src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  94     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  95     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  96     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  97     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  98     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  99     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 100     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 101     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 102     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 103     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 104     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 105     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 106     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 107     real             *charge;
 108     int              nvdwtype;
 109     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 110     int              *vdwtype;
 111     real             *vdwparam;
 112     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 113     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 114     __m128           c6grid_00;
 115     __m128           c6grid_01;
 116     __m128           c6grid_02;
 117     __m128           c6grid_10;
 118     __m128           c6grid_11;
 119     __m128           c6grid_12;
 120     __m128           c6grid_20;
 121     __m128           c6grid_21;
 122     __m128           c6grid_22;
 123     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 124     real             *vdwgridparam;
 125     __m128           one_half = _mm_set1_ps(0.5);
 126     __m128           minus_one = _mm_set1_ps(-1.0);
 127     __m128i          ewitab;
 128     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 129     real             *ewtab;
 130     __m128           dummy_mask,cutoff_mask;
 131     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 132     __m128           one     = _mm_set1_ps(1.0);
 133     __m128           two     = _mm_set1_ps(2.0);
 134     x                = xx[0];
 135     f                = ff[0];
 136
 137     nri              = nlist->nri;
 138     iinr             = nlist->iinr;
 139     jindex           = nlist->jindex;
 140     jjnr             = nlist->jjnr;
 141     shiftidx         = nlist->shift;
 142     gid              = nlist->gid;
 143     shiftvec         = fr->shift_vec[0];
 144     fshift           = fr->fshift[0];
 145     facel            = _mm_set1_ps(fr->epsfac);
 146     charge           = mdatoms->chargeA;
 147     nvdwtype         = fr->ntype;
 148     vdwparam         = fr->nbfp;
 149     vdwtype          = mdatoms->typeA;
 150     vdwgridparam     = fr->ljpme_c6grid;
 151     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 152     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 153     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 154
 155     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 156     ewtab            = fr->ic->tabq_coul_FDV0;
 157     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 158     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 159
 160     /* Setup water-specific parameters */
 161     inr              = nlist->iinr[0];
 162     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 163     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 164     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 165     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 166
 167     jq0              = _mm_set1_ps(charge[inr+0]);
 168     jq1              = _mm_set1_ps(charge[inr+1]);
 169     jq2              = _mm_set1_ps(charge[inr+2]);
 170     vdwjidx0A        = 2*vdwtype[inr+0];
 171     qq00             = _mm_mul_ps(iq0,jq0);
 172     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 173     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 174     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
 175     qq01             = _mm_mul_ps(iq0,jq1);
 176     qq02             = _mm_mul_ps(iq0,jq2);
 177     qq10             = _mm_mul_ps(iq1,jq0);
 178     qq11             = _mm_mul_ps(iq1,jq1);
 179     qq12             = _mm_mul_ps(iq1,jq2);
 180     qq20             = _mm_mul_ps(iq2,jq0);
 181     qq21             = _mm_mul_ps(iq2,jq1);
 182     qq22             = _mm_mul_ps(iq2,jq2);
 183
 184     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 185     rcutoff_scalar   = fr->rcoulomb;
 186     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 187     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 188
 189     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 190     rvdw             = _mm_set1_ps(fr->rvdw);
 191
 192     /* Avoid stupid compiler warnings */
 193     jnrA = jnrB = jnrC = jnrD = 0;
 194     j_coord_offsetA = 0;
 195     j_coord_offsetB = 0;
 196     j_coord_offsetC = 0;
 197     j_coord_offsetD = 0;
 198
 199     outeriter        = 0;
 200     inneriter        = 0;
 201
 202     for(iidx=0;iidx<4*DIM;iidx++)
 203     {
 204         scratch[iidx] = 0.0;
 205     }
 206
 207     /* Start outer loop over neighborlists */
 208     for(iidx=0; iidx<nri; iidx++)
 209     {
 210         /* Load shift vector for this list */
 211         i_shift_offset   = DIM*shiftidx[iidx];
 212
 213         /* Load limits for loop over neighbors */
 214         j_index_start    = jindex[iidx];
 215         j_index_end      = jindex[iidx+1];
 216
 217         /* Get outer coordinate index */
 218         inr              = iinr[iidx];
 219         i_coord_offset   = DIM*inr;
 220
 221         /* Load i particle coords and add shift vector */
 222         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 223                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 224
 225         fix0             = _mm_setzero_ps();
 226         fiy0             = _mm_setzero_ps();
 227         fiz0             = _mm_setzero_ps();
 228         fix1             = _mm_setzero_ps();
 229         fiy1             = _mm_setzero_ps();
 230         fiz1             = _mm_setzero_ps();
 231         fix2             = _mm_setzero_ps();
 232         fiy2             = _mm_setzero_ps();
 233         fiz2             = _mm_setzero_ps();
 234
 235         /* Reset potential sums */
 236         velecsum         = _mm_setzero_ps();
 237         vvdwsum          = _mm_setzero_ps();
 238
 239         /* Start inner kernel loop */
 240         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 241         {
 242
 243             /* Get j neighbor index, and coordinate index */
 244             jnrA             = jjnr[jidx];
 245             jnrB             = jjnr[jidx+1];
 246             jnrC             = jjnr[jidx+2];
 247             jnrD             = jjnr[jidx+3];
 248             j_coord_offsetA  = DIM*jnrA;
 249             j_coord_offsetB  = DIM*jnrB;
 250             j_coord_offsetC  = DIM*jnrC;
 251             j_coord_offsetD  = DIM*jnrD;
 252
 253             /* load j atom coordinates */
 254             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 255                                               x+j_coord_offsetC,x+j_coord_offsetD,
 256                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm_sub_ps(ix0,jx0);
 260             dy00             = _mm_sub_ps(iy0,jy0);
 261             dz00             = _mm_sub_ps(iz0,jz0);
 262             dx01             = _mm_sub_ps(ix0,jx1);
 263             dy01             = _mm_sub_ps(iy0,jy1);
 264             dz01             = _mm_sub_ps(iz0,jz1);
 265             dx02             = _mm_sub_ps(ix0,jx2);
 266             dy02             = _mm_sub_ps(iy0,jy2);
 267             dz02             = _mm_sub_ps(iz0,jz2);
 268             dx10             = _mm_sub_ps(ix1,jx0);
 269             dy10             = _mm_sub_ps(iy1,jy0);
 270             dz10             = _mm_sub_ps(iz1,jz0);
 271             dx11             = _mm_sub_ps(ix1,jx1);
 272             dy11             = _mm_sub_ps(iy1,jy1);
 273             dz11             = _mm_sub_ps(iz1,jz1);
 274             dx12             = _mm_sub_ps(ix1,jx2);
 275             dy12             = _mm_sub_ps(iy1,jy2);
 276             dz12             = _mm_sub_ps(iz1,jz2);
 277             dx20             = _mm_sub_ps(ix2,jx0);
 278             dy20             = _mm_sub_ps(iy2,jy0);
 279             dz20             = _mm_sub_ps(iz2,jz0);
 280             dx21             = _mm_sub_ps(ix2,jx1);
 281             dy21             = _mm_sub_ps(iy2,jy1);
 282             dz21             = _mm_sub_ps(iz2,jz1);
 283             dx22             = _mm_sub_ps(ix2,jx2);
 284             dy22             = _mm_sub_ps(iy2,jy2);
 285             dz22             = _mm_sub_ps(iz2,jz2);
 286
 287             /* Calculate squared distance and things based on it */
 288             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 289             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 290             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 291             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 292             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 293             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 294             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 295             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 296             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 297
 298             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 299             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 300             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 301             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 302             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 303             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 304             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 305             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 306             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 307
 308             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 309             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 310             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 311             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 312             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 313             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 314             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 315             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 316             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 317
 318             fjx0             = _mm_setzero_ps();
 319             fjy0             = _mm_setzero_ps();
 320             fjz0             = _mm_setzero_ps();
 321             fjx1             = _mm_setzero_ps();
 322             fjy1             = _mm_setzero_ps();
 323             fjz1             = _mm_setzero_ps();
 324             fjx2             = _mm_setzero_ps();
 325             fjy2             = _mm_setzero_ps();
 326             fjz2             = _mm_setzero_ps();
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             if (gmx_mm_any_lt(rsq00,rcutoff2))
 333             {
 334
 335             r00              = _mm_mul_ps(rsq00,rinv00);
 336
 337             /* EWALD ELECTROSTATICS */
 338
 339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 340             ewrt             = _mm_mul_ps(r00,ewtabscale);
 341             ewitab           = _mm_cvttps_epi32(ewrt);
 342             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 343             ewitab           = _mm_slli_epi32(ewitab,2);
 344             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 345             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 346             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 347             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 348             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 349             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 350             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 351             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 352             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 353
 354             /* Analytical LJ-PME */
 355             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 356             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 357             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 358             exponent         = gmx_simd_exp_r(ewcljrsq);
 359             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 360             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 361             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 362             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 363             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 364             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) ,
 365                                           _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));
 366             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 367             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);
 368
 369             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velec            = _mm_and_ps(velec,cutoff_mask);
