vdwtype[inr+0];
 177     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 178     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 179     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
 180     qq11             = _mm_mul_ps(iq1,jq1);
 181     qq12             = _mm_mul_ps(iq1,jq2);
 182     qq13             = _mm_mul_ps(iq1,jq3);
 183     qq21             = _mm_mul_ps(iq2,jq1);
 184     qq22             = _mm_mul_ps(iq2,jq2);
 185     qq23             = _mm_mul_ps(iq2,jq3);
 186     qq31             = _mm_mul_ps(iq3,jq1);
 187     qq32             = _mm_mul_ps(iq3,jq2);
 188     qq33             = _mm_mul_ps(iq3,jq3);
 189
 190     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 191     rcutoff_scalar   = fr->rcoulomb;
 192     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 193     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 194
 195     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 196     rvdw             = _mm_set1_ps(fr->rvdw);
 197
 198     /* Avoid stupid compiler warnings */
 199     jnrA = jnrB = jnrC = jnrD = 0;
 200     j_coord_offsetA = 0;
 201     j_coord_offsetB = 0;
 202     j_coord_offsetC = 0;
 203     j_coord_offsetD = 0;
 204
 205     outeriter        = 0;
 206     inneriter        = 0;
 207
 208     for(iidx=0;iidx<4*DIM;iidx++)
 209     {
 210         scratch[iidx] = 0.0;
 211     }
 212
 213     /* Start outer loop over neighborlists */
 214     for(iidx=0; iidx<nri; iidx++)
 215     {
 216         /* Load shift vector for this list */
 217         i_shift_offset   = DIM*shiftidx[iidx];
 218
 219         /* Load limits for loop over neighbors */
 220         j_index_start    = jindex[iidx];
 221         j_index_end      = jindex[iidx+1];
 222
 223         /* Get outer coordinate index */
 224         inr              = iinr[iidx];
 225         i_coord_offset   = DIM*inr;
 226
 227         /* Load i particle coords and add shift vector */
 228         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 229                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 230
 231         fix0             = _mm_setzero_ps();
 232         fiy0             = _mm_setzero_ps();
 233         fiz0             = _mm_setzero_ps();
 234         fix1             = _mm_setzero_ps();
 235         fiy1             = _mm_setzero_ps();
 236         fiz1             = _mm_setzero_ps();
 237         fix2             = _mm_setzero_ps();
 238         fiy2             = _mm_setzero_ps();
 239         fiz2             = _mm_setzero_ps();
 240         fix3             = _mm_setzero_ps();
 241         fiy3             = _mm_setzero_ps();
 242         fiz3             = _mm_setzero_ps();
 243
 244         /* Reset potential sums */
 245         velecsum         = _mm_setzero_ps();
 246         vvdwsum          = _mm_setzero_ps();
 247
 248         /* Start inner kernel loop */
 249         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 250         {
 251
 252             /* Get j neighbor index, and coordinate index */
 253             jnrA             = jjnr[jidx];
 254             jnrB             = jjnr[jidx+1];
 255             jnrC             = jjnr[jidx+2];
 256             jnrD             = jjnr[jidx+3];
 257             j_coord_offsetA  = DIM*jnrA;
 258             j_coord_offsetB  = DIM*jnrB;
 259             j_coord_offsetC  = DIM*jnrC;
 260             j_coord_offsetD  = DIM*jnrD;
 261
 262             /* load j atom coordinates */
 263             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 264                                               x+j_coord_offsetC,x+j_coord_offsetD,
 265                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 266                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 267
 268             /* Calculate displacement vector */
 269             dx00             = _mm_sub_ps(ix0,jx0);
 270             dy00             = _mm_sub_ps(iy0,jy0);
 271             dz00             = _mm_sub_ps(iz0,jz0);
 272             dx11             = _mm_sub_ps(ix1,jx1);
 273             dy11             = _mm_sub_ps(iy1,jy1);
 274             dz11             = _mm_sub_ps(iz1,jz1);
 275             dx12             = _mm_sub_ps(ix1,jx2);
 276             dy12             = _mm_sub_ps(iy1,jy2);
 277             dz12             = _mm_sub_ps(iz1,jz2);
 278             dx13             = _mm_sub_ps(ix1,jx3);
 279             dy13             = _mm_sub_ps(iy1,jy3);
 280             dz13             = _mm_sub_ps(iz1,jz3);
 281             dx21             = _mm_sub_ps(ix2,jx1);
 282             dy21             = _mm_sub_ps(iy2,jy1);
 283             dz21             = _mm_sub_ps(iz2,jz1);
 284             dx22             = _mm_sub_ps(ix2,jx2);
 285             dy22             = _mm_sub_ps(iy2,jy2);
 286             dz22             = _mm_sub_ps(iz2,jz2);
 287             dx23             = _mm_sub_ps(ix2,jx3);
 288             dy23             = _mm_sub_ps(iy2,jy3);
 289             dz23             = _mm_sub_ps(iz2,jz3);
 290             dx31             = _mm_sub_ps(ix3,jx1);
 291             dy31             = _mm_sub_ps(iy3,jy1);
 292             dz31             = _mm_sub_ps(iz3,jz1);
 293             dx32             = _mm_sub_ps(ix3,jx2);
 294             dy32             = _mm_sub_ps(iy3,jy2);
 295             dz32             = _mm_sub_ps(iz3,jz2);
 296             dx33             = _mm_sub_ps(ix3,jx3);
 297             dy33             = _mm_sub_ps(iy3,jy3);
 298             dz33             = _mm_sub_ps(iz3,jz3);
 299
 300             /* Calculate squared distance and things based on it */
 301             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 302             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 303             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 304             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 305             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 306             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 307             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 308             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 309             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 310             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 311
 312             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 313             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 314             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 315             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 316             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 317             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 318             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 319             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 320             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 321             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 322
 323             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 324             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 325             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 326             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 327             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 328             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 329             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 330             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 331             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 332             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 333
 334             fjx0             = _mm_setzero_ps();
 335             fjy0             = _mm_setzero_ps();
 336             fjz0             = _mm_setzero_ps();
 337             fjx1             = _mm_setzero_ps();
 338             fjy1             = _mm_setzero_ps();
 339             fjz1             = _mm_setzero_ps();
 340             fjx2             = _mm_setzero_ps();
 341             fjy2             = _mm_setzero_ps();
 342             fjz2             = _mm_setzero_ps();
 343             fjx3             = _mm_setzero_ps();
 344             fjy3             = _mm_setzero_ps();
 345             fjz3             = _mm_setzero_ps();
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             if (gmx_mm_any_lt(rsq00,rcutoff2))
 352             {
 353
 354             r00              = _mm_mul_ps(rsq00,rinv00);
 355
 356             /* Analytical LJ-PME */
 357             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 358             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 359             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 360             exponent         = gmx_simd_exp_r(ewcljrsq);
 361             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 362             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 363             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 364             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 365             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 366             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 367                                _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 368             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 369             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);
 370
 371             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 375             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 376
 377             fscal            = fvdw;
 378
 379             fscal            = _mm_and_ps(fscal,cutoff_mask);
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_ps(fscal,dx00);
 383             ty               = _mm_mul_ps(fscal,dy00);
 384             tz               = _mm_mul_ps(fscal,dz00);
 385
 386             /* Update vectorial force */
 387             fix0             = _mm_add_ps(fix0,tx);
 388             fiy0             = _mm_add_ps(fiy0,ty);
 389             fiz0             = _mm_add_ps(fiz0,tz);
 390
 391             fjx0             = _mm_add_ps(fjx0,tx);
 392             fjy0             = _mm_add_ps(fjy0,ty);
 393             fjz0             = _mm_add_ps(fjz0,tz);
 394
 395             }
 396
 397             /**************************
 398              * CALCULATE INTERACTIONS *
 399              **************************/
 400
 401             if (gmx_mm_any_lt(rsq11,rcutoff2))
 402             {
 403
 404             r11              = _mm_mul_ps(rsq11,rinv11);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 409             ewrt             = _mm_mul_ps(r11,ewtabscale);
 410             ewitab           = _mm_cvttps_epi32(ewrt);
 411             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 412             ewitab           = _mm_slli_epi32(ewitab,2);
 413             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 414             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 415             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 416             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 417             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 418             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 419             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 420             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec));
 421             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
 422
 423             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velec            = _mm_and_ps(velec,cutoff_mask);
 427             velecsum         = _mm_add_ps(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             fscal            = _mm_and_ps(fscal,cutoff_mask);
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm_mul_ps(fscal,dx11);
 435             ty               = _mm_mul_ps(fscal,dy11);
 436             tz               = _mm_mul_ps(fscal,dz11);
 437
 438             /* Update vectorial force */
 439             fix1             = _mm_add_ps(fix1,tx);
 440             fiy1             = _mm_add_ps(fiy1,ty);
 441             fiz1             = _mm_add_ps(fiz1,tz);
 442
 443             fjx1             = _mm_add_ps(fjx1,tx);
 444             fjy1             = _mm_add_ps(fjy1,ty);
 445             fjz1             = _mm_add_ps(fjz1,tz);
 446
 447             }
 448
 449             /**************************
 450              * CALCULATE INTERACTIONS *
 451              **************************/
 452
 453             if (gmx_mm_any_lt(rsq12,rcutoff2))
 454             {
 455
 456             r12              = _mm_mul_ps(rsq12,rinv12);
 457
 458             /* EWALD ELECTROSTATICS */
 459
 460             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 461             ewrt             = _mm_mul_ps(r12,ewtabscale);
 462             ewitab           = _mm_cvttps_epi32(ewrt);
 463             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 464             ewitab           = _mm_slli_epi32(ewitab,2);
 465             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 466             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 467             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 468             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 469             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 470             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 471             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 472             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec));
 473             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
 474
 475             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             velec            = _mm_and_ps(velec,cutoff_mask);
 479             velecsum         = _mm_add_ps(velecsum,velec);
 480
 481             fscal            = felec;
 482
 483             fscal            = _mm_and_ps(fscal,cutoff_mask);
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm_mul_ps(fscal,dx12);
 487             ty               = _mm_mul_ps(fscal,dy12);
 488             tz               = _mm_mul_ps(fscal,dz12);
 489
 490             /* Update vectorial force */
 491             fix1             = _mm_add_ps(fix1,tx);
 492             fiy1             = _mm_add_ps(fiy1,ty);
 493             fiz1             = _mm_add_ps(fiz1,tz);
 494
 495             fjx2             = _mm_add_ps(fjx2,tx);
 496             fjy2             = _mm_add_ps(fjy2,ty);
 497             fjz2             = _mm_add_ps(fjz2,tz);
 498
 499             }
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             if (gmx_mm_any_lt(rsq13,rcutoff2))
 506             {
 507
 508             r13              = _mm_mul_ps(rsq13,rinv13);
 509
 510             /* EWALD ELECTROSTATICS */
 511
 512             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 513             ewrt             = _mm_mul_ps(r13,ewtabscale);
 514             ewitab           = _mm_cvttps_epi32(ewrt);
 515             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 516             ewitab           = _mm_slli_epi32(ewitab,2);
 517             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 518             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 519             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 520             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 521             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 522             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 523             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 524             velec            = _mm_mul_ps(qq13,_mm_sub_ps(_mm_sub_ps(rinv13,sh_ewald),velec));
