src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 avx_128_fma_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_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_VF_avx_128_fma_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 AVX_128, 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           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  96     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  97     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  98     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  99     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 100     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 101     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 102     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 105     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 108     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 109     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 110     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 111     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 112     real             *charge;
 113     int              nvdwtype;
 114     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 115     int              *vdwtype;
 116     real             *vdwparam;
 117     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 118     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 119     __m128i          ewitab;
 120     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 121     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 122     real             *ewtab;
 123     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 124     real             rswitch_scalar,d_scalar;
 125     __m128           dummy_mask,cutoff_mask;
 126     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 127     __m128           one     = _mm_set1_ps(1.0);
 128     __m128           two     = _mm_set1_ps(2.0);
 129     x                = xx[0];
 130     f                = ff[0];
 131
 132     nri              = nlist->nri;
 133     iinr             = nlist->iinr;
 134     jindex           = nlist->jindex;
 135     jjnr             = nlist->jjnr;
 136     shiftidx         = nlist->shift;
 137     gid              = nlist->gid;
 138     shiftvec         = fr->shift_vec[0];
 139     fshift           = fr->fshift[0];
 140     facel            = _mm_set1_ps(fr->epsfac);
 141     charge           = mdatoms->chargeA;
 142     nvdwtype         = fr->ntype;
 143     vdwparam         = fr->nbfp;
 144     vdwtype          = mdatoms->typeA;
 145
 146     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 147     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 148     beta2            = _mm_mul_ps(beta,beta);
 149     beta3            = _mm_mul_ps(beta,beta2);
 150     ewtab            = fr->ic->tabq_coul_FDV0;
 151     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 152     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 153
 154     /* Setup water-specific parameters */
 155     inr              = nlist->iinr[0];
 156     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 157     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 158     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 159     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161     jq1              = _mm_set1_ps(charge[inr+1]);
 162     jq2              = _mm_set1_ps(charge[inr+2]);
 163     jq3              = _mm_set1_ps(charge[inr+3]);
 164     vdwjidx0A        = 2*vdwtype[inr+0];
 165     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 166     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 167     qq11             = _mm_mul_ps(iq1,jq1);
 168     qq12             = _mm_mul_ps(iq1,jq2);
 169     qq13             = _mm_mul_ps(iq1,jq3);
 170     qq21             = _mm_mul_ps(iq2,jq1);
 171     qq22             = _mm_mul_ps(iq2,jq2);
 172     qq23             = _mm_mul_ps(iq2,jq3);
 173     qq31             = _mm_mul_ps(iq3,jq1);
 174     qq32             = _mm_mul_ps(iq3,jq2);
 175     qq33             = _mm_mul_ps(iq3,jq3);
 176
 177     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 178     rcutoff_scalar   = fr->rcoulomb;
 179     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 180     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 181
 182     rswitch_scalar   = fr->rcoulomb_switch;
 183     rswitch          = _mm_set1_ps(rswitch_scalar);
 184     /* Setup switch parameters */
 185     d_scalar         = rcutoff_scalar-rswitch_scalar;
 186     d                = _mm_set1_ps(d_scalar);
 187     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 188     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 189     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 190     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 191     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 192     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 193
 194     /* Avoid stupid compiler warnings */
 195     jnrA = jnrB = jnrC = jnrD = 0;
 196     j_coord_offsetA = 0;
 197     j_coord_offsetB = 0;
 198     j_coord_offsetC = 0;
 199     j_coord_offsetD = 0;
 200
 201     outeriter        = 0;
 202     inneriter        = 0;
 203
 204     for(iidx=0;iidx<4*DIM;iidx++)
 205     {
 206         scratch[iidx] = 0.0;
 207     }
 208
 209     /* Start outer loop over neighborlists */
 210     for(iidx=0; iidx<nri; iidx++)
 211     {
 212         /* Load shift vector for this list */
 213         i_shift_offset   = DIM*shiftidx[iidx];
 214
 215         /* Load limits for loop over neighbors */
 216         j_index_start    = jindex[iidx];
 217         j_index_end      = jindex[iidx+1];
 218
 219         /* Get outer coordinate index */
 220         inr              = iinr[iidx];
 221         i_coord_offset   = DIM*inr;
 222
 223         /* Load i particle coords and add shift vector */
 224         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 225                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 226
 227         fix0             = _mm_setzero_ps();
 228         fiy0             = _mm_setzero_ps();
 229         fiz0             = _mm_setzero_ps();
 230         fix1             = _mm_setzero_ps();
 231         fiy1             = _mm_setzero_ps();
 232         fiz1             = _mm_setzero_ps();
 233         fix2             = _mm_setzero_ps();
 234         fiy2             = _mm_setzero_ps();
 235         fiz2             = _mm_setzero_ps();
 236         fix3             = _mm_setzero_ps();
 237         fiy3             = _mm_setzero_ps();
 238         fiz3             = _mm_setzero_ps();
 239
 240         /* Reset potential sums */
 241         velecsum         = _mm_setzero_ps();
 242         vvdwsum          = _mm_setzero_ps();
 243
 244         /* Start inner kernel loop */
 245         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 246         {
 247
 248             /* Get j neighbor index, and coordinate index */
 249             jnrA             = jjnr[jidx];
 250             jnrB             = jjnr[jidx+1];
 251             jnrC             = jjnr[jidx+2];
 252             jnrD             = jjnr[jidx+3];
 253             j_coord_offsetA  = DIM*jnrA;
 254             j_coord_offsetB  = DIM*jnrB;
 255             j_coord_offsetC  = DIM*jnrC;
 256             j_coord_offsetD  = DIM*jnrD;
 257
 258             /* load j atom coordinates */
 259             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 260                                               x+j_coord_offsetC,x+j_coord_offsetD,
 261                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 262                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 263
 264             /* Calculate displacement vector */
 265             dx00             = _mm_sub_ps(ix0,jx0);
 266             dy00             = _mm_sub_ps(iy0,jy0);
 267             dz00             = _mm_sub_ps(iz0,jz0);
 268             dx11             = _mm_sub_ps(ix1,jx1);
 269             dy11             = _mm_sub_ps(iy1,jy1);
 270             dz11             = _mm_sub_ps(iz1,jz1);
 271             dx12             = _mm_sub_ps(ix1,jx2);
 272             dy12             = _mm_sub_ps(iy1,jy2);
 273             dz12             = _mm_sub_ps(iz1,jz2);
 274             dx13             = _mm_sub_ps(ix1,jx3);
 275             dy13             = _mm_sub_ps(iy1,jy3);
 276             dz13             = _mm_sub_ps(iz1,jz3);
 277             dx21             = _mm_sub_ps(ix2,jx1);
 278             dy21             = _mm_sub_ps(iy2,jy1);
 279             dz21             = _mm_sub_ps(iz2,jz1);
 280             dx22             = _mm_sub_ps(ix2,jx2);
 281             dy22             = _mm_sub_ps(iy2,jy2);
 282             dz22             = _mm_sub_ps(iz2,jz2);
 283             dx23             = _mm_sub_ps(ix2,jx3);
 284             dy23             = _mm_sub_ps(iy2,jy3);
 285             dz23             = _mm_sub_ps(iz2,jz3);
 286             dx31             = _mm_sub_ps(ix3,jx1);
 287             dy31             = _mm_sub_ps(iy3,jy1);
 288             dz31             = _mm_sub_ps(iz3,jz1);
 289             dx32             = _mm_sub_ps(ix3,jx2);
 290             dy32             = _mm_sub_ps(iy3,jy2);
 291             dz32             = _mm_sub_ps(iz3,jz2);
 292             dx33             = _mm_sub_ps(ix3,jx3);
 293             dy33             = _mm_sub_ps(iy3,jy3);
 294             dz33             = _mm_sub_ps(iz3,jz3);
 295
 296             /* Calculate squared distance and things based on it */
 297             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 298             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 299             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 300             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 301             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 302             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 303             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 304             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 305             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 306             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 307
 308             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 309             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 310             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 311             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 312             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 313             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 314             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 315             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 316             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 317             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 318
 319             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 320             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 321             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 322             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 323             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 324             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 325             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 326             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 327             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 328             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 329
 330             fjx0             = _mm_setzero_ps();
 331             fjy0             = _mm_setzero_ps();
 332             fjz0             = _mm_setzero_ps();
 333             fjx1             = _mm_setzero_ps();
 334             fjy1             = _mm_setzero_ps();
 335             fjz1             = _mm_setzero_ps();
 336             fjx2             = _mm_setzero_ps();
 337             fjy2             = _mm_setzero_ps();
 338             fjz2             = _mm_setzero_ps();
 339             fjx3             = _mm_setzero_ps();
 340             fjy3             = _mm_setzero_ps();
 341             fjz3             = _mm_setzero_ps();
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             if (gmx_mm_any_lt(rsq00,rcutoff2))
 348             {
 349
 350             r00              = _mm_mul_ps(rsq00,rinv00);
 351
 352             /* LENNARD-JONES DISPERSION/REPULSION */
 353
 354             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 355             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 356             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 357             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 358             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 359
 360             d                = _mm_sub_ps(r00,rswitch);
 361             d                = _mm_max_ps(d,_mm_setzero_ps());
 362             d2               = _mm_mul_ps(d,d);
 363             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 364
 365             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 366
 367             /* Evaluate switch function */
 368             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 369             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 370             vvdw             = _mm_mul_ps(vvdw,sw);
 371             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 375             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 376
 377             fscal            = fvdw;
 378
 379             fscal            = _mm_and_ps(fscal,cutoff_mask);
 380
 381              /* Update vectorial force */
 382             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 383             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 384             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 385
 386             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 387             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 388             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 389
 390             }
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             if (gmx_mm_any_lt(rsq11,rcutoff2))
 397             {
 398
 399             r11              = _mm_mul_ps(rsq11,rinv11);
 400
 401             /* EWALD ELECTROSTATICS */
 402
 403             /* Analytical PME correction */
 404             zeta2            = _mm_mul_ps(beta2,rsq11);
 405             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 406             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 407             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 408             felec            = _mm_mul_ps(qq11,felec);
 409             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 410             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 411             velec            = _mm_mul_ps(qq11,velec);
 412
 413             d                = _mm_sub_ps(r11,rswitch);
 414             d                = _mm_max_ps(d,_mm_setzero_ps());
 415             d2               = _mm_mul_ps(d,d);
 416             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 417
 418             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 419
 420             /* Evaluate switch function */
 421             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 422             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
 423             velec            = _mm_mul_ps(velec,sw);
 424             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velec            = _mm_and_ps(velec,cutoff_mask);
 428             velecsum         = _mm_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             fscal            = _mm_and_ps(fscal,cutoff_mask);
 433
 434              /* Update vectorial force */
 435             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 436             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 437             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 438
 439             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 440             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 441             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 442
 443             }
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             if (gmx_mm_any_lt(rsq12,rcutoff2))
 450             {
 451
 452             r12              = _mm_mul_ps(rsq12,rinv12);
 453
 454             /* EWALD ELECTROSTATICS */
 455
 456             /* Analytical PME correction */
 457             zeta2            = _mm_mul_ps(beta2,rsq12);
 458             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 459             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 460             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 461             felec            = _mm_mul_ps(qq12,felec);
 462             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 463             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 464             velec            = _mm_mul_ps(qq12,velec);
 465
 466             d                = _mm_sub_ps(r12,rswitch);
 467             d                = _mm_max_ps(d,_mm_setzero_ps());
 468             d2               = _mm_mul_ps(d,d);
 469             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 470
 471             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 472
 473             /* Evaluate switch function */
 474             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 475             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
