src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  80     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  81     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  82     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  83     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  84     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  85     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  87     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  88     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  89     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  90     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  91     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  92     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  93     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  94     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  95     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 102     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 103     __m128i          ewitab;
 104     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 106     real             *ewtab;
 107     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 108     real             rswitch_scalar,d_scalar;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 131     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 132     beta2            = _mm_mul_ps(beta,beta);
 133     beta3            = _mm_mul_ps(beta,beta2);
 134     ewtab            = fr->ic->tabq_coul_FDV0;
 135     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 136     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 141     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 142     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 143     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145     jq1              = _mm_set1_ps(charge[inr+1]);
 146     jq2              = _mm_set1_ps(charge[inr+2]);
 147     jq3              = _mm_set1_ps(charge[inr+3]);
 148     vdwjidx0A        = 2*vdwtype[inr+0];
 149     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 150     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 151     qq11             = _mm_mul_ps(iq1,jq1);
 152     qq12             = _mm_mul_ps(iq1,jq2);
 153     qq13             = _mm_mul_ps(iq1,jq3);
 154     qq21             = _mm_mul_ps(iq2,jq1);
 155     qq22             = _mm_mul_ps(iq2,jq2);
 156     qq23             = _mm_mul_ps(iq2,jq3);
 157     qq31             = _mm_mul_ps(iq3,jq1);
 158     qq32             = _mm_mul_ps(iq3,jq2);
 159     qq33             = _mm_mul_ps(iq3,jq3);
 160
 161     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 162     rcutoff_scalar   = fr->rcoulomb;
 163     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 164     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 165
 166     rswitch_scalar   = fr->rcoulomb_switch;
 167     rswitch          = _mm_set1_ps(rswitch_scalar);
 168     /* Setup switch parameters */
 169     d_scalar         = rcutoff_scalar-rswitch_scalar;
 170     d                = _mm_set1_ps(d_scalar);
 171     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 172     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 173     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 174     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 175     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 176     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 177
 178     /* Avoid stupid compiler warnings */
 179     jnrA = jnrB = jnrC = jnrD = 0;
 180     j_coord_offsetA = 0;
 181     j_coord_offsetB = 0;
 182     j_coord_offsetC = 0;
 183     j_coord_offsetD = 0;
 184
 185     outeriter        = 0;
 186     inneriter        = 0;
 187
 188     for(iidx=0;iidx<4*DIM;iidx++)
 189     {
 190         scratch[iidx] = 0.0;
 191     }
 192
 193     /* Start outer loop over neighborlists */
 194     for(iidx=0; iidx<nri; iidx++)
 195     {
 196         /* Load shift vector for this list */
 197         i_shift_offset   = DIM*shiftidx[iidx];
 198
 199         /* Load limits for loop over neighbors */
 200         j_index_start    = jindex[iidx];
 201         j_index_end      = jindex[iidx+1];
 202
 203         /* Get outer coordinate index */
 204         inr              = iinr[iidx];
 205         i_coord_offset   = DIM*inr;
 206
 207         /* Load i particle coords and add shift vector */
 208         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 209                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 210
 211         fix0             = _mm_setzero_ps();
 212         fiy0             = _mm_setzero_ps();
 213         fiz0             = _mm_setzero_ps();
 214         fix1             = _mm_setzero_ps();
 215         fiy1             = _mm_setzero_ps();
 216         fiz1             = _mm_setzero_ps();
 217         fix2             = _mm_setzero_ps();
 218         fiy2             = _mm_setzero_ps();
 219         fiz2             = _mm_setzero_ps();
 220         fix3             = _mm_setzero_ps();
 221         fiy3             = _mm_setzero_ps();
 222         fiz3             = _mm_setzero_ps();
 223
 224         /* Reset potential sums */
 225         velecsum         = _mm_setzero_ps();
 226         vvdwsum          = _mm_setzero_ps();
 227
 228         /* Start inner kernel loop */
 229         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 230         {
 231
 232             /* Get j neighbor index, and coordinate index */
 233             jnrA             = jjnr[jidx];
 234             jnrB             = jjnr[jidx+1];
 235             jnrC             = jjnr[jidx+2];
 236             jnrD             = jjnr[jidx+3];
 237             j_coord_offsetA  = DIM*jnrA;
 238             j_coord_offsetB  = DIM*jnrB;
 239             j_coord_offsetC  = DIM*jnrC;
 240             j_coord_offsetD  = DIM*jnrD;
 241
 242             /* load j atom coordinates */
 243             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 244                                               x+j_coord_offsetC,x+j_coord_offsetD,
 245                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 246                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 247
 248             /* Calculate displacement vector */
 249             dx00             = _mm_sub_ps(ix0,jx0);
 250             dy00             = _mm_sub_ps(iy0,jy0);
 251             dz00             = _mm_sub_ps(iz0,jz0);
 252             dx11             = _mm_sub_ps(ix1,jx1);
 253             dy11             = _mm_sub_ps(iy1,jy1);
 254             dz11             = _mm_sub_ps(iz1,jz1);
 255             dx12             = _mm_sub_ps(ix1,jx2);
 256             dy12             = _mm_sub_ps(iy1,jy2);
 257             dz12             = _mm_sub_ps(iz1,jz2);
 258             dx13             = _mm_sub_ps(ix1,jx3);
 259             dy13             = _mm_sub_ps(iy1,jy3);
 260             dz13             = _mm_sub_ps(iz1,jz3);
 261             dx21             = _mm_sub_ps(ix2,jx1);
 262             dy21             = _mm_sub_ps(iy2,jy1);
 263             dz21             = _mm_sub_ps(iz2,jz1);
 264             dx22             = _mm_sub_ps(ix2,jx2);
 265             dy22             = _mm_sub_ps(iy2,jy2);
 266             dz22             = _mm_sub_ps(iz2,jz2);
 267             dx23             = _mm_sub_ps(ix2,jx3);
 268             dy23             = _mm_sub_ps(iy2,jy3);
 269             dz23             = _mm_sub_ps(iz2,jz3);
 270             dx31             = _mm_sub_ps(ix3,jx1);
 271             dy31             = _mm_sub_ps(iy3,jy1);
 272             dz31             = _mm_sub_ps(iz3,jz1);
 273             dx32             = _mm_sub_ps(ix3,jx2);
 274             dy32             = _mm_sub_ps(iy3,jy2);
 275             dz32             = _mm_sub_ps(iz3,jz2);
 276             dx33             = _mm_sub_ps(ix3,jx3);
 277             dy33             = _mm_sub_ps(iy3,jy3);
 278             dz33             = _mm_sub_ps(iz3,jz3);
 279
 280             /* Calculate squared distance and things based on it */
 281             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 282             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 283             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 284             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 285             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 286             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 287             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 288             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 289             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 290             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 291
 292             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 293             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 294             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 295             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 296             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 297             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 298             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 299             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 300             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 301             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 302
 303             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 304             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 305             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 306             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 307             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 308             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 309             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 310             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 311             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 312             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 313
 314             fjx0             = _mm_setzero_ps();
 315             fjy0             = _mm_setzero_ps();
 316             fjz0             = _mm_setzero_ps();
 317             fjx1             = _mm_setzero_ps();
 318             fjy1             = _mm_setzero_ps();
 319             fjz1             = _mm_setzero_ps();
 320             fjx2             = _mm_setzero_ps();
 321             fjy2             = _mm_setzero_ps();
 322             fjz2             = _mm_setzero_ps();
 323             fjx3             = _mm_setzero_ps();
 324             fjy3             = _mm_setzero_ps();
 325             fjz3             = _mm_setzero_ps();
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             if (gmx_mm_any_lt(rsq00,rcutoff2))
 332             {
 333
 334             r00              = _mm_mul_ps(rsq00,rinv00);
 335
 336             /* LENNARD-JONES DISPERSION/REPULSION */
 337
 338             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 339             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 340             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 341             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 342             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 343
 344             d                = _mm_sub_ps(r00,rswitch);
 345             d                = _mm_max_ps(d,_mm_setzero_ps());
 346             d2               = _mm_mul_ps(d,d);
 347             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 348
 349             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 350
 351             /* Evaluate switch function */
 352             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 353             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 354             vvdw             = _mm_mul_ps(vvdw,sw);
 355             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 359             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 360
 361             fscal            = fvdw;
 362
 363             fscal            = _mm_and_ps(fscal,cutoff_mask);
 364
 365              /* Update vectorial force */
 366             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 367             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 368             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 369
 370             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 371             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 372             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 373
 374             }
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm_any_lt(rsq11,rcutoff2))
 381             {
 382
 383             r11              = _mm_mul_ps(rsq11,rinv11);
 384
 385             /* EWALD ELECTROSTATICS */
 386
 387             /* Analytical PME correction */
 388             zeta2            = _mm_mul_ps(beta2,rsq11);
 389             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 390             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 391             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 392             felec            = _mm_mul_ps(qq11,felec);
 393             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 394             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 395             velec            = _mm_mul_ps(qq11,velec);
 396
 397             d                = _mm_sub_ps(r11,rswitch);
 398             d                = _mm_max_ps(d,_mm_setzero_ps());
 399             d2               = _mm_mul_ps(d,d);
 400             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 401
 402             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 403
 404             /* Evaluate switch function */
 405             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 406             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
 407             velec            = _mm_mul_ps(velec,sw);
 408             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velec            = _mm_and_ps(velec,cutoff_mask);
 412             velecsum         = _mm_add_ps(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             fscal            = _mm_and_ps(fscal,cutoff_mask);
 417
 418              /* Update vectorial force */
 419             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 420             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 421             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 422
 423             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 424             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 425             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 426
 427             }
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             if (gmx_mm_any_lt(rsq12,rcutoff2))
 434             {
 435
 436             r12              = _mm_mul_ps(rsq12,rinv12);
 437
 438             /* EWALD ELECTROSTATICS */
 439
 440             /* Analytical PME correction */
 441             zeta2            = _mm_mul_ps(beta2,rsq12);
 442             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 443             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 444             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 445             felec            = _mm_mul_ps(qq12,felec);
 446             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 447             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 448             velec            = _mm_mul_ps(qq12,velec);
 449
 450             d                = _mm_sub_ps(r12,rswitch);
 451             d                = _mm_max_ps(d,_mm_setzero_ps());
 452             d2               = _mm_mul_ps(d,d);
 453             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 454
 455             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 456
 457             /* Evaluate switch function */
 458             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 459             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
 460             velec            = _mm_mul_ps(velec,sw);
 461             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 462
 463             /* Update potential sum for this i atom from the interaction with this j atom. */
 464             velec            = _mm_and_ps(velec,cutoff_mask);
 465             velecsum         = _mm_add_ps(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469             fscal            = _mm_and_ps(fscal,cutoff_mask);
 470
 471              /* Update vectorial force */
 472             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 473             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 474             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 475