 373             velecsum         = _mm_add_ps(velecsum,velec);
 374             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 375             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 376
 377             fscal            = _mm_add_ps(felec,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(rsq01,rcutoff2))
 402             {
 403
 404             r01              = _mm_mul_ps(rsq01,rinv01);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 409             ewrt             = _mm_mul_ps(r01,ewtabscale);
 410             ewitab           = _mm_cvttps_epi32(ewrt);
 411             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq01,_mm_sub_ps(_mm_sub_ps(rinv01,sh_ewald),velec));
 421             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
 422
 423             cutoff_mask      = _mm_cmplt_ps(rsq01,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,dx01);
 435             ty               = _mm_mul_ps(fscal,dy01);
 436             tz               = _mm_mul_ps(fscal,dz01);
 437
 438             /* Update vectorial force */
 439             fix0             = _mm_add_ps(fix0,tx);
 440             fiy0             = _mm_add_ps(fiy0,ty);
 441             fiz0             = _mm_add_ps(fiz0,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(rsq02,rcutoff2))
 454             {
 455
 456             r02              = _mm_mul_ps(rsq02,rinv02);
 457
 458             /* EWALD ELECTROSTATICS */
 459
 460             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 461             ewrt             = _mm_mul_ps(r02,ewtabscale);
 462             ewitab           = _mm_cvttps_epi32(ewrt);
 463             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq02,_mm_sub_ps(_mm_sub_ps(rinv02,sh_ewald),velec));
 473             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
 474
 475             cutoff_mask      = _mm_cmplt_ps(rsq02,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,dx02);
 487             ty               = _mm_mul_ps(fscal,dy02);
 488             tz               = _mm_mul_ps(fscal,dz02);
 489
 490             /* Update vectorial force */
 491             fix0             = _mm_add_ps(fix0,tx);
 492             fiy0             = _mm_add_ps(fiy0,ty);
 493             fiz0             = _mm_add_ps(fiz0,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(rsq10,rcutoff2))
 506             {
 507
 508             r10              = _mm_mul_ps(rsq10,rinv10);
 509
 510             /* EWALD ELECTROSTATICS */
 511
 512             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 513             ewrt             = _mm_mul_ps(r10,ewtabscale);
 514             ewitab           = _mm_cvttps_epi32(ewrt);
 515             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 525             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 526
 527             cutoff_mask      = _mm_cmplt_ps(rsq10,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,dx10);
 539             ty               = _mm_mul_ps(fscal,dy10);
 540             tz               = _mm_mul_ps(fscal,dz10);
 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             fjx0             = _mm_add_ps(fjx0,tx);
 548             fjy0             = _mm_add_ps(fjy0,ty);
 549             fjz0             = _mm_add_ps(fjz0,tz);
 550
 551             }
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm_any_lt(rsq11,rcutoff2))
 558             {
 559
 560             r11              = _mm_mul_ps(rsq11,rinv11);
 561
 562             /* EWALD ELECTROSTATICS */
 563
 564             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 565             ewrt             = _mm_mul_ps(r11,ewtabscale);
 566             ewitab           = _mm_cvttps_epi32(ewrt);
 567             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec));
 577             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
 578
 579             cutoff_mask      = _mm_cmplt_ps(rsq11,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,dx11);
 591             ty               = _mm_mul_ps(fscal,dy11);
 592             tz               = _mm_mul_ps(fscal,dz11);
 593
 594             /* Update vectorial force */
 595             fix1             = _mm_add_ps(fix1,tx);
 596             fiy1             = _mm_add_ps(fiy1,ty);
 597             fiz1             = _mm_add_ps(fiz1,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(rsq12,rcutoff2))
 610             {
 611
 612             r12              = _mm_mul_ps(rsq12,rinv12);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 617             ewrt             = _mm_mul_ps(r12,ewtabscale);
 618             ewitab           = _mm_cvttps_epi32(ewrt);
 619             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec));
 629             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
 630
 631             cutoff_mask      = _mm_cmplt_ps(rsq12,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,dx12);
 643             ty               = _mm_mul_ps(fscal,dy12);
 644             tz               = _mm_mul_ps(fscal,dz12);
 645
 646             /* Update vectorial force */
 647             fix1             = _mm_add_ps(fix1,tx);
 648             fiy1             = _mm_add_ps(fiy1,ty);
 649             fiz1             = _mm_add_ps(fiz1,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(rsq20,rcutoff2))
 662             {
 663
 664             r20              = _mm_mul_ps(rsq20,rinv20);
 665
 666             /* EWALD ELECTROSTATICS */
 667
 668             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 669             ewrt             = _mm_mul_ps(r20,ewtabscale);
 670             ewitab           = _mm_cvttps_epi32(ewrt);
 671             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 681             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 682
 683             cutoff_mask      = _mm_cmplt_ps(rsq20,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,dx20);
 695             ty               = _mm_mul_ps(fscal,dy20);
 696             tz               = _mm_mul_ps(fscal,dz20);
 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             fjx0             = _mm_add_ps(fjx0,tx);
 704             fjy0             = _mm_add_ps(fjy0,ty);
 705             fjz0             = _mm_add_ps(fjz0,tz);
 706
 707             }
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             if (gmx_mm_any_lt(rsq21,rcutoff2))
 714             {
 715
 716             r21              = _mm_mul_ps(rsq21,rinv21);
 717
 718             /* EWALD ELECTROSTATICS */
 719
 720             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 721             ewrt             = _mm_mul_ps(r21,ewtabscale);
 722             ewitab           = _mm_cvttps_epi32(ewrt);
 723             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec));
 733             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
 734
 735             cutoff_mask      = _mm_cmplt_ps(rsq21,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,dx21);
 747             ty               = _mm_mul_ps(fscal,dy21);
 748             tz               = _mm_mul_ps(fscal,dz21);
 749
 750             /* Update vectorial force */
 751             fix2             = _mm_add_ps(fix2,tx);
 752             fiy2             = _mm_add_ps(fiy2,ty);
 753             fiz2             = _mm_add_ps(fiz2,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(rsq22,rcutoff2))
 766             {
 767
 768             r22              = _mm_mul_ps(rsq22,rinv22);
 769
 770             /* EWALD ELECTROSTATICS */
 771
 772             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 773             ewrt             = _mm_mul_ps(r22,ewtabscale);
 774             ewitab           = _mm_cvttps_epi32(ewrt);
 775             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec));
 785             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
 786
 787             cutoff_mask      = _mm_cmplt_ps(rsq22,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,dx22);
 799             ty               = _mm_mul_ps(fscal,dy22);
 800             tz               = _mm_mul_ps(fscal,dz22);
 801
 802             /* Update vectorial force */
 803             fix2             = _mm_add_ps(fix2,tx);
 804             fiy2             = _mm_add_ps(fiy2,ty);
 805             fiz2             = _mm_add_ps(fiz2,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             fjptrA             = f+j_coord_offsetA;
 814             fjptrB             = f+j_coord_offsetB;
 815             fjptrC             = f+j_coord_offsetC;
 816             fjptrD             = f+j_coord_offsetD;
 817
 818             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 819                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 820
 821             /* Inner loop uses 450 flops */
 822         }
 823
 824         if(jidx<j_index_end)
 825         {
 826
 827             /* Get j neighbor index, and coordinate index */
 828             jnrlistA         = jjnr[jidx];
 829             jnrlistB         = jjnr[jidx+1];
 830             jnrlistC         = jjnr[jidx+2];
 831             jnrlistD         = jjnr[jidx+3];
 832             /* Sign of each element will be negative for non-real atoms.
 833              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 834              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 835              */
 836             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 837             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 838             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 839             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 840             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843             j_coord_offsetC  = DIM*jnrC;
 844             j_coord_offsetD  = DIM*jnrD;
 845
 846             /* load j atom coordinates */
 847             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                               x+j_coord_offsetC,x+j_coord_offsetD,
 849                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_ps(ix0,jx0);
 853             dy00             = _mm_sub_ps(iy0,jy0);
 854             dz00             = _mm_sub_ps(iz0,jz0);
 855             dx01             = _mm_sub_ps(ix0,jx1);
 856             dy01             = _mm_sub_ps(iy0,jy1);
 857             dz01             = _mm_sub_ps(iz0,jz1);
 858             dx02             = _mm_sub_ps(ix0,jx2);
 859             dy02             = _mm_sub_ps(iy0,jy2);
 860             dz02             = _mm_sub_ps(iz0,jz2);
 861             dx10             = _mm_sub_ps(ix1,jx0);
 862             dy10             = _mm_sub_ps(iy1,jy0);
 863             dz10             = _mm_sub_ps(iz1,jz0);
 864             dx11             = _mm_sub_ps(ix1,jx1);
 865             dy11             = _mm_sub_ps(iy1,jy1);
 866             dz11             = _mm_sub_ps(iz1,jz1);
 867             dx12             = _mm_sub_ps(ix1,jx2);
 868             dy12             = _mm_sub_ps(iy1,jy2);
 869             dz12             = _mm_sub_ps(iz1,jz2);
 870             dx20             = _mm_sub_ps(ix2,jx0);
 871             dy20             = _mm_sub_ps(iy2,jy0);
 872             dz20             = _mm_sub_ps(iz2,jz0);
 873             dx21             = _mm_sub_ps(ix2,jx1);
 874             dy21             = _mm_sub_ps(iy2,jy1);
 875             dz21             = _mm_sub_ps(iz2,jz1);
 876             dx22             = _mm_sub_ps(ix2,jx2);
 877             dy22             = _mm_sub_ps(iy2,jy2);
 878             dz22             = _mm_sub_ps(iz2,jz2);
 879
 880             /* Calculate squared distance and things based on it */
 881             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 882             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 883             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 884             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 885             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 886             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 887             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 888             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 889             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 890
 891             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 892             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 893             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 894             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 895             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 896             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 897             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 898             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 899             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 900
 901             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 902             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 903             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 904             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 905             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 906             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 907             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 908             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 909             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 910