 476             velec            = _mm_mul_ps(velec,sw);
 477             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velec            = _mm_and_ps(velec,cutoff_mask);
 481             velecsum         = _mm_add_ps(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485             fscal            = _mm_and_ps(fscal,cutoff_mask);
 486
 487              /* Update vectorial force */
 488             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 489             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 490             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 491
 492             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 493             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 494             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 495
 496             }
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             if (gmx_mm_any_lt(rsq13,rcutoff2))
 503             {
 504
 505             r13              = _mm_mul_ps(rsq13,rinv13);
 506
 507             /* EWALD ELECTROSTATICS */
 508
 509             /* Analytical PME correction */
 510             zeta2            = _mm_mul_ps(beta2,rsq13);
 511             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
 512             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 513             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 514             felec            = _mm_mul_ps(qq13,felec);
 515             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 516             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
 517             velec            = _mm_mul_ps(qq13,velec);
 518
 519             d                = _mm_sub_ps(r13,rswitch);
 520             d                = _mm_max_ps(d,_mm_setzero_ps());
 521             d2               = _mm_mul_ps(d,d);
 522             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 523
 524             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 525
 526             /* Evaluate switch function */
 527             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 528             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
 529             velec            = _mm_mul_ps(velec,sw);
 530             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm_and_ps(velec,cutoff_mask);
 534             velecsum         = _mm_add_ps(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm_and_ps(fscal,cutoff_mask);
 539
 540              /* Update vectorial force */
 541             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 542             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 543             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 544
 545             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 546             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 547             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 548
 549             }
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             if (gmx_mm_any_lt(rsq21,rcutoff2))
 556             {
 557
 558             r21              = _mm_mul_ps(rsq21,rinv21);
 559
 560             /* EWALD ELECTROSTATICS */
 561
 562             /* Analytical PME correction */
 563             zeta2            = _mm_mul_ps(beta2,rsq21);
 564             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 565             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 566             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 567             felec            = _mm_mul_ps(qq21,felec);
 568             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 569             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
 570             velec            = _mm_mul_ps(qq21,velec);
 571
 572             d                = _mm_sub_ps(r21,rswitch);
 573             d                = _mm_max_ps(d,_mm_setzero_ps());
 574             d2               = _mm_mul_ps(d,d);
 575             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 576
 577             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 578
 579             /* Evaluate switch function */
 580             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 581             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
 582             velec            = _mm_mul_ps(velec,sw);
 583             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velec            = _mm_and_ps(velec,cutoff_mask);
 587             velecsum         = _mm_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_and_ps(fscal,cutoff_mask);
 592
 593              /* Update vectorial force */
 594             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 595             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 596             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 597
 598             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 599             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 600             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 601
 602             }
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_mm_any_lt(rsq22,rcutoff2))
 609             {
 610
 611             r22              = _mm_mul_ps(rsq22,rinv22);
 612
 613             /* EWALD ELECTROSTATICS */
 614
 615             /* Analytical PME correction */
 616             zeta2            = _mm_mul_ps(beta2,rsq22);
 617             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
 618             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 619             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 620             felec            = _mm_mul_ps(qq22,felec);
 621             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 622             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
 623             velec            = _mm_mul_ps(qq22,velec);
 624
 625             d                = _mm_sub_ps(r22,rswitch);
 626             d                = _mm_max_ps(d,_mm_setzero_ps());
 627             d2               = _mm_mul_ps(d,d);
 628             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 629
 630             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 631
 632             /* Evaluate switch function */
 633             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 634             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
 635             velec            = _mm_mul_ps(velec,sw);
 636             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
 637
 638             /* Update potential sum for this i atom from the interaction with this j atom. */
 639             velec            = _mm_and_ps(velec,cutoff_mask);
 640             velecsum         = _mm_add_ps(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm_and_ps(fscal,cutoff_mask);
 645
 646              /* Update vectorial force */
 647             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 648             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 649             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 650
 651             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 652             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 653             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 654
 655             }
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             if (gmx_mm_any_lt(rsq23,rcutoff2))
 662             {
 663
 664             r23              = _mm_mul_ps(rsq23,rinv23);
 665
 666             /* EWALD ELECTROSTATICS */
 667
 668             /* Analytical PME correction */
 669             zeta2            = _mm_mul_ps(beta2,rsq23);
 670             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
 671             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 672             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 673             felec            = _mm_mul_ps(qq23,felec);
 674             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 675             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
 676             velec            = _mm_mul_ps(qq23,velec);
 677
 678             d                = _mm_sub_ps(r23,rswitch);
 679             d                = _mm_max_ps(d,_mm_setzero_ps());
 680             d2               = _mm_mul_ps(d,d);
 681             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 682
 683             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 684
 685             /* Evaluate switch function */
 686             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 687             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
 688             velec            = _mm_mul_ps(velec,sw);
 689             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
 690
 691             /* Update potential sum for this i atom from the interaction with this j atom. */
 692             velec            = _mm_and_ps(velec,cutoff_mask);
 693             velecsum         = _mm_add_ps(velecsum,velec);
 694
 695             fscal            = felec;
 696
 697             fscal            = _mm_and_ps(fscal,cutoff_mask);
 698
 699              /* Update vectorial force */
 700             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 701             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 702             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 703
 704             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 705             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 706             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 707
 708             }
 709
 710             /**************************
 711              * CALCULATE INTERACTIONS *
 712              **************************/
 713
 714             if (gmx_mm_any_lt(rsq31,rcutoff2))
 715             {
 716
 717             r31              = _mm_mul_ps(rsq31,rinv31);
 718
 719             /* EWALD ELECTROSTATICS */
 720
 721             /* Analytical PME correction */
 722             zeta2            = _mm_mul_ps(beta2,rsq31);
 723             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
 724             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 725             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 726             felec            = _mm_mul_ps(qq31,felec);
 727             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 728             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
 729             velec            = _mm_mul_ps(qq31,velec);
 730
 731             d                = _mm_sub_ps(r31,rswitch);
 732             d                = _mm_max_ps(d,_mm_setzero_ps());
 733             d2               = _mm_mul_ps(d,d);
 734             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 735
 736             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 737
 738             /* Evaluate switch function */
 739             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 740             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
 741             velec            = _mm_mul_ps(velec,sw);
 742             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
 743
 744             /* Update potential sum for this i atom from the interaction with this j atom. */
 745             velec            = _mm_and_ps(velec,cutoff_mask);
 746             velecsum         = _mm_add_ps(velecsum,velec);
 747
 748             fscal            = felec;
 749
 750             fscal            = _mm_and_ps(fscal,cutoff_mask);
 751
 752              /* Update vectorial force */
 753             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 754             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 755             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 756
 757             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 758             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 759             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 760
 761             }
 762
 763             /**************************
 764              * CALCULATE INTERACTIONS *
 765              **************************/
 766
 767             if (gmx_mm_any_lt(rsq32,rcutoff2))
 768             {
 769
 770             r32              = _mm_mul_ps(rsq32,rinv32);
 771
 772             /* EWALD ELECTROSTATICS */
 773
 774             /* Analytical PME correction */
 775             zeta2            = _mm_mul_ps(beta2,rsq32);
 776             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
 777             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 778             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 779             felec            = _mm_mul_ps(qq32,felec);
 780             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 781             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
 782             velec            = _mm_mul_ps(qq32,velec);
 783
 784             d                = _mm_sub_ps(r32,rswitch);
 785             d                = _mm_max_ps(d,_mm_setzero_ps());
 786             d2               = _mm_mul_ps(d,d);
 787             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 788
 789             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 790
 791             /* Evaluate switch function */
 792             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 793             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
 794             velec            = _mm_mul_ps(velec,sw);
 795             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
 796
 797             /* Update potential sum for this i atom from the interaction with this j atom. */
 798             velec            = _mm_and_ps(velec,cutoff_mask);
 799             velecsum         = _mm_add_ps(velecsum,velec);
 800
 801             fscal            = felec;
 802
 803             fscal            = _mm_and_ps(fscal,cutoff_mask);
 804
 805              /* Update vectorial force */
 806             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 807             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 808             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 809
 810             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 811             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 812             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 813
 814             }
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             if (gmx_mm_any_lt(rsq33,rcutoff2))
 821             {
 822
 823             r33              = _mm_mul_ps(rsq33,rinv33);
 824
 825             /* EWALD ELECTROSTATICS */
 826
 827             /* Analytical PME correction */
 828             zeta2            = _mm_mul_ps(beta2,rsq33);
 829             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
 830             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 831             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 832             felec            = _mm_mul_ps(qq33,felec);
 833             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 834             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
 835             velec            = _mm_mul_ps(qq33,velec);
 836
 837             d                = _mm_sub_ps(r33,rswitch);
 838             d                = _mm_max_ps(d,_mm_setzero_ps());
 839             d2               = _mm_mul_ps(d,d);
 840             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 841
 842             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 843
 844             /* Evaluate switch function */
 845             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 846             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
 847             velec            = _mm_mul_ps(velec,sw);
 848             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
 849
 850             /* Update potential sum for this i atom from the interaction with this j atom. */
 851             velec            = _mm_and_ps(velec,cutoff_mask);
 852             velecsum         = _mm_add_ps(velecsum,velec);
 853
 854             fscal            = felec;
 855
 856             fscal            = _mm_and_ps(fscal,cutoff_mask);
 857
 858              /* Update vectorial force */
 859             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 860             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 861             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 862
 863             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 864             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 865             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 866
 867             }
 868
 869             fjptrA             = f+j_coord_offsetA;
 870             fjptrB             = f+j_coord_offsetB;
 871             fjptrC             = f+j_coord_offsetC;
 872             fjptrD             = f+j_coord_offsetD;
 873
 874             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 875                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 876                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 877
 878             /* Inner loop uses 542 flops */
 879         }
 880
 881         if(jidx<j_index_end)
 882         {
 883
 884             /* Get j neighbor index, and coordinate index */
 885             jnrlistA         = jjnr[jidx];
 886             jnrlistB         = jjnr[jidx+1];
 887             jnrlistC         = jjnr[jidx+2];
 888             jnrlistD         = jjnr[jidx+3];
 889             /* Sign of each element will be negative for non-real atoms.