 476             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 477             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 478             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 479
 480             }
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             if (gmx_mm_any_lt(rsq13,rcutoff2))
 487             {
 488
 489             r13              = _mm_mul_ps(rsq13,rinv13);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Analytical PME correction */
 494             zeta2            = _mm_mul_ps(beta2,rsq13);
 495             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
 496             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 497             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 498             felec            = _mm_mul_ps(qq13,felec);
 499             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 500             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
 501             velec            = _mm_mul_ps(qq13,velec);
 502
 503             d                = _mm_sub_ps(r13,rswitch);
 504             d                = _mm_max_ps(d,_mm_setzero_ps());
 505             d2               = _mm_mul_ps(d,d);
 506             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 507
 508             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 509
 510             /* Evaluate switch function */
 511             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 512             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
 513             velec            = _mm_mul_ps(velec,sw);
 514             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm_and_ps(velec,cutoff_mask);
 518             velecsum         = _mm_add_ps(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             fscal            = _mm_and_ps(fscal,cutoff_mask);
 523
 524              /* Update vectorial force */
 525             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 526             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 527             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 528
 529             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 530             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 531             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 532
 533             }
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             if (gmx_mm_any_lt(rsq21,rcutoff2))
 540             {
 541
 542             r21              = _mm_mul_ps(rsq21,rinv21);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Analytical PME correction */
 547             zeta2            = _mm_mul_ps(beta2,rsq21);
 548             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 549             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 550             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 551             felec            = _mm_mul_ps(qq21,felec);
 552             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 553             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
 554             velec            = _mm_mul_ps(qq21,velec);
 555
 556             d                = _mm_sub_ps(r21,rswitch);
 557             d                = _mm_max_ps(d,_mm_setzero_ps());
 558             d2               = _mm_mul_ps(d,d);
 559             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 560
 561             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 562
 563             /* Evaluate switch function */
 564             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 565             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
 566             velec            = _mm_mul_ps(velec,sw);
 567             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_and_ps(velec,cutoff_mask);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_and_ps(fscal,cutoff_mask);
 576
 577              /* Update vectorial force */
 578             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 579             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 580             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 581
 582             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 583             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 584             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 585
 586             }
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             if (gmx_mm_any_lt(rsq22,rcutoff2))
 593             {
 594
 595             r22              = _mm_mul_ps(rsq22,rinv22);
 596
 597             /* EWALD ELECTROSTATICS */
 598
 599             /* Analytical PME correction */
 600             zeta2            = _mm_mul_ps(beta2,rsq22);
 601             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
 602             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 603             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 604             felec            = _mm_mul_ps(qq22,felec);
 605             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 606             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
 607             velec            = _mm_mul_ps(qq22,velec);
 608
 609             d                = _mm_sub_ps(r22,rswitch);
 610             d                = _mm_max_ps(d,_mm_setzero_ps());
 611             d2               = _mm_mul_ps(d,d);
 612             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 613
 614             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 615
 616             /* Evaluate switch function */
 617             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 618             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
 619             velec            = _mm_mul_ps(velec,sw);
 620             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velec            = _mm_and_ps(velec,cutoff_mask);
 624             velecsum         = _mm_add_ps(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm_and_ps(fscal,cutoff_mask);
 629
 630              /* Update vectorial force */
 631             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 632             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 633             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 634
 635             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 636             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 637             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 638
 639             }
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             if (gmx_mm_any_lt(rsq23,rcutoff2))
 646             {
 647
 648             r23              = _mm_mul_ps(rsq23,rinv23);
 649
 650             /* EWALD ELECTROSTATICS */
 651
 652             /* Analytical PME correction */
 653             zeta2            = _mm_mul_ps(beta2,rsq23);
 654             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
 655             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 656             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 657             felec            = _mm_mul_ps(qq23,felec);
 658             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 659             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
 660             velec            = _mm_mul_ps(qq23,velec);
 661
 662             d                = _mm_sub_ps(r23,rswitch);
 663             d                = _mm_max_ps(d,_mm_setzero_ps());
 664             d2               = _mm_mul_ps(d,d);
 665             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 666
 667             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 668
 669             /* Evaluate switch function */
 670             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 671             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
 672             velec            = _mm_mul_ps(velec,sw);
 673             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
 674
 675             /* Update potential sum for this i atom from the interaction with this j atom. */
 676             velec            = _mm_and_ps(velec,cutoff_mask);
 677             velecsum         = _mm_add_ps(velecsum,velec);
 678
 679             fscal            = felec;
 680
 681             fscal            = _mm_and_ps(fscal,cutoff_mask);
 682
 683              /* Update vectorial force */
 684             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 685             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 686             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 687
 688             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 689             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 690             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 691
 692             }
 693
 694             /**************************
 695              * CALCULATE INTERACTIONS *
 696              **************************/
 697
 698             if (gmx_mm_any_lt(rsq31,rcutoff2))
 699             {
 700
 701             r31              = _mm_mul_ps(rsq31,rinv31);
 702
 703             /* EWALD ELECTROSTATICS */
 704
 705             /* Analytical PME correction */
 706             zeta2            = _mm_mul_ps(beta2,rsq31);
 707             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
 708             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 709             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 710             felec            = _mm_mul_ps(qq31,felec);
 711             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 712             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
 713             velec            = _mm_mul_ps(qq31,velec);
 714
 715             d                = _mm_sub_ps(r31,rswitch);
 716             d                = _mm_max_ps(d,_mm_setzero_ps());
 717             d2               = _mm_mul_ps(d,d);
 718             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 719
 720             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 721
 722             /* Evaluate switch function */
 723             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 724             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
 725             velec            = _mm_mul_ps(velec,sw);
 726             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
 727
 728             /* Update potential sum for this i atom from the interaction with this j atom. */
 729             velec            = _mm_and_ps(velec,cutoff_mask);
 730             velecsum         = _mm_add_ps(velecsum,velec);
 731
 732             fscal            = felec;
 733
 734             fscal            = _mm_and_ps(fscal,cutoff_mask);
 735
 736              /* Update vectorial force */
 737             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 738             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 739             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 740
 741             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 742             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 743             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 744
 745             }
 746
 747             /**************************
 748              * CALCULATE INTERACTIONS *
 749              **************************/
 750
 751             if (gmx_mm_any_lt(rsq32,rcutoff2))
 752             {
 753
 754             r32              = _mm_mul_ps(rsq32,rinv32);
 755
 756             /* EWALD ELECTROSTATICS */
 757
 758             /* Analytical PME correction */
 759             zeta2            = _mm_mul_ps(beta2,rsq32);
 760             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
 761             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 762             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 763             felec            = _mm_mul_ps(qq32,felec);
 764             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 765             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
 766             velec            = _mm_mul_ps(qq32,velec);
 767
 768             d                = _mm_sub_ps(r32,rswitch);
 769             d                = _mm_max_ps(d,_mm_setzero_ps());
 770             d2               = _mm_mul_ps(d,d);
 771             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 772
 773             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 774
 775             /* Evaluate switch function */
 776             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 777             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
 778             velec            = _mm_mul_ps(velec,sw);
 779             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
 780
 781             /* Update potential sum for this i atom from the interaction with this j atom. */
 782             velec            = _mm_and_ps(velec,cutoff_mask);
 783             velecsum         = _mm_add_ps(velecsum,velec);
 784
 785             fscal            = felec;
 786
 787             fscal            = _mm_and_ps(fscal,cutoff_mask);
 788
 789              /* Update vectorial force */
 790             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 791             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 792             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 793
 794             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 795             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 796             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 797
 798             }
 799
 800             /**************************
 801              * CALCULATE INTERACTIONS *
 802              **************************/
 803
 804             if (gmx_mm_any_lt(rsq33,rcutoff2))
 805             {
 806
 807             r33              = _mm_mul_ps(rsq33,rinv33);
 808
 809             /* EWALD ELECTROSTATICS */
 810
 811             /* Analytical PME correction */
 812             zeta2            = _mm_mul_ps(beta2,rsq33);
 813             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
 814             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 815             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 816             felec            = _mm_mul_ps(qq33,felec);
 817             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 818             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
 819             velec            = _mm_mul_ps(qq33,velec);
 820
 821             d                = _mm_sub_ps(r33,rswitch);
 822             d                = _mm_max_ps(d,_mm_setzero_ps());
 823             d2               = _mm_mul_ps(d,d);
 824             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 825
 826             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 827
 828             /* Evaluate switch function */
 829             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 830             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
 831             velec            = _mm_mul_ps(velec,sw);
 832             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
 833
 834             /* Update potential sum for this i atom from the interaction with this j atom. */
 835             velec            = _mm_and_ps(velec,cutoff_mask);
 836             velecsum         = _mm_add_ps(velecsum,velec);
 837
 838             fscal            = felec;
 839
 840             fscal            = _mm_and_ps(fscal,cutoff_mask);
 841
 842              /* Update vectorial force */
 843             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 844             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 845             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 846
 847             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 848             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 849             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 850
 851             }
 852
 853             fjptrA             = f+j_coord_offsetA;
 854             fjptrB             = f+j_coord_offsetB;
 855             fjptrC             = f+j_coord_offsetC;
 856             fjptrD             = f+j_coord_offsetD;
 857
 858             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 859                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 860                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 861
 862             /* Inner loop uses 542 flops */
 863         }
 864
 865         if(jidx<j_index_end)
 866         {
 867
 868             /* Get j neighbor index, and coordinate index */
 869             jnrlistA         = jjnr[jidx];
 870             jnrlistB         = jjnr[jidx+1];
 871             jnrlistC         = jjnr[jidx+2];
 872             jnrlistD         = jjnr[jidx+3];
 873             /* Sign of each element will be negative for non-real atoms.