 525             felec            = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
 526
 527             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm_and_ps(velec,cutoff_mask);
 531             velecsum         = _mm_add_ps(velecsum,velec);
 532
 533             fscal            = felec;
 534
 535             fscal            = _mm_and_ps(fscal,cutoff_mask);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm_mul_ps(fscal,dx13);
 539             ty               = _mm_mul_ps(fscal,dy13);
 540             tz               = _mm_mul_ps(fscal,dz13);
 541
 542             /* Update vectorial force */
 543             fix1             = _mm_add_ps(fix1,tx);
 544             fiy1             = _mm_add_ps(fiy1,ty);
 545             fiz1             = _mm_add_ps(fiz1,tz);
 546
 547             fjx3             = _mm_add_ps(fjx3,tx);
 548             fjy3             = _mm_add_ps(fjy3,ty);
 549             fjz3             = _mm_add_ps(fjz3,tz);
 550
 551             }
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm_any_lt(rsq21,rcutoff2))
 558             {
 559
 560             r21              = _mm_mul_ps(rsq21,rinv21);
 561
 562             /* EWALD ELECTROSTATICS */
 563
 564             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 565             ewrt             = _mm_mul_ps(r21,ewtabscale);
 566             ewitab           = _mm_cvttps_epi32(ewrt);
 567             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 568             ewitab           = _mm_slli_epi32(ewitab,2);
 569             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 570             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 571             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 572             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 573             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 574             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 575             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 576             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec));
 577             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
 578
 579             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_and_ps(velec,cutoff_mask);
 583             velecsum         = _mm_add_ps(velecsum,velec);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm_and_ps(fscal,cutoff_mask);
 588
 589             /* Calculate temporary vectorial force */
 590             tx               = _mm_mul_ps(fscal,dx21);
 591             ty               = _mm_mul_ps(fscal,dy21);
 592             tz               = _mm_mul_ps(fscal,dz21);
 593
 594             /* Update vectorial force */
 595             fix2             = _mm_add_ps(fix2,tx);
 596             fiy2             = _mm_add_ps(fiy2,ty);
 597             fiz2             = _mm_add_ps(fiz2,tz);
 598
 599             fjx1             = _mm_add_ps(fjx1,tx);
 600             fjy1             = _mm_add_ps(fjy1,ty);
 601             fjz1             = _mm_add_ps(fjz1,tz);
 602
 603             }
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_mm_any_lt(rsq22,rcutoff2))
 610             {
 611
 612             r22              = _mm_mul_ps(rsq22,rinv22);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 617             ewrt             = _mm_mul_ps(r22,ewtabscale);
 618             ewitab           = _mm_cvttps_epi32(ewrt);
 619             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 620             ewitab           = _mm_slli_epi32(ewitab,2);
 621             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 622             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 623             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 624             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 625             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 626             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 627             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 628             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec));
 629             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
 630
 631             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
 632
 633             /* Update potential sum for this i atom from the interaction with this j atom. */
 634             velec            = _mm_and_ps(velec,cutoff_mask);
 635             velecsum         = _mm_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             fscal            = _mm_and_ps(fscal,cutoff_mask);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm_mul_ps(fscal,dx22);
 643             ty               = _mm_mul_ps(fscal,dy22);
 644             tz               = _mm_mul_ps(fscal,dz22);
 645
 646             /* Update vectorial force */
 647             fix2             = _mm_add_ps(fix2,tx);
 648             fiy2             = _mm_add_ps(fiy2,ty);
 649             fiz2             = _mm_add_ps(fiz2,tz);
 650
 651             fjx2             = _mm_add_ps(fjx2,tx);
 652             fjy2             = _mm_add_ps(fjy2,ty);
 653             fjz2             = _mm_add_ps(fjz2,tz);
 654
 655             }
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             if (gmx_mm_any_lt(rsq23,rcutoff2))
 662             {
 663
 664             r23              = _mm_mul_ps(rsq23,rinv23);
 665
 666             /* EWALD ELECTROSTATICS */
 667
 668             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 669             ewrt             = _mm_mul_ps(r23,ewtabscale);
 670             ewitab           = _mm_cvttps_epi32(ewrt);
 671             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 672             ewitab           = _mm_slli_epi32(ewitab,2);
 673             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 674             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 675             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 676             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 677             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 678             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 679             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 680             velec            = _mm_mul_ps(qq23,_mm_sub_ps(_mm_sub_ps(rinv23,sh_ewald),velec));
 681             felec            = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
 682
 683             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
 684
 685             /* Update potential sum for this i atom from the interaction with this j atom. */
 686             velec            = _mm_and_ps(velec,cutoff_mask);
 687             velecsum         = _mm_add_ps(velecsum,velec);
 688
 689             fscal            = felec;
 690
 691             fscal            = _mm_and_ps(fscal,cutoff_mask);
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm_mul_ps(fscal,dx23);
 695             ty               = _mm_mul_ps(fscal,dy23);
 696             tz               = _mm_mul_ps(fscal,dz23);
 697
 698             /* Update vectorial force */
 699             fix2             = _mm_add_ps(fix2,tx);
 700             fiy2             = _mm_add_ps(fiy2,ty);
 701             fiz2             = _mm_add_ps(fiz2,tz);
 702
 703             fjx3             = _mm_add_ps(fjx3,tx);
 704             fjy3             = _mm_add_ps(fjy3,ty);
 705             fjz3             = _mm_add_ps(fjz3,tz);
 706
 707             }
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             if (gmx_mm_any_lt(rsq31,rcutoff2))
 714             {
 715
 716             r31              = _mm_mul_ps(rsq31,rinv31);
 717
 718             /* EWALD ELECTROSTATICS */
 719
 720             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 721             ewrt             = _mm_mul_ps(r31,ewtabscale);
 722             ewitab           = _mm_cvttps_epi32(ewrt);
 723             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 724             ewitab           = _mm_slli_epi32(ewitab,2);
 725             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 726             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 727             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 728             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 729             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 730             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 731             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 732             velec            = _mm_mul_ps(qq31,_mm_sub_ps(_mm_sub_ps(rinv31,sh_ewald),velec));
 733             felec            = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
 734
 735             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
 736
 737             /* Update potential sum for this i atom from the interaction with this j atom. */
 738             velec            = _mm_and_ps(velec,cutoff_mask);
 739             velecsum         = _mm_add_ps(velecsum,velec);
 740
 741             fscal            = felec;
 742
 743             fscal            = _mm_and_ps(fscal,cutoff_mask);
 744
 745             /* Calculate temporary vectorial force */
 746             tx               = _mm_mul_ps(fscal,dx31);
 747             ty               = _mm_mul_ps(fscal,dy31);
 748             tz               = _mm_mul_ps(fscal,dz31);
 749
 750             /* Update vectorial force */
 751             fix3             = _mm_add_ps(fix3,tx);
 752             fiy3             = _mm_add_ps(fiy3,ty);
 753             fiz3             = _mm_add_ps(fiz3,tz);
 754
 755             fjx1             = _mm_add_ps(fjx1,tx);
 756             fjy1             = _mm_add_ps(fjy1,ty);
 757             fjz1             = _mm_add_ps(fjz1,tz);
 758
 759             }
 760
 761             /**************************
 762              * CALCULATE INTERACTIONS *
 763              **************************/
 764
 765             if (gmx_mm_any_lt(rsq32,rcutoff2))
 766             {
 767
 768             r32              = _mm_mul_ps(rsq32,rinv32);
 769
 770             /* EWALD ELECTROSTATICS */
 771
 772             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 773             ewrt             = _mm_mul_ps(r32,ewtabscale);
 774             ewitab           = _mm_cvttps_epi32(ewrt);
 775             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 776             ewitab           = _mm_slli_epi32(ewitab,2);
 777             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 778             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 779             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 780             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 781             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 782             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 783             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 784             velec            = _mm_mul_ps(qq32,_mm_sub_ps(_mm_sub_ps(rinv32,sh_ewald),velec));
 785             felec            = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
 786
 787             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
 788
 789             /* Update potential sum for this i atom from the interaction with this j atom. */
 790             velec            = _mm_and_ps(velec,cutoff_mask);
 791             velecsum         = _mm_add_ps(velecsum,velec);
 792
 793             fscal            = felec;
 794
 795             fscal            = _mm_and_ps(fscal,cutoff_mask);
 796
 797             /* Calculate temporary vectorial force */
 798             tx               = _mm_mul_ps(fscal,dx32);
 799             ty               = _mm_mul_ps(fscal,dy32);
 800             tz               = _mm_mul_ps(fscal,dz32);
 801
 802             /* Update vectorial force */
 803             fix3             = _mm_add_ps(fix3,tx);
 804             fiy3             = _mm_add_ps(fiy3,ty);
 805             fiz3             = _mm_add_ps(fiz3,tz);
 806
 807             fjx2             = _mm_add_ps(fjx2,tx);
 808             fjy2             = _mm_add_ps(fjy2,ty);
 809             fjz2             = _mm_add_ps(fjz2,tz);
 810
 811             }
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             if (gmx_mm_any_lt(rsq33,rcutoff2))
 818             {
 819
 820             r33              = _mm_mul_ps(rsq33,rinv33);
 821
 822             /* EWALD ELECTROSTATICS */
 823
 824             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 825             ewrt             = _mm_mul_ps(r33,ewtabscale);
 826             ewitab           = _mm_cvttps_epi32(ewrt);
 827             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 828             ewitab           = _mm_slli_epi32(ewitab,2);
 829             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 830             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 831             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 832             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 833             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 834             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 835             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 836             velec            = _mm_mul_ps(qq33,_mm_sub_ps(_mm_sub_ps(rinv33,sh_ewald),velec));
 837             felec            = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
 838
 839             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
 840
 841             /* Update potential sum for this i atom from the interaction with this j atom. */
 842             velec            = _mm_and_ps(velec,cutoff_mask);
 843             velecsum         = _mm_add_ps(velecsum,velec);
 844
 845             fscal            = felec;
 846
 847             fscal            = _mm_and_ps(fscal,cutoff_mask);
 848
 849             /* Calculate temporary vectorial force */
 850             tx               = _mm_mul_ps(fscal,dx33);
 851             ty               = _mm_mul_ps(fscal,dy33);
 852             tz               = _mm_mul_ps(fscal,dz33);
 853
 854             /* Update vectorial force */
 855             fix3             = _mm_add_ps(fix3,tx);
 856             fiy3             = _mm_add_ps(fiy3,ty);
 857             fiz3             = _mm_add_ps(fiz3,tz);
 858
 859             fjx3             = _mm_add_ps(fjx3,tx);
 860             fjy3             = _mm_add_ps(fjy3,ty);
 861             fjz3             = _mm_add_ps(fjz3,tz);
 862
 863             }
 864
 865             fjptrA             = f+j_coord_offsetA;
 866             fjptrB             = f+j_coord_offsetB;
 867             fjptrC             = f+j_coord_offsetC;
 868             fjptrD             = f+j_coord_offsetD;
 869
 870             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 871                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 872                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 873
 874             /* Inner loop uses 479 flops */
 875         }
 876
 877         if(jidx<j_index_end)
 878         {
 879
 880             /* Get j neighbor index, and coordinate index */
 881             jnrlistA         = jjnr[jidx];
 882             jnrlistB         = jjnr[jidx+1];
 883             jnrlistC         = jjnr[jidx+2];
 884             jnrlistD         = jjnr[jidx+3];
 885             /* Sign of each element will be negative for non-real atoms.