 911             fjx0             = _mm_setzero_ps();
 912             fjy0             = _mm_setzero_ps();
 913             fjz0             = _mm_setzero_ps();
 914             fjx1             = _mm_setzero_ps();
 915             fjy1             = _mm_setzero_ps();
 916             fjz1             = _mm_setzero_ps();
 917             fjx2             = _mm_setzero_ps();
 918             fjy2             = _mm_setzero_ps();
 919             fjz2             = _mm_setzero_ps();
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             if (gmx_mm_any_lt(rsq00,rcutoff2))
 926             {
 927
 928             r00              = _mm_mul_ps(rsq00,rinv00);
 929             r00              = _mm_andnot_ps(dummy_mask,r00);
 930
 931             /* EWALD ELECTROSTATICS */
 932
 933             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 934             ewrt             = _mm_mul_ps(r00,ewtabscale);
 935             ewitab           = _mm_cvttps_epi32(ewrt);
 936             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 937             ewitab           = _mm_slli_epi32(ewitab,2);
 938             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 939             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 940             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 941             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 942             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 943             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 944             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 945             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 946             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 947
 948             /* Analytical LJ-PME */
 949             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 950             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 951             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 952             exponent         = gmx_simd_exp_r(ewcljrsq);
 953             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 954             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 955             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 956             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 957             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 958             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) ,
 959                                           _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));
 960             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 961             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);
 962
 963             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 964
 965             /* Update potential sum for this i atom from the interaction with this j atom. */
 966             velec            = _mm_and_ps(velec,cutoff_mask);
 967             velec            = _mm_andnot_ps(dummy_mask,velec);
 968             velecsum         = _mm_add_ps(velecsum,velec);
 969             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 970             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 971             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 972
 973             fscal            = _mm_add_ps(felec,fvdw);
 974
 975             fscal            = _mm_and_ps(fscal,cutoff_mask);
 976
 977             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 978
 979             /* Calculate temporary vectorial force */
 980             tx               = _mm_mul_ps(fscal,dx00);
 981             ty               = _mm_mul_ps(fscal,dy00);
 982             tz               = _mm_mul_ps(fscal,dz00);
 983
 984             /* Update vectorial force */
 985             fix0             = _mm_add_ps(fix0,tx);
 986             fiy0             = _mm_add_ps(fiy0,ty);
 987             fiz0             = _mm_add_ps(fiz0,tz);
 988
 989             fjx0             = _mm_add_ps(fjx0,tx);
 990             fjy0             = _mm_add_ps(fjy0,ty);
 991             fjz0             = _mm_add_ps(fjz0,tz);
 992
 993             }
 994
 995             /**************************
 996              * CALCULATE INTERACTIONS *
 997              **************************/
 998
 999             if (gmx_mm_any_lt(rsq01,rcutoff2))
1000             {
1001
1002             r01              = _mm_mul_ps(rsq01,rinv01);
1003             r01              = _mm_andnot_ps(dummy_mask,r01);
1004
1005             /* EWALD ELECTROSTATICS */
1006
1007             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1008             ewrt             = _mm_mul_ps(r01,ewtabscale);
1009             ewitab           = _mm_cvttps_epi32(ewrt);
1010             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1011             ewitab           = _mm_slli_epi32(ewitab,2);
1012             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1013             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1014             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1015             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1016             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1017             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1018             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1019             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_sub_ps(rinv01,sh_ewald),velec));
1020             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
1021
1022             cutoff_mask      = _mm_cmplt_ps(rsq01,rcutoff2);
1023
1024             /* Update potential sum for this i atom from the interaction with this j atom. */
1025             velec            = _mm_and_ps(velec,cutoff_mask);
1026             velec            = _mm_andnot_ps(dummy_mask,velec);
1027             velecsum         = _mm_add_ps(velecsum,velec);
1028
1029             fscal            = felec;
1030
1031             fscal            = _mm_and_ps(fscal,cutoff_mask);
1032
1033             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm_mul_ps(fscal,dx01);
1037             ty               = _mm_mul_ps(fscal,dy01);
1038             tz               = _mm_mul_ps(fscal,dz01);
1039
1040             /* Update vectorial force */
1041             fix0             = _mm_add_ps(fix0,tx);
1042             fiy0             = _mm_add_ps(fiy0,ty);
1043             fiz0             = _mm_add_ps(fiz0,tz);
1044
1045             fjx1             = _mm_add_ps(fjx1,tx);
1046             fjy1             = _mm_add_ps(fjy1,ty);
1047             fjz1             = _mm_add_ps(fjz1,tz);
1048
1049             }
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             if (gmx_mm_any_lt(rsq02,rcutoff2))
1056             {
1057
1058             r02              = _mm_mul_ps(rsq02,rinv02);
1059             r02              = _mm_andnot_ps(dummy_mask,r02);
1060
1061             /* EWALD ELECTROSTATICS */
1062
1063             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1064             ewrt             = _mm_mul_ps(r02,ewtabscale);
1065             ewitab           = _mm_cvttps_epi32(ewrt);
1066             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1067             ewitab           = _mm_slli_epi32(ewitab,2);
1068             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1069             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1070             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1071             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1072             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1073             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1074             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1075             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_sub_ps(rinv02,sh_ewald),velec));
1076             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
1077
1078             cutoff_mask      = _mm_cmplt_ps(rsq02,rcutoff2);
1079
1080             /* Update potential sum for this i atom from the interaction with this j atom. */
1081             velec            = _mm_and_ps(velec,cutoff_mask);
1082             velec            = _mm_andnot_ps(dummy_mask,velec);
1083             velecsum         = _mm_add_ps(velecsum,velec);
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm_and_ps(fscal,cutoff_mask);
1088
1089             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1090
1091             /* Calculate temporary vectorial force */
1092             tx               = _mm_mul_ps(fscal,dx02);
1093             ty               = _mm_mul_ps(fscal,dy02);
1094             tz               = _mm_mul_ps(fscal,dz02);
1095
1096             /* Update vectorial force */
1097             fix0             = _mm_add_ps(fix0,tx);
1098             fiy0             = _mm_add_ps(fiy0,ty);
1099             fiz0             = _mm_add_ps(fiz0,tz);
1100
1101             fjx2             = _mm_add_ps(fjx2,tx);
1102             fjy2             = _mm_add_ps(fjy2,ty);
1103             fjz2             = _mm_add_ps(fjz2,tz);
1104
1105             }
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             if (gmx_mm_any_lt(rsq10,rcutoff2))
1112             {
1113
1114             r10              = _mm_mul_ps(rsq10,rinv10);
1115             r10              = _mm_andnot_ps(dummy_mask,r10);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120             ewrt             = _mm_mul_ps(r10,ewtabscale);
1121             ewitab           = _mm_cvttps_epi32(ewrt);
1122             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1123             ewitab           = _mm_slli_epi32(ewitab,2);
1124             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1125             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1126             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1127             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1128             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1129             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1130             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1131             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
1132             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1133
1134             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1135
1136             /* Update potential sum for this i atom from the interaction with this j atom. */
1137             velec            = _mm_and_ps(velec,cutoff_mask);
1138             velec            = _mm_andnot_ps(dummy_mask,velec);
1139             velecsum         = _mm_add_ps(velecsum,velec);
1140
1141             fscal            = felec;
1142
1143             fscal            = _mm_and_ps(fscal,cutoff_mask);
1144
1145             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1146
1147             /* Calculate temporary vectorial force */
1148             tx               = _mm_mul_ps(fscal,dx10);
1149             ty               = _mm_mul_ps(fscal,dy10);
1150             tz               = _mm_mul_ps(fscal,dz10);
1151
1152             /* Update vectorial force */
1153             fix1             = _mm_add_ps(fix1,tx);
1154             fiy1             = _mm_add_ps(fiy1,ty);
1155             fiz1             = _mm_add_ps(fiz1,tz);
1156
1157             fjx0             = _mm_add_ps(fjx0,tx);
1158             fjy0             = _mm_add_ps(fjy0,ty);
1159             fjz0             = _mm_add_ps(fjz0,tz);
1160
1161             }
1162
1163             /**************************
1164              * CALCULATE INTERACTIONS *
1165              **************************/
1166
1167             if (gmx_mm_any_lt(rsq11,rcutoff2))
1168             {
1169
1170             r11              = _mm_mul_ps(rsq11,rinv11);
1171             r11              = _mm_andnot_ps(dummy_mask,r11);
1172
1173             /* EWALD ELECTROSTATICS */
1174
1175             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1176             ewrt             = _mm_mul_ps(r11,ewtabscale);
1177             ewitab           = _mm_cvttps_epi32(ewrt);
1178             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1179             ewitab           = _mm_slli_epi32(ewitab,2);
1180             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1181             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1182             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1183             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1184             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1185             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1186             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1187             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec));
1188             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1189
1190             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1191
1192             /* Update potential sum for this i atom from the interaction with this j atom. */
1193             velec            = _mm_and_ps(velec,cutoff_mask);
1194             velec            = _mm_andnot_ps(dummy_mask,velec);
1195             velecsum         = _mm_add_ps(velecsum,velec);