 874              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 875              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 876              */
 877             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 878             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 879             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 880             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 881             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 882             j_coord_offsetA  = DIM*jnrA;
 883             j_coord_offsetB  = DIM*jnrB;
 884             j_coord_offsetC  = DIM*jnrC;
 885             j_coord_offsetD  = DIM*jnrD;
 886
 887             /* load j atom coordinates */
 888             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 889                                               x+j_coord_offsetC,x+j_coord_offsetD,
 890                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 891                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 892
 893             /* Calculate displacement vector */
 894             dx00             = _mm_sub_ps(ix0,jx0);
 895             dy00             = _mm_sub_ps(iy0,jy0);
 896             dz00             = _mm_sub_ps(iz0,jz0);
 897             dx11             = _mm_sub_ps(ix1,jx1);
 898             dy11             = _mm_sub_ps(iy1,jy1);
 899             dz11             = _mm_sub_ps(iz1,jz1);
 900             dx12             = _mm_sub_ps(ix1,jx2);
 901             dy12             = _mm_sub_ps(iy1,jy2);
 902             dz12             = _mm_sub_ps(iz1,jz2);
 903             dx13             = _mm_sub_ps(ix1,jx3);
 904             dy13             = _mm_sub_ps(iy1,jy3);
 905             dz13             = _mm_sub_ps(iz1,jz3);
 906             dx21             = _mm_sub_ps(ix2,jx1);
 907             dy21             = _mm_sub_ps(iy2,jy1);
 908             dz21             = _mm_sub_ps(iz2,jz1);
 909             dx22             = _mm_sub_ps(ix2,jx2);
 910             dy22             = _mm_sub_ps(iy2,jy2);
 911             dz22             = _mm_sub_ps(iz2,jz2);
 912             dx23             = _mm_sub_ps(ix2,jx3);
 913             dy23             = _mm_sub_ps(iy2,jy3);
 914             dz23             = _mm_sub_ps(iz2,jz3);
 915             dx31             = _mm_sub_ps(ix3,jx1);
 916             dy31             = _mm_sub_ps(iy3,jy1);
 917             dz31             = _mm_sub_ps(iz3,jz1);
 918             dx32             = _mm_sub_ps(ix3,jx2);
 919             dy32             = _mm_sub_ps(iy3,jy2);
 920             dz32             = _mm_sub_ps(iz3,jz2);
 921             dx33             = _mm_sub_ps(ix3,jx3);
 922             dy33             = _mm_sub_ps(iy3,jy3);
 923             dz33             = _mm_sub_ps(iz3,jz3);
 924
 925             /* Calculate squared distance and things based on it */
 926             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 927             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 928             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 929             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 930             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 931             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 932             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 933             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 934             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 935             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 936
 937             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 938             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 939             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 940             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 941             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 942             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 943             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 944             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 945             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 946             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 947
 948             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 949             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 950             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 951             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 952             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 953             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 954             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 955             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 956             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 957             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 958
 959             fjx0             = _mm_setzero_ps();
 960             fjy0             = _mm_setzero_ps();
 961             fjz0             = _mm_setzero_ps();
 962             fjx1             = _mm_setzero_ps();
 963             fjy1             = _mm_setzero_ps();
 964             fjz1             = _mm_setzero_ps();
 965             fjx2             = _mm_setzero_ps();
 966             fjy2             = _mm_setzero_ps();
 967             fjz2             = _mm_setzero_ps();
 968             fjx3             = _mm_setzero_ps();
 969             fjy3             = _mm_setzero_ps();
 970             fjz3             = _mm_setzero_ps();
 971
 972             /**************************
 973              * CALCULATE INTERACTIONS *
 974              **************************/
 975
 976             if (gmx_mm_any_lt(rsq00,rcutoff2))
 977             {
 978
 979             r00              = _mm_mul_ps(rsq00,rinv00);
 980             r00              = _mm_andnot_ps(dummy_mask,r00);
 981
 982             /* LENNARD-JONES DISPERSION/REPULSION */
 983
 984             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 985             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 986             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 987             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 988             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 989
 990             d                = _mm_sub_ps(r00,rswitch);
 991             d                = _mm_max_ps(d,_mm_setzero_ps());
 992             d2               = _mm_mul_ps(d,d);
 993             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 994
 995             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 996
 997             /* Evaluate switch function */
 998             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 999             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1000             vvdw             = _mm_mul_ps(vvdw,sw);
1001             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1002
1003             /* Update potential sum for this i atom from the interaction with this j atom. */
1004             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
1005             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
1006             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
1007
1008             fscal            = fvdw;
1009
1010             fscal            = _mm_and_ps(fscal,cutoff_mask);
1011
1012             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1013
1014              /* Update vectorial force */
1015             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1016             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1017             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1018
1019             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1020             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1021             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1022
1023             }
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             if (gmx_mm_any_lt(rsq11,rcutoff2))
1030             {
1031
1032             r11              = _mm_mul_ps(rsq11,rinv11);
1033             r11              = _mm_andnot_ps(dummy_mask,r11);
1034
1035             /* EWALD ELECTROSTATICS */
1036
1037             /* Analytical PME correction */
1038             zeta2            = _mm_mul_ps(beta2,rsq11);
1039             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1040             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1041             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1042             felec            = _mm_mul_ps(qq11,felec);
1043             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1044             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
1045             velec            = _mm_mul_ps(qq11,velec);
1046
1047             d                = _mm_sub_ps(r11,rswitch);
1048             d                = _mm_max_ps(d,_mm_setzero_ps());
1049             d2               = _mm_mul_ps(d,d);
1050             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1051
1052             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1053
1054             /* Evaluate switch function */
1055             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1056             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
1057             velec            = _mm_mul_ps(velec,sw);
1058             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1059
1060             /* Update potential sum for this i atom from the interaction with this j atom. */
1061             velec            = _mm_and_ps(velec,cutoff_mask);
1062             velec            = _mm_andnot_ps(dummy_mask,velec);
1063             velecsum         = _mm_add_ps(velecsum,velec);
1064
1065             fscal            = felec;
1066
1067             fscal            = _mm_and_ps(fscal,cutoff_mask);
1068
1069             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1070
1071              /* Update vectorial force */
1072             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1073             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1074             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1075
1076             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1077             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1078             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1079
1080             }
1081
1082             /**************************
1083              * CALCULATE INTERACTIONS *
1084              **************************/
1085
1086             if (gmx_mm_any_lt(rsq12,rcutoff2))
1087             {
1088
1089             r12              = _mm_mul_ps(rsq12,rinv12);
1090             r12              = _mm_andnot_ps(dummy_mask,r12);
1091
1092             /* EWALD ELECTROSTATICS */
1093
1094             /* Analytical PME correction */
1095             zeta2            = _mm_mul_ps(beta2,rsq12);
1096             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1097             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1098             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1099             felec            = _mm_mul_ps(qq12,felec);
1100             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1101             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
1102             velec            = _mm_mul_ps(qq12,velec);
1103
1104             d                = _mm_sub_ps(r12,rswitch);
1105             d                = _mm_max_ps(d,_mm_setzero_ps());
1106             d2               = _mm_mul_ps(d,d);
1107             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1108
1109             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1110
1111             /* Evaluate switch function */
1112             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1113             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
1114             velec            = _mm_mul_ps(velec,sw);
1115             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1116
1117             /* Update potential sum for this i atom from the interaction with this j atom. */
1118             velec            = _mm_and_ps(velec,cutoff_mask);
1119             velec            = _mm_andnot_ps(dummy_mask,velec);
1120             velecsum         = _mm_add_ps(velecsum,velec);
1121
1122             fscal            = felec;
1123
1124             fscal            = _mm_and_ps(fscal,cutoff_mask);
1125
1126             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1127
1128              /* Update vectorial force */
1129             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1130             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1131             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1132
1133             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1134             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1135             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1136
1137             }
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             if (gmx_mm_any_lt(rsq13,rcutoff2))
1144             {
1145
1146             r13              = _mm_mul_ps(rsq13,rinv13);
1147             r13              = _mm_andnot_ps(dummy_mask,r13);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Analytical PME correction */
1152             zeta2            = _mm_mul_ps(beta2,rsq13);
1153             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
1154             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1155             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1156             felec            = _mm_mul_ps(qq13,felec);