 886              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 887              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 888              */
 889             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 890             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 891             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 892             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 893             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 894             j_coord_offsetA  = DIM*jnrA;
 895             j_coord_offsetB  = DIM*jnrB;
 896             j_coord_offsetC  = DIM*jnrC;
 897             j_coord_offsetD  = DIM*jnrD;
 898
 899             /* load j atom coordinates */
 900             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 901                                               x+j_coord_offsetC,x+j_coord_offsetD,
 902                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 903                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 904
 905             /* Calculate displacement vector */
 906             dx00             = _mm_sub_ps(ix0,jx0);
 907             dy00             = _mm_sub_ps(iy0,jy0);
 908             dz00             = _mm_sub_ps(iz0,jz0);
 909             dx11             = _mm_sub_ps(ix1,jx1);
 910             dy11             = _mm_sub_ps(iy1,jy1);
 911             dz11             = _mm_sub_ps(iz1,jz1);
 912             dx12             = _mm_sub_ps(ix1,jx2);
 913             dy12             = _mm_sub_ps(iy1,jy2);
 914             dz12             = _mm_sub_ps(iz1,jz2);
 915             dx13             = _mm_sub_ps(ix1,jx3);
 916             dy13             = _mm_sub_ps(iy1,jy3);
 917             dz13             = _mm_sub_ps(iz1,jz3);
 918             dx21             = _mm_sub_ps(ix2,jx1);
 919             dy21             = _mm_sub_ps(iy2,jy1);
 920             dz21             = _mm_sub_ps(iz2,jz1);
 921             dx22             = _mm_sub_ps(ix2,jx2);
 922             dy22             = _mm_sub_ps(iy2,jy2);
 923             dz22             = _mm_sub_ps(iz2,jz2);
 924             dx23             = _mm_sub_ps(ix2,jx3);
 925             dy23             = _mm_sub_ps(iy2,jy3);
 926             dz23             = _mm_sub_ps(iz2,jz3);
 927             dx31             = _mm_sub_ps(ix3,jx1);
 928             dy31             = _mm_sub_ps(iy3,jy1);
 929             dz31             = _mm_sub_ps(iz3,jz1);
 930             dx32             = _mm_sub_ps(ix3,jx2);
 931             dy32             = _mm_sub_ps(iy3,jy2);
 932             dz32             = _mm_sub_ps(iz3,jz2);
 933             dx33             = _mm_sub_ps(ix3,jx3);
 934             dy33             = _mm_sub_ps(iy3,jy3);
 935             dz33             = _mm_sub_ps(iz3,jz3);
 936
 937             /* Calculate squared distance and things based on it */
 938             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 939             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 940             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 941             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 942             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 943             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 944             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 945             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 946             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 947             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 948
 949             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 950             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 951             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 952             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 953             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 954             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 955             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 956             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 957             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 958             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 959
 960             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 961             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 962             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 963             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 964             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 965             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 966             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 967             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 968             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 969             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 970
 971             fjx0             = _mm_setzero_ps();
 972             fjy0             = _mm_setzero_ps();
 973             fjz0             = _mm_setzero_ps();
 974             fjx1             = _mm_setzero_ps();
 975             fjy1             = _mm_setzero_ps();
 976             fjz1             = _mm_setzero_ps();
 977             fjx2             = _mm_setzero_ps();
 978             fjy2             = _mm_setzero_ps();
 979             fjz2             = _mm_setzero_ps();
 980             fjx3             = _mm_setzero_ps();
 981             fjy3             = _mm_setzero_ps();
 982             fjz3             = _mm_setzero_ps();
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             if (gmx_mm_any_lt(rsq00,rcutoff2))
 989             {
 990
 991             r00              = _mm_mul_ps(rsq00,rinv00);
 992             r00              = _mm_andnot_ps(dummy_mask,r00);
 993
 994             /* Analytical LJ-PME */
 995             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 996             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 997             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 998             exponent         = gmx_simd_exp_r(ewcljrsq);
 999             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1000             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1001             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
1002             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
1003             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1004             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
1005                                _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
1006             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
1007             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);
1008
1009             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1010
1011             /* Update potential sum for this i atom from the interaction with this j atom. */
1012             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
1013             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
1014             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
1015
1016             fscal            = fvdw;
1017
1018             fscal            = _mm_and_ps(fscal,cutoff_mask);
1019
1020             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm_mul_ps(fscal,dx00);
1024             ty               = _mm_mul_ps(fscal,dy00);
1025             tz               = _mm_mul_ps(fscal,dz00);
1026
1027             /* Update vectorial force */
1028             fix0             = _mm_add_ps(fix0,tx);
1029             fiy0             = _mm_add_ps(fiy0,ty);
1030             fiz0             = _mm_add_ps(fiz0,tz);
1031
1032             fjx0             = _mm_add_ps(fjx0,tx);
1033             fjy0             = _mm_add_ps(fjy0,ty);
1034             fjz0             = _mm_add_ps(fjz0,tz);
1035
1036             }
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             if (gmx_mm_any_lt(rsq11,rcutoff2))
1043             {
1044
1045             r11              = _mm_mul_ps(rsq11,rinv11);
1046             r11              = _mm_andnot_ps(dummy_mask,r11);
1047
1048             /* EWALD ELECTROSTATICS */
1049
1050             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1051             ewrt             = _mm_mul_ps(r11,ewtabscale);
1052             ewitab           = _mm_cvttps_epi32(ewrt);
1053             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1054             ewitab           = _mm_slli_epi32(ewitab,2);
1055             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1056             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1057             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1058             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1059             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1060             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1061             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1062             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec));
1063             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1064
1065             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1066
1067             /* Update potential sum for this i atom from the interaction with this j atom. */
1068             velec            = _mm_and_ps(velec,cutoff_mask);
1069             velec            = _mm_andnot_ps(dummy_mask,velec);
1070             velecsum         = _mm_add_ps(velecsum,velec);
1071
1072             fscal            = felec;
1073
1074             fscal            = _mm_and_ps(fscal,cutoff_mask);
1075
1076             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm_mul_ps(fscal,dx11);
1080             ty               = _mm_mul_ps(fscal,dy11);
1081             tz               = _mm_mul_ps(fscal,dz11);
1082
1083             /* Update vectorial force */
1084             fix1             = _mm_add_ps(fix1,tx);
1085             fiy1             = _mm_add_ps(fiy1,ty);
1086             fiz1             = _mm_add_ps(fiz1,tz);
1087
1088             fjx1             = _mm_add_ps(fjx1,tx);
1089             fjy1             = _mm_add_ps(fjy1,ty);
1090             fjz1             = _mm_add_ps(fjz1,tz);
1091
1092             }
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             if (gmx_mm_any_lt(rsq12,rcutoff2))
1099             {
1100
1101             r12              = _mm_mul_ps(rsq12,rinv12);
1102             r12              = _mm_andnot_ps(dummy_mask,r12);
1103
1104             /* EWALD ELECTROSTATICS */
1105
1106             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1107             ewrt             = _mm_mul_ps(r12,ewtabscale);
1108             ewitab           = _mm_cvttps_epi32(ewrt);
1109             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1110             ewitab           = _mm_slli_epi32(ewitab,2);
1111             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1112             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1113             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1114             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1115             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1116             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1117             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1118             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec));
1119             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1120
1121             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1122
1123             /* Update potential sum for this i atom from the interaction with this j atom. */
1124             velec            = _mm_and_ps(velec,cutoff_mask);
1125             velec            = _mm_andnot_ps(dummy_mask,velec);
1126             velecsum         = _mm_add_ps(velecsum,velec);
1127
1128             fscal            = felec;
1129
1130             fscal            = _mm_and_ps(fscal,cutoff_mask);
1131
1132             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm_mul_ps(fscal,dx12);
1136             ty               = _mm_mul_ps(fscal,dy12);
1137             tz               = _mm_mul_ps(fscal,dz12);
1138
1139             /* Update vectorial force */
1140             fix1             = _mm_add_ps(fix1,tx);
1141             fiy1             = _mm_add_ps(fiy1,ty);
1142             fiz1             = _mm_add_ps(fiz1,tz);
1143
1144             fjx2             = _mm_add_ps(fjx2,tx);
1145             fjy2             = _mm_add_ps(fjy2,ty);
1146             fjz2             = _mm_add_ps(fjz2,tz);
1147
1148             }
1149
1150             /**************************
1151              * CALCULATE INTERACTIONS *
1152              **************************/
1153
1154             if (gmx_mm_any_lt(rsq13,rcutoff2))
1155             {
1156
1157             r13              = _mm_mul_ps(rsq13,rinv13);
1158             r13              = _mm_andnot_ps(dummy_mask,r13);
1159
1160             /* EWALD ELECTROSTATICS */
1161
1162             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1163             ewrt             = _mm_mul_ps(r13,ewtabscale);
1164             ewitab           = _mm_cvttps_epi32(ewrt);
1165             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1166             ewitab           = _mm_slli_epi32(ewitab,2);
1167             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1168             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1169             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1170             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1171             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1172             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1173             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1174             velec            = _mm_mul_ps(qq13,_mm_sub_ps(_mm_sub_ps(rinv13,sh_ewald),velec));
1175             felec            = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
1176
1177             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1178
1179             /* Update potential sum for this i atom from the interaction with this j atom. */
1180             velec            = _mm_and_ps(velec,cutoff_mask);
1181             velec            = _mm_andnot_ps(dummy_mask,velec);
1182             velecsum         = _mm_add_ps(velecsum,velec);
1183
1184             fscal            = felec;
1185
1186             fscal            = _mm_and_ps(fscal,cutoff_mask);
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm_mul_ps(fscal,dx13);
1192             ty               = _mm_mul_ps(fscal,dy13);
1193             tz               = _mm_mul_ps(fscal,dz13);
1194
1195             /* Update vectorial force */
1196             fix1             = _mm_add_ps(fix1,tx);
1197             fiy1             = _mm_add_ps(fiy1,ty);
1198             fiz1             = _mm_add_ps(fiz1,tz);
1199
1200             fjx3             = _mm_add_ps(fjx3,tx);
1201             fjy3             = _mm_add_ps(fjy3,ty);
1202             fjz3             = _mm_add_ps(fjz3,tz);
1203
1204             }
1205
1206             /**************************
1207              * CALCULATE INTERACTIONS *
1208              **************************/
1209
1210             if (gmx_mm_any_lt(rsq21,rcutoff2))
1211             {
1212
1213             r21              = _mm_mul_ps(rsq21,rinv21);
1214             r21              = _mm_andnot_ps(dummy_mask,r21);
1215
1216             /* EWALD ELECTROSTATICS */
1217
1218             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1219             ewrt             = _mm_mul_ps(r21,ewtabscale);
1220             ewitab           = _mm_cvttps_epi32(ewrt);
1221             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1222             ewitab           = _mm_slli_epi32(ewitab,2);
1223             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1224             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1225             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1226             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1227             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1228             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1229             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1230             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec));
1231             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1232