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm_and_ps(fscal,cutoff_mask);
1200
1201             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm_mul_ps(fscal,dx11);
1205             ty               = _mm_mul_ps(fscal,dy11);
1206             tz               = _mm_mul_ps(fscal,dz11);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm_add_ps(fix1,tx);
1210             fiy1             = _mm_add_ps(fiy1,ty);
1211             fiz1             = _mm_add_ps(fiz1,tz);
1212
1213             fjx1             = _mm_add_ps(fjx1,tx);
1214             fjy1             = _mm_add_ps(fjy1,ty);
1215             fjz1             = _mm_add_ps(fjz1,tz);
1216
1217             }
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             if (gmx_mm_any_lt(rsq12,rcutoff2))
1224             {
1225
1226             r12              = _mm_mul_ps(rsq12,rinv12);
1227             r12              = _mm_andnot_ps(dummy_mask,r12);
1228
1229             /* EWALD ELECTROSTATICS */
1230
1231             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1232             ewrt             = _mm_mul_ps(r12,ewtabscale);
1233             ewitab           = _mm_cvttps_epi32(ewrt);
1234             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1235             ewitab           = _mm_slli_epi32(ewitab,2);
1236             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1237             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1238             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1239             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1240             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1241             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1242             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1243             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec));
1244             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1245
1246             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1247
1248             /* Update potential sum for this i atom from the interaction with this j atom. */
1249             velec            = _mm_and_ps(velec,cutoff_mask);
1250             velec            = _mm_andnot_ps(dummy_mask,velec);
1251             velecsum         = _mm_add_ps(velecsum,velec);
1252
1253             fscal            = felec;
1254
1255             fscal            = _mm_and_ps(fscal,cutoff_mask);
1256
1257             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1258
1259             /* Calculate temporary vectorial force */
1260             tx               = _mm_mul_ps(fscal,dx12);
1261             ty               = _mm_mul_ps(fscal,dy12);
1262             tz               = _mm_mul_ps(fscal,dz12);
1263
1264             /* Update vectorial force */
1265             fix1             = _mm_add_ps(fix1,tx);
1266             fiy1             = _mm_add_ps(fiy1,ty);
1267             fiz1             = _mm_add_ps(fiz1,tz);
1268
1269             fjx2             = _mm_add_ps(fjx2,tx);
1270             fjy2             = _mm_add_ps(fjy2,ty);
1271             fjz2             = _mm_add_ps(fjz2,tz);
1272
1273             }
1274
1275             /**************************
1276              * CALCULATE INTERACTIONS *
1277              **************************/
1278
1279             if (gmx_mm_any_lt(rsq20,rcutoff2))
1280             {
1281
1282             r20              = _mm_mul_ps(rsq20,rinv20);
1283             r20              = _mm_andnot_ps(dummy_mask,r20);
1284
1285             /* EWALD ELECTROSTATICS */
1286
1287             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1288             ewrt             = _mm_mul_ps(r20,ewtabscale);
1289             ewitab           = _mm_cvttps_epi32(ewrt);
1290             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1291             ewitab           = _mm_slli_epi32(ewitab,2);
1292             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1293             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1294             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1295             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1296             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1297             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1298             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1299             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
1300             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1301
1302             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1303
1304             /* Update potential sum for this i atom from the interaction with this j atom. */
1305             velec            = _mm_and_ps(velec,cutoff_mask);
1306             velec            = _mm_andnot_ps(dummy_mask,velec);
1307             velecsum         = _mm_add_ps(velecsum,velec);
1308
1309             fscal            = felec;
1310
1311             fscal            = _mm_and_ps(fscal,cutoff_mask);
1312
1313             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1314
1315             /* Calculate temporary vectorial force */
1316             tx               = _mm_mul_ps(fscal,dx20);
1317             ty               = _mm_mul_ps(fscal,dy20);
1318             tz               = _mm_mul_ps(fscal,dz20);
1319
1320             /* Update vectorial force */
1321             fix2             = _mm_add_ps(fix2,tx);
1322             fiy2             = _mm_add_ps(fiy2,ty);
1323             fiz2             = _mm_add_ps(fiz2,tz);
1324
1325             fjx0             = _mm_add_ps(fjx0,tx);
1326             fjy0             = _mm_add_ps(fjy0,ty);
1327             fjz0             = _mm_add_ps(fjz0,tz);
1328
1329             }
1330
1331             /**************************
1332              * CALCULATE INTERACTIONS *
1333              **************************/
1334
1335             if (gmx_mm_any_lt(rsq21,rcutoff2))
1336             {
1337
1338             r21              = _mm_mul_ps(rsq21,rinv21);
1339             r21              = _mm_andnot_ps(dummy_mask,r21);
1340
1341             /* EWALD ELECTROSTATICS */
1342
1343             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1344             ewrt             = _mm_mul_ps(r21,ewtabscale);
1345             ewitab           = _mm_cvttps_epi32(ewrt);
1346             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1347             ewitab           = _mm_slli_epi32(ewitab,2);
1348             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1349             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1350             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1351             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1352             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1353             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1354             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1355             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec));
1356             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1357
1358             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1359
1360             /* Update potential sum for this i atom from the interaction with this j atom. */
1361             velec            = _mm_and_ps(velec,cutoff_mask);
1362             velec            = _mm_andnot_ps(dummy_mask,velec);
1363             velecsum         = _mm_add_ps(velecsum,velec);
1364
1365             fscal            = felec;
1366
1367             fscal            = _mm_and_ps(fscal,cutoff_mask);
1368
1369             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1370
1371             /* Calculate temporary vectorial force */
1372             tx               = _mm_mul_ps(fscal,dx21);
1373             ty               = _mm_mul_ps(fscal,dy21);
1374             tz               = _mm_mul_ps(fscal,dz21);
1375
1376             /* Update vectorial force */
1377             fix2             = _mm_add_ps(fix2,tx);
1378             fiy2             = _mm_add_ps(fiy2,ty);
1379             fiz2             = _mm_add_ps(fiz2,tz);
1380
1381             fjx1             = _mm_add_ps(fjx1,tx);
1382             fjy1             = _mm_add_ps(fjy1,ty);
1383             fjz1             = _mm_add_ps(fjz1,tz);
1384
1385             }
1386
1387             /**************************
1388              * CALCULATE INTERACTIONS *
1389              **************************/
1390
1391             if (gmx_mm_any_lt(rsq22,rcutoff2))
1392             {
1393
1394             r22              = _mm_mul_ps(rsq22,rinv22);
1395             r22              = _mm_andnot_ps(dummy_mask,r22);
1396
1397             /* EWALD ELECTROSTATICS */
1398
1399             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1400             ewrt             = _mm_mul_ps(r22,ewtabscale);
1401             ewitab           = _mm_cvttps_epi32(ewrt);
1402             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1403             ewitab           = _mm_slli_epi32(ewitab,2);
1404             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1405             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1406             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1407             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1408             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1409             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1410             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1411             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec));
1412             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1413
1414             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1415
1416             /* Update potential sum for this i atom from the interaction with this j atom. */
1417             velec            = _mm_and_ps(velec,cutoff_mask);
1418             velec            = _mm_andnot_ps(dummy_mask,velec);
1419             velecsum         = _mm_add_ps(velecsum,velec);
1420
1421             fscal            = felec;
1422
1423             fscal            = _mm_and_ps(fscal,cutoff_mask);
1424
1425             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1426
1427             /* Calculate temporary vectorial force */
1428             tx               = _mm_mul_ps(fscal,dx22);
1429             ty               = _mm_mul_ps(fscal,dy22);
1430             tz               = _mm_mul_ps(fscal,dz22);
1431
1432             /* Update vectorial force */
1433             fix2             = _mm_add_ps(fix2,tx);
1434             fiy2             = _mm_add_ps(fiy2,ty);
1435             fiz2             = _mm_add_ps(fiz2,tz);
1436
1437             fjx2             = _mm_add_ps(fjx2,tx);
1438             fjy2             = _mm_add_ps(fjy2,ty);
1439             fjz2             = _mm_add_ps(fjz2,tz);
1440
1441             }
1442
1443             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1444             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1445             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1446             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1447
1448             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1449                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1450
1451             /* Inner loop uses 459 flops */
1452         }
1453
1454         /* End of innermost loop */
1455
1456         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1457                                               f+i_coord_offset,fshift+i_shift_offset);
1458
1459         ggid                        = gid[iidx];
1460         /* Update potential energies */
1461         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1462         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1463
1464         /* Increment number of inner iterations */
1465         inneriter                  += j_index_end - j_index_start;
1466
1467         /* Outer loop uses 20 flops */
1468     }
1469
1470     /* Increment number of outer iterations */
1471     outeriter        += nri;
1472
1473     /* Update outer/inner flops */
1474
1475     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*459);
1476 }
1477 /*
1478  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_F_sse2_single
1479  * Electrostatics interaction: Ewald
1480  * VdW interaction:            LJEwald
1481  * Geometry:                   Water3-Water3
1482  * Calculate force/pot:        Force
1483  */
1484 void
1485 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_F_sse2_single
1486                     (t_nblist                    * gmx_restrict       nlist,
1487                      rvec                        * gmx_restrict          xx,
1488                      rvec                        * gmx_restrict          ff,
1489                      t_forcerec                  * gmx_restrict          fr,
1490                      t_mdatoms                   * gmx_restrict     mdatoms,
1491                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1492                      t_nrnb                      * gmx_restrict        nrnb)
1493 {
1494     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1495      * just 0 for non-waters.
1496      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1497      * jnr indices corresponding to data put in the four positions in the SIMD register.