1157             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1158             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
1159             velec            = _mm_mul_ps(qq13,velec);
1160
1161             d                = _mm_sub_ps(r13,rswitch);
1162             d                = _mm_max_ps(d,_mm_setzero_ps());
1163             d2               = _mm_mul_ps(d,d);
1164             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1165
1166             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1167
1168             /* Evaluate switch function */
1169             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1170             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
1171             velec            = _mm_mul_ps(velec,sw);
1172             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1173
1174             /* Update potential sum for this i atom from the interaction with this j atom. */
1175             velec            = _mm_and_ps(velec,cutoff_mask);
1176             velec            = _mm_andnot_ps(dummy_mask,velec);
1177             velecsum         = _mm_add_ps(velecsum,velec);
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm_and_ps(fscal,cutoff_mask);
1182
1183             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1184
1185              /* Update vectorial force */
1186             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1187             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1188             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1189
1190             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1191             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1192             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1193
1194             }
1195
1196             /**************************
1197              * CALCULATE INTERACTIONS *
1198              **************************/
1199
1200             if (gmx_mm_any_lt(rsq21,rcutoff2))
1201             {
1202
1203             r21              = _mm_mul_ps(rsq21,rinv21);
1204             r21              = _mm_andnot_ps(dummy_mask,r21);
1205
1206             /* EWALD ELECTROSTATICS */
1207
1208             /* Analytical PME correction */
1209             zeta2            = _mm_mul_ps(beta2,rsq21);
1210             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1211             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1212             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1213             felec            = _mm_mul_ps(qq21,felec);
1214             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1215             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
1216             velec            = _mm_mul_ps(qq21,velec);
1217
1218             d                = _mm_sub_ps(r21,rswitch);
1219             d                = _mm_max_ps(d,_mm_setzero_ps());
1220             d2               = _mm_mul_ps(d,d);
1221             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1222
1223             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1224
1225             /* Evaluate switch function */
1226             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1227             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
1228             velec            = _mm_mul_ps(velec,sw);
1229             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1230
1231             /* Update potential sum for this i atom from the interaction with this j atom. */
1232             velec            = _mm_and_ps(velec,cutoff_mask);
1233             velec            = _mm_andnot_ps(dummy_mask,velec);
1234             velecsum         = _mm_add_ps(velecsum,velec);
1235
1236             fscal            = felec;
1237
1238             fscal            = _mm_and_ps(fscal,cutoff_mask);
1239
1240             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1241
1242              /* Update vectorial force */
1243             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1244             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1245             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1246
1247             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1248             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1249             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1250
1251             }
1252
1253             /**************************
1254              * CALCULATE INTERACTIONS *
1255              **************************/
1256
1257             if (gmx_mm_any_lt(rsq22,rcutoff2))
1258             {
1259
1260             r22              = _mm_mul_ps(rsq22,rinv22);
1261             r22              = _mm_andnot_ps(dummy_mask,r22);
1262
1263             /* EWALD ELECTROSTATICS */
1264
1265             /* Analytical PME correction */
1266             zeta2            = _mm_mul_ps(beta2,rsq22);
1267             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1268             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1269             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1270             felec            = _mm_mul_ps(qq22,felec);
1271             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1272             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
1273             velec            = _mm_mul_ps(qq22,velec);
1274
1275             d                = _mm_sub_ps(r22,rswitch);
1276             d                = _mm_max_ps(d,_mm_setzero_ps());
1277             d2               = _mm_mul_ps(d,d);
1278             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1279
1280             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1281
1282             /* Evaluate switch function */
1283             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1284             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
1285             velec            = _mm_mul_ps(velec,sw);
1286             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1287
1288             /* Update potential sum for this i atom from the interaction with this j atom. */
1289             velec            = _mm_and_ps(velec,cutoff_mask);
1290             velec            = _mm_andnot_ps(dummy_mask,velec);
1291             velecsum         = _mm_add_ps(velecsum,velec);
1292
1293             fscal            = felec;
1294
1295             fscal            = _mm_and_ps(fscal,cutoff_mask);
1296
1297             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1298
1299              /* Update vectorial force */
1300             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1301             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1302             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1303
1304             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1305             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1306             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1307
1308             }
1309
1310             /**************************
1311              * CALCULATE INTERACTIONS *
1312              **************************/
1313
1314             if (gmx_mm_any_lt(rsq23,rcutoff2))
1315             {
1316
1317             r23              = _mm_mul_ps(rsq23,rinv23);
1318             r23              = _mm_andnot_ps(dummy_mask,r23);
1319
1320             /* EWALD ELECTROSTATICS */
1321
1322             /* Analytical PME correction */
1323             zeta2            = _mm_mul_ps(beta2,rsq23);
1324             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
1325             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1326             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1327             felec            = _mm_mul_ps(qq23,felec);
1328             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1329             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
1330             velec            = _mm_mul_ps(qq23,velec);
1331
1332             d                = _mm_sub_ps(r23,rswitch);
1333             d                = _mm_max_ps(d,_mm_setzero_ps());
1334             d2               = _mm_mul_ps(d,d);
1335             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1336
1337             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1338
1339             /* Evaluate switch function */
1340             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1341             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
1342             velec            = _mm_mul_ps(velec,sw);
1343             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1344
1345             /* Update potential sum for this i atom from the interaction with this j atom. */
1346             velec            = _mm_and_ps(velec,cutoff_mask);
1347             velec            = _mm_andnot_ps(dummy_mask,velec);
1348             velecsum         = _mm_add_ps(velecsum,velec);
1349
1350             fscal            = felec;
1351
1352             fscal            = _mm_and_ps(fscal,cutoff_mask);
1353
1354             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1355
1356              /* Update vectorial force */
1357             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1358             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1359             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1360
1361             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1362             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1363             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1364
1365             }
1366
1367             /**************************
1368              * CALCULATE INTERACTIONS *
1369              **************************/
1370
1371             if (gmx_mm_any_lt(rsq31,rcutoff2))
1372             {
1373
1374             r31              = _mm_mul_ps(rsq31,rinv31);
1375             r31              = _mm_andnot_ps(dummy_mask,r31);
1376
1377             /* EWALD ELECTROSTATICS */
1378
1379             /* Analytical PME correction */
1380             zeta2            = _mm_mul_ps(beta2,rsq31);
1381             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
1382             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1383             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1384             felec            = _mm_mul_ps(qq31,felec);
1385             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1386             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
1387             velec            = _mm_mul_ps(qq31,velec);
1388
1389             d                = _mm_sub_ps(r31,rswitch);
1390             d                = _mm_max_ps(d,_mm_setzero_ps());
1391             d2               = _mm_mul_ps(d,d);
1392             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1393
1394             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1395
1396             /* Evaluate switch function */
1397             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1398             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
1399             velec            = _mm_mul_ps(velec,sw);
1400             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1401
1402             /* Update potential sum for this i atom from the interaction with this j atom. */
1403             velec            = _mm_and_ps(velec,cutoff_mask);
1404             velec            = _mm_andnot_ps(dummy_mask,velec);
1405             velecsum         = _mm_add_ps(velecsum,velec);
1406
1407             fscal            = felec;
1408
1409             fscal            = _mm_and_ps(fscal,cutoff_mask);
1410
1411             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1412
1413              /* Update vectorial force */
1414             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1415             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1416             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1417
1418             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1419             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1420             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1421
1422             }
1423
1424             /**************************
1425              * CALCULATE INTERACTIONS *
1426              **************************/
1427
1428             if (gmx_mm_any_lt(rsq32,rcutoff2))
1429             {
1430
1431             r32              = _mm_mul_ps(rsq32,rinv32);
1432             r32              = _mm_andnot_ps(dummy_mask,r32);
1433
1434             /* EWALD ELECTROSTATICS */
1435
1436             /* Analytical PME correction */
1437             zeta2            = _mm_mul_ps(beta2,rsq32);
1438             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
1439             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1440             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1441             felec            = _mm_mul_ps(qq32,felec);
1442             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1443             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
1444             velec            = _mm_mul_ps(qq32,velec);
1445
1446             d                = _mm_sub_ps(r32,rswitch);
1447             d                = _mm_max_ps(d,_mm_setzero_ps());
1448             d2               = _mm_mul_ps(d,d);
1449             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1450
1451             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1452
1453             /* Evaluate switch function */
1454             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1455             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