1233             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1234
1235             /* Update potential sum for this i atom from the interaction with this j atom. */
1236             velec            = _mm_and_ps(velec,cutoff_mask);
1237             velec            = _mm_andnot_ps(dummy_mask,velec);
1238             velecsum         = _mm_add_ps(velecsum,velec);
1239
1240             fscal            = felec;
1241
1242             fscal            = _mm_and_ps(fscal,cutoff_mask);
1243
1244             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1245
1246             /* Calculate temporary vectorial force */
1247             tx               = _mm_mul_ps(fscal,dx21);
1248             ty               = _mm_mul_ps(fscal,dy21);
1249             tz               = _mm_mul_ps(fscal,dz21);
1250
1251             /* Update vectorial force */
1252             fix2             = _mm_add_ps(fix2,tx);
1253             fiy2             = _mm_add_ps(fiy2,ty);
1254             fiz2             = _mm_add_ps(fiz2,tz);
1255
1256             fjx1             = _mm_add_ps(fjx1,tx);
1257             fjy1             = _mm_add_ps(fjy1,ty);
1258             fjz1             = _mm_add_ps(fjz1,tz);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm_any_lt(rsq22,rcutoff2))
1267             {
1268
1269             r22              = _mm_mul_ps(rsq22,rinv22);
1270             r22              = _mm_andnot_ps(dummy_mask,r22);
1271
1272             /* EWALD ELECTROSTATICS */
1273
1274             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1275             ewrt             = _mm_mul_ps(r22,ewtabscale);
1276             ewitab           = _mm_cvttps_epi32(ewrt);
1277             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1278             ewitab           = _mm_slli_epi32(ewitab,2);
1279             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1280             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1281             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1282             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1283             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1284             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1285             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1286             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec));
1287             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1288
1289             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1290
1291             /* Update potential sum for this i atom from the interaction with this j atom. */
1292             velec            = _mm_and_ps(velec,cutoff_mask);
1293             velec            = _mm_andnot_ps(dummy_mask,velec);
1294             velecsum         = _mm_add_ps(velecsum,velec);
1295
1296             fscal            = felec;
1297
1298             fscal            = _mm_and_ps(fscal,cutoff_mask);
1299
1300             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1301
1302             /* Calculate temporary vectorial force */
1303             tx               = _mm_mul_ps(fscal,dx22);
1304             ty               = _mm_mul_ps(fscal,dy22);
1305             tz               = _mm_mul_ps(fscal,dz22);
1306
1307             /* Update vectorial force */
1308             fix2             = _mm_add_ps(fix2,tx);
1309             fiy2             = _mm_add_ps(fiy2,ty);
1310             fiz2             = _mm_add_ps(fiz2,tz);
1311
1312             fjx2             = _mm_add_ps(fjx2,tx);
1313             fjy2             = _mm_add_ps(fjy2,ty);
1314             fjz2             = _mm_add_ps(fjz2,tz);
1315
1316             }
1317
1318             /**************************
1319              * CALCULATE INTERACTIONS *
1320              **************************/
1321
1322             if (gmx_mm_any_lt(rsq23,rcutoff2))
1323             {
1324
1325             r23              = _mm_mul_ps(rsq23,rinv23);
1326             r23              = _mm_andnot_ps(dummy_mask,r23);
1327
1328             /* EWALD ELECTROSTATICS */
1329
1330             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1331             ewrt             = _mm_mul_ps(r23,ewtabscale);
1332             ewitab           = _mm_cvttps_epi32(ewrt);
1333             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1334             ewitab           = _mm_slli_epi32(ewitab,2);
1335             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1336             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1337             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1338             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1339             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1340             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1341             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1342             velec            = _mm_mul_ps(qq23,_mm_sub_ps(_mm_sub_ps(rinv23,sh_ewald),velec));
1343             felec            = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
1344
1345             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1346
1347             /* Update potential sum for this i atom from the interaction with this j atom. */
1348             velec            = _mm_and_ps(velec,cutoff_mask);
1349             velec            = _mm_andnot_ps(dummy_mask,velec);
1350             velecsum         = _mm_add_ps(velecsum,velec);
1351
1352             fscal            = felec;
1353
1354             fscal            = _mm_and_ps(fscal,cutoff_mask);
1355
1356             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1357
1358             /* Calculate temporary vectorial force */
1359             tx               = _mm_mul_ps(fscal,dx23);
1360             ty               = _mm_mul_ps(fscal,dy23);
1361             tz               = _mm_mul_ps(fscal,dz23);
1362
1363             /* Update vectorial force */
1364             fix2             = _mm_add_ps(fix2,tx);
1365             fiy2             = _mm_add_ps(fiy2,ty);
1366             fiz2             = _mm_add_ps(fiz2,tz);
1367
1368             fjx3             = _mm_add_ps(fjx3,tx);
1369             fjy3             = _mm_add_ps(fjy3,ty);
1370             fjz3             = _mm_add_ps(fjz3,tz);
1371
1372             }
1373
1374             /**************************
1375              * CALCULATE INTERACTIONS *
1376              **************************/
1377
1378             if (gmx_mm_any_lt(rsq31,rcutoff2))
1379             {
1380
1381             r31              = _mm_mul_ps(rsq31,rinv31);
1382             r31              = _mm_andnot_ps(dummy_mask,r31);
1383
1384             /* EWALD ELECTROSTATICS */
1385
1386             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1387             ewrt             = _mm_mul_ps(r31,ewtabscale);
1388             ewitab           = _mm_cvttps_epi32(ewrt);
1389             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1390             ewitab           = _mm_slli_epi32(ewitab,2);
1391             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1392             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1393             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1394             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1395             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1396             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1397             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1398             velec            = _mm_mul_ps(qq31,_mm_sub_ps(_mm_sub_ps(rinv31,sh_ewald),velec));
1399             felec            = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
1400
1401             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1402
1403             /* Update potential sum for this i atom from the interaction with this j atom. */
1404             velec            = _mm_and_ps(velec,cutoff_mask);
1405             velec            = _mm_andnot_ps(dummy_mask,velec);
1406             velecsum         = _mm_add_ps(velecsum,velec);
1407
1408             fscal            = felec;
1409
1410             fscal            = _mm_and_ps(fscal,cutoff_mask);
1411
1412             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1413
1414             /* Calculate temporary vectorial force */
1415             tx               = _mm_mul_ps(fscal,dx31);
1416             ty               = _mm_mul_ps(fscal,dy31);
1417             tz               = _mm_mul_ps(fscal,dz31);
1418
1419             /* Update vectorial force */
1420             fix3             = _mm_add_ps(fix3,tx);
1421             fiy3             = _mm_add_ps(fiy3,ty);
1422             fiz3             = _mm_add_ps(fiz3,tz);
1423
1424             fjx1             = _mm_add_ps(fjx1,tx);
1425             fjy1             = _mm_add_ps(fjy1,ty);
1426             fjz1             = _mm_add_ps(fjz1,tz);
1427
1428             }
1429
1430             /**************************
1431              * CALCULATE INTERACTIONS *
1432              **************************/
1433
1434             if (gmx_mm_any_lt(rsq32,rcutoff2))
1435             {
1436
1437             r32              = _mm_mul_ps(rsq32,rinv32);
1438             r32              = _mm_andnot_ps(dummy_mask,r32);
1439
1440             /* EWALD ELECTROSTATICS */
1441
1442             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1443             ewrt             = _mm_mul_ps(r32,ewtabscale);
1444             ewitab           = _mm_cvttps_epi32(ewrt);
1445             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1446             ewitab           = _mm_slli_epi32(ewitab,2);
1447             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1448             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1449             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1450             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1451             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1452             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1453             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1454             velec            = _mm_mul_ps(qq32,_mm_sub_ps(_mm_sub_ps(rinv32,sh_ewald),velec));
1455             felec            = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
1456
1457             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1458
1459             /* Update potential sum for this i atom from the interaction with this j atom. */
1460             velec            = _mm_and_ps(velec,cutoff_mask);
1461             velec            = _mm_andnot_ps(dummy_mask,velec);
1462             velecsum         = _mm_add_ps(velecsum,velec);
1463
1464             fscal            = felec;
1465
1466             fscal            = _mm_and_ps(fscal,cutoff_mask);
1467
1468             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1469
1470             /* Calculate temporary vectorial force */
1471             tx               = _mm_mul_ps(fscal,dx32);
1472             ty               = _mm_mul_ps(fscal,dy32);
1473             tz               = _mm_mul_ps(fscal,dz32);
1474
1475             /* Update vectorial force */
1476             fix3             = _mm_add_ps(fix3,tx);
1477             fiy3             = _mm_add_ps(fiy3,ty);
1478             fiz3             = _mm_add_ps(fiz3,tz);
1479
1480             fjx2             = _mm_add_ps(fjx2,tx);
1481             fjy2             = _mm_add_ps(fjy2,ty);
1482             fjz2             = _mm_add_ps(fjz2,tz);
1483
1484             }
1485
1486             /**************************
1487              * CALCULATE INTERACTIONS *
1488              **************************/
1489
1490             if (gmx_mm_any_lt(rsq33,rcutoff2))
1491             {
1492
1493             r33              = _mm_mul_ps(rsq33,rinv33);
1494             r33              = _mm_andnot_ps(dummy_mask,r33);
1495
1496             /* EWALD ELECTROSTATICS */
1497
1498             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1499             ewrt             = _mm_mul_ps(r33,ewtabscale);
1500             ewitab           = _mm_cvttps_epi32(ewrt);
1501             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1502             ewitab           = _mm_slli_epi32(ewitab,2);
1503             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1504             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1505             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1506             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1507             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1508             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1509             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1510             velec            = _mm_mul_ps(qq33,_mm_sub_ps(_mm_sub_ps(rinv33,sh_ewald),velec));
1511             felec            = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
1512
1513             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
1514
1515             /* Update potential sum for this i atom from the interaction with this j atom. */
1516             velec            = _mm_and_ps(velec,cutoff_mask);
1517             velec            = _mm_andnot_ps(dummy_mask,velec);
1518             velecsum         = _mm_add_ps(velecsum,velec);
1519
1520             fscal            = felec;
1521
1522             fscal            = _mm_and_ps(fscal,cutoff_mask);
1523
1524             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1525
1526             /* Calculate temporary vectorial force */
1527             tx               = _mm_mul_ps(fscal,dx33);
1528             ty               = _mm_mul_ps(fscal,dy33);
1529             tz               = _mm_mul_ps(fscal,dz33);
1530
1531             /* Update vectorial force */
1532             fix3             = _mm_add_ps(fix3,tx);
1533             fiy3             = _mm_add_ps(fiy3,ty);
1534             fiz3             = _mm_add_ps(fiz3,tz);
1535
1536             fjx3             = _mm_add_ps(fjx3,tx);
1537             fjy3             = _mm_add_ps(fjy3,ty);
1538             fjz3             = _mm_add_ps(fjz3,tz);
1539
1540             }
1541
1542             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1543             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1544             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1545             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1546
1547             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1548                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1549                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1550
1551             /* Inner loop uses 489 flops */
1552         }
1553
1554         /* End of innermost loop */
1555
1556         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1557                                               f+i_coord_offset,fshift+i_shift_offset);
1558
1559         ggid                        = gid[iidx];
1560         /* Update potential energies */
1561         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1562         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1563
1564         /* Increment number of inner iterations */
1565         inneriter                  += j_index_end - j_index_start;
1566
1567         /* Outer loop uses 26 flops */
1568     }
1569
1570     /* Increment number of outer iterations */
1571     outeriter        += nri;
1572
1573     /* Update outer/inner flops */
1574
1575     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*489);
1576 }
1577 /*
1578  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_sse4_1_single
1579  * Electrostatics interaction: Ewald
1580  * VdW interaction:            LJEwald
1581  * Geometry:                   Water4-Water4
1582  * Calculate force/pot:        Force
1583  */
1584 void
1585 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_sse4_1_single
1586                     (t_nblist                    * gmx_restrict       nlist,
1587                      rvec                        * gmx_restrict          xx,
1588                      rvec                        * gmx_restrict          ff,
1589                      t_forcerec                  * gmx_restrict          fr,
1590                      t_mdatoms                   * gmx_restrict     mdatoms,
1591                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1592                      t_nrnb                      * gmx_restrict        nrnb)
1593 {
1594     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1595      * just 0 for non-waters.