 890              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 891              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 892              */
 893             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 894             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 895             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 896             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 897             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 898             j_coord_offsetA  = DIM*jnrA;
 899             j_coord_offsetB  = DIM*jnrB;
 900             j_coord_offsetC  = DIM*jnrC;
 901             j_coord_offsetD  = DIM*jnrD;
 902
 903             /* load j atom coordinates */
 904             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 905                                               x+j_coord_offsetC,x+j_coord_offsetD,
 906                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 907                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 908
 909             /* Calculate displacement vector */
 910             dx00             = _mm_sub_ps(ix0,jx0);
 911             dy00             = _mm_sub_ps(iy0,jy0);
 912             dz00             = _mm_sub_ps(iz0,jz0);
 913             dx11             = _mm_sub_ps(ix1,jx1);
 914             dy11             = _mm_sub_ps(iy1,jy1);
 915             dz11             = _mm_sub_ps(iz1,jz1);
 916             dx12             = _mm_sub_ps(ix1,jx2);
 917             dy12             = _mm_sub_ps(iy1,jy2);
 918             dz12             = _mm_sub_ps(iz1,jz2);
 919             dx13             = _mm_sub_ps(ix1,jx3);
 920             dy13             = _mm_sub_ps(iy1,jy3);
 921             dz13             = _mm_sub_ps(iz1,jz3);
 922             dx21             = _mm_sub_ps(ix2,jx1);
 923             dy21             = _mm_sub_ps(iy2,jy1);
 924             dz21             = _mm_sub_ps(iz2,jz1);
 925             dx22             = _mm_sub_ps(ix2,jx2);
 926             dy22             = _mm_sub_ps(iy2,jy2);
 927             dz22             = _mm_sub_ps(iz2,jz2);
 928             dx23             = _mm_sub_ps(ix2,jx3);
 929             dy23             = _mm_sub_ps(iy2,jy3);
 930             dz23             = _mm_sub_ps(iz2,jz3);
 931             dx31             = _mm_sub_ps(ix3,jx1);
 932             dy31             = _mm_sub_ps(iy3,jy1);
 933             dz31             = _mm_sub_ps(iz3,jz1);
 934             dx32             = _mm_sub_ps(ix3,jx2);
 935             dy32             = _mm_sub_ps(iy3,jy2);
 936             dz32             = _mm_sub_ps(iz3,jz2);
 937             dx33             = _mm_sub_ps(ix3,jx3);
 938             dy33             = _mm_sub_ps(iy3,jy3);
 939             dz33             = _mm_sub_ps(iz3,jz3);
 940
 941             /* Calculate squared distance and things based on it */
 942             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 943             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 944             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 945             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 946             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 947             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 948             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 949             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 950             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 951             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 952
 953             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 954             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 955             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 956             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 957             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 958             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 959             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 960             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 961             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 962             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 963
 964             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 965             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 966             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 967             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 968             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 969             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 970             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 971             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 972             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 973             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 974
 975             fjx0             = _mm_setzero_ps();
 976             fjy0             = _mm_setzero_ps();
 977             fjz0             = _mm_setzero_ps();
 978             fjx1             = _mm_setzero_ps();
 979             fjy1             = _mm_setzero_ps();
 980             fjz1             = _mm_setzero_ps();
 981             fjx2             = _mm_setzero_ps();
 982             fjy2             = _mm_setzero_ps();
 983             fjz2             = _mm_setzero_ps();
 984             fjx3             = _mm_setzero_ps();
 985             fjy3             = _mm_setzero_ps();
 986             fjz3             = _mm_setzero_ps();
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             if (gmx_mm_any_lt(rsq00,rcutoff2))
 993             {
 994
 995             r00              = _mm_mul_ps(rsq00,rinv00);
 996             r00              = _mm_andnot_ps(dummy_mask,r00);
 997
 998             /* LENNARD-JONES DISPERSION/REPULSION */
 999
1000             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1001             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1002             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1003             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1004             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1005
1006             d                = _mm_sub_ps(r00,rswitch);
1007             d                = _mm_max_ps(d,_mm_setzero_ps());
1008             d2               = _mm_mul_ps(d,d);
1009             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1010
1011             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1012
1013             /* Evaluate switch function */
1014             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1015             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1016             vvdw             = _mm_mul_ps(vvdw,sw);
1017             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1018
1019             /* Update potential sum for this i atom from the interaction with this j atom. */
1020             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
1021             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
1022             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
1023
1024             fscal            = fvdw;
1025
1026             fscal            = _mm_and_ps(fscal,cutoff_mask);
1027
1028             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1029
1030              /* Update vectorial force */
1031             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1032             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1033             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1034
1035             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1036             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1037             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1038
1039             }
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             if (gmx_mm_any_lt(rsq11,rcutoff2))
1046             {
1047
1048             r11              = _mm_mul_ps(rsq11,rinv11);
1049             r11              = _mm_andnot_ps(dummy_mask,r11);
1050
1051             /* EWALD ELECTROSTATICS */
1052
1053             /* Analytical PME correction */
1054             zeta2            = _mm_mul_ps(beta2,rsq11);
1055             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1056             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1057             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1058             felec            = _mm_mul_ps(qq11,felec);
1059             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1060             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
1061             velec            = _mm_mul_ps(qq11,velec);
1062
1063             d                = _mm_sub_ps(r11,rswitch);
1064             d                = _mm_max_ps(d,_mm_setzero_ps());
1065             d2               = _mm_mul_ps(d,d);
1066             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1067
1068             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1069
1070             /* Evaluate switch function */
1071             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1072             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
1073             velec            = _mm_mul_ps(velec,sw);
1074             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1075
1076             /* Update potential sum for this i atom from the interaction with this j atom. */
1077             velec            = _mm_and_ps(velec,cutoff_mask);
1078             velec            = _mm_andnot_ps(dummy_mask,velec);
1079             velecsum         = _mm_add_ps(velecsum,velec);
1080
1081             fscal            = felec;
1082
1083             fscal            = _mm_and_ps(fscal,cutoff_mask);
1084
1085             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1086
1087              /* Update vectorial force */
1088             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1089             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1090             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1091
1092             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1093             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1094             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1095
1096             }
1097
1098             /**************************
1099              * CALCULATE INTERACTIONS *
1100              **************************/
1101
1102             if (gmx_mm_any_lt(rsq12,rcutoff2))
1103             {
1104
1105             r12              = _mm_mul_ps(rsq12,rinv12);
1106             r12              = _mm_andnot_ps(dummy_mask,r12);
1107
1108             /* EWALD ELECTROSTATICS */
1109
1110             /* Analytical PME correction */
1111             zeta2            = _mm_mul_ps(beta2,rsq12);
1112             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1113             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1114             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1115             felec            = _mm_mul_ps(qq12,felec);
1116             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1117             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
1118             velec            = _mm_mul_ps(qq12,velec);
1119
1120             d                = _mm_sub_ps(r12,rswitch);
1121             d                = _mm_max_ps(d,_mm_setzero_ps());
1122             d2               = _mm_mul_ps(d,d);
1123             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1124
1125             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1126
1127             /* Evaluate switch function */
1128             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1129             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
1130             velec            = _mm_mul_ps(velec,sw);
1131             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1132
1133             /* Update potential sum for this i atom from the interaction with this j atom. */
1134             velec            = _mm_and_ps(velec,cutoff_mask);
1135             velec            = _mm_andnot_ps(dummy_mask,velec);
1136             velecsum         = _mm_add_ps(velecsum,velec);
1137
1138             fscal            = felec;
1139
1140             fscal            = _mm_and_ps(fscal,cutoff_mask);
1141
1142             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1143
1144              /* Update vectorial force */
1145             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1146             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1147             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1148
1149             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1150             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1151             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1152
1153             }
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             if (gmx_mm_any_lt(rsq13,rcutoff2))
1160             {
1161
1162             r13              = _mm_mul_ps(rsq13,rinv13);
1163             r13              = _mm_andnot_ps(dummy_mask,r13);
1164
1165             /* EWALD ELECTROSTATICS */
1166
1167             /* Analytical PME correction */
1168             zeta2            = _mm_mul_ps(beta2,rsq13);
1169             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
1170             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1171             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1172             felec            = _mm_mul_ps(qq13,felec);
1173             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1174             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
1175             velec            = _mm_mul_ps(qq13,velec);
1176
1177             d                = _mm_sub_ps(r13,rswitch);
1178             d                = _mm_max_ps(d,_mm_setzero_ps());
1179             d2               = _mm_mul_ps(d,d);
1180             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1181
1182             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1183
1184             /* Evaluate switch function */
1185             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1186             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
1187             velec            = _mm_mul_ps(velec,sw);
1188             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1189
1190             /* Update potential sum for this i atom from the interaction with this j atom. */
1191             velec            = _mm_and_ps(velec,cutoff_mask);
1192             velec            = _mm_andnot_ps(dummy_mask,velec);
1193             velecsum         = _mm_add_ps(velecsum,velec);
1194
1195             fscal            = felec;
1196
1197             fscal            = _mm_and_ps(fscal,cutoff_mask);
1198
1199             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1200
1201              /* Update vectorial force */
1202             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1203             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1204             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1205
1206             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1207             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1208             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1209
1210             }
1211
1212             /**************************
1213              * CALCULATE INTERACTIONS *
1214              **************************/
1215
1216             if (gmx_mm_any_lt(rsq21,rcutoff2))
1217             {
1218
1219             r21              = _mm_mul_ps(rsq21,rinv21);
1220             r21              = _mm_andnot_ps(dummy_mask,r21);
1221
1222             /* EWALD ELECTROSTATICS */
1223
1224             /* Analytical PME correction */
1225             zeta2            = _mm_mul_ps(beta2,rsq21);
1226             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1227             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1228             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1229             felec            = _mm_mul_ps(qq21,felec);
1230             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1231             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
1232             velec            = _mm_mul_ps(qq21,velec);
1233