1498      */
1499     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1500     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1501     int              jnrA,jnrB,jnrC,jnrD;
1502     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1503     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1504     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1505     real             rcutoff_scalar;
1506     real             *shiftvec,*fshift,*x,*f;
1507     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1508     real             scratch[4*DIM];
1509     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1510     int              vdwioffset0;
1511     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1512     int              vdwioffset1;
1513     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1514     int              vdwioffset2;
1515     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1516     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1517     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1518     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1519     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1520     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1521     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1522     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1523     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1524     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1525     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1526     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1527     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1528     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1529     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1530     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1531     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1532     real             *charge;
1533     int              nvdwtype;
1534     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1535     int              *vdwtype;
1536     real             *vdwparam;
1537     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1538     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1539     __m128           c6grid_00;
1540     __m128           c6grid_01;
1541     __m128           c6grid_02;
1542     __m128           c6grid_10;
1543     __m128           c6grid_11;
1544     __m128           c6grid_12;
1545     __m128           c6grid_20;
1546     __m128           c6grid_21;
1547     __m128           c6grid_22;
1548     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1549     real             *vdwgridparam;
1550     __m128           one_half = _mm_set1_ps(0.5);
1551     __m128           minus_one = _mm_set1_ps(-1.0);
1552     __m128i          ewitab;
1553     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1554     real             *ewtab;
1555     __m128           dummy_mask,cutoff_mask;
1556     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1557     __m128           one     = _mm_set1_ps(1.0);
1558     __m128           two     = _mm_set1_ps(2.0);
1559     x                = xx[0];
1560     f                = ff[0];
1561
1562     nri              = nlist->nri;
1563     iinr             = nlist->iinr;
1564     jindex           = nlist->jindex;
1565     jjnr             = nlist->jjnr;
1566     shiftidx         = nlist->shift;
1567     gid              = nlist->gid;
1568     shiftvec         = fr->shift_vec[0];
1569     fshift           = fr->fshift[0];
1570     facel            = _mm_set1_ps(fr->epsfac);
1571     charge           = mdatoms->chargeA;
1572     nvdwtype         = fr->ntype;
1573     vdwparam         = fr->nbfp;
1574     vdwtype          = mdatoms->typeA;
1575     vdwgridparam     = fr->ljpme_c6grid;
1576     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
1577     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
1578     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
1579
1580     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1581     ewtab            = fr->ic->tabq_coul_F;
1582     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1583     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1584
1585     /* Setup water-specific parameters */
1586     inr              = nlist->iinr[0];
1587     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1588     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1589     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1590     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1591
1592     jq0              = _mm_set1_ps(charge[inr+0]);
1593     jq1              = _mm_set1_ps(charge[inr+1]);
1594     jq2              = _mm_set1_ps(charge[inr+2]);
1595     vdwjidx0A        = 2*vdwtype[inr+0];
1596     qq00             = _mm_mul_ps(iq0,jq0);
1597     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1598     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1599     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
1600     qq01             = _mm_mul_ps(iq0,jq1);
1601     qq02             = _mm_mul_ps(iq0,jq2);
1602     qq10             = _mm_mul_ps(iq1,jq0);
1603     qq11             = _mm_mul_ps(iq1,jq1);
1604     qq12             = _mm_mul_ps(iq1,jq2);
1605     qq20             = _mm_mul_ps(iq2,jq0);
1606     qq21             = _mm_mul_ps(iq2,jq1);
1607     qq22             = _mm_mul_ps(iq2,jq2);
1608
1609     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1610     rcutoff_scalar   = fr->rcoulomb;
1611     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1612     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1613
1614     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
1615     rvdw             = _mm_set1_ps(fr->rvdw);
1616
1617     /* Avoid stupid compiler warnings */
1618     jnrA = jnrB = jnrC = jnrD = 0;
1619     j_coord_offsetA = 0;
1620     j_coord_offsetB = 0;
1621     j_coord_offsetC = 0;
1622     j_coord_offsetD = 0;
1623
1624     outeriter        = 0;
1625     inneriter        = 0;
1626
1627     for(iidx=0;iidx<4*DIM;iidx++)
1628     {
1629         scratch[iidx] = 0.0;
1630     }
1631
1632     /* Start outer loop over neighborlists */
1633     for(iidx=0; iidx<nri; iidx++)
1634     {
1635         /* Load shift vector for this list */
1636         i_shift_offset   = DIM*shiftidx[iidx];
1637
1638         /* Load limits for loop over neighbors */
1639         j_index_start    = jindex[iidx];
1640         j_index_end      = jindex[iidx+1];
1641
1642         /* Get outer coordinate index */
1643         inr              = iinr[iidx];
1644         i_coord_offset   = DIM*inr;
1645
1646         /* Load i particle coords and add shift vector */
1647         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1648                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1649
1650         fix0             = _mm_setzero_ps();
1651         fiy0             = _mm_setzero_ps();
1652         fiz0             = _mm_setzero_ps();
1653         fix1             = _mm_setzero_ps();
1654         fiy1             = _mm_setzero_ps();
1655         fiz1             = _mm_setzero_ps();
1656         fix2             = _mm_setzero_ps();
1657         fiy2             = _mm_setzero_ps();
1658         fiz2             = _mm_setzero_ps();
1659
1660         /* Start inner kernel loop */
1661         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1662         {
1663
1664             /* Get j neighbor index, and coordinate index */
1665             jnrA             = jjnr[jidx];
1666             jnrB             = jjnr[jidx+1];
1667             jnrC             = jjnr[jidx+2];
1668             jnrD             = jjnr[jidx+3];
1669             j_coord_offsetA  = DIM*jnrA;
1670             j_coord_offsetB  = DIM*jnrB;
1671             j_coord_offsetC  = DIM*jnrC;
1672             j_coord_offsetD  = DIM*jnrD;
1673
1674             /* load j atom coordinates */
1675             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1676                                               x+j_coord_offsetC,x+j_coord_offsetD,
1677                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1678
1679             /* Calculate displacement vector */
1680             dx00             = _mm_sub_ps(ix0,jx0);
1681             dy00             = _mm_sub_ps(iy0,jy0);
1682             dz00             = _mm_sub_ps(iz0,jz0);
1683             dx01             = _mm_sub_ps(ix0,jx1);
1684             dy01             = _mm_sub_ps(iy0,jy1);
1685             dz01             = _mm_sub_ps(iz0,jz1);
1686             dx02             = _mm_sub_ps(ix0,jx2);
1687             dy02             = _mm_sub_ps(iy0,jy2);
1688             dz02             = _mm_sub_ps(iz0,jz2);
1689             dx10             = _mm_sub_ps(ix1,jx0);
1690             dy10             = _mm_sub_ps(iy1,jy0);
1691             dz10             = _mm_sub_ps(iz1,jz0);
1692             dx11             = _mm_sub_ps(ix1,jx1);
1693             dy11             = _mm_sub_ps(iy1,jy1);
1694             dz11             = _mm_sub_ps(iz1,jz1);
1695             dx12             = _mm_sub_ps(ix1,jx2);
1696             dy12             = _mm_sub_ps(iy1,jy2);
1697             dz12             = _mm_sub_ps(iz1,jz2);
1698             dx20             = _mm_sub_ps(ix2,jx0);
1699             dy20             = _mm_sub_ps(iy2,jy0);
1700             dz20             = _mm_sub_ps(iz2,jz0);
1701             dx21             = _mm_sub_ps(ix2,jx1);
1702             dy21             = _mm_sub_ps(iy2,jy1);
1703             dz21             = _mm_sub_ps(iz2,jz1);
1704             dx22             = _mm_sub_ps(ix2,jx2);
1705             dy22             = _mm_sub_ps(iy2,jy2);
1706             dz22             = _mm_sub_ps(iz2,jz2);
1707
1708             /* Calculate squared distance and things based on it */
1709             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1710             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1711             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1712             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1713             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1714             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1715             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1716             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1717             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1718
1719             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1720             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1721             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1722             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1723             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1724             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1725             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1726             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1727             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1728
1729             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1730             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1731             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1732             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1733             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1734             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1735             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1736             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1737             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1738
1739             fjx0             = _mm_setzero_ps();
1740             fjy0             = _mm_setzero_ps();
1741             fjz0             = _mm_setzero_ps();
1742             fjx1             = _mm_setzero_ps();
1743             fjy1             = _mm_setzero_ps();
1744             fjz1             = _mm_setzero_ps();
1745             fjx2             = _mm_setzero_ps();
1746             fjy2             = _mm_setzero_ps();
1747             fjz2             = _mm_setzero_ps();
1748
1749             /**************************
1750              * CALCULATE INTERACTIONS *
1751              **************************/
1752
1753             if (gmx_mm_any_lt(rsq00,rcutoff2))
1754             {
1755
1756             r00              = _mm_mul_ps(rsq00,rinv00);
1757
1758             /* EWALD ELECTROSTATICS */
1759
1760             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1761             ewrt             = _mm_mul_ps(r00,ewtabscale);
1762             ewitab           = _mm_cvttps_epi32(ewrt);
1763             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1764             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1765                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1766                                          &ewtabF,&ewtabFn);
1767             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1768             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1769
1770             /* Analytical LJ-PME */
1771             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1772             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1773             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1774             exponent         = gmx_simd_exp_r(ewcljrsq);
1775             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1776             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1777             /* f6A = 6 * C6grid * (1 - poly) */
1778             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1779             /* f6B = C6grid * exponent * beta^6 */
1780             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1781             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1782             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);
1783
1784             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1785
1786             fscal            = _mm_add_ps(felec,fvdw);
1787
1788             fscal            = _mm_and_ps(fscal,cutoff_mask);
1789
1790             /* Calculate temporary vectorial force */
1791             tx               = _mm_mul_ps(fscal,dx00);
1792             ty               = _mm_mul_ps(fscal,dy00);
1793             tz               = _mm_mul_ps(fscal,dz00);
1794
1795             /* Update vectorial force */
1796             fix0             = _mm_add_ps(fix0,tx);
1797             fiy0             = _mm_add_ps(fiy0,ty);
1798             fiz0             = _mm_add_ps(fiz0,tz);
1799
1800             fjx0             = _mm_add_ps(fjx0,tx);
1801             fjy0             = _mm_add_ps(fjy0,ty);
1802             fjz0             = _mm_add_ps(fjz0,tz);
1803
1804             }
1805
1806             /**************************
1807              * CALCULATE INTERACTIONS *
1808              **************************/
1809
1810             if (gmx_mm_any_lt(rsq01,rcutoff2))