1456             velec            = _mm_mul_ps(velec,sw);
1457             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1458
1459             /* Update potential sum for this i atom from the interaction with this j atom. */
1460             velec            = _mm_and_ps(velec,cutoff_mask);
1461             velec            = _mm_andnot_ps(dummy_mask,velec);
1462             velecsum         = _mm_add_ps(velecsum,velec);
1463
1464             fscal            = felec;
1465
1466             fscal            = _mm_and_ps(fscal,cutoff_mask);
1467
1468             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1469
1470              /* Update vectorial force */
1471             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1472             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1473             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1474
1475             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1476             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1477             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1478
1479             }
1480
1481             /**************************
1482              * CALCULATE INTERACTIONS *
1483              **************************/
1484
1485             if (gmx_mm_any_lt(rsq33,rcutoff2))
1486             {
1487
1488             r33              = _mm_mul_ps(rsq33,rinv33);
1489             r33              = _mm_andnot_ps(dummy_mask,r33);
1490
1491             /* EWALD ELECTROSTATICS */
1492
1493             /* Analytical PME correction */
1494             zeta2            = _mm_mul_ps(beta2,rsq33);
1495             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
1496             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1497             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1498             felec            = _mm_mul_ps(qq33,felec);
1499             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1500             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
1501             velec            = _mm_mul_ps(qq33,velec);
1502
1503             d                = _mm_sub_ps(r33,rswitch);
1504             d                = _mm_max_ps(d,_mm_setzero_ps());
1505             d2               = _mm_mul_ps(d,d);
1506             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1507
1508             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1509
1510             /* Evaluate switch function */
1511             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1512             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
1513             velec            = _mm_mul_ps(velec,sw);
1514             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
1515
1516             /* Update potential sum for this i atom from the interaction with this j atom. */
1517             velec            = _mm_and_ps(velec,cutoff_mask);
1518             velec            = _mm_andnot_ps(dummy_mask,velec);
1519             velecsum         = _mm_add_ps(velecsum,velec);
1520
1521             fscal            = felec;
1522
1523             fscal            = _mm_and_ps(fscal,cutoff_mask);
1524
1525             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1526
1527              /* Update vectorial force */
1528             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1529             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1530             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1531
1532             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1533             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1534             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1535
1536             }
1537
1538             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1539             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1540             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1541             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1542
1543             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1544                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1545                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1546
1547             /* Inner loop uses 552 flops */
1548         }
1549
1550         /* End of innermost loop */
1551
1552         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1553                                               f+i_coord_offset,fshift+i_shift_offset);
1554
1555         ggid                        = gid[iidx];
1556         /* Update potential energies */
1557         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1558         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1559
1560         /* Increment number of inner iterations */
1561         inneriter                  += j_index_end - j_index_start;
1562
1563         /* Outer loop uses 26 flops */
1564     }
1565
1566     /* Increment number of outer iterations */
1567     outeriter        += nri;
1568
1569     /* Update outer/inner flops */
1570
1571     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*552);
1572 }
1573 /*
1574  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_F_avx_128_fma_single
1575  * Electrostatics interaction: Ewald
1576  * VdW interaction:            LennardJones
1577  * Geometry:                   Water4-Water4
1578  * Calculate force/pot:        Force
1579  */
1580 void
1581 nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_F_avx_128_fma_single
1582                     (t_nblist * gmx_restrict                nlist,
1583                      rvec * gmx_restrict                    xx,
1584                      rvec * gmx_restrict                    ff,
1585                      t_forcerec * gmx_restrict              fr,
1586                      t_mdatoms * gmx_restrict               mdatoms,
1587                      nb_kernel_data_t * gmx_restrict        kernel_data,
1588                      t_nrnb * gmx_restrict                  nrnb)
1589 {
1590     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1591      * just 0 for non-waters.
1592      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1593      * jnr indices corresponding to data put in the four positions in the SIMD register.
1594      */
1595     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1596     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1597     int              jnrA,jnrB,jnrC,jnrD;
1598     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1599     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1600     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1601     real             rcutoff_scalar;
1602     real             *shiftvec,*fshift,*x,*f;
1603     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1604     real             scratch[4*DIM];
1605     __m128           fscal,rcutoff,rcutoff2,jidxall;
1606     int              vdwioffset0;
1607     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1608     int              vdwioffset1;
1609     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1610     int              vdwioffset2;
1611     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1612     int              vdwioffset3;
1613     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1614     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1615     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1616     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1617     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1618     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1619     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1620     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1621     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1622     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1623     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1624     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1625     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1626     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1627     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1628     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1629     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1630     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1631     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1632     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1633     real             *charge;
1634     int              nvdwtype;
1635     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1636     int              *vdwtype;
1637     real             *vdwparam;
1638     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1639     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1640     __m128i          ewitab;
1641     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1642     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1643     real             *ewtab;
1644     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1645     real             rswitch_scalar,d_scalar;
1646     __m128           dummy_mask,cutoff_mask;
1647     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1648     __m128           one     = _mm_set1_ps(1.0);
1649     __m128           two     = _mm_set1_ps(2.0);
1650     x                = xx[0];
1651     f                = ff[0];
1652
1653     nri              = nlist->nri;
1654     iinr             = nlist->iinr;
1655     jindex           = nlist->jindex;
1656     jjnr             = nlist->jjnr;
1657     shiftidx         = nlist->shift;
1658     gid              = nlist->gid;
1659     shiftvec         = fr->shift_vec[0];
1660     fshift           = fr->fshift[0];
1661     facel            = _mm_set1_ps(fr->epsfac);
1662     charge           = mdatoms->chargeA;
1663     nvdwtype         = fr->ntype;
1664     vdwparam         = fr->nbfp;
1665     vdwtype          = mdatoms->typeA;
1666
1667     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1668     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
1669     beta2            = _mm_mul_ps(beta,beta);
1670     beta3            = _mm_mul_ps(beta,beta2);
1671     ewtab            = fr->ic->tabq_coul_FDV0;
1672     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1673     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1674
1675     /* Setup water-specific parameters */
1676     inr              = nlist->iinr[0];
1677     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1678     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1679     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1680     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1681
1682     jq1              = _mm_set1_ps(charge[inr+1]);
1683     jq2              = _mm_set1_ps(charge[inr+2]);
1684     jq3              = _mm_set1_ps(charge[inr+3]);
1685     vdwjidx0A        = 2*vdwtype[inr+0];
1686     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1687     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1688     qq11             = _mm_mul_ps(iq1,jq1);
1689     qq12             = _mm_mul_ps(iq1,jq2);
1690     qq13             = _mm_mul_ps(iq1,jq3);
1691     qq21             = _mm_mul_ps(iq2,jq1);
1692     qq22             = _mm_mul_ps(iq2,jq2);
1693     qq23             = _mm_mul_ps(iq2,jq3);
1694     qq31             = _mm_mul_ps(iq3,jq1);
1695     qq32             = _mm_mul_ps(iq3,jq2);
1696     qq33             = _mm_mul_ps(iq3,jq3);
1697
1698     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1699     rcutoff_scalar   = fr->rcoulomb;
1700     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1701     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1702
1703     rswitch_scalar   = fr->rcoulomb_switch;
1704     rswitch          = _mm_set1_ps(rswitch_scalar);
1705     /* Setup switch parameters */
1706     d_scalar         = rcutoff_scalar-rswitch_scalar;
1707     d                = _mm_set1_ps(d_scalar);
1708     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
1709     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1710     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1711     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
1712     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1713     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1714
1715     /* Avoid stupid compiler warnings */
1716     jnrA = jnrB = jnrC = jnrD = 0;
1717     j_coord_offsetA = 0;
1718     j_coord_offsetB = 0;
1719     j_coord_offsetC = 0;
1720     j_coord_offsetD = 0;
1721
1722     outeriter        = 0;
1723     inneriter        = 0;
1724
1725     for(iidx=0;iidx<4*DIM;iidx++)
1726     {
1727         scratch[iidx] = 0.0;
1728     }
1729
1730     /* Start outer loop over neighborlists */
1731     for(iidx=0; iidx<nri; iidx++)
1732     {
1733         /* Load shift vector for this list */
1734         i_shift_offset   = DIM*shiftidx[iidx];
1735
1736         /* Load limits for loop over neighbors */
1737         j_index_start    = jindex[iidx];
1738         j_index_end      = jindex[iidx+1];
1739
1740         /* Get outer coordinate index */
1741         inr              = iinr[iidx];
1742         i_coord_offset   = DIM*inr;
1743
1744         /* Load i particle coords and add shift vector */
1745         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1746                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1747
1748         fix0             = _mm_setzero_ps();
1749         fiy0             = _mm_setzero_ps();
1750         fiz0             = _mm_setzero_ps();
1751         fix1             = _mm_setzero_ps();
1752         fiy1             = _mm_setzero_ps();
1753         fiz1             = _mm_setzero_ps();
1754         fix2             = _mm_setzero_ps();
1755         fiy2             = _mm_setzero_ps();