1596      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1597      * jnr indices corresponding to data put in the four positions in the SIMD register.
1598      */
1599     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1600     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1601     int              jnrA,jnrB,jnrC,jnrD;
1602     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1603     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1604     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1605     real             rcutoff_scalar;
1606     real             *shiftvec,*fshift,*x,*f;
1607     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1608     real             scratch[4*DIM];
1609     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1610     int              vdwioffset0;
1611     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1612     int              vdwioffset1;
1613     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1614     int              vdwioffset2;
1615     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1616     int              vdwioffset3;
1617     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1618     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1619     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1620     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1621     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1622     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1623     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1624     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1625     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1626     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1627     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1628     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1629     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1630     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1631     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1632     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1633     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1634     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1635     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1636     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1637     real             *charge;
1638     int              nvdwtype;
1639     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1640     int              *vdwtype;
1641     real             *vdwparam;
1642     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1643     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1644     __m128           c6grid_00;
1645     __m128           c6grid_11;
1646     __m128           c6grid_12;
1647     __m128           c6grid_13;
1648     __m128           c6grid_21;
1649     __m128           c6grid_22;
1650     __m128           c6grid_23;
1651     __m128           c6grid_31;
1652     __m128           c6grid_32;
1653     __m128           c6grid_33;
1654     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1655     real             *vdwgridparam;
1656     __m128           one_half  = _mm_set1_ps(0.5);
1657     __m128           minus_one = _mm_set1_ps(-1.0);
1658     __m128i          ewitab;
1659     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1660     real             *ewtab;
1661     __m128           dummy_mask,cutoff_mask;
1662     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1663     __m128           one     = _mm_set1_ps(1.0);
1664     __m128           two     = _mm_set1_ps(2.0);
1665     x                = xx[0];
1666     f                = ff[0];
1667
1668     nri              = nlist->nri;
1669     iinr             = nlist->iinr;
1670     jindex           = nlist->jindex;
1671     jjnr             = nlist->jjnr;
1672     shiftidx         = nlist->shift;
1673     gid              = nlist->gid;
1674     shiftvec         = fr->shift_vec[0];
1675     fshift           = fr->fshift[0];
1676     facel            = _mm_set1_ps(fr->epsfac);
1677     charge           = mdatoms->chargeA;
1678     nvdwtype         = fr->ntype;
1679     vdwparam         = fr->nbfp;
1680     vdwtype          = mdatoms->typeA;
1681     vdwgridparam     = fr->ljpme_c6grid;
1682     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
1683     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
1684     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
1685
1686     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1687     ewtab            = fr->ic->tabq_coul_F;
1688     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1689     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1690
1691     /* Setup water-specific parameters */
1692     inr              = nlist->iinr[0];
1693     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1694     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1695     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1696     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1697
1698     jq1              = _mm_set1_ps(charge[inr+1]);
1699     jq2              = _mm_set1_ps(charge[inr+2]);
1700     jq3              = _mm_set1_ps(charge[inr+3]);
1701     vdwjidx0A        = 2*vdwtype[inr+0];
1702     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1703     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1704     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
1705     qq11             = _mm_mul_ps(iq1,jq1);
1706     qq12             = _mm_mul_ps(iq1,jq2);
1707     qq13             = _mm_mul_ps(iq1,jq3);
1708     qq21             = _mm_mul_ps(iq2,jq1);
1709     qq22             = _mm_mul_ps(iq2,jq2);
1710     qq23             = _mm_mul_ps(iq2,jq3);
1711     qq31             = _mm_mul_ps(iq3,jq1);
1712     qq32             = _mm_mul_ps(iq3,jq2);
1713     qq33             = _mm_mul_ps(iq3,jq3);
1714
1715     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1716     rcutoff_scalar   = fr->rcoulomb;
1717     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1718     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1719
1720     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
1721     rvdw             = _mm_set1_ps(fr->rvdw);
1722
1723     /* Avoid stupid compiler warnings */
1724     jnrA = jnrB = jnrC = jnrD = 0;
1725     j_coord_offsetA = 0;
1726     j_coord_offsetB = 0;
1727     j_coord_offsetC = 0;
1728     j_coord_offsetD = 0;
1729
1730     outeriter        = 0;
1731     inneriter        = 0;
1732
1733     for(iidx=0;iidx<4*DIM;iidx++)
1734     {
1735         scratch[iidx] = 0.0;
1736     }
1737
1738     /* Start outer loop over neighborlists */
1739     for(iidx=0; iidx<nri; iidx++)
1740     {
1741         /* Load shift vector for this list */
1742         i_shift_offset   = DIM*shiftidx[iidx];
1743
1744         /* Load limits for loop over neighbors */
1745         j_index_start    = jindex[iidx];
1746         j_index_end      = jindex[iidx+1];
1747
1748         /* Get outer coordinate index */
1749         inr              = iinr[iidx];
1750         i_coord_offset   = DIM*inr;
1751
1752         /* Load i particle coords and add shift vector */
1753         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1754                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1755
1756         fix0             = _mm_setzero_ps();
1757         fiy0             = _mm_setzero_ps();
1758         fiz0             = _mm_setzero_ps();
1759         fix1             = _mm_setzero_ps();
1760         fiy1             = _mm_setzero_ps();
1761         fiz1             = _mm_setzero_ps();
1762         fix2             = _mm_setzero_ps();
1763         fiy2             = _mm_setzero_ps();
1764         fiz2             = _mm_setzero_ps();
1765         fix3             = _mm_setzero_ps();
1766         fiy3             = _mm_setzero_ps();
1767         fiz3             = _mm_setzero_ps();
1768
1769         /* Start inner kernel loop */
1770         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1771         {
1772
1773             /* Get j neighbor index, and coordinate index */
1774             jnrA             = jjnr[jidx];
1775             jnrB             = jjnr[jidx+1];
1776             jnrC             = jjnr[jidx+2];
1777             jnrD             = jjnr[jidx+3];
1778             j_coord_offsetA  = DIM*jnrA;
1779             j_coord_offsetB  = DIM*jnrB;
1780             j_coord_offsetC  = DIM*jnrC;
1781             j_coord_offsetD  = DIM*jnrD;
1782
1783             /* load j atom coordinates */
1784             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1785                                               x+j_coord_offsetC,x+j_coord_offsetD,
1786                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1787                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1788
1789             /* Calculate displacement vector */
1790             dx00             = _mm_sub_ps(ix0,jx0);
1791             dy00             = _mm_sub_ps(iy0,jy0);
1792             dz00             = _mm_sub_ps(iz0,jz0);
1793             dx11             = _mm_sub_ps(ix1,jx1);
1794             dy11             = _mm_sub_ps(iy1,jy1);
1795             dz11             = _mm_sub_ps(iz1,jz1);
1796             dx12             = _mm_sub_ps(ix1,jx2);
1797             dy12             = _mm_sub_ps(iy1,jy2);
1798             dz12             = _mm_sub_ps(iz1,jz2);
1799             dx13             = _mm_sub_ps(ix1,jx3);
1800             dy13             = _mm_sub_ps(iy1,jy3);
1801             dz13             = _mm_sub_ps(iz1,jz3);
1802             dx21             = _mm_sub_ps(ix2,jx1);
1803             dy21             = _mm_sub_ps(iy2,jy1);
1804             dz21             = _mm_sub_ps(iz2,jz1);
1805             dx22             = _mm_sub_ps(ix2,jx2);
1806             dy22             = _mm_sub_ps(iy2,jy2);
1807             dz22             = _mm_sub_ps(iz2,jz2);
1808             dx23             = _mm_sub_ps(ix2,jx3);
1809             dy23             = _mm_sub_ps(iy2,jy3);
1810             dz23             = _mm_sub_ps(iz2,jz3);
1811             dx31             = _mm_sub_ps(ix3,jx1);
1812             dy31             = _mm_sub_ps(iy3,jy1);
1813             dz31             = _mm_sub_ps(iz3,jz1);
1814             dx32             = _mm_sub_ps(ix3,jx2);
1815             dy32             = _mm_sub_ps(iy3,jy2);
1816             dz32             = _mm_sub_ps(iz3,jz2);
1817             dx33             = _mm_sub_ps(ix3,jx3);
1818             dy33             = _mm_sub_ps(iy3,jy3);
1819             dz33             = _mm_sub_ps(iz3,jz3);
1820
1821             /* Calculate squared distance and things based on it */
1822             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1823             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1824             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1825             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1826             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1827             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1828             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1829             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1830             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1831             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1832
1833             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1834             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1835             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1836             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1837             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1838             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1839             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1840             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1841             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1842             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1843
1844             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1845             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1846             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1847             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1848             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1849             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1850             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1851             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1852             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1853             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1854
1855             fjx0             = _mm_setzero_ps();
1856             fjy0             = _mm_setzero_ps();
1857             fjz0             = _mm_setzero_ps();
1858             fjx1             = _mm_setzero_ps();
1859             fjy1             = _mm_setzero_ps();
1860             fjz1             = _mm_setzero_ps();
1861             fjx2             = _mm_setzero_ps();
1862             fjy2             = _mm_setzero_ps();
1863             fjz2             = _mm_setzero_ps();
1864             fjx3             = _mm_setzero_ps();
1865             fjy3             = _mm_setzero_ps();
1866             fjz3             = _mm_setzero_ps();
1867
1868             /**************************
1869              * CALCULATE INTERACTIONS *
1870              **************************/
1871
1872             if (gmx_mm_any_lt(rsq00,rcutoff2))
1873             {
1874
1875             r00              = _mm_mul_ps(rsq00,rinv00);
1876
1877             /* Analytical LJ-PME */
1878             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1879             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1880             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1881             exponent         = gmx_simd_exp_r(ewcljrsq);
1882             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1883             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1884             /* f6A = 6 * C6grid * (1 - poly) */
1885             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1886             /* f6B = C6grid * exponent * beta^6 */
1887             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1888             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1889             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);
1890
1891             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1892
1893             fscal            = fvdw;
1894
1895             fscal            = _mm_and_ps(fscal,cutoff_mask);
1896
1897             /* Calculate temporary vectorial force */
1898             tx               = _mm_mul_ps(fscal,dx00);
1899             ty               = _mm_mul_ps(fscal,dy00);
1900             tz               = _mm_mul_ps(fscal,dz00);
1901
1902             /* Update vectorial force */
1903             fix0             = _mm_add_ps(fix0,tx);
1904             fiy0             = _mm_add_ps(fiy0,ty);
1905             fiz0             = _mm_add_ps(fiz0,tz);
1906
1907             fjx0             = _mm_add_ps(fjx0,tx);
1908             fjy0             = _mm_add_ps(fjy0,ty);
1909             fjz0             = _mm_add_ps(fjz0,tz);
1910
1911             }
1912
1913             /**************************
1914              * CALCULATE INTERACTIONS *
1915              **************************/
1916
1917             if (gmx_mm_any_lt(rsq11,rcutoff2))
1918             {
1919
1920             r11              = _mm_mul_ps(rsq11,rinv11);
1921
1922             /* EWALD ELECTROSTATICS */
1923
1924             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1925             ewrt             = _mm_mul_ps(r11,ewtabscale);
1926             ewitab           = _mm_cvttps_epi32(ewrt);
1927             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1928             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1929                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1930                                          &ewtabF,&ewtabFn);
1931             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1932             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1933