1811             {
1812
1813             r01              = _mm_mul_ps(rsq01,rinv01);
1814
1815             /* EWALD ELECTROSTATICS */
1816
1817             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1818             ewrt             = _mm_mul_ps(r01,ewtabscale);
1819             ewitab           = _mm_cvttps_epi32(ewrt);
1820             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1821             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1822                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1823                                          &ewtabF,&ewtabFn);
1824             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1825             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
1826
1827             cutoff_mask      = _mm_cmplt_ps(rsq01,rcutoff2);
1828
1829             fscal            = felec;
1830
1831             fscal            = _mm_and_ps(fscal,cutoff_mask);
1832
1833             /* Calculate temporary vectorial force */
1834             tx               = _mm_mul_ps(fscal,dx01);
1835             ty               = _mm_mul_ps(fscal,dy01);
1836             tz               = _mm_mul_ps(fscal,dz01);
1837
1838             /* Update vectorial force */
1839             fix0             = _mm_add_ps(fix0,tx);
1840             fiy0             = _mm_add_ps(fiy0,ty);
1841             fiz0             = _mm_add_ps(fiz0,tz);
1842
1843             fjx1             = _mm_add_ps(fjx1,tx);
1844             fjy1             = _mm_add_ps(fjy1,ty);
1845             fjz1             = _mm_add_ps(fjz1,tz);
1846
1847             }
1848
1849             /**************************
1850              * CALCULATE INTERACTIONS *
1851              **************************/
1852
1853             if (gmx_mm_any_lt(rsq02,rcutoff2))
1854             {
1855
1856             r02              = _mm_mul_ps(rsq02,rinv02);
1857
1858             /* EWALD ELECTROSTATICS */
1859
1860             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1861             ewrt             = _mm_mul_ps(r02,ewtabscale);
1862             ewitab           = _mm_cvttps_epi32(ewrt);
1863             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1864             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1865                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1866                                          &ewtabF,&ewtabFn);
1867             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1868             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
1869
1870             cutoff_mask      = _mm_cmplt_ps(rsq02,rcutoff2);
1871
1872             fscal            = felec;
1873
1874             fscal            = _mm_and_ps(fscal,cutoff_mask);
1875
1876             /* Calculate temporary vectorial force */
1877             tx               = _mm_mul_ps(fscal,dx02);
1878             ty               = _mm_mul_ps(fscal,dy02);
1879             tz               = _mm_mul_ps(fscal,dz02);
1880
1881             /* Update vectorial force */
1882             fix0             = _mm_add_ps(fix0,tx);
1883             fiy0             = _mm_add_ps(fiy0,ty);
1884             fiz0             = _mm_add_ps(fiz0,tz);
1885
1886             fjx2             = _mm_add_ps(fjx2,tx);
1887             fjy2             = _mm_add_ps(fjy2,ty);
1888             fjz2             = _mm_add_ps(fjz2,tz);
1889
1890             }
1891
1892             /**************************
1893              * CALCULATE INTERACTIONS *
1894              **************************/
1895
1896             if (gmx_mm_any_lt(rsq10,rcutoff2))
1897             {
1898
1899             r10              = _mm_mul_ps(rsq10,rinv10);
1900
1901             /* EWALD ELECTROSTATICS */
1902
1903             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1904             ewrt             = _mm_mul_ps(r10,ewtabscale);
1905             ewitab           = _mm_cvttps_epi32(ewrt);
1906             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1907             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1908                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1909                                          &ewtabF,&ewtabFn);
1910             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1911             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1912
1913             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1914
1915             fscal            = felec;
1916
1917             fscal            = _mm_and_ps(fscal,cutoff_mask);
1918
1919             /* Calculate temporary vectorial force */
1920             tx               = _mm_mul_ps(fscal,dx10);
1921             ty               = _mm_mul_ps(fscal,dy10);
1922             tz               = _mm_mul_ps(fscal,dz10);
1923
1924             /* Update vectorial force */
1925             fix1             = _mm_add_ps(fix1,tx);
1926             fiy1             = _mm_add_ps(fiy1,ty);
1927             fiz1             = _mm_add_ps(fiz1,tz);
1928
1929             fjx0             = _mm_add_ps(fjx0,tx);
1930             fjy0             = _mm_add_ps(fjy0,ty);
1931             fjz0             = _mm_add_ps(fjz0,tz);
1932
1933             }
1934
1935             /**************************
1936              * CALCULATE INTERACTIONS *
1937              **************************/
1938
1939             if (gmx_mm_any_lt(rsq11,rcutoff2))
1940             {
1941
1942             r11              = _mm_mul_ps(rsq11,rinv11);
1943
1944             /* EWALD ELECTROSTATICS */
1945
1946             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1947             ewrt             = _mm_mul_ps(r11,ewtabscale);
1948             ewitab           = _mm_cvttps_epi32(ewrt);
1949             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1950             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1951                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1952                                          &ewtabF,&ewtabFn);
1953             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1954             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1955
1956             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1957
1958             fscal            = felec;
1959
1960             fscal            = _mm_and_ps(fscal,cutoff_mask);
1961
1962             /* Calculate temporary vectorial force */
1963             tx               = _mm_mul_ps(fscal,dx11);
1964             ty               = _mm_mul_ps(fscal,dy11);
1965             tz               = _mm_mul_ps(fscal,dz11);
1966
1967             /* Update vectorial force */
1968             fix1             = _mm_add_ps(fix1,tx);
1969             fiy1             = _mm_add_ps(fiy1,ty);
1970             fiz1             = _mm_add_ps(fiz1,tz);
1971
1972             fjx1             = _mm_add_ps(fjx1,tx);
1973             fjy1             = _mm_add_ps(fjy1,ty);
1974             fjz1             = _mm_add_ps(fjz1,tz);
1975
1976             }
1977
1978             /**************************
1979              * CALCULATE INTERACTIONS *
1980              **************************/
1981
1982             if (gmx_mm_any_lt(rsq12,rcutoff2))
1983             {
1984
1985             r12              = _mm_mul_ps(rsq12,rinv12);
1986
1987             /* EWALD ELECTROSTATICS */
1988
1989             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1990             ewrt             = _mm_mul_ps(r12,ewtabscale);
1991             ewitab           = _mm_cvttps_epi32(ewrt);
1992             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1993             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1994                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1995                                          &ewtabF,&ewtabFn);
1996             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1997             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1998
1999             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
2000
2001             fscal            = felec;
2002
2003             fscal            = _mm_and_ps(fscal,cutoff_mask);
2004
2005             /* Calculate temporary vectorial force */
2006             tx               = _mm_mul_ps(fscal,dx12);
2007             ty               = _mm_mul_ps(fscal,dy12);
2008             tz               = _mm_mul_ps(fscal,dz12);
2009
2010             /* Update vectorial force */
2011             fix1             = _mm_add_ps(fix1,tx);
2012             fiy1             = _mm_add_ps(fiy1,ty);
2013             fiz1             = _mm_add_ps(fiz1,tz);
2014
2015             fjx2             = _mm_add_ps(fjx2,tx);
2016             fjy2             = _mm_add_ps(fjy2,ty);
2017             fjz2             = _mm_add_ps(fjz2,tz);
2018
2019             }
2020
2021             /**************************
2022              * CALCULATE INTERACTIONS *
2023              **************************/
2024
2025             if (gmx_mm_any_lt(rsq20,rcutoff2))
2026             {
2027
2028             r20              = _mm_mul_ps(rsq20,rinv20);
2029
2030             /* EWALD ELECTROSTATICS */
2031
2032             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2033             ewrt             = _mm_mul_ps(r20,ewtabscale);
2034             ewitab           = _mm_cvttps_epi32(ewrt);
2035             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2036             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2037                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2038                                          &ewtabF,&ewtabFn);
2039             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2040             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
2041
2042             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
2043
2044             fscal            = felec;
2045
2046             fscal            = _mm_and_ps(fscal,cutoff_mask);
2047
2048             /* Calculate temporary vectorial force */
2049             tx               = _mm_mul_ps(fscal,dx20);
2050             ty               = _mm_mul_ps(fscal,dy20);
2051             tz               = _mm_mul_ps(fscal,dz20);
2052
2053             /* Update vectorial force */
2054             fix2             = _mm_add_ps(fix2,tx);
2055             fiy2             = _mm_add_ps(fiy2,ty);
2056             fiz2             = _mm_add_ps(fiz2,tz);
2057
2058             fjx0             = _mm_add_ps(fjx0,tx);
2059             fjy0             = _mm_add_ps(fjy0,ty);
2060             fjz0             = _mm_add_ps(fjz0,tz);
2061
2062             }
2063
2064             /**************************
2065              * CALCULATE INTERACTIONS *
2066              **************************/
2067
2068             if (gmx_mm_any_lt(rsq21,rcutoff2))
2069             {
2070
2071             r21              = _mm_mul_ps(rsq21,rinv21);
2072
2073             /* EWALD ELECTROSTATICS */
2074
2075             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2076             ewrt             = _mm_mul_ps(r21,ewtabscale);
2077             ewitab           = _mm_cvttps_epi32(ewrt);
2078             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2079             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2080                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2081                                          &ewtabF,&ewtabFn);
2082             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2083             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2084
2085             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2086
2087             fscal            = felec;
2088
2089             fscal            = _mm_and_ps(fscal,cutoff_mask);
2090
2091             /* Calculate temporary vectorial force */
2092             tx               = _mm_mul_ps(fscal,dx21);
2093             ty               = _mm_mul_ps(fscal,dy21);
2094             tz               = _mm_mul_ps(fscal,dz21);
2095
2096             /* Update vectorial force */
2097             fix2             = _mm_add_ps(fix2,tx);
2098             fiy2             = _mm_add_ps(fiy2,ty);
2099             fiz2             = _mm_add_ps(fiz2,tz);
2100
2101             fjx1             = _mm_add_ps(fjx1,tx);
2102             fjy1             = _mm_add_ps(fjy1,ty);
2103             fjz1             = _mm_add_ps(fjz1,tz);
2104
2105             }
2106
2107             /**************************
2108              * CALCULATE INTERACTIONS *
2109              **************************/
2110
2111             if (gmx_mm_any_lt(rsq22,rcutoff2))
2112             {
2113
2114             r22              = _mm_mul_ps(rsq22,rinv22);
2115
2116             /* EWALD ELECTROSTATICS */
2117
2118             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2119             ewrt             = _mm_mul_ps(r22,ewtabscale);
2120             ewitab           = _mm_cvttps_epi32(ewrt);
2121             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2122             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2123                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2124                                          &ewtabF,&ewtabFn);
2125             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2126             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2127
2128             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2129
2130             fscal            = felec;
2131
2132             fscal            = _mm_and_ps(fscal,cutoff_mask);
2133
2134             /* Calculate temporary vectorial force */
2135             tx               = _mm_mul_ps(fscal,dx22);
2136             ty               = _mm_mul_ps(fscal,dy22);
2137             tz               = _mm_mul_ps(fscal,dz22);
2138
2139             /* Update vectorial force */
2140             fix2             = _mm_add_ps(fix2,tx);
2141             fiy2             = _mm_add_ps(fiy2,ty);
2142             fiz2             = _mm_add_ps(fiz2,tz);
2143
2144             fjx2             = _mm_add_ps(fjx2,tx);
2145             fjy2             = _mm_add_ps(fjy2,ty);
2146             fjz2             = _mm_add_ps(fjz2,tz);
2147
2148             }
2149
2150             fjptrA             = f+j_coord_offsetA;
2151             fjptrB             = f+j_coord_offsetB;
2152             fjptrC             = f+j_coord_offsetC;
2153             fjptrD             = f+j_coord_offsetD;
2154
2155             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2156                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2157
2158             /* Inner loop uses 374 flops */
2159         }
2160
2161         if(jidx<j_index_end)
2162         {
2163
2164             /* Get j neighbor index, and coordinate index */
2165             jnrlistA         = jjnr[jidx];
2166             jnrlistB         = jjnr[jidx+1];
2167             jnrlistC         = jjnr[jidx+2];
2168             jnrlistD         = jjnr[jidx+3];
2169             /* Sign of each element will be negative for non-real atoms.