1234             d                = _mm_sub_ps(r21,rswitch);
1235             d                = _mm_max_ps(d,_mm_setzero_ps());
1236             d2               = _mm_mul_ps(d,d);
1237             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1238
1239             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1240
1241             /* Evaluate switch function */
1242             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1243             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
1244             velec            = _mm_mul_ps(velec,sw);
1245             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1246
1247             /* Update potential sum for this i atom from the interaction with this j atom. */
1248             velec            = _mm_and_ps(velec,cutoff_mask);
1249             velec            = _mm_andnot_ps(dummy_mask,velec);
1250             velecsum         = _mm_add_ps(velecsum,velec);
1251
1252             fscal            = felec;
1253
1254             fscal            = _mm_and_ps(fscal,cutoff_mask);
1255
1256             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1257
1258              /* Update vectorial force */
1259             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1260             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1261             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1262
1263             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1264             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1265             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1266
1267             }
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             if (gmx_mm_any_lt(rsq22,rcutoff2))
1274             {
1275
1276             r22              = _mm_mul_ps(rsq22,rinv22);
1277             r22              = _mm_andnot_ps(dummy_mask,r22);
1278
1279             /* EWALD ELECTROSTATICS */
1280
1281             /* Analytical PME correction */
1282             zeta2            = _mm_mul_ps(beta2,rsq22);
1283             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1284             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1285             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1286             felec            = _mm_mul_ps(qq22,felec);
1287             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1288             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
1289             velec            = _mm_mul_ps(qq22,velec);
1290
1291             d                = _mm_sub_ps(r22,rswitch);
1292             d                = _mm_max_ps(d,_mm_setzero_ps());
1293             d2               = _mm_mul_ps(d,d);
1294             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1295
1296             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1297
1298             /* Evaluate switch function */
1299             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1300             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
1301             velec            = _mm_mul_ps(velec,sw);
1302             cutoff_mask      = _mm_cmplt_ps(rsq22,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              /* Update vectorial force */
1316             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1317             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1318             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1319
1320             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1321             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1322             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1323
1324             }
1325
1326             /**************************
1327              * CALCULATE INTERACTIONS *
1328              **************************/
1329
1330             if (gmx_mm_any_lt(rsq23,rcutoff2))
1331             {
1332
1333             r23              = _mm_mul_ps(rsq23,rinv23);
1334             r23              = _mm_andnot_ps(dummy_mask,r23);
1335
1336             /* EWALD ELECTROSTATICS */
1337
1338             /* Analytical PME correction */
1339             zeta2            = _mm_mul_ps(beta2,rsq23);
1340             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
1341             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1342             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1343             felec            = _mm_mul_ps(qq23,felec);
1344             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1345             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
1346             velec            = _mm_mul_ps(qq23,velec);
1347
1348             d                = _mm_sub_ps(r23,rswitch);
1349             d                = _mm_max_ps(d,_mm_setzero_ps());
1350             d2               = _mm_mul_ps(d,d);
1351             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1352
1353             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1354
1355             /* Evaluate switch function */
1356             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1357             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
1358             velec            = _mm_mul_ps(velec,sw);
1359             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1360
1361             /* Update potential sum for this i atom from the interaction with this j atom. */
1362             velec            = _mm_and_ps(velec,cutoff_mask);
1363             velec            = _mm_andnot_ps(dummy_mask,velec);
1364             velecsum         = _mm_add_ps(velecsum,velec);
1365
1366             fscal            = felec;
1367
1368             fscal            = _mm_and_ps(fscal,cutoff_mask);
1369
1370             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1371
1372              /* Update vectorial force */
1373             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1374             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1375             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1376
1377             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1378             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1379             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1380
1381             }
1382
1383             /**************************
1384              * CALCULATE INTERACTIONS *
1385              **************************/
1386
1387             if (gmx_mm_any_lt(rsq31,rcutoff2))
1388             {
1389
1390             r31              = _mm_mul_ps(rsq31,rinv31);
1391             r31              = _mm_andnot_ps(dummy_mask,r31);
1392
1393             /* EWALD ELECTROSTATICS */
1394
1395             /* Analytical PME correction */
1396             zeta2            = _mm_mul_ps(beta2,rsq31);
1397             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
1398             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1399             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1400             felec            = _mm_mul_ps(qq31,felec);
1401             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1402             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
1403             velec            = _mm_mul_ps(qq31,velec);
1404
1405             d                = _mm_sub_ps(r31,rswitch);
1406             d                = _mm_max_ps(d,_mm_setzero_ps());
1407             d2               = _mm_mul_ps(d,d);
1408             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1409
1410             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1411
1412             /* Evaluate switch function */
1413             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1414             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
1415             velec            = _mm_mul_ps(velec,sw);
1416             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1417
1418             /* Update potential sum for this i atom from the interaction with this j atom. */
1419             velec            = _mm_and_ps(velec,cutoff_mask);
1420             velec            = _mm_andnot_ps(dummy_mask,velec);
1421             velecsum         = _mm_add_ps(velecsum,velec);
1422
1423             fscal            = felec;
1424
1425             fscal            = _mm_and_ps(fscal,cutoff_mask);
1426
1427             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1428
1429              /* Update vectorial force */
1430             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1431             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1432             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1433
1434             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1435             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1436             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1437
1438             }
1439
1440             /**************************
1441              * CALCULATE INTERACTIONS *
1442              **************************/
1443
1444             if (gmx_mm_any_lt(rsq32,rcutoff2))
1445             {
1446
1447             r32              = _mm_mul_ps(rsq32,rinv32);
1448             r32              = _mm_andnot_ps(dummy_mask,r32);
1449
1450             /* EWALD ELECTROSTATICS */
1451
1452             /* Analytical PME correction */
1453             zeta2            = _mm_mul_ps(beta2,rsq32);
1454             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
1455             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1456             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1457             felec            = _mm_mul_ps(qq32,felec);
1458             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1459             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
1460             velec            = _mm_mul_ps(qq32,velec);
1461
1462             d                = _mm_sub_ps(r32,rswitch);
1463             d                = _mm_max_ps(d,_mm_setzero_ps());
1464             d2               = _mm_mul_ps(d,d);
1465             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1466
1467             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1468
1469             /* Evaluate switch function */
1470             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1471             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
1472             velec            = _mm_mul_ps(velec,sw);
1473             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1474
1475             /* Update potential sum for this i atom from the interaction with this j atom. */
1476             velec            = _mm_and_ps(velec,cutoff_mask);
1477             velec            = _mm_andnot_ps(dummy_mask,velec);
1478             velecsum         = _mm_add_ps(velecsum,velec);
1479
1480             fscal            = felec;
1481
1482             fscal            = _mm_and_ps(fscal,cutoff_mask);
1483
1484             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1485
1486              /* Update vectorial force */
1487             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1488             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1489             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1490
1491             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1492             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1493             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1494
1495             }
1496
1497             /**************************
1498              * CALCULATE INTERACTIONS *
1499              **************************/
1500
1501             if (gmx_mm_any_lt(rsq33,rcutoff2))
1502             {
1503
1504             r33              = _mm_mul_ps(rsq33,rinv33);
1505             r33              = _mm_andnot_ps(dummy_mask,r33);
1506
1507             /* EWALD ELECTROSTATICS */
1508
1509             /* Analytical PME correction */
1510             zeta2            = _mm_mul_ps(beta2,rsq33);
1511             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
1512             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1513             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1514             felec            = _mm_mul_ps(qq33,felec);
1515             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1516             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
1517             velec            = _mm_mul_ps(qq33,velec);
1518
1519             d                = _mm_sub_ps(r33,rswitch);
1520             d                = _mm_max_ps(d,_mm_setzero_ps());
1521             d2               = _mm_mul_ps(d,d);
1522             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1523
1524             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1525
1526             /* Evaluate switch function */
1527             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1528             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
1529             velec            = _mm_mul_ps(velec,sw);
1530             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
1531
1532             /* Update potential sum for this i atom from the interaction with this j atom. */
1533             velec            = _mm_and_ps(velec,cutoff_mask);
1534             velec            = _mm_andnot_ps(dummy_mask,velec);
1535             velecsum         = _mm_add_ps(velecsum,velec);
1536
1537             fscal            = felec;
1538
1539             fscal            = _mm_and_ps(fscal,cutoff_mask);
1540
1541             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1542
1543              /* Update vectorial force */
1544             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1545             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1546             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1547
1548             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1549             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1550             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1551
1552             }
1553
1554             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1555             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1556             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1557             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1558
1559             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1560                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1561                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1562
1563             /* Inner loop uses 552 flops */
1564         }
1565
1566         /* End of innermost loop */
1567
1568         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1569                                               f+i_coord_offset,fshift+i_shift_offset);
1570
1571         ggid                        = gid[iidx];
1572         /* Update potential energies */
1573         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1574         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1575
1576         /* Increment number of inner iterations */
1577         inneriter                  += j_index_end - j_index_start;
1578
1579         /* Outer loop uses 26 flops */
1580     }
1581
1582     /* Increment number of outer iterations */
1583     outeriter        += nri;
1584
1585     /* Update outer/inner flops */
1586
1587     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*552);
1588 }
1589 /*
1590  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_F_avx_128_fma_single
1591  * Electrostatics interaction: Ewald
1592  * VdW interaction:            LennardJones
1593  * Geometry:                   Water4-Water4
1594  * Calculate force/pot:        Force
1595  */
1596 void
1597 nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_F_avx_128_fma_single
1598                     (t_nblist                    * gmx_restrict       nlist,
1599                      rvec                        * gmx_restrict          xx,
1600                      rvec                        * gmx_restrict          ff,
1601                      t_forcerec                  * gmx_restrict          fr,
1602                      t_mdatoms                   * gmx_restrict     mdatoms,
1603                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1604                      t_nrnb                      * gmx_restrict        nrnb)
1605 {
1606     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1607      * just 0 for non-waters.