1934             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1935
1936             fscal            = felec;
1937
1938             fscal            = _mm_and_ps(fscal,cutoff_mask);
1939
1940             /* Calculate temporary vectorial force */
1941             tx               = _mm_mul_ps(fscal,dx11);
1942             ty               = _mm_mul_ps(fscal,dy11);
1943             tz               = _mm_mul_ps(fscal,dz11);
1944
1945             /* Update vectorial force */
1946             fix1             = _mm_add_ps(fix1,tx);
1947             fiy1             = _mm_add_ps(fiy1,ty);
1948             fiz1             = _mm_add_ps(fiz1,tz);
1949
1950             fjx1             = _mm_add_ps(fjx1,tx);
1951             fjy1             = _mm_add_ps(fjy1,ty);
1952             fjz1             = _mm_add_ps(fjz1,tz);
1953
1954             }
1955
1956             /**************************
1957              * CALCULATE INTERACTIONS *
1958              **************************/
1959
1960             if (gmx_mm_any_lt(rsq12,rcutoff2))
1961             {
1962
1963             r12              = _mm_mul_ps(rsq12,rinv12);
1964
1965             /* EWALD ELECTROSTATICS */
1966
1967             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1968             ewrt             = _mm_mul_ps(r12,ewtabscale);
1969             ewitab           = _mm_cvttps_epi32(ewrt);
1970             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1971             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1972                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1973                                          &ewtabF,&ewtabFn);
1974             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1975             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1976
1977             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1978
1979             fscal            = felec;
1980
1981             fscal            = _mm_and_ps(fscal,cutoff_mask);
1982
1983             /* Calculate temporary vectorial force */
1984             tx               = _mm_mul_ps(fscal,dx12);
1985             ty               = _mm_mul_ps(fscal,dy12);
1986             tz               = _mm_mul_ps(fscal,dz12);
1987
1988             /* Update vectorial force */
1989             fix1             = _mm_add_ps(fix1,tx);
1990             fiy1             = _mm_add_ps(fiy1,ty);
1991             fiz1             = _mm_add_ps(fiz1,tz);
1992
1993             fjx2             = _mm_add_ps(fjx2,tx);
1994             fjy2             = _mm_add_ps(fjy2,ty);
1995             fjz2             = _mm_add_ps(fjz2,tz);
1996
1997             }
1998
1999             /**************************
2000              * CALCULATE INTERACTIONS *
2001              **************************/
2002
2003             if (gmx_mm_any_lt(rsq13,rcutoff2))
2004             {
2005
2006             r13              = _mm_mul_ps(rsq13,rinv13);
2007
2008             /* EWALD ELECTROSTATICS */
2009
2010             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2011             ewrt             = _mm_mul_ps(r13,ewtabscale);
2012             ewitab           = _mm_cvttps_epi32(ewrt);
2013             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2014             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2015                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2016                                          &ewtabF,&ewtabFn);
2017             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2018             felec            = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
2019
2020             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
2021
2022             fscal            = felec;
2023
2024             fscal            = _mm_and_ps(fscal,cutoff_mask);
2025
2026             /* Calculate temporary vectorial force */
2027             tx               = _mm_mul_ps(fscal,dx13);
2028             ty               = _mm_mul_ps(fscal,dy13);
2029             tz               = _mm_mul_ps(fscal,dz13);
2030
2031             /* Update vectorial force */
2032             fix1             = _mm_add_ps(fix1,tx);
2033             fiy1             = _mm_add_ps(fiy1,ty);
2034             fiz1             = _mm_add_ps(fiz1,tz);
2035
2036             fjx3             = _mm_add_ps(fjx3,tx);
2037             fjy3             = _mm_add_ps(fjy3,ty);
2038             fjz3             = _mm_add_ps(fjz3,tz);
2039
2040             }
2041
2042             /**************************
2043              * CALCULATE INTERACTIONS *
2044              **************************/
2045
2046             if (gmx_mm_any_lt(rsq21,rcutoff2))
2047             {
2048
2049             r21              = _mm_mul_ps(rsq21,rinv21);
2050
2051             /* EWALD ELECTROSTATICS */
2052
2053             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2054             ewrt             = _mm_mul_ps(r21,ewtabscale);
2055             ewitab           = _mm_cvttps_epi32(ewrt);
2056             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2057             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2058                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2059                                          &ewtabF,&ewtabFn);
2060             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2061             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2062
2063             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2064
2065             fscal            = felec;
2066
2067             fscal            = _mm_and_ps(fscal,cutoff_mask);
2068
2069             /* Calculate temporary vectorial force */
2070             tx               = _mm_mul_ps(fscal,dx21);
2071             ty               = _mm_mul_ps(fscal,dy21);
2072             tz               = _mm_mul_ps(fscal,dz21);
2073
2074             /* Update vectorial force */
2075             fix2             = _mm_add_ps(fix2,tx);
2076             fiy2             = _mm_add_ps(fiy2,ty);
2077             fiz2             = _mm_add_ps(fiz2,tz);
2078
2079             fjx1             = _mm_add_ps(fjx1,tx);
2080             fjy1             = _mm_add_ps(fjy1,ty);
2081             fjz1             = _mm_add_ps(fjz1,tz);
2082
2083             }
2084
2085             /**************************
2086              * CALCULATE INTERACTIONS *
2087              **************************/
2088
2089             if (gmx_mm_any_lt(rsq22,rcutoff2))
2090             {
2091
2092             r22              = _mm_mul_ps(rsq22,rinv22);
2093
2094             /* EWALD ELECTROSTATICS */
2095
2096             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2097             ewrt             = _mm_mul_ps(r22,ewtabscale);
2098             ewitab           = _mm_cvttps_epi32(ewrt);
2099             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2100             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2101                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2102                                          &ewtabF,&ewtabFn);
2103             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2104             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2105
2106             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2107
2108             fscal            = felec;
2109
2110             fscal            = _mm_and_ps(fscal,cutoff_mask);
2111
2112             /* Calculate temporary vectorial force */
2113             tx               = _mm_mul_ps(fscal,dx22);
2114             ty               = _mm_mul_ps(fscal,dy22);
2115             tz               = _mm_mul_ps(fscal,dz22);
2116
2117             /* Update vectorial force */
2118             fix2             = _mm_add_ps(fix2,tx);
2119             fiy2             = _mm_add_ps(fiy2,ty);
2120             fiz2             = _mm_add_ps(fiz2,tz);
2121
2122             fjx2             = _mm_add_ps(fjx2,tx);
2123             fjy2             = _mm_add_ps(fjy2,ty);
2124             fjz2             = _mm_add_ps(fjz2,tz);
2125
2126             }
2127
2128             /**************************
2129              * CALCULATE INTERACTIONS *
2130              **************************/
2131
2132             if (gmx_mm_any_lt(rsq23,rcutoff2))
2133             {
2134
2135             r23              = _mm_mul_ps(rsq23,rinv23);
2136
2137             /* EWALD ELECTROSTATICS */
2138
2139             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2140             ewrt             = _mm_mul_ps(r23,ewtabscale);
2141             ewitab           = _mm_cvttps_epi32(ewrt);
2142             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2143             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2144                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2145                                          &ewtabF,&ewtabFn);
2146             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2147             felec            = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
2148
2149             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
2150
2151             fscal            = felec;
2152
2153             fscal            = _mm_and_ps(fscal,cutoff_mask);
2154
2155             /* Calculate temporary vectorial force */
2156             tx               = _mm_mul_ps(fscal,dx23);
2157             ty               = _mm_mul_ps(fscal,dy23);
2158             tz               = _mm_mul_ps(fscal,dz23);
2159
2160             /* Update vectorial force */
2161             fix2             = _mm_add_ps(fix2,tx);
2162             fiy2             = _mm_add_ps(fiy2,ty);
2163             fiz2             = _mm_add_ps(fiz2,tz);
2164
2165             fjx3             = _mm_add_ps(fjx3,tx);
2166             fjy3             = _mm_add_ps(fjy3,ty);
2167             fjz3             = _mm_add_ps(fjz3,tz);
2168
2169             }
2170
2171             /**************************
2172              * CALCULATE INTERACTIONS *
2173              **************************/
2174
2175             if (gmx_mm_any_lt(rsq31,rcutoff2))
2176             {
2177
2178             r31              = _mm_mul_ps(rsq31,rinv31);
2179
2180             /* EWALD ELECTROSTATICS */
2181
2182             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2183             ewrt             = _mm_mul_ps(r31,ewtabscale);
2184             ewitab           = _mm_cvttps_epi32(ewrt);
2185             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2186             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2187                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2188                                          &ewtabF,&ewtabFn);
2189             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2190             felec            = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
2191
2192             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
2193
2194             fscal            = felec;
2195
2196             fscal            = _mm_and_ps(fscal,cutoff_mask);
2197
2198             /* Calculate temporary vectorial force */
2199             tx               = _mm_mul_ps(fscal,dx31);
2200             ty               = _mm_mul_ps(fscal,dy31);
2201             tz               = _mm_mul_ps(fscal,dz31);
2202
2203             /* Update vectorial force */
2204             fix3             = _mm_add_ps(fix3,tx);
2205             fiy3             = _mm_add_ps(fiy3,ty);
2206             fiz3             = _mm_add_ps(fiz3,tz);
2207
2208             fjx1             = _mm_add_ps(fjx1,tx);
2209             fjy1             = _mm_add_ps(fjy1,ty);
2210             fjz1             = _mm_add_ps(fjz1,tz);
2211
2212             }
2213
2214             /**************************
2215              * CALCULATE INTERACTIONS *
2216              **************************/
2217
2218             if (gmx_mm_any_lt(rsq32,rcutoff2))
2219             {
2220
2221             r32              = _mm_mul_ps(rsq32,rinv32);
2222
2223             /* EWALD ELECTROSTATICS */
2224
2225             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2226             ewrt             = _mm_mul_ps(r32,ewtabscale);
2227             ewitab           = _mm_cvttps_epi32(ewrt);
2228             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2229             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2230                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2231                                          &ewtabF,&ewtabFn);
2232             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2233             felec            = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
2234
2235             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
2236
2237             fscal            = felec;
2238
2239             fscal            = _mm_and_ps(fscal,cutoff_mask);
2240
2241             /* Calculate temporary vectorial force */
2242             tx               = _mm_mul_ps(fscal,dx32);
2243             ty               = _mm_mul_ps(fscal,dy32);
2244             tz               = _mm_mul_ps(fscal,dz32);
2245
2246             /* Update vectorial force */
2247             fix3             = _mm_add_ps(fix3,tx);
2248             fiy3             = _mm_add_ps(fiy3,ty);
2249             fiz3             = _mm_add_ps(fiz3,tz);
2250
2251             fjx2             = _mm_add_ps(fjx2,tx);
2252             fjy2             = _mm_add_ps(fjy2,ty);
2253             fjz2             = _mm_add_ps(fjz2,tz);
2254
2255             }
2256
2257             /**************************
2258              * CALCULATE INTERACTIONS *
2259              **************************/
2260
2261             if (gmx_mm_any_lt(rsq33,rcutoff2))
2262             {
2263
2264             r33              = _mm_mul_ps(rsq33,rinv33);
2265
2266             /* EWALD ELECTROSTATICS */
2267
2268             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2269             ewrt             = _mm_mul_ps(r33,ewtabscale);
2270             ewitab           = _mm_cvttps_epi32(ewrt);
2271             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2272             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2273                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2274                                          &ewtabF,&ewtabFn);
2275             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2276             felec            = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
2277
2278             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2279
2280             fscal            = felec;
2281
2282             fscal            = _mm_and_ps(fscal,cutoff_mask);
2283
2284             /* Calculate temporary vectorial force */
2285             tx               = _mm_mul_ps(fscal,dx33);
2286             ty               = _mm_mul_ps(fscal,dy33);
2287             tz               = _mm_mul_ps(fscal,dz33);
2288
2289             /* Update vectorial force */
2290             fix3             = _mm_add_ps(fix3,tx);
2291             fiy3             = _mm_add_ps(fiy3,ty);
2292             fiz3             = _mm_add_ps(fiz3,tz);
2293
2294             fjx3             = _mm_add_ps(fjx3,tx);
2295             fjy3             = _mm_add_ps(fjy3,ty);
2296             fjz3             = _mm_add_ps(fjz3,tz);
2297
2298             }
2299
2300             fjptrA             = f+j_coord_offsetA;
2301             fjptrB             = f+j_coord_offsetB;
2302             fjptrC             = f+j_coord_offsetC;
2303             fjptrD             = f+j_coord_offsetD;
2304
2305             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2306                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2307                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2308
2309             /* Inner loop uses 403 flops */
2310         }
2311
2312         if(jidx<j_index_end)
2313         {
2314
2315             /* Get j neighbor index, and coordinate index */
2316             jnrlistA         = jjnr[jidx];
2317             jnrlistB         = jjnr[jidx+1];
2318             jnrlistC         = jjnr[jidx+2];
2319             jnrlistD         = jjnr[jidx+3];
2320             /* Sign of each element will be negative for non-real atoms.