2170              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2171              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2172              */
2173             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2174             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2175             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2176             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2177             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2178             j_coord_offsetA  = DIM*jnrA;
2179             j_coord_offsetB  = DIM*jnrB;
2180             j_coord_offsetC  = DIM*jnrC;
2181             j_coord_offsetD  = DIM*jnrD;
2182
2183             /* load j atom coordinates */
2184             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2185                                               x+j_coord_offsetC,x+j_coord_offsetD,
2186                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2187
2188             /* Calculate displacement vector */
2189             dx00             = _mm_sub_ps(ix0,jx0);
2190             dy00             = _mm_sub_ps(iy0,jy0);
2191             dz00             = _mm_sub_ps(iz0,jz0);
2192             dx01             = _mm_sub_ps(ix0,jx1);
2193             dy01             = _mm_sub_ps(iy0,jy1);
2194             dz01             = _mm_sub_ps(iz0,jz1);
2195             dx02             = _mm_sub_ps(ix0,jx2);
2196             dy02             = _mm_sub_ps(iy0,jy2);
2197             dz02             = _mm_sub_ps(iz0,jz2);
2198             dx10             = _mm_sub_ps(ix1,jx0);
2199             dy10             = _mm_sub_ps(iy1,jy0);
2200             dz10             = _mm_sub_ps(iz1,jz0);
2201             dx11             = _mm_sub_ps(ix1,jx1);
2202             dy11             = _mm_sub_ps(iy1,jy1);
2203             dz11             = _mm_sub_ps(iz1,jz1);
2204             dx12             = _mm_sub_ps(ix1,jx2);
2205             dy12             = _mm_sub_ps(iy1,jy2);
2206             dz12             = _mm_sub_ps(iz1,jz2);
2207             dx20             = _mm_sub_ps(ix2,jx0);
2208             dy20             = _mm_sub_ps(iy2,jy0);
2209             dz20             = _mm_sub_ps(iz2,jz0);
2210             dx21             = _mm_sub_ps(ix2,jx1);
2211             dy21             = _mm_sub_ps(iy2,jy1);
2212             dz21             = _mm_sub_ps(iz2,jz1);
2213             dx22             = _mm_sub_ps(ix2,jx2);
2214             dy22             = _mm_sub_ps(iy2,jy2);
2215             dz22             = _mm_sub_ps(iz2,jz2);
2216
2217             /* Calculate squared distance and things based on it */
2218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
2219             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
2220             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
2221             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
2222             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2223             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2224             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
2225             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2226             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2227
2228             rinv00           = gmx_mm_invsqrt_ps(rsq00);
2229             rinv01           = gmx_mm_invsqrt_ps(rsq01);
2230             rinv02           = gmx_mm_invsqrt_ps(rsq02);
2231             rinv10           = gmx_mm_invsqrt_ps(rsq10);
2232             rinv11           = gmx_mm_invsqrt_ps(rsq11);
2233             rinv12           = gmx_mm_invsqrt_ps(rsq12);
2234             rinv20           = gmx_mm_invsqrt_ps(rsq20);
2235             rinv21           = gmx_mm_invsqrt_ps(rsq21);
2236             rinv22           = gmx_mm_invsqrt_ps(rsq22);
2237
2238             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
2239             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
2240             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
2241             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
2242             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
2243             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
2244             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
2245             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
2246             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
2247
2248             fjx0             = _mm_setzero_ps();
2249             fjy0             = _mm_setzero_ps();
2250             fjz0             = _mm_setzero_ps();
2251             fjx1             = _mm_setzero_ps();
2252             fjy1             = _mm_setzero_ps();
2253             fjz1             = _mm_setzero_ps();
2254             fjx2             = _mm_setzero_ps();
2255             fjy2             = _mm_setzero_ps();
2256             fjz2             = _mm_setzero_ps();
2257
2258             /**************************
2259              * CALCULATE INTERACTIONS *
2260              **************************/
2261
2262             if (gmx_mm_any_lt(rsq00,rcutoff2))
2263             {
2264
2265             r00              = _mm_mul_ps(rsq00,rinv00);
2266             r00              = _mm_andnot_ps(dummy_mask,r00);
2267
2268             /* EWALD ELECTROSTATICS */
2269
2270             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2271             ewrt             = _mm_mul_ps(r00,ewtabscale);
2272             ewitab           = _mm_cvttps_epi32(ewrt);
2273             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2274             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2275                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2276                                          &ewtabF,&ewtabFn);
2277             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2278             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
2279
2280             /* Analytical LJ-PME */
2281             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2282             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
2283             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
2284             exponent         = gmx_simd_exp_r(ewcljrsq);
2285             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2286             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
2287             /* f6A = 6 * C6grid * (1 - poly) */
2288             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
2289             /* f6B = C6grid * exponent * beta^6 */
2290             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
2291             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2292             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);
2293
2294             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
2295
2296             fscal            = _mm_add_ps(felec,fvdw);
2297
2298             fscal            = _mm_and_ps(fscal,cutoff_mask);
2299
2300             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2301
2302             /* Calculate temporary vectorial force */
2303             tx               = _mm_mul_ps(fscal,dx00);
2304             ty               = _mm_mul_ps(fscal,dy00);
2305             tz               = _mm_mul_ps(fscal,dz00);
2306
2307             /* Update vectorial force */
2308             fix0             = _mm_add_ps(fix0,tx);
2309             fiy0             = _mm_add_ps(fiy0,ty);
2310             fiz0             = _mm_add_ps(fiz0,tz);
2311
2312             fjx0             = _mm_add_ps(fjx0,tx);
2313             fjy0             = _mm_add_ps(fjy0,ty);
2314             fjz0             = _mm_add_ps(fjz0,tz);
2315
2316             }
2317
2318             /**************************
2319              * CALCULATE INTERACTIONS *
2320              **************************/
2321
2322             if (gmx_mm_any_lt(rsq01,rcutoff2))
2323             {
2324
2325             r01              = _mm_mul_ps(rsq01,rinv01);
2326             r01              = _mm_andnot_ps(dummy_mask,r01);
2327
2328             /* EWALD ELECTROSTATICS */
2329
2330             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2331             ewrt             = _mm_mul_ps(r01,ewtabscale);
2332             ewitab           = _mm_cvttps_epi32(ewrt);
2333             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2334             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2335                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2336                                          &ewtabF,&ewtabFn);
2337             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2338             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
2339
2340             cutoff_mask      = _mm_cmplt_ps(rsq01,rcutoff2);
2341
2342             fscal            = felec;
2343
2344             fscal            = _mm_and_ps(fscal,cutoff_mask);
2345
2346             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2347
2348             /* Calculate temporary vectorial force */
2349             tx               = _mm_mul_ps(fscal,dx01);
2350             ty               = _mm_mul_ps(fscal,dy01);
2351             tz               = _mm_mul_ps(fscal,dz01);
2352
2353             /* Update vectorial force */
2354             fix0             = _mm_add_ps(fix0,tx);
2355             fiy0             = _mm_add_ps(fiy0,ty);
2356             fiz0             = _mm_add_ps(fiz0,tz);
2357
2358             fjx1             = _mm_add_ps(fjx1,tx);
2359             fjy1             = _mm_add_ps(fjy1,ty);
2360             fjz1             = _mm_add_ps(fjz1,tz);
2361
2362             }
2363
2364             /**************************
2365              * CALCULATE INTERACTIONS *
2366              **************************/
2367
2368             if (gmx_mm_any_lt(rsq02,rcutoff2))
2369             {
2370
2371             r02              = _mm_mul_ps(rsq02,rinv02);
2372             r02              = _mm_andnot_ps(dummy_mask,r02);
2373
2374             /* EWALD ELECTROSTATICS */
2375
2376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2377             ewrt             = _mm_mul_ps(r02,ewtabscale);
2378             ewitab           = _mm_cvttps_epi32(ewrt);
2379             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2380             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2381                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2382                                          &ewtabF,&ewtabFn);
2383             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2384             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
2385
2386             cutoff_mask      = _mm_cmplt_ps(rsq02,rcutoff2);
2387
2388             fscal            = felec;
2389
2390             fscal            = _mm_and_ps(fscal,cutoff_mask);
2391
2392             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2393
2394             /* Calculate temporary vectorial force */
2395             tx               = _mm_mul_ps(fscal,dx02);
2396             ty               = _mm_mul_ps(fscal,dy02);
2397             tz               = _mm_mul_ps(fscal,dz02);
2398
2399             /* Update vectorial force */
2400             fix0             = _mm_add_ps(fix0,tx);
2401             fiy0             = _mm_add_ps(fiy0,ty);
2402             fiz0             = _mm_add_ps(fiz0,tz);
2403
2404             fjx2             = _mm_add_ps(fjx2,tx);
2405             fjy2             = _mm_add_ps(fjy2,ty);
2406             fjz2             = _mm_add_ps(fjz2,tz);
2407
2408             }
2409
2410             /**************************
2411              * CALCULATE INTERACTIONS *
2412              **************************/
2413
2414             if (gmx_mm_any_lt(rsq10,rcutoff2))
2415             {
2416
2417             r10              = _mm_mul_ps(rsq10,rinv10);
2418             r10              = _mm_andnot_ps(dummy_mask,r10);
2419
2420             /* EWALD ELECTROSTATICS */
2421
2422             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2423             ewrt             = _mm_mul_ps(r10,ewtabscale);
2424             ewitab           = _mm_cvttps_epi32(ewrt);