1608      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1609      * jnr indices corresponding to data put in the four positions in the SIMD register.
1610      */
1611     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1612     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1613     int              jnrA,jnrB,jnrC,jnrD;
1614     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1615     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1616     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1617     real             rcutoff_scalar;
1618     real             *shiftvec,*fshift,*x,*f;
1619     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1620     real             scratch[4*DIM];
1621     __m128           fscal,rcutoff,rcutoff2,jidxall;
1622     int              vdwioffset0;
1623     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1624     int              vdwioffset1;
1625     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1626     int              vdwioffset2;
1627     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1628     int              vdwioffset3;
1629     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1630     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1631     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1632     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1633     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1634     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1635     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1636     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1637     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1638     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1639     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1640     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1641     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1642     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1643     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1644     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1645     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1646     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1647     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1648     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1649     real             *charge;
1650     int              nvdwtype;
1651     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1652     int              *vdwtype;
1653     real             *vdwparam;
1654     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1655     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1656     __m128i          ewitab;
1657     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1658     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1659     real             *ewtab;
1660     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1661     real             rswitch_scalar,d_scalar;
1662     __m128           dummy_mask,cutoff_mask;
1663     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1664     __m128           one     = _mm_set1_ps(1.0);
1665     __m128           two     = _mm_set1_ps(2.0);
1666     x                = xx[0];
1667     f                = ff[0];
1668
1669     nri              = nlist->nri;
1670     iinr             = nlist->iinr;
1671     jindex           = nlist->jindex;
1672     jjnr             = nlist->jjnr;
1673     shiftidx         = nlist->shift;
1674     gid              = nlist->gid;
1675     shiftvec         = fr->shift_vec[0];
1676     fshift           = fr->fshift[0];
1677     facel            = _mm_set1_ps(fr->epsfac);
1678     charge           = mdatoms->chargeA;
1679     nvdwtype         = fr->ntype;
1680     vdwparam         = fr->nbfp;
1681     vdwtype          = mdatoms->typeA;
1682
1683     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1684     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
1685     beta2            = _mm_mul_ps(beta,beta);
1686     beta3            = _mm_mul_ps(beta,beta2);
1687     ewtab            = fr->ic->tabq_coul_FDV0;
1688     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1689     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1690
1691     /* Setup water-specific parameters */
1692     inr              = nlist->iinr[0];
1693     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1694     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1695     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1696     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1697
1698     jq1              = _mm_set1_ps(charge[inr+1]);
1699     jq2              = _mm_set1_ps(charge[inr+2]);
1700     jq3              = _mm_set1_ps(charge[inr+3]);
1701     vdwjidx0A        = 2*vdwtype[inr+0];
1702     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1703     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1704     qq11             = _mm_mul_ps(iq1,jq1);
1705     qq12             = _mm_mul_ps(iq1,jq2);
1706     qq13             = _mm_mul_ps(iq1,jq3);
1707     qq21             = _mm_mul_ps(iq2,jq1);
1708     qq22             = _mm_mul_ps(iq2,jq2);
1709     qq23             = _mm_mul_ps(iq2,jq3);
1710     qq31             = _mm_mul_ps(iq3,jq1);
1711     qq32             = _mm_mul_ps(iq3,jq2);
1712     qq33             = _mm_mul_ps(iq3,jq3);
1713
1714     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1715     rcutoff_scalar   = fr->rcoulomb;
1716     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1717     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1718
1719     rswitch_scalar   = fr->rcoulomb_switch;
1720     rswitch          = _mm_set1_ps(rswitch_scalar);
1721     /* Setup switch parameters */
1722     d_scalar         = rcutoff_scalar-rswitch_scalar;
1723     d                = _mm_set1_ps(d_scalar);
1724     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
1725     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1726     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1727     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
1728     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1729     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1730
1731     /* Avoid stupid compiler warnings */
1732     jnrA = jnrB = jnrC = jnrD = 0;
1733     j_coord_offsetA = 0;
1734     j_coord_offsetB = 0;
1735     j_coord_offsetC = 0;
1736     j_coord_offsetD = 0;
1737
1738     outeriter        = 0;
1739     inneriter        = 0;
1740
1741     for(iidx=0;iidx<4*DIM;iidx++)
1742     {
1743         scratch[iidx] = 0.0;
1744     }
1745
1746     /* Start outer loop over neighborlists */
1747     for(iidx=0; iidx<nri; iidx++)
1748     {
1749         /* Load shift vector for this list */
1750         i_shift_offset   = DIM*shiftidx[iidx];
1751
1752         /* Load limits for loop over neighbors */
1753         j_index_start    = jindex[iidx];
1754         j_index_end      = jindex[iidx+1];
1755
1756         /* Get outer coordinate index */
1757         inr              = iinr[iidx];
1758         i_coord_offset   = DIM*inr;
1759
1760         /* Load i particle coords and add shift vector */
1761         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1762                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1763
1764         fix0             = _mm_setzero_ps();
1765         fiy0             = _mm_setzero_ps();
1766         fiz0             = _mm_setzero_ps();
1767         fix1             = _mm_setzero_ps();
1768         fiy1             = _mm_setzero_ps();
1769         fiz1             = _mm_setzero_ps();
1770         fix2             = _mm_setzero_ps();
1771         fiy2             = _mm_setzero_ps();
1772         fiz2             = _mm_setzero_ps();
1773         fix3             = _mm_setzero_ps();
1774         fiy3             = _mm_setzero_ps();
1775         fiz3             = _mm_setzero_ps();
1776
1777         /* Start inner kernel loop */
1778         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1779         {
1780
1781             /* Get j neighbor index, and coordinate index */
1782             jnrA             = jjnr[jidx];
1783             jnrB             = jjnr[jidx+1];
1784             jnrC             = jjnr[jidx+2];
1785             jnrD             = jjnr[jidx+3];
1786             j_coord_offsetA  = DIM*jnrA;
1787             j_coord_offsetB  = DIM*jnrB;
1788             j_coord_offsetC  = DIM*jnrC;
1789             j_coord_offsetD  = DIM*jnrD;
1790
1791             /* load j atom coordinates */
1792             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1793                                               x+j_coord_offsetC,x+j_coord_offsetD,
1794                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1795                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1796
1797             /* Calculate displacement vector */
1798             dx00             = _mm_sub_ps(ix0,jx0);
1799             dy00             = _mm_sub_ps(iy0,jy0);
1800             dz00             = _mm_sub_ps(iz0,jz0);
1801             dx11             = _mm_sub_ps(ix1,jx1);
1802             dy11             = _mm_sub_ps(iy1,jy1);
1803             dz11             = _mm_sub_ps(iz1,jz1);
1804             dx12             = _mm_sub_ps(ix1,jx2);
1805             dy12             = _mm_sub_ps(iy1,jy2);
1806             dz12             = _mm_sub_ps(iz1,jz2);
1807             dx13             = _mm_sub_ps(ix1,jx3);
1808             dy13             = _mm_sub_ps(iy1,jy3);
1809             dz13             = _mm_sub_ps(iz1,jz3);
1810             dx21             = _mm_sub_ps(ix2,jx1);
1811             dy21             = _mm_sub_ps(iy2,jy1);
1812             dz21             = _mm_sub_ps(iz2,jz1);
1813             dx22             = _mm_sub_ps(ix2,jx2);
1814             dy22             = _mm_sub_ps(iy2,jy2);
1815             dz22             = _mm_sub_ps(iz2,jz2);
1816             dx23             = _mm_sub_ps(ix2,jx3);
1817             dy23             = _mm_sub_ps(iy2,jy3);
1818             dz23             = _mm_sub_ps(iz2,jz3);
1819             dx31             = _mm_sub_ps(ix3,jx1);
1820             dy31             = _mm_sub_ps(iy3,jy1);
1821             dz31             = _mm_sub_ps(iz3,jz1);
1822             dx32             = _mm_sub_ps(ix3,jx2);
1823             dy32             = _mm_sub_ps(iy3,jy2);
1824             dz32             = _mm_sub_ps(iz3,jz2);
1825             dx33             = _mm_sub_ps(ix3,jx3);
1826             dy33             = _mm_sub_ps(iy3,jy3);
1827             dz33             = _mm_sub_ps(iz3,jz3);
1828
1829             /* Calculate squared distance and things based on it */
1830             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1831             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1832             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1833             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1834             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1835             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1836             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1837             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1838             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1839             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1840
1841             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1842             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1843             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1844             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1845             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1846             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1847             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1848             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1849             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1850             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1851
1852             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1853             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1854             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1855             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1856             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1857             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1858             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1859             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1860             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1861             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1862
1863             fjx0             = _mm_setzero_ps();
1864             fjy0             = _mm_setzero_ps();
1865             fjz0             = _mm_setzero_ps();
1866             fjx1             = _mm_setzero_ps();
1867             fjy1             = _mm_setzero_ps();
1868             fjz1             = _mm_setzero_ps();
1869             fjx2             = _mm_setzero_ps();
1870             fjy2             = _mm_setzero_ps();
1871             fjz2             = _mm_setzero_ps();
1872             fjx3             = _mm_setzero_ps();
1873             fjy3             = _mm_setzero_ps();
1874             fjz3             = _mm_setzero_ps();
1875
1876             /**************************
1877              * CALCULATE INTERACTIONS *
1878              **************************/
1879
1880             if (gmx_mm_any_lt(rsq00,rcutoff2))
1881             {
1882
1883             r00              = _mm_mul_ps(rsq00,rinv00);
1884
1885             /* LENNARD-JONES DISPERSION/REPULSION */
1886
1887             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1888             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1889             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1890             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1891             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1892
1893             d                = _mm_sub_ps(r00,rswitch);
1894             d                = _mm_max_ps(d,_mm_setzero_ps());
1895             d2               = _mm_mul_ps(d,d);
1896             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1897
1898             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1899
1900             /* Evaluate switch function */
1901             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1902             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1903             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1904
1905             fscal            = fvdw;
1906
1907             fscal            = _mm_and_ps(fscal,cutoff_mask);
1908
1909              /* Update vectorial force */
1910             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1911             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1912             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1913
1914             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1915             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1916             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1917
1918             }
1919
1920             /**************************
1921              * CALCULATE INTERACTIONS *
1922              **************************/
1923
1924             if (gmx_mm_any_lt(rsq11,rcutoff2))
1925             {
1926
1927             r11              = _mm_mul_ps(rsq11,rinv11);
1928
1929             /* EWALD ELECTROSTATICS */
1930
1931             /* Analytical PME correction */
1932             zeta2            = _mm_mul_ps(beta2,rsq11);
1933             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1934             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1935             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1936             felec            = _mm_mul_ps(qq11,felec);
1937             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1938             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
1939             velec            = _mm_mul_ps(qq11,velec);
1940
1941             d                = _mm_sub_ps(r11,rswitch);
1942             d                = _mm_max_ps(d,_mm_setzero_ps());
1943             d2               = _mm_mul_ps(d,d);
1944             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1945
1946             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1947
1948             /* Evaluate switch function */
1949             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1950             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
1951             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1952
1953             fscal            = felec;
1954
1955             fscal            = _mm_and_ps(fscal,cutoff_mask);
1956
1957              /* Update vectorial force */
1958             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1959             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1960             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1961
1962             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1963             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1964             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1965
1966             }
1967
1968             /**************************
1969              * CALCULATE INTERACTIONS *