2321              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2322              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2323              */
2324             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2325             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2326             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2327             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2328             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2329             j_coord_offsetA  = DIM*jnrA;
2330             j_coord_offsetB  = DIM*jnrB;
2331             j_coord_offsetC  = DIM*jnrC;
2332             j_coord_offsetD  = DIM*jnrD;
2333
2334             /* load j atom coordinates */
2335             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2336                                               x+j_coord_offsetC,x+j_coord_offsetD,
2337                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2338                                               &jy2,&jz2,&jx3,&jy3,&jz3);
2339
2340             /* Calculate displacement vector */
2341             dx00             = _mm_sub_ps(ix0,jx0);
2342             dy00             = _mm_sub_ps(iy0,jy0);
2343             dz00             = _mm_sub_ps(iz0,jz0);
2344             dx11             = _mm_sub_ps(ix1,jx1);
2345             dy11             = _mm_sub_ps(iy1,jy1);
2346             dz11             = _mm_sub_ps(iz1,jz1);
2347             dx12             = _mm_sub_ps(ix1,jx2);
2348             dy12             = _mm_sub_ps(iy1,jy2);
2349             dz12             = _mm_sub_ps(iz1,jz2);
2350             dx13             = _mm_sub_ps(ix1,jx3);
2351             dy13             = _mm_sub_ps(iy1,jy3);
2352             dz13             = _mm_sub_ps(iz1,jz3);
2353             dx21             = _mm_sub_ps(ix2,jx1);
2354             dy21             = _mm_sub_ps(iy2,jy1);
2355             dz21             = _mm_sub_ps(iz2,jz1);
2356             dx22             = _mm_sub_ps(ix2,jx2);
2357             dy22             = _mm_sub_ps(iy2,jy2);
2358             dz22             = _mm_sub_ps(iz2,jz2);
2359             dx23             = _mm_sub_ps(ix2,jx3);
2360             dy23             = _mm_sub_ps(iy2,jy3);
2361             dz23             = _mm_sub_ps(iz2,jz3);
2362             dx31             = _mm_sub_ps(ix3,jx1);
2363             dy31             = _mm_sub_ps(iy3,jy1);
2364             dz31             = _mm_sub_ps(iz3,jz1);
2365             dx32             = _mm_sub_ps(ix3,jx2);
2366             dy32             = _mm_sub_ps(iy3,jy2);
2367             dz32             = _mm_sub_ps(iz3,jz2);
2368             dx33             = _mm_sub_ps(ix3,jx3);
2369             dy33             = _mm_sub_ps(iy3,jy3);
2370             dz33             = _mm_sub_ps(iz3,jz3);
2371
2372             /* Calculate squared distance and things based on it */
2373             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
2374             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2375             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2376             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
2377             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2378             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2379             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
2380             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
2381             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
2382             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2383
2384             rinv00           = gmx_mm_invsqrt_ps(rsq00);
2385             rinv11           = gmx_mm_invsqrt_ps(rsq11);
2386             rinv12           = gmx_mm_invsqrt_ps(rsq12);
2387             rinv13           = gmx_mm_invsqrt_ps(rsq13);
2388             rinv21           = gmx_mm_invsqrt_ps(rsq21);
2389             rinv22           = gmx_mm_invsqrt_ps(rsq22);
2390             rinv23           = gmx_mm_invsqrt_ps(rsq23);
2391             rinv31           = gmx_mm_invsqrt_ps(rsq31);
2392             rinv32           = gmx_mm_invsqrt_ps(rsq32);
2393             rinv33           = gmx_mm_invsqrt_ps(rsq33);
2394
2395             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
2396             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
2397             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
2398             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
2399             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
2400             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
2401             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
2402             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
2403             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
2404             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
2405
2406             fjx0             = _mm_setzero_ps();
2407             fjy0             = _mm_setzero_ps();
2408             fjz0             = _mm_setzero_ps();
2409             fjx1             = _mm_setzero_ps();
2410             fjy1             = _mm_setzero_ps();
2411             fjz1             = _mm_setzero_ps();
2412             fjx2             = _mm_setzero_ps();
2413             fjy2             = _mm_setzero_ps();
2414             fjz2             = _mm_setzero_ps();
2415             fjx3             = _mm_setzero_ps();
2416             fjy3             = _mm_setzero_ps();
2417             fjz3             = _mm_setzero_ps();
2418
2419             /**************************
2420              * CALCULATE INTERACTIONS *
2421              **************************/
2422
2423             if (gmx_mm_any_lt(rsq00,rcutoff2))
2424             {
2425
2426             r00              = _mm_mul_ps(rsq00,rinv00);
2427             r00              = _mm_andnot_ps(dummy_mask,r00);
2428
2429             /* Analytical LJ-PME */
2430             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2431             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
2432             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
2433             exponent         = gmx_simd_exp_r(ewcljrsq);
2434             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2435             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
2436             /* f6A = 6 * C6grid * (1 - poly) */
2437             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
2438             /* f6B = C6grid * exponent * beta^6 */
2439             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
2440             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2441             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);
2442
2443             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
2444
2445             fscal            = fvdw;
2446
2447             fscal            = _mm_and_ps(fscal,cutoff_mask);
2448
2449             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2450
2451             /* Calculate temporary vectorial force */
2452             tx               = _mm_mul_ps(fscal,dx00);
2453             ty               = _mm_mul_ps(fscal,dy00);
2454             tz               = _mm_mul_ps(fscal,dz00);
2455
2456             /* Update vectorial force */
2457             fix0             = _mm_add_ps(fix0,tx);
2458             fiy0             = _mm_add_ps(fiy0,ty);
2459             fiz0             = _mm_add_ps(fiz0,tz);
2460
2461             fjx0             = _mm_add_ps(fjx0,tx);
2462             fjy0             = _mm_add_ps(fjy0,ty);
2463             fjz0             = _mm_add_ps(fjz0,tz);
2464
2465             }
2466
2467             /**************************
2468              * CALCULATE INTERACTIONS *
2469              **************************/
2470
2471             if (gmx_mm_any_lt(rsq11,rcutoff2))
2472             {
2473
2474             r11              = _mm_mul_ps(rsq11,rinv11);
2475             r11              = _mm_andnot_ps(dummy_mask,r11);
2476
2477             /* EWALD ELECTROSTATICS */
2478
2479             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2480             ewrt             = _mm_mul_ps(r11,ewtabscale);
2481             ewitab           = _mm_cvttps_epi32(ewrt);
2482             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2483             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2484                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2485                                          &ewtabF,&ewtabFn);
2486             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2487             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
2488
2489             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
2490
2491             fscal            = felec;
2492
2493             fscal            = _mm_and_ps(fscal,cutoff_mask);
2494
2495             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2496
2497             /* Calculate temporary vectorial force */
2498             tx               = _mm_mul_ps(fscal,dx11);
2499             ty               = _mm_mul_ps(fscal,dy11);
2500             tz               = _mm_mul_ps(fscal,dz11);
2501
2502             /* Update vectorial force */
2503             fix1             = _mm_add_ps(fix1,tx);
2504             fiy1             = _mm_add_ps(fiy1,ty);
2505             fiz1             = _mm_add_ps(fiz1,tz);
2506
2507             fjx1             = _mm_add_ps(fjx1,tx);
2508             fjy1             = _mm_add_ps(fjy1,ty);
2509             fjz1             = _mm_add_ps(fjz1,tz);
2510
2511             }
2512
2513             /**************************
2514              * CALCULATE INTERACTIONS *
2515              **************************/
2516
2517             if (gmx_mm_any_lt(rsq12,rcutoff2))
2518             {
2519
2520             r12              = _mm_mul_ps(rsq12,rinv12);
2521             r12              = _mm_andnot_ps(dummy_mask,r12);
2522
2523             /* EWALD ELECTROSTATICS */
2524
2525             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2526             ewrt             = _mm_mul_ps(r12,ewtabscale);
2527             ewitab           = _mm_cvttps_epi32(ewrt);
2528             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2529             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2530                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2531                                          &ewtabF,&ewtabFn);
2532             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2533             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
2534
2535             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
2536
2537             fscal            = felec;
2538
2539             fscal            = _mm_and_ps(fscal,cutoff_mask);
2540
2541             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2542
2543             /* Calculate temporary vectorial force */
2544             tx               = _mm_mul_ps(fscal,dx12);
2545             ty               = _mm_mul_ps(fscal,dy12);
2546             tz               = _mm_mul_ps(fscal,dz12);
2547
2548             /* Update vectorial force */
2549             fix1             = _mm_add_ps(fix1,tx);
2550             fiy1             = _mm_add_ps(fiy1,ty);
2551             fiz1             = _mm_add_ps(fiz1,tz);
2552
2553             fjx2             = _mm_add_ps(fjx2,tx);
2554             fjy2             = _mm_add_ps(fjy2,ty);
2555             fjz2             = _mm_add_ps(fjz2,tz);
2556
2557             }
2558
2559             /**************************
2560              * CALCULATE INTERACTIONS *
2561              **************************/
2562
2563             if (gmx_mm_any_lt(rsq13,rcutoff2))
2564             {
2565
2566             r13              = _mm_mul_ps(rsq13,rinv13);
2567             r13              = _mm_andnot_ps(dummy_mask,r13);
2568
2569             /* EWALD ELECTROSTATICS */
2570
2571             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2572             ewrt             = _mm_mul_ps(r13,ewtabscale);
2573             ewitab           = _mm_cvttps_epi32(ewrt);
2574             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2575             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2576                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2577                                          &ewtabF,&ewtabFn);
2578             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2579             felec            = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
2580
2581             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
2582
2583             fscal            = felec;
2584
2585             fscal            = _mm_and_ps(fscal,cutoff_mask);
2586
2587             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2588
2589             /* Calculate temporary vectorial force */
2590             tx               = _mm_mul_ps(fscal,dx13);
2591             ty               = _mm_mul_ps(fscal,dy13);
2592             tz               = _mm_mul_ps(fscal,dz13);
2593
2594             /* Update vectorial force */
2595             fix1             = _mm_add_ps(fix1,tx);