2425             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2426             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2427                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2428                                          &ewtabF,&ewtabFn);
2429             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2430             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
2431
2432             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
2433
2434             fscal            = felec;
2435
2436             fscal            = _mm_and_ps(fscal,cutoff_mask);
2437
2438             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2439
2440             /* Calculate temporary vectorial force */
2441             tx               = _mm_mul_ps(fscal,dx10);
2442             ty               = _mm_mul_ps(fscal,dy10);
2443             tz               = _mm_mul_ps(fscal,dz10);
2444
2445             /* Update vectorial force */
2446             fix1             = _mm_add_ps(fix1,tx);
2447             fiy1             = _mm_add_ps(fiy1,ty);
2448             fiz1             = _mm_add_ps(fiz1,tz);
2449
2450             fjx0             = _mm_add_ps(fjx0,tx);
2451             fjy0             = _mm_add_ps(fjy0,ty);
2452             fjz0             = _mm_add_ps(fjz0,tz);
2453
2454             }
2455
2456             /**************************
2457              * CALCULATE INTERACTIONS *
2458              **************************/
2459
2460             if (gmx_mm_any_lt(rsq11,rcutoff2))
2461             {
2462
2463             r11              = _mm_mul_ps(rsq11,rinv11);
2464             r11              = _mm_andnot_ps(dummy_mask,r11);
2465
2466             /* EWALD ELECTROSTATICS */
2467
2468             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2469             ewrt             = _mm_mul_ps(r11,ewtabscale);
2470             ewitab           = _mm_cvttps_epi32(ewrt);
2471             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2472             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2473                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2474                                          &ewtabF,&ewtabFn);
2475             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2476             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
2477
2478             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
2479
2480             fscal            = felec;
2481
2482             fscal            = _mm_and_ps(fscal,cutoff_mask);
2483
2484             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2485
2486             /* Calculate temporary vectorial force */
2487             tx               = _mm_mul_ps(fscal,dx11);
2488             ty               = _mm_mul_ps(fscal,dy11);
2489             tz               = _mm_mul_ps(fscal,dz11);
2490
2491             /* Update vectorial force */
2492             fix1             = _mm_add_ps(fix1,tx);
2493             fiy1             = _mm_add_ps(fiy1,ty);
2494             fiz1             = _mm_add_ps(fiz1,tz);
2495
2496             fjx1             = _mm_add_ps(fjx1,tx);
2497             fjy1             = _mm_add_ps(fjy1,ty);
2498             fjz1             = _mm_add_ps(fjz1,tz);
2499
2500             }
2501
2502             /**************************
2503              * CALCULATE INTERACTIONS *
2504              **************************/
2505
2506             if (gmx_mm_any_lt(rsq12,rcutoff2))
2507             {
2508
2509             r12              = _mm_mul_ps(rsq12,rinv12);
2510             r12              = _mm_andnot_ps(dummy_mask,r12);
2511
2512             /* EWALD ELECTROSTATICS */
2513
2514             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2515             ewrt             = _mm_mul_ps(r12,ewtabscale);
2516             ewitab           = _mm_cvttps_epi32(ewrt);
2517             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2518             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2519                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2520                                          &ewtabF,&ewtabFn);
2521             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2522             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
2523
2524             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
2525
2526             fscal            = felec;
2527
2528             fscal            = _mm_and_ps(fscal,cutoff_mask);
2529
2530             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2531
2532             /* Calculate temporary vectorial force */
2533             tx               = _mm_mul_ps(fscal,dx12);
2534             ty               = _mm_mul_ps(fscal,dy12);
2535             tz               = _mm_mul_ps(fscal,dz12);
2536
2537             /* Update vectorial force */
2538             fix1             = _mm_add_ps(fix1,tx);
2539             fiy1             = _mm_add_ps(fiy1,ty);
2540             fiz1             = _mm_add_ps(fiz1,tz);
2541
2542             fjx2             = _mm_add_ps(fjx2,tx);
2543             fjy2             = _mm_add_ps(fjy2,ty);
2544             fjz2             = _mm_add_ps(fjz2,tz);
2545
2546             }
2547
2548             /**************************
2549              * CALCULATE INTERACTIONS *
2550              **************************/
2551
2552             if (gmx_mm_any_lt(rsq20,rcutoff2))
2553             {
2554
2555             r20              = _mm_mul_ps(rsq20,rinv20);
2556             r20              = _mm_andnot_ps(dummy_mask,r20);
2557
2558             /* EWALD ELECTROSTATICS */
2559
2560             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2561             ewrt             = _mm_mul_ps(r20,ewtabscale);
2562             ewitab           = _mm_cvttps_epi32(ewrt);
2563             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2564             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2565                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2566                                          &ewtabF,&ewtabFn);
2567             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2568             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
2569
2570             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
2571
2572             fscal            = felec;
2573
2574             fscal            = _mm_and_ps(fscal,cutoff_mask);
2575
2576             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2577
2578             /* Calculate temporary vectorial force */
2579             tx               = _mm_mul_ps(fscal,dx20);
2580             ty               = _mm_mul_ps(fscal,dy20);
2581             tz               = _mm_mul_ps(fscal,dz20);
2582
2583             /* Update vectorial force */
2584             fix2             = _mm_add_ps(fix2,tx);
2585             fiy2             = _mm_add_ps(fiy2,ty);
2586             fiz2             = _mm_add_ps(fiz2,tz);
2587
2588             fjx0             = _mm_add_ps(fjx0,tx);
2589             fjy0             = _mm_add_ps(fjy0,ty);
2590             fjz0             = _mm_add_ps(fjz0,tz);
2591
2592             }
2593
2594             /**************************
2595              * CALCULATE INTERACTIONS *
2596              **************************/
2597
2598             if (gmx_mm_any_lt(rsq21,rcutoff2))
2599             {
2600
2601             r21              = _mm_mul_ps(rsq21,rinv21);
2602             r21              = _mm_andnot_ps(dummy_mask,r21);
2603
2604             /* EWALD ELECTROSTATICS */
2605
2606             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2607             ewrt             = _mm_mul_ps(r21,ewtabscale);
2608             ewitab           = _mm_cvttps_epi32(ewrt);
2609             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2610             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2611                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2612                                          &ewtabF,&ewtabFn);
2613             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2614             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2615
2616             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2617
2618             fscal            = felec;
2619
2620             fscal            = _mm_and_ps(fscal,cutoff_mask);
2621
2622             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2623
2624             /* Calculate temporary vectorial force */
2625             tx               = _mm_mul_ps(fscal,dx21);
2626             ty               = _mm_mul_ps(fscal,dy21);
2627             tz               = _mm_mul_ps(fscal,dz21);
2628
2629             /* Update vectorial force */
2630             fix2             = _mm_add_ps(fix2,tx);
2631             fiy2             = _mm_add_ps(fiy2,ty);
2632             fiz2             = _mm_add_ps(fiz2,tz);
2633
2634             fjx1             = _mm_add_ps(fjx1,tx);
2635             fjy1             = _mm_add_ps(fjy1,ty);
2636             fjz1             = _mm_add_ps(fjz1,tz);
2637
2638             }
2639
2640             /**************************
2641              * CALCULATE INTERACTIONS *
2642              **************************/
2643
2644             if (gmx_mm_any_lt(rsq22,rcutoff2))
2645             {
2646
2647             r22              = _mm_mul_ps(rsq22,rinv22);
2648             r22              = _mm_andnot_ps(dummy_mask,r22);
2649
2650             /* EWALD ELECTROSTATICS */
2651
2652             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2653             ewrt             = _mm_mul_ps(r22,ewtabscale);
2654             ewitab           = _mm_cvttps_epi32(ewrt);
2655             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2656             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2657                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2658                                          &ewtabF,&ewtabFn);
2659             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2660             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2661
2662             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2663
2664             fscal            = felec;
2665
2666             fscal            = _mm_and_ps(fscal,cutoff_mask);
2667
2668             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2669
2670             /* Calculate temporary vectorial force */
2671             tx               = _mm_mul_ps(fscal,dx22);
2672             ty               = _mm_mul_ps(fscal,dy22);
2673             tz               = _mm_mul_ps(fscal,dz22);
2674
2675             /* Update vectorial force */
2676             fix2             = _mm_add_ps(fix2,tx);
2677             fiy2             = _mm_add_ps(fiy2,ty);
2678             fiz2             = _mm_add_ps(fiz2,tz);
2679
2680             fjx2             = _mm_add_ps(fjx2,tx);
2681             fjy2             = _mm_add_ps(fjy2,ty);
2682             fjz2             = _mm_add_ps(fjz2,tz);
2683
2684             }
2685
2686             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2687             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2688             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2689             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2690
2691             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2692                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2693
2694             /* Inner loop uses 383 flops */
2695         }
2696
2697         /* End of innermost loop */
2698
2699         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2700                                               f+i_coord_offset,fshift+i_shift_offset);
2701
2702         /* Increment number of inner iterations */
2703         inneriter                  += j_index_end - j_index_start;
2704
2705         /* Outer loop uses 18 flops */
2706     }
2707
2708     /* Increment number of outer iterations */
2709     outeriter        += nri;
2710
2711     /* Update outer/inner flops */
2712
2713     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*383);
2714 }