1756         fiz2             = _mm_setzero_ps();
1757         fix3             = _mm_setzero_ps();
1758         fiy3             = _mm_setzero_ps();
1759         fiz3             = _mm_setzero_ps();
1760
1761         /* Start inner kernel loop */
1762         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1763         {
1764
1765             /* Get j neighbor index, and coordinate index */
1766             jnrA             = jjnr[jidx];
1767             jnrB             = jjnr[jidx+1];
1768             jnrC             = jjnr[jidx+2];
1769             jnrD             = jjnr[jidx+3];
1770             j_coord_offsetA  = DIM*jnrA;
1771             j_coord_offsetB  = DIM*jnrB;
1772             j_coord_offsetC  = DIM*jnrC;
1773             j_coord_offsetD  = DIM*jnrD;
1774
1775             /* load j atom coordinates */
1776             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1777                                               x+j_coord_offsetC,x+j_coord_offsetD,
1778                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1779                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1780
1781             /* Calculate displacement vector */
1782             dx00             = _mm_sub_ps(ix0,jx0);
1783             dy00             = _mm_sub_ps(iy0,jy0);
1784             dz00             = _mm_sub_ps(iz0,jz0);
1785             dx11             = _mm_sub_ps(ix1,jx1);
1786             dy11             = _mm_sub_ps(iy1,jy1);
1787             dz11             = _mm_sub_ps(iz1,jz1);
1788             dx12             = _mm_sub_ps(ix1,jx2);
1789             dy12             = _mm_sub_ps(iy1,jy2);
1790             dz12             = _mm_sub_ps(iz1,jz2);
1791             dx13             = _mm_sub_ps(ix1,jx3);
1792             dy13             = _mm_sub_ps(iy1,jy3);
1793             dz13             = _mm_sub_ps(iz1,jz3);
1794             dx21             = _mm_sub_ps(ix2,jx1);
1795             dy21             = _mm_sub_ps(iy2,jy1);
1796             dz21             = _mm_sub_ps(iz2,jz1);
1797             dx22             = _mm_sub_ps(ix2,jx2);
1798             dy22             = _mm_sub_ps(iy2,jy2);
1799             dz22             = _mm_sub_ps(iz2,jz2);
1800             dx23             = _mm_sub_ps(ix2,jx3);
1801             dy23             = _mm_sub_ps(iy2,jy3);
1802             dz23             = _mm_sub_ps(iz2,jz3);
1803             dx31             = _mm_sub_ps(ix3,jx1);
1804             dy31             = _mm_sub_ps(iy3,jy1);
1805             dz31             = _mm_sub_ps(iz3,jz1);
1806             dx32             = _mm_sub_ps(ix3,jx2);
1807             dy32             = _mm_sub_ps(iy3,jy2);
1808             dz32             = _mm_sub_ps(iz3,jz2);
1809             dx33             = _mm_sub_ps(ix3,jx3);
1810             dy33             = _mm_sub_ps(iy3,jy3);
1811             dz33             = _mm_sub_ps(iz3,jz3);
1812
1813             /* Calculate squared distance and things based on it */
1814             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1815             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1816             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1817             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1818             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1819             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1820             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1821             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1822             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1823             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1824
1825             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1826             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1827             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1828             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1829             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1830             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1831             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1832             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1833             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1834             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1835
1836             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1837             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1838             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1839             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1840             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1841             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1842             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1843             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1844             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1845             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1846
1847             fjx0             = _mm_setzero_ps();
1848             fjy0             = _mm_setzero_ps();
1849             fjz0             = _mm_setzero_ps();
1850             fjx1             = _mm_setzero_ps();
1851             fjy1             = _mm_setzero_ps();
1852             fjz1             = _mm_setzero_ps();
1853             fjx2             = _mm_setzero_ps();
1854             fjy2             = _mm_setzero_ps();
1855             fjz2             = _mm_setzero_ps();
1856             fjx3             = _mm_setzero_ps();
1857             fjy3             = _mm_setzero_ps();
1858             fjz3             = _mm_setzero_ps();
1859
1860             /**************************
1861              * CALCULATE INTERACTIONS *
1862              **************************/
1863
1864             if (gmx_mm_any_lt(rsq00,rcutoff2))
1865             {
1866
1867             r00              = _mm_mul_ps(rsq00,rinv00);
1868
1869             /* LENNARD-JONES DISPERSION/REPULSION */
1870
1871             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1872             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1873             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1874             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1875             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1876
1877             d                = _mm_sub_ps(r00,rswitch);
1878             d                = _mm_max_ps(d,_mm_setzero_ps());
1879             d2               = _mm_mul_ps(d,d);
1880             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1881
1882             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1883
1884             /* Evaluate switch function */
1885             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1886             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1887             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1888
1889             fscal            = fvdw;
1890
1891             fscal            = _mm_and_ps(fscal,cutoff_mask);
1892
1893              /* Update vectorial force */
1894             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1895             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1896             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1897
1898             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1899             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1900             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1901
1902             }
1903
1904             /**************************
1905              * CALCULATE INTERACTIONS *
1906              **************************/
1907
1908             if (gmx_mm_any_lt(rsq11,rcutoff2))
1909             {
1910
1911             r11              = _mm_mul_ps(rsq11,rinv11);
1912
1913             /* EWALD ELECTROSTATICS */
1914
1915             /* Analytical PME correction */
1916             zeta2            = _mm_mul_ps(beta2,rsq11);
1917             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1918             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1919             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1920             felec            = _mm_mul_ps(qq11,felec);
1921             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1922             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
1923             velec            = _mm_mul_ps(qq11,velec);
1924
1925             d                = _mm_sub_ps(r11,rswitch);
1926             d                = _mm_max_ps(d,_mm_setzero_ps());
1927             d2               = _mm_mul_ps(d,d);
1928             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1929
1930             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1931
1932             /* Evaluate switch function */
1933             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1934             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
1935             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1936
1937             fscal            = felec;
1938
1939             fscal            = _mm_and_ps(fscal,cutoff_mask);
1940
1941              /* Update vectorial force */
1942             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1943             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1944             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1945
1946             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1947             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1948             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1949
1950             }
1951
1952             /**************************
1953              * CALCULATE INTERACTIONS *
1954              **************************/
1955
1956             if (gmx_mm_any_lt(rsq12,rcutoff2))
1957             {
1958
1959             r12              = _mm_mul_ps(rsq12,rinv12);
1960
1961             /* EWALD ELECTROSTATICS */
1962
1963             /* Analytical PME correction */
1964             zeta2            = _mm_mul_ps(beta2,rsq12);
1965             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1966             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1967             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1968             felec            = _mm_mul_ps(qq12,felec);
1969             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1970             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
1971             velec            = _mm_mul_ps(qq12,velec);
1972
1973             d                = _mm_sub_ps(r12,rswitch);
1974             d                = _mm_max_ps(d,_mm_setzero_ps());
1975             d2               = _mm_mul_ps(d,d);
1976             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1977
1978             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1979
1980             /* Evaluate switch function */
1981             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1982             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
1983             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1984
1985             fscal            = felec;
1986
1987             fscal            = _mm_and_ps(fscal,cutoff_mask);
1988
1989              /* Update vectorial force */
1990             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1991             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1992             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1993
1994             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1995             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1996             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1997
1998             }
1999
2000             /**************************
2001              * CALCULATE INTERACTIONS *
2002              **************************/
2003
2004             if (gmx_mm_any_lt(rsq13,rcutoff2))
2005             {
2006
2007             r13              = _mm_mul_ps(rsq13,rinv13);
2008
2009             /* EWALD ELECTROSTATICS */
2010
2011             /* Analytical PME correction */
2012             zeta2            = _mm_mul_ps(beta2,rsq13);
2013             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
2014             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2015             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2016             felec            = _mm_mul_ps(qq13,felec);
2017             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2018             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
2019             velec            = _mm_mul_ps(qq13,velec);
2020
2021             d                = _mm_sub_ps(r13,rswitch);
2022             d                = _mm_max_ps(d,_mm_setzero_ps());
2023             d2               = _mm_mul_ps(d,d);
2024             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2025
2026             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2027
2028             /* Evaluate switch function */
2029             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2030             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
2031             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
2032
2033             fscal            = felec;
2034
2035             fscal            = _mm_and_ps(fscal,cutoff_mask);
2036
2037              /* Update vectorial force */
2038             fix1             = _mm_macc_ps(dx13,fscal,fix1);
2039             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
2040             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
2041
2042             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