1970              **************************/
1971
1972             if (gmx_mm_any_lt(rsq12,rcutoff2))
1973             {
1974
1975             r12              = _mm_mul_ps(rsq12,rinv12);
1976
1977             /* EWALD ELECTROSTATICS */
1978
1979             /* Analytical PME correction */
1980             zeta2            = _mm_mul_ps(beta2,rsq12);
1981             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1982             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1983             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1984             felec            = _mm_mul_ps(qq12,felec);
1985             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1986             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
1987             velec            = _mm_mul_ps(qq12,velec);
1988
1989             d                = _mm_sub_ps(r12,rswitch);
1990             d                = _mm_max_ps(d,_mm_setzero_ps());
1991             d2               = _mm_mul_ps(d,d);
1992             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1993
1994             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1995
1996             /* Evaluate switch function */
1997             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1998             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
1999             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
2000
2001             fscal            = felec;
2002
2003             fscal            = _mm_and_ps(fscal,cutoff_mask);
2004
2005              /* Update vectorial force */
2006             fix1             = _mm_macc_ps(dx12,fscal,fix1);
2007             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
2008             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
2009
2010             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
2011             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
2012             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
2013
2014             }
2015
2016             /**************************
2017              * CALCULATE INTERACTIONS *
2018              **************************/
2019
2020             if (gmx_mm_any_lt(rsq13,rcutoff2))
2021             {
2022
2023             r13              = _mm_mul_ps(rsq13,rinv13);
2024
2025             /* EWALD ELECTROSTATICS */
2026
2027             /* Analytical PME correction */
2028             zeta2            = _mm_mul_ps(beta2,rsq13);
2029             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
2030             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2031             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2032             felec            = _mm_mul_ps(qq13,felec);
2033             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2034             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
2035             velec            = _mm_mul_ps(qq13,velec);
2036
2037             d                = _mm_sub_ps(r13,rswitch);
2038             d                = _mm_max_ps(d,_mm_setzero_ps());
2039             d2               = _mm_mul_ps(d,d);
2040             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2041
2042             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2043
2044             /* Evaluate switch function */
2045             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2046             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
2047             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
2048
2049             fscal            = felec;
2050
2051             fscal            = _mm_and_ps(fscal,cutoff_mask);
2052
2053              /* Update vectorial force */
2054             fix1             = _mm_macc_ps(dx13,fscal,fix1);
2055             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
2056             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
2057
2058             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
2059             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
2060             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
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             /* Analytical PME correction */
2076             zeta2            = _mm_mul_ps(beta2,rsq21);
2077             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
2078             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2079             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2080             felec            = _mm_mul_ps(qq21,felec);
2081             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2082             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
2083             velec            = _mm_mul_ps(qq21,velec);
2084
2085             d                = _mm_sub_ps(r21,rswitch);
2086             d                = _mm_max_ps(d,_mm_setzero_ps());
2087             d2               = _mm_mul_ps(d,d);
2088             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2089
2090             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2091
2092             /* Evaluate switch function */
2093             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2094             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
2095             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2096
2097             fscal            = felec;
2098
2099             fscal            = _mm_and_ps(fscal,cutoff_mask);
2100
2101              /* Update vectorial force */
2102             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2103             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2104             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2105
2106             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2107             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2108             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2109
2110             }
2111
2112             /**************************
2113              * CALCULATE INTERACTIONS *
2114              **************************/
2115
2116             if (gmx_mm_any_lt(rsq22,rcutoff2))
2117             {
2118
2119             r22              = _mm_mul_ps(rsq22,rinv22);
2120
2121             /* EWALD ELECTROSTATICS */
2122
2123             /* Analytical PME correction */
2124             zeta2            = _mm_mul_ps(beta2,rsq22);
2125             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
2126             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2127             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2128             felec            = _mm_mul_ps(qq22,felec);
2129             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2130             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
2131             velec            = _mm_mul_ps(qq22,velec);
2132
2133             d                = _mm_sub_ps(r22,rswitch);
2134             d                = _mm_max_ps(d,_mm_setzero_ps());
2135             d2               = _mm_mul_ps(d,d);
2136             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2137
2138             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2139
2140             /* Evaluate switch function */
2141             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2142             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
2143             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2144
2145             fscal            = felec;
2146
2147             fscal            = _mm_and_ps(fscal,cutoff_mask);
2148
2149              /* Update vectorial force */
2150             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2151             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2152             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2153
2154             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2155             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2156             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2157
2158             }
2159
2160             /**************************
2161              * CALCULATE INTERACTIONS *
2162              **************************/
2163
2164             if (gmx_mm_any_lt(rsq23,rcutoff2))
2165             {
2166
2167             r23              = _mm_mul_ps(rsq23,rinv23);
2168
2169             /* EWALD ELECTROSTATICS */
2170
2171             /* Analytical PME correction */
2172             zeta2            = _mm_mul_ps(beta2,rsq23);
2173             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
2174             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2175             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2176             felec            = _mm_mul_ps(qq23,felec);
2177             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2178             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
2179             velec            = _mm_mul_ps(qq23,velec);
2180
2181             d                = _mm_sub_ps(r23,rswitch);
2182             d                = _mm_max_ps(d,_mm_setzero_ps());
2183             d2               = _mm_mul_ps(d,d);
2184             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2185
2186             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2187
2188             /* Evaluate switch function */
2189             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2190             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
2191             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
2192
2193             fscal            = felec;
2194
2195             fscal            = _mm_and_ps(fscal,cutoff_mask);
2196
2197              /* Update vectorial force */
2198             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2199             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2200             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2201
2202             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2203             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2204             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2205
2206             }
2207
2208             /**************************
2209              * CALCULATE INTERACTIONS *
2210              **************************/
2211
2212             if (gmx_mm_any_lt(rsq31,rcutoff2))
2213             {
2214
2215             r31              = _mm_mul_ps(rsq31,rinv31);
2216
2217             /* EWALD ELECTROSTATICS */
2218
2219             /* Analytical PME correction */
2220             zeta2            = _mm_mul_ps(beta2,rsq31);
2221             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
2222             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2223             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2224             felec            = _mm_mul_ps(qq31,felec);
2225             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2226             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
2227             velec            = _mm_mul_ps(qq31,velec);
2228
2229             d                = _mm_sub_ps(r31,rswitch);
2230             d                = _mm_max_ps(d,_mm_setzero_ps());
2231             d2               = _mm_mul_ps(d,d);
2232             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2233
2234             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2235
2236             /* Evaluate switch function */
2237             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2238             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
2239             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
2240
2241             fscal            = felec;
2242
2243             fscal            = _mm_and_ps(fscal,cutoff_mask);
2244
2245              /* Update vectorial force */
2246             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2247             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2248             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2249
2250             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2251             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2252             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2253
2254             }
2255
2256             /**************************
2257              * CALCULATE INTERACTIONS *
2258              **************************/
2259
2260             if (gmx_mm_any_lt(rsq32,rcutoff2))
2261             {
2262
2263             r32              = _mm_mul_ps(rsq32,rinv32);
2264
2265             /* EWALD ELECTROSTATICS */
2266
2267             /* Analytical PME correction */
2268             zeta2            = _mm_mul_ps(beta2,rsq32);
2269             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
2270             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2271             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2272             felec            = _mm_mul_ps(qq32,felec);
2273             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2274             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
2275             velec            = _mm_mul_ps(qq32,velec);
2276
2277             d                = _mm_sub_ps(r32,rswitch);
2278             d                = _mm_max_ps(d,_mm_setzero_ps());
2279             d2               = _mm_mul_ps(d,d);
2280             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2281
2282             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2283
2284             /* Evaluate switch function */
2285             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2286             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
2287             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
2288
2289             fscal            = felec;
2290
2291             fscal            = _mm_and_ps(fscal,cutoff_mask);
2292
2293              /* Update vectorial force */
2294             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2295             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2296             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2297
2298             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2299             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2300             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2301
2302             }
2303
2304             /**************************
2305              * CALCULATE INTERACTIONS *
2306              **************************/
2307
2308             if (gmx_mm_any_lt(rsq33,rcutoff2))
2309             {
2310
2311             r33              = _mm_mul_ps(rsq33,rinv33);
2312
2313             /* EWALD ELECTROSTATICS */
2314
2315             /* Analytical PME correction */
2316             zeta2            = _mm_mul_ps(beta2,rsq33);
2317             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
2318             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2319             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2320             felec            = _mm_mul_ps(qq33,felec);
2321             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2322             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
2323             velec            = _mm_mul_ps(qq33,velec);
2324
2325             d                = _mm_sub_ps(r33,rswitch);
2326             d                = _mm_max_ps(d,_mm_setzero_ps());
2327             d2               = _mm_mul_ps(d,d);
2328             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2329
2330             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2331
2332             /* Evaluate switch function */
2333             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2334             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
2335             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2336
2337             fscal            = felec;
2338
2339             fscal            = _mm_and_ps(fscal,cutoff_mask);
2340
2341              /* Update vectorial force */
2342             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2343             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2344             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2345
2346             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2347             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2348             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2349
2350             }
2351
2352             fjptrA             = f+j_coord_offsetA;
2353             fjptrB             = f+j_coord_offsetB;
2354             fjptrC             = f+j_coord_offsetC;
2355             fjptrD             = f+j_coord_offsetD;
2356
2357             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2358                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2359                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2360
2361             /* Inner loop uses 512 flops */
2362         }
2363
2364         if(jidx<j_index_end)
2365         {
2366
2367             /* Get j neighbor index, and coordinate index */
2368             jnrlistA         = jjnr[jidx];
2369             jnrlistB         = jjnr[jidx+1];
2370             jnrlistC         = jjnr[jidx+2];
2371             jnrlistD         = jjnr[jidx+3];
2372             /* Sign of each element will be negative for non-real atoms.