2596             fiy1             = _mm_add_ps(fiy1,ty);
2597             fiz1             = _mm_add_ps(fiz1,tz);
2598
2599             fjx3             = _mm_add_ps(fjx3,tx);
2600             fjy3             = _mm_add_ps(fjy3,ty);
2601             fjz3             = _mm_add_ps(fjz3,tz);
2602
2603             }
2604
2605             /**************************
2606              * CALCULATE INTERACTIONS *
2607              **************************/
2608
2609             if (gmx_mm_any_lt(rsq21,rcutoff2))
2610             {
2611
2612             r21              = _mm_mul_ps(rsq21,rinv21);
2613             r21              = _mm_andnot_ps(dummy_mask,r21);
2614
2615             /* EWALD ELECTROSTATICS */
2616
2617             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2618             ewrt             = _mm_mul_ps(r21,ewtabscale);
2619             ewitab           = _mm_cvttps_epi32(ewrt);
2620             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2621             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2622                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2623                                          &ewtabF,&ewtabFn);
2624             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2625             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2626
2627             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2628
2629             fscal            = felec;
2630
2631             fscal            = _mm_and_ps(fscal,cutoff_mask);
2632
2633             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2634
2635             /* Calculate temporary vectorial force */
2636             tx               = _mm_mul_ps(fscal,dx21);
2637             ty               = _mm_mul_ps(fscal,dy21);
2638             tz               = _mm_mul_ps(fscal,dz21);
2639
2640             /* Update vectorial force */
2641             fix2             = _mm_add_ps(fix2,tx);
2642             fiy2             = _mm_add_ps(fiy2,ty);
2643             fiz2             = _mm_add_ps(fiz2,tz);
2644
2645             fjx1             = _mm_add_ps(fjx1,tx);
2646             fjy1             = _mm_add_ps(fjy1,ty);
2647             fjz1             = _mm_add_ps(fjz1,tz);
2648
2649             }
2650
2651             /**************************
2652              * CALCULATE INTERACTIONS *
2653              **************************/
2654
2655             if (gmx_mm_any_lt(rsq22,rcutoff2))
2656             {
2657
2658             r22              = _mm_mul_ps(rsq22,rinv22);
2659             r22              = _mm_andnot_ps(dummy_mask,r22);
2660
2661             /* EWALD ELECTROSTATICS */
2662
2663             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2664             ewrt             = _mm_mul_ps(r22,ewtabscale);
2665             ewitab           = _mm_cvttps_epi32(ewrt);
2666             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2667             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2668                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2669                                          &ewtabF,&ewtabFn);
2670             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2671             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2672
2673             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2674
2675             fscal            = felec;
2676
2677             fscal            = _mm_and_ps(fscal,cutoff_mask);
2678
2679             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2680
2681             /* Calculate temporary vectorial force */
2682             tx               = _mm_mul_ps(fscal,dx22);
2683             ty               = _mm_mul_ps(fscal,dy22);
2684             tz               = _mm_mul_ps(fscal,dz22);
2685
2686             /* Update vectorial force */
2687             fix2             = _mm_add_ps(fix2,tx);
2688             fiy2             = _mm_add_ps(fiy2,ty);
2689             fiz2             = _mm_add_ps(fiz2,tz);
2690
2691             fjx2             = _mm_add_ps(fjx2,tx);
2692             fjy2             = _mm_add_ps(fjy2,ty);
2693             fjz2             = _mm_add_ps(fjz2,tz);
2694
2695             }
2696
2697             /**************************
2698              * CALCULATE INTERACTIONS *
2699              **************************/
2700
2701             if (gmx_mm_any_lt(rsq23,rcutoff2))
2702             {
2703
2704             r23              = _mm_mul_ps(rsq23,rinv23);
2705             r23              = _mm_andnot_ps(dummy_mask,r23);
2706
2707             /* EWALD ELECTROSTATICS */
2708
2709             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2710             ewrt             = _mm_mul_ps(r23,ewtabscale);
2711             ewitab           = _mm_cvttps_epi32(ewrt);
2712             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2713             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2714                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2715                                          &ewtabF,&ewtabFn);
2716             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2717             felec            = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
2718
2719             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
2720
2721             fscal            = felec;
2722
2723             fscal            = _mm_and_ps(fscal,cutoff_mask);
2724
2725             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2726
2727             /* Calculate temporary vectorial force */
2728             tx               = _mm_mul_ps(fscal,dx23);
2729             ty               = _mm_mul_ps(fscal,dy23);
2730             tz               = _mm_mul_ps(fscal,dz23);
2731
2732             /* Update vectorial force */
2733             fix2             = _mm_add_ps(fix2,tx);
2734             fiy2             = _mm_add_ps(fiy2,ty);
2735             fiz2             = _mm_add_ps(fiz2,tz);
2736
2737             fjx3             = _mm_add_ps(fjx3,tx);
2738             fjy3             = _mm_add_ps(fjy3,ty);
2739             fjz3             = _mm_add_ps(fjz3,tz);
2740
2741             }
2742
2743             /**************************
2744              * CALCULATE INTERACTIONS *
2745              **************************/
2746
2747             if (gmx_mm_any_lt(rsq31,rcutoff2))
2748             {
2749
2750             r31              = _mm_mul_ps(rsq31,rinv31);
2751             r31              = _mm_andnot_ps(dummy_mask,r31);
2752
2753             /* EWALD ELECTROSTATICS */
2754
2755             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2756             ewrt             = _mm_mul_ps(r31,ewtabscale);
2757             ewitab           = _mm_cvttps_epi32(ewrt);
2758             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2759             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2760                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2761                                          &ewtabF,&ewtabFn);
2762             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2763             felec            = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
2764
2765             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
2766
2767             fscal            = felec;
2768
2769             fscal            = _mm_and_ps(fscal,cutoff_mask);
2770
2771             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2772
2773             /* Calculate temporary vectorial force */
2774             tx               = _mm_mul_ps(fscal,dx31);
2775             ty               = _mm_mul_ps(fscal,dy31);
2776             tz               = _mm_mul_ps(fscal,dz31);
2777
2778             /* Update vectorial force */
2779             fix3             = _mm_add_ps(fix3,tx);
2780             fiy3             = _mm_add_ps(fiy3,ty);
2781             fiz3             = _mm_add_ps(fiz3,tz);
2782
2783             fjx1             = _mm_add_ps(fjx1,tx);
2784             fjy1             = _mm_add_ps(fjy1,ty);
2785             fjz1             = _mm_add_ps(fjz1,tz);
2786
2787             }
2788
2789             /**************************
2790              * CALCULATE INTERACTIONS *
2791              **************************/
2792
2793             if (gmx_mm_any_lt(rsq32,rcutoff2))
2794             {
2795
2796             r32              = _mm_mul_ps(rsq32,rinv32);
2797             r32              = _mm_andnot_ps(dummy_mask,r32);
2798
2799             /* EWALD ELECTROSTATICS */
2800
2801             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2802             ewrt             = _mm_mul_ps(r32,ewtabscale);
2803             ewitab           = _mm_cvttps_epi32(ewrt);
2804             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2805             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2806                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2807                                          &ewtabF,&ewtabFn);
2808             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2809             felec            = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
2810
2811             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
2812
2813             fscal            = felec;
2814
2815             fscal            = _mm_and_ps(fscal,cutoff_mask);
2816
2817             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2818
2819             /* Calculate temporary vectorial force */
2820             tx               = _mm_mul_ps(fscal,dx32);
2821             ty               = _mm_mul_ps(fscal,dy32);
2822             tz               = _mm_mul_ps(fscal,dz32);
2823
2824             /* Update vectorial force */
2825             fix3             = _mm_add_ps(fix3,tx);
2826             fiy3             = _mm_add_ps(fiy3,ty);
2827             fiz3             = _mm_add_ps(fiz3,tz);
2828
2829             fjx2             = _mm_add_ps(fjx2,tx);
2830             fjy2             = _mm_add_ps(fjy2,ty);
2831             fjz2             = _mm_add_ps(fjz2,tz);
2832
2833             }
2834
2835             /**************************
2836              * CALCULATE INTERACTIONS *
2837              **************************/
2838
2839             if (gmx_mm_any_lt(rsq33,rcutoff2))
2840             {
2841
2842             r33              = _mm_mul_ps(rsq33,rinv33);
2843             r33              = _mm_andnot_ps(dummy_mask,r33);
2844
2845             /* EWALD ELECTROSTATICS */
2846
2847             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2848             ewrt             = _mm_mul_ps(r33,ewtabscale);
2849             ewitab           = _mm_cvttps_epi32(ewrt);
2850             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2851             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2852                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2853                                          &ewtabF,&ewtabFn);
2854             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2855             felec            = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
2856
2857             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2858
2859             fscal            = felec;
2860
2861             fscal            = _mm_and_ps(fscal,cutoff_mask);
2862
2863             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2864
2865             /* Calculate temporary vectorial force */
2866             tx               = _mm_mul_ps(fscal,dx33);
2867             ty               = _mm_mul_ps(fscal,dy33);
2868             tz               = _mm_mul_ps(fscal,dz33);
2869
2870             /* Update vectorial force */
2871             fix3             = _mm_add_ps(fix3,tx);
2872             fiy3             = _mm_add_ps(fiy3,ty);
2873             fiz3             = _mm_add_ps(fiz3,tz);
2874
2875             fjx3             = _mm_add_ps(fjx3,tx);
2876             fjy3             = _mm_add_ps(fjy3,ty);
2877             fjz3             = _mm_add_ps(fjz3,tz);
2878
2879             }
2880
2881             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2882             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2883             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2884             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2885
2886             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2887                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2888                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2889
2890             /* Inner loop uses 413 flops */
2891         }
2892
2893         /* End of innermost loop */
2894
2895         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2896                                               f+i_coord_offset,fshift+i_shift_offset);
2897
2898         /* Increment number of inner iterations */
2899         inneriter                  += j_index_end - j_index_start;
2900
2901         /* Outer loop uses 24 flops */
2902     }
2903
2904     /* Increment number of outer iterations */
2905     outeriter        += nri;
2906
2907     /* Update outer/inner flops */
2908
2909     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*413);
2910 }