2043             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
2044             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
2045
2046             }
2047
2048             /**************************
2049              * CALCULATE INTERACTIONS *
2050              **************************/
2051
2052             if (gmx_mm_any_lt(rsq21,rcutoff2))
2053             {
2054
2055             r21              = _mm_mul_ps(rsq21,rinv21);
2056
2057             /* EWALD ELECTROSTATICS */
2058
2059             /* Analytical PME correction */
2060             zeta2            = _mm_mul_ps(beta2,rsq21);
2061             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
2062             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2063             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2064             felec            = _mm_mul_ps(qq21,felec);
2065             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2066             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
2067             velec            = _mm_mul_ps(qq21,velec);
2068
2069             d                = _mm_sub_ps(r21,rswitch);
2070             d                = _mm_max_ps(d,_mm_setzero_ps());
2071             d2               = _mm_mul_ps(d,d);
2072             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2073
2074             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2075
2076             /* Evaluate switch function */
2077             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2078             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
2079             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2080
2081             fscal            = felec;
2082
2083             fscal            = _mm_and_ps(fscal,cutoff_mask);
2084
2085              /* Update vectorial force */
2086             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2087             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2088             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2089
2090             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2091             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2092             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2093
2094             }
2095
2096             /**************************
2097              * CALCULATE INTERACTIONS *
2098              **************************/
2099
2100             if (gmx_mm_any_lt(rsq22,rcutoff2))
2101             {
2102
2103             r22              = _mm_mul_ps(rsq22,rinv22);
2104
2105             /* EWALD ELECTROSTATICS */
2106
2107             /* Analytical PME correction */
2108             zeta2            = _mm_mul_ps(beta2,rsq22);
2109             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
2110             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2111             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2112             felec            = _mm_mul_ps(qq22,felec);
2113             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2114             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
2115             velec            = _mm_mul_ps(qq22,velec);
2116
2117             d                = _mm_sub_ps(r22,rswitch);
2118             d                = _mm_max_ps(d,_mm_setzero_ps());
2119             d2               = _mm_mul_ps(d,d);
2120             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2121
2122             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2123
2124             /* Evaluate switch function */
2125             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2126             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
2127             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2128
2129             fscal            = felec;
2130
2131             fscal            = _mm_and_ps(fscal,cutoff_mask);
2132
2133              /* Update vectorial force */
2134             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2135             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2136             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2137
2138             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2139             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2140             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2141
2142             }
2143
2144             /**************************
2145              * CALCULATE INTERACTIONS *
2146              **************************/
2147
2148             if (gmx_mm_any_lt(rsq23,rcutoff2))
2149             {
2150
2151             r23              = _mm_mul_ps(rsq23,rinv23);
2152
2153             /* EWALD ELECTROSTATICS */
2154
2155             /* Analytical PME correction */
2156             zeta2            = _mm_mul_ps(beta2,rsq23);
2157             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
2158             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2159             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2160             felec            = _mm_mul_ps(qq23,felec);
2161             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2162             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
2163             velec            = _mm_mul_ps(qq23,velec);
2164
2165             d                = _mm_sub_ps(r23,rswitch);
2166             d                = _mm_max_ps(d,_mm_setzero_ps());
2167             d2               = _mm_mul_ps(d,d);
2168             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2169
2170             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2171
2172             /* Evaluate switch function */
2173             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2174             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
2175             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
2176
2177             fscal            = felec;
2178
2179             fscal            = _mm_and_ps(fscal,cutoff_mask);
2180
2181              /* Update vectorial force */
2182             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2183             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2184             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2185
2186             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2187             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2188             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2189
2190             }
2191
2192             /**************************
2193              * CALCULATE INTERACTIONS *
2194              **************************/
2195
2196             if (gmx_mm_any_lt(rsq31,rcutoff2))
2197             {
2198
2199             r31              = _mm_mul_ps(rsq31,rinv31);
2200
2201             /* EWALD ELECTROSTATICS */
2202
2203             /* Analytical PME correction */
2204             zeta2            = _mm_mul_ps(beta2,rsq31);
2205             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
2206             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2207             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2208             felec            = _mm_mul_ps(qq31,felec);
2209             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2210             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
2211             velec            = _mm_mul_ps(qq31,velec);
2212
2213             d                = _mm_sub_ps(r31,rswitch);
2214             d                = _mm_max_ps(d,_mm_setzero_ps());
2215             d2               = _mm_mul_ps(d,d);
2216             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2217
2218             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2219
2220             /* Evaluate switch function */
2221             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2222             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
2223             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
2224
2225             fscal            = felec;
2226
2227             fscal            = _mm_and_ps(fscal,cutoff_mask);
2228
2229              /* Update vectorial force */
2230             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2231             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2232             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2233
2234             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2235             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2236             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2237
2238             }
2239
2240             /**************************
2241              * CALCULATE INTERACTIONS *
2242              **************************/
2243
2244             if (gmx_mm_any_lt(rsq32,rcutoff2))
2245             {
2246
2247             r32              = _mm_mul_ps(rsq32,rinv32);
2248
2249             /* EWALD ELECTROSTATICS */
2250
2251             /* Analytical PME correction */
2252             zeta2            = _mm_mul_ps(beta2,rsq32);
2253             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
2254             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2255             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2256             felec            = _mm_mul_ps(qq32,felec);
2257             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2258             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
2259             velec            = _mm_mul_ps(qq32,velec);
2260
2261             d                = _mm_sub_ps(r32,rswitch);
2262             d                = _mm_max_ps(d,_mm_setzero_ps());
2263             d2               = _mm_mul_ps(d,d);
2264             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2265
2266             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2267
2268             /* Evaluate switch function */
2269             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2270             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
2271             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
2272
2273             fscal            = felec;
2274
2275             fscal            = _mm_and_ps(fscal,cutoff_mask);
2276
2277              /* Update vectorial force */
2278             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2279             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2280             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2281
2282             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2283             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2284             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2285
2286             }
2287
2288             /**************************
2289              * CALCULATE INTERACTIONS *
2290              **************************/
2291
2292             if (gmx_mm_any_lt(rsq33,rcutoff2))
2293             {
2294
2295             r33              = _mm_mul_ps(rsq33,rinv33);
2296
2297             /* EWALD ELECTROSTATICS */
2298
2299             /* Analytical PME correction */
2300             zeta2            = _mm_mul_ps(beta2,rsq33);
2301             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
2302             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2303             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2304             felec            = _mm_mul_ps(qq33,felec);
2305             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2306             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
2307             velec            = _mm_mul_ps(qq33,velec);
2308
2309             d                = _mm_sub_ps(r33,rswitch);
2310             d                = _mm_max_ps(d,_mm_setzero_ps());
2311             d2               = _mm_mul_ps(d,d);
2312             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2313
2314             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2315
2316             /* Evaluate switch function */
2317             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2318             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
2319             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2320
2321             fscal            = felec;
2322
2323             fscal            = _mm_and_ps(fscal,cutoff_mask);
2324
2325              /* Update vectorial force */
2326             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2327             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2328             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2329
2330             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2331             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2332             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2333
2334             }
2335
2336             fjptrA             = f+j_coord_offsetA;
2337             fjptrB             = f+j_coord_offsetB;
2338             fjptrC             = f+j_coord_offsetC;
2339             fjptrD             = f+j_coord_offsetD;
2340
2341             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2342                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2343                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2344
2345             /* Inner loop uses 512 flops */
2346         }
2347
2348         if(jidx<j_index_end)
2349         {
2350
2351             /* Get j neighbor index, and coordinate index */
2352             jnrlistA         = jjnr[jidx];
2353             jnrlistB         = jjnr[jidx+1];
2354             jnrlistC         = jjnr[jidx+2];
2355             jnrlistD         = jjnr[jidx+3];
2356             /* Sign of each element will be negative for non-real atoms.