2373              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2374              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2375              */
2376             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2377             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2378             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2379             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2380             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2381             j_coord_offsetA  = DIM*jnrA;
2382             j_coord_offsetB  = DIM*jnrB;
2383             j_coord_offsetC  = DIM*jnrC;
2384             j_coord_offsetD  = DIM*jnrD;
2385
2386             /* load j atom coordinates */
2387             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2388                                               x+j_coord_offsetC,x+j_coord_offsetD,
2389                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2390                                               &jy2,&jz2,&jx3,&jy3,&jz3);
2391
2392             /* Calculate displacement vector */
2393             dx00             = _mm_sub_ps(ix0,jx0);
2394             dy00             = _mm_sub_ps(iy0,jy0);
2395             dz00             = _mm_sub_ps(iz0,jz0);
2396             dx11             = _mm_sub_ps(ix1,jx1);
2397             dy11             = _mm_sub_ps(iy1,jy1);
2398             dz11             = _mm_sub_ps(iz1,jz1);
2399             dx12             = _mm_sub_ps(ix1,jx2);
2400             dy12             = _mm_sub_ps(iy1,jy2);
2401             dz12             = _mm_sub_ps(iz1,jz2);
2402             dx13             = _mm_sub_ps(ix1,jx3);
2403             dy13             = _mm_sub_ps(iy1,jy3);
2404             dz13             = _mm_sub_ps(iz1,jz3);
2405             dx21             = _mm_sub_ps(ix2,jx1);
2406             dy21             = _mm_sub_ps(iy2,jy1);
2407             dz21             = _mm_sub_ps(iz2,jz1);
2408             dx22             = _mm_sub_ps(ix2,jx2);
2409             dy22             = _mm_sub_ps(iy2,jy2);
2410             dz22             = _mm_sub_ps(iz2,jz2);
2411             dx23             = _mm_sub_ps(ix2,jx3);
2412             dy23             = _mm_sub_ps(iy2,jy3);
2413             dz23             = _mm_sub_ps(iz2,jz3);
2414             dx31             = _mm_sub_ps(ix3,jx1);
2415             dy31             = _mm_sub_ps(iy3,jy1);
2416             dz31             = _mm_sub_ps(iz3,jz1);
2417             dx32             = _mm_sub_ps(ix3,jx2);
2418             dy32             = _mm_sub_ps(iy3,jy2);
2419             dz32             = _mm_sub_ps(iz3,jz2);
2420             dx33             = _mm_sub_ps(ix3,jx3);
2421             dy33             = _mm_sub_ps(iy3,jy3);
2422             dz33             = _mm_sub_ps(iz3,jz3);
2423
2424             /* Calculate squared distance and things based on it */
2425             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
2426             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2427             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2428             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
2429             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2430             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2431             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
2432             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
2433             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
2434             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2435
2436             rinv00           = gmx_mm_invsqrt_ps(rsq00);
2437             rinv11           = gmx_mm_invsqrt_ps(rsq11);
2438             rinv12           = gmx_mm_invsqrt_ps(rsq12);
2439             rinv13           = gmx_mm_invsqrt_ps(rsq13);
2440             rinv21           = gmx_mm_invsqrt_ps(rsq21);
2441             rinv22           = gmx_mm_invsqrt_ps(rsq22);
2442             rinv23           = gmx_mm_invsqrt_ps(rsq23);
2443             rinv31           = gmx_mm_invsqrt_ps(rsq31);
2444             rinv32           = gmx_mm_invsqrt_ps(rsq32);
2445             rinv33           = gmx_mm_invsqrt_ps(rsq33);
2446
2447             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
2448             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
2449             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
2450             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
2451             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
2452             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
2453             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
2454             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
2455             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
2456             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
2457
2458             fjx0             = _mm_setzero_ps();
2459             fjy0             = _mm_setzero_ps();
2460             fjz0             = _mm_setzero_ps();
2461             fjx1             = _mm_setzero_ps();
2462             fjy1             = _mm_setzero_ps();
2463             fjz1             = _mm_setzero_ps();
2464             fjx2             = _mm_setzero_ps();
2465             fjy2             = _mm_setzero_ps();
2466             fjz2             = _mm_setzero_ps();
2467             fjx3             = _mm_setzero_ps();
2468             fjy3             = _mm_setzero_ps();
2469             fjz3             = _mm_setzero_ps();
2470
2471             /**************************
2472              * CALCULATE INTERACTIONS *
2473              **************************/
2474
2475             if (gmx_mm_any_lt(rsq00,rcutoff2))
2476             {
2477
2478             r00              = _mm_mul_ps(rsq00,rinv00);
2479             r00              = _mm_andnot_ps(dummy_mask,r00);
2480
2481             /* LENNARD-JONES DISPERSION/REPULSION */
2482
2483             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2484             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
2485             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
2486             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
2487             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
2488
2489             d                = _mm_sub_ps(r00,rswitch);
2490             d                = _mm_max_ps(d,_mm_setzero_ps());
2491             d2               = _mm_mul_ps(d,d);
2492             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2493
2494             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2495
2496             /* Evaluate switch function */
2497             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2498             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
2499             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
2500
2501             fscal            = fvdw;
2502
2503             fscal            = _mm_and_ps(fscal,cutoff_mask);
2504
2505             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2506
2507              /* Update vectorial force */
2508             fix0             = _mm_macc_ps(dx00,fscal,fix0);
2509             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
2510             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
2511
2512             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
2513             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
2514             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
2515
2516             }
2517
2518             /**************************
2519              * CALCULATE INTERACTIONS *
2520              **************************/
2521
2522             if (gmx_mm_any_lt(rsq11,rcutoff2))
2523             {
2524
2525             r11              = _mm_mul_ps(rsq11,rinv11);
2526             r11              = _mm_andnot_ps(dummy_mask,r11);
2527
2528             /* EWALD ELECTROSTATICS */
2529
2530             /* Analytical PME correction */
2531             zeta2            = _mm_mul_ps(beta2,rsq11);
2532             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
2533             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2534             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2535             felec            = _mm_mul_ps(qq11,felec);
2536             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2537             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
2538             velec            = _mm_mul_ps(qq11,velec);
2539
2540             d                = _mm_sub_ps(r11,rswitch);
2541             d                = _mm_max_ps(d,_mm_setzero_ps());
2542             d2               = _mm_mul_ps(d,d);
2543             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2544
2545             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2546
2547             /* Evaluate switch function */
2548             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2549             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
2550             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
2551
2552             fscal            = felec;
2553
2554             fscal            = _mm_and_ps(fscal,cutoff_mask);
2555
2556             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2557
2558              /* Update vectorial force */
2559             fix1             = _mm_macc_ps(dx11,fscal,fix1);
2560             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
2561             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
2562
2563             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
2564             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
2565             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
2566
2567             }
2568
2569             /**************************
2570              * CALCULATE INTERACTIONS *
2571              **************************/
2572
2573             if (gmx_mm_any_lt(rsq12,rcutoff2))
2574             {
2575
2576             r12              = _mm_mul_ps(rsq12,rinv12);
2577             r12              = _mm_andnot_ps(dummy_mask,r12);
2578
2579             /* EWALD ELECTROSTATICS */
2580
2581             /* Analytical PME correction */
2582             zeta2            = _mm_mul_ps(beta2,rsq12);
2583             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
2584             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2585             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2586             felec            = _mm_mul_ps(qq12,felec);
2587             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2588             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
2589             velec            = _mm_mul_ps(qq12,velec);
2590
2591             d                = _mm_sub_ps(r12,rswitch);
2592             d                = _mm_max_ps(d,_mm_setzero_ps());
2593             d2               = _mm_mul_ps(d,d);
2594             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2595
2596             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2597
2598             /* Evaluate switch function */
2599             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2600             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
2601             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
2602
2603             fscal            = felec;
2604
2605             fscal            = _mm_and_ps(fscal,cutoff_mask);
2606
2607             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2608
2609              /* Update vectorial force */
2610             fix1             = _mm_macc_ps(dx12,fscal,fix1);
2611             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
2612             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
2613
2614             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
2615             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
2616             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
2617
2618             }
2619
2620             /**************************
2621              * CALCULATE INTERACTIONS *
2622              **************************/
2623
2624             if (gmx_mm_any_lt(rsq13,rcutoff2))
2625             {
2626
2627             r13              = _mm_mul_ps(rsq13,rinv13);
2628             r13              = _mm_andnot_ps(dummy_mask,r13);
2629
2630             /* EWALD ELECTROSTATICS */
2631
2632             /* Analytical PME correction */
2633             zeta2            = _mm_mul_ps(beta2,rsq13);
2634             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
2635             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2636             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2637             felec            = _mm_mul_ps(qq13,felec);
2638             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2639             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
2640             velec            = _mm_mul_ps(qq13,velec);
2641
2642             d                = _mm_sub_ps(r13,rswitch);
2643             d                = _mm_max_ps(d,_mm_setzero_ps());
2644             d2               = _mm_mul_ps(d,d);
2645             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2646
2647             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2648
2649             /* Evaluate switch function */
2650             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2651             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
2652             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
2653
2654             fscal            = felec;
2655
2656             fscal            = _mm_and_ps(fscal,cutoff_mask);
2657
2658             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2659
2660              /* Update vectorial force */
2661             fix1             = _mm_macc_ps(dx13,fscal,fix1);
2662             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
2663             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
2664
2665             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
2666             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
2667             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
2668
2669             }
2670
2671             /**************************
2672              * CALCULATE INTERACTIONS *
2673              **************************/
2674
2675             if (gmx_mm_any_lt(rsq21,rcutoff2))
2676             {
2677
2678             r21              = _mm_mul_ps(rsq21,rinv21);
2679             r21              = _mm_andnot_ps(dummy_mask,r21);
2680
2681             /* EWALD ELECTROSTATICS */
2682
2683             /* Analytical PME correction */
2684             zeta2            = _mm_mul_ps(beta2,rsq21);
2685             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
2686             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2687             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2688             felec            = _mm_mul_ps(qq21,felec);
2689             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2690             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
2691             velec            = _mm_mul_ps(qq21,velec);
2692
2693             d                = _mm_sub_ps(r21,rswitch);
2694             d                = _mm_max_ps(d,_mm_setzero_ps());
2695             d2               = _mm_mul_ps(d,d);
2696             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2697
2698             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2699
2700             /* Evaluate switch function */
2701             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2702             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
2703             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2704
2705             fscal            = felec;
2706
2707             fscal            = _mm_and_ps(fscal,cutoff_mask);
2708
2709             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2710
2711              /* Update vectorial force */
2712             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2713             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2714             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2715
2716             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2717             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2718             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2719
2720             }
2721
2722             /**************************
2723              * CALCULATE INTERACTIONS *
2724              **************************/
2725
2726             if (gmx_mm_any_lt(rsq22,rcutoff2))
2727             {
2728
2729             r22              = _mm_mul_ps(rsq22,rinv22);
2730             r22              = _mm_andnot_ps(dummy_mask,r22);
2731
2732             /* EWALD ELECTROSTATICS */
2733
2734             /* Analytical PME correction */
2735             zeta2            = _mm_mul_ps(beta2,rsq22);
2736             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
2737             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2738             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2739             felec            = _mm_mul_ps(qq22,felec);
2740             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2741             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
2742             velec            = _mm_mul_ps(qq22,velec);
2743
2744             d                = _mm_sub_ps(r22,rswitch);
2745             d                = _mm_max_ps(d,_mm_setzero_ps());
2746             d2               = _mm_mul_ps(d,d);
2747             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2748
2749             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2750
2751             /* Evaluate switch function */
2752             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2753             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
2754             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2755
2756             fscal            = felec;
2757
2758             fscal            = _mm_and_ps(fscal,cutoff_mask);
2759
2760             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2761
2762              /* Update vectorial force */
2763             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2764             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2765             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2766
2767             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2768             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2769             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2770
2771             }
2772
2773             /**************************
2774              * CALCULATE INTERACTIONS *
2775              **************************/
2776
2777             if (gmx_mm_any_lt(rsq23,rcutoff2))
2778             {
2779
2780             r23              = _mm_mul_ps(rsq23,rinv23);
2781             r23              = _mm_andnot_ps(dummy_mask,r23);
2782
2783             /* EWALD ELECTROSTATICS */
2784
2785             /* Analytical PME correction */
2786             zeta2            = _mm_mul_ps(beta2,rsq23);
2787             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
2788             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2789             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2790             felec            = _mm_mul_ps(qq23,felec);
2791             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2792             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
2793             velec            = _mm_mul_ps(qq23,velec);
2794
2795             d                = _mm_sub_ps(r23,rswitch);
2796             d                = _mm_max_ps(d,_mm_setzero_ps());
2797             d2               = _mm_mul_ps(d,d);
2798             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2799
2800             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2801
2802             /* Evaluate switch function */
2803             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2804             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
2805             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
2806
2807             fscal            = felec;
2808
2809             fscal            = _mm_and_ps(fscal,cutoff_mask);
2810
2811             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2812
2813              /* Update vectorial force */
2814             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2815             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2816             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2817
2818             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2819             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2820             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2821
2822             }
2823
2824             /**************************
2825              * CALCULATE INTERACTIONS *
2826              **************************/
2827
2828             if (gmx_mm_any_lt(rsq31,rcutoff2))
2829             {
2830
2831             r31              = _mm_mul_ps(rsq31,rinv31);
2832             r31              = _mm_andnot_ps(dummy_mask,r31);
2833
2834             /* EWALD ELECTROSTATICS */
2835
2836             /* Analytical PME correction */
2837             zeta2            = _mm_mul_ps(beta2,rsq31);
2838             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
2839             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2840             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2841             felec            = _mm_mul_ps(qq31,felec);
2842             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2843             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
2844             velec            = _mm_mul_ps(qq31,velec);
2845
2846             d                = _mm_sub_ps(r31,rswitch);
2847             d                = _mm_max_ps(d,_mm_setzero_ps());
2848             d2               = _mm_mul_ps(d,d);
2849             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2850
2851             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2852
2853             /* Evaluate switch function */
2854             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2855             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
2856             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
2857
2858             fscal            = felec;
2859
2860             fscal            = _mm_and_ps(fscal,cutoff_mask);
2861
2862             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2863
2864              /* Update vectorial force */
2865             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2866             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2867             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2868
2869             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2870             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2871             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2872
2873             }
2874
2875             /**************************
2876              * CALCULATE INTERACTIONS *
2877              **************************/
2878
2879             if (gmx_mm_any_lt(rsq32,rcutoff2))
2880             {
2881
2882             r32              = _mm_mul_ps(rsq32,rinv32);
2883             r32              = _mm_andnot_ps(dummy_mask,r32);
2884
2885             /* EWALD ELECTROSTATICS */
2886
2887             /* Analytical PME correction */
2888             zeta2            = _mm_mul_ps(beta2,rsq32);
2889             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
2890             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2891             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2892             felec            = _mm_mul_ps(qq32,felec);
2893             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2894             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
2895             velec            = _mm_mul_ps(qq32,velec);
2896
2897             d                = _mm_sub_ps(r32,rswitch);
2898             d                = _mm_max_ps(d,_mm_setzero_ps());
2899             d2               = _mm_mul_ps(d,d);
2900             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2901
2902             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2903
2904             /* Evaluate switch function */
2905             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2906             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
2907             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
2908
2909             fscal            = felec;
2910
2911             fscal            = _mm_and_ps(fscal,cutoff_mask);
2912
2913             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2914
2915              /* Update vectorial force */
2916             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2917             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2918             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2919
2920             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2921             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2922             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2923
2924             }
2925
2926             /**************************
2927              * CALCULATE INTERACTIONS *
2928              **************************/
2929
2930             if (gmx_mm_any_lt(rsq33,rcutoff2))
2931             {
2932
2933             r33              = _mm_mul_ps(rsq33,rinv33);
2934             r33              = _mm_andnot_ps(dummy_mask,r33);
2935
2936             /* EWALD ELECTROSTATICS */
2937
2938             /* Analytical PME correction */
2939             zeta2            = _mm_mul_ps(beta2,rsq33);
2940             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
2941             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2942             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2943             felec            = _mm_mul_ps(qq33,felec);
2944             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2945             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
2946             velec            = _mm_mul_ps(qq33,velec);
2947
2948             d                = _mm_sub_ps(r33,rswitch);
2949             d                = _mm_max_ps(d,_mm_setzero_ps());
2950             d2               = _mm_mul_ps(d,d);
2951             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2952
2953             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2954
2955             /* Evaluate switch function */
2956             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2957             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
2958             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2959
2960             fscal            = felec;
2961
2962             fscal            = _mm_and_ps(fscal,cutoff_mask);
2963
2964             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2965
2966              /* Update vectorial force */
2967             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2968             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2969             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2970
2971             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2972             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2973             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2974
2975             }
2976
2977             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2978             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2979             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2980             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2981
2982             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2983                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2984                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2985
2986             /* Inner loop uses 522 flops */
2987         }
2988
2989         /* End of innermost loop */
2990
2991         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2992                                               f+i_coord_offset,fshift+i_shift_offset);
2993
2994         /* Increment number of inner iterations */
2995         inneriter                  += j_index_end - j_index_start;
2996
2997         /* Outer loop uses 24 flops */
2998     }
2999
3000     /* Increment number of outer iterations */
3001     outeriter        += nri;
3002
3003     /* Update outer/inner flops */
3004
3005     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*522);
3006 }