2357              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2358              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2359              */
2360             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2361             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2362             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2363             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2364             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2365             j_coord_offsetA  = DIM*jnrA;
2366             j_coord_offsetB  = DIM*jnrB;
2367             j_coord_offsetC  = DIM*jnrC;
2368             j_coord_offsetD  = DIM*jnrD;
2369
2370             /* load j atom coordinates */
2371             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2372                                               x+j_coord_offsetC,x+j_coord_offsetD,
2373                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2374                                               &jy2,&jz2,&jx3,&jy3,&jz3);
2375
2376             /* Calculate displacement vector */
2377             dx00             = _mm_sub_ps(ix0,jx0);
2378             dy00             = _mm_sub_ps(iy0,jy0);
2379             dz00             = _mm_sub_ps(iz0,jz0);
2380             dx11             = _mm_sub_ps(ix1,jx1);
2381             dy11             = _mm_sub_ps(iy1,jy1);
2382             dz11             = _mm_sub_ps(iz1,jz1);
2383             dx12             = _mm_sub_ps(ix1,jx2);
2384             dy12             = _mm_sub_ps(iy1,jy2);
2385             dz12             = _mm_sub_ps(iz1,jz2);
2386             dx13             = _mm_sub_ps(ix1,jx3);
2387             dy13             = _mm_sub_ps(iy1,jy3);
2388             dz13             = _mm_sub_ps(iz1,jz3);
2389             dx21             = _mm_sub_ps(ix2,jx1);
2390             dy21             = _mm_sub_ps(iy2,jy1);
2391             dz21             = _mm_sub_ps(iz2,jz1);
2392             dx22             = _mm_sub_ps(ix2,jx2);
2393             dy22             = _mm_sub_ps(iy2,jy2);
2394             dz22             = _mm_sub_ps(iz2,jz2);
2395             dx23             = _mm_sub_ps(ix2,jx3);
2396             dy23             = _mm_sub_ps(iy2,jy3);
2397             dz23             = _mm_sub_ps(iz2,jz3);
2398             dx31             = _mm_sub_ps(ix3,jx1);
2399             dy31             = _mm_sub_ps(iy3,jy1);
2400             dz31             = _mm_sub_ps(iz3,jz1);
2401             dx32             = _mm_sub_ps(ix3,jx2);
2402             dy32             = _mm_sub_ps(iy3,jy2);
2403             dz32             = _mm_sub_ps(iz3,jz2);
2404             dx33             = _mm_sub_ps(ix3,jx3);
2405             dy33             = _mm_sub_ps(iy3,jy3);
2406             dz33             = _mm_sub_ps(iz3,jz3);
2407
2408             /* Calculate squared distance and things based on it */
2409             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
2410             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2411             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2412             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
2413             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2414             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2415             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
2416             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
2417             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
2418             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2419
2420             rinv00           = gmx_mm_invsqrt_ps(rsq00);
2421             rinv11           = gmx_mm_invsqrt_ps(rsq11);
2422             rinv12           = gmx_mm_invsqrt_ps(rsq12);
2423             rinv13           = gmx_mm_invsqrt_ps(rsq13);
2424             rinv21           = gmx_mm_invsqrt_ps(rsq21);
2425             rinv22           = gmx_mm_invsqrt_ps(rsq22);
2426             rinv23           = gmx_mm_invsqrt_ps(rsq23);
2427             rinv31           = gmx_mm_invsqrt_ps(rsq31);
2428             rinv32           = gmx_mm_invsqrt_ps(rsq32);
2429             rinv33           = gmx_mm_invsqrt_ps(rsq33);
2430
2431             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
2432             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
2433             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
2434             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
2435             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
2436             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
2437             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
2438             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
2439             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
2440             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
2441
2442             fjx0             = _mm_setzero_ps();
2443             fjy0             = _mm_setzero_ps();
2444             fjz0             = _mm_setzero_ps();
2445             fjx1             = _mm_setzero_ps();
2446             fjy1             = _mm_setzero_ps();
2447             fjz1             = _mm_setzero_ps();
2448             fjx2             = _mm_setzero_ps();
2449             fjy2             = _mm_setzero_ps();
2450             fjz2             = _mm_setzero_ps();
2451             fjx3             = _mm_setzero_ps();
2452             fjy3             = _mm_setzero_ps();
2453             fjz3             = _mm_setzero_ps();
2454
2455             /**************************
2456              * CALCULATE INTERACTIONS *
2457              **************************/
2458
2459             if (gmx_mm_any_lt(rsq00,rcutoff2))
2460             {
2461
2462             r00              = _mm_mul_ps(rsq00,rinv00);
2463             r00              = _mm_andnot_ps(dummy_mask,r00);
2464
2465             /* LENNARD-JONES DISPERSION/REPULSION */
2466
2467             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2468             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
2469             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
2470             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
2471             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
2472
2473             d                = _mm_sub_ps(r00,rswitch);
2474             d                = _mm_max_ps(d,_mm_setzero_ps());
2475             d2               = _mm_mul_ps(d,d);
2476             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2477
2478             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2479
2480             /* Evaluate switch function */
2481             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2482             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
2483             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
2484
2485             fscal            = fvdw;
2486
2487             fscal            = _mm_and_ps(fscal,cutoff_mask);
2488
2489             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2490
2491              /* Update vectorial force */
2492             fix0             = _mm_macc_ps(dx00,fscal,fix0);
2493             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
2494             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
2495
2496             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
2497             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
2498             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
2499
2500             }
2501
2502             /**************************
2503              * CALCULATE INTERACTIONS *
2504              **************************/
2505
2506             if (gmx_mm_any_lt(rsq11,rcutoff2))
2507             {
2508
2509             r11              = _mm_mul_ps(rsq11,rinv11);
2510             r11              = _mm_andnot_ps(dummy_mask,r11);
2511
2512             /* EWALD ELECTROSTATICS */
2513
2514             /* Analytical PME correction */
2515             zeta2            = _mm_mul_ps(beta2,rsq11);
2516             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
2517             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2518             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2519             felec            = _mm_mul_ps(qq11,felec);
2520             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2521             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
2522             velec            = _mm_mul_ps(qq11,velec);
2523
2524             d                = _mm_sub_ps(r11,rswitch);
2525             d                = _mm_max_ps(d,_mm_setzero_ps());
2526             d2               = _mm_mul_ps(d,d);
2527             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2528
2529             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2530
2531             /* Evaluate switch function */
2532             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2533             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
2534             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
2535
2536             fscal            = felec;
2537
2538             fscal            = _mm_and_ps(fscal,cutoff_mask);
2539
2540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2541
2542              /* Update vectorial force */
2543             fix1             = _mm_macc_ps(dx11,fscal,fix1);
2544             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
2545             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
2546
2547             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
2548             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
2549             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
2550
2551             }
2552
2553             /**************************
2554              * CALCULATE INTERACTIONS *
2555              **************************/
2556
2557             if (gmx_mm_any_lt(rsq12,rcutoff2))
2558             {
2559
2560             r12              = _mm_mul_ps(rsq12,rinv12);
2561             r12              = _mm_andnot_ps(dummy_mask,r12);
2562
2563             /* EWALD ELECTROSTATICS */
2564
2565             /* Analytical PME correction */
2566             zeta2            = _mm_mul_ps(beta2,rsq12);
2567             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
2568             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2569             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2570             felec            = _mm_mul_ps(qq12,felec);
2571             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2572             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
2573             velec            = _mm_mul_ps(qq12,velec);
2574
2575             d                = _mm_sub_ps(r12,rswitch);
2576             d                = _mm_max_ps(d,_mm_setzero_ps());
2577             d2               = _mm_mul_ps(d,d);
2578             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2579
2580             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2581
2582             /* Evaluate switch function */
2583             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2584             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
2585             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
2586
2587             fscal            = felec;
2588
2589             fscal            = _mm_and_ps(fscal,cutoff_mask);
2590
2591             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2592
2593              /* Update vectorial force */
2594             fix1             = _mm_macc_ps(dx12,fscal,fix1);
2595             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
2596             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
2597
2598             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
2599             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
2600             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
2601
2602             }
2603
2604             /**************************
2605              * CALCULATE INTERACTIONS *
2606              **************************/
2607
2608             if (gmx_mm_any_lt(rsq13,rcutoff2))
2609             {
2610
2611             r13              = _mm_mul_ps(rsq13,rinv13);
2612             r13              = _mm_andnot_ps(dummy_mask,r13);
2613
2614             /* EWALD ELECTROSTATICS */
2615
2616             /* Analytical PME correction */
2617             zeta2            = _mm_mul_ps(beta2,rsq13);
2618             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
2619             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2620             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2621             felec            = _mm_mul_ps(qq13,felec);
2622             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2623             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv13);
2624             velec            = _mm_mul_ps(qq13,velec);
2625
2626             d                = _mm_sub_ps(r13,rswitch);
2627             d                = _mm_max_ps(d,_mm_setzero_ps());
2628             d2               = _mm_mul_ps(d,d);
2629             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2630
2631             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2632
2633             /* Evaluate switch function */
2634             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2635             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
2636             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
2637
2638             fscal            = felec;
2639
2640             fscal            = _mm_and_ps(fscal,cutoff_mask);
2641
2642             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2643
2644              /* Update vectorial force */
2645             fix1             = _mm_macc_ps(dx13,fscal,fix1);
2646             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
2647             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
2648
2649             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
2650             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
2651             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
2652
2653             }
2654
2655             /**************************
2656              * CALCULATE INTERACTIONS *
2657              **************************/
2658
2659             if (gmx_mm_any_lt(rsq21,rcutoff2))
2660             {
2661
2662             r21              = _mm_mul_ps(rsq21,rinv21);
2663             r21              = _mm_andnot_ps(dummy_mask,r21);
2664
2665             /* EWALD ELECTROSTATICS */
2666
2667             /* Analytical PME correction */
2668             zeta2            = _mm_mul_ps(beta2,rsq21);
2669             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
2670             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2671             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2672             felec            = _mm_mul_ps(qq21,felec);
2673             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2674             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
2675             velec            = _mm_mul_ps(qq21,velec);
2676
2677             d                = _mm_sub_ps(r21,rswitch);
2678             d                = _mm_max_ps(d,_mm_setzero_ps());
2679             d2               = _mm_mul_ps(d,d);
2680             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2681
2682             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2683
2684             /* Evaluate switch function */
2685             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2686             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
2687             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2688
2689             fscal            = felec;
2690
2691             fscal            = _mm_and_ps(fscal,cutoff_mask);
2692
2693             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2694
2695              /* Update vectorial force */
2696             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2697             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2698             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2699
2700             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2701             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2702             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2703
2704             }
2705
2706             /**************************
2707              * CALCULATE INTERACTIONS *
2708              **************************/
2709
2710             if (gmx_mm_any_lt(rsq22,rcutoff2))
2711             {
2712
2713             r22              = _mm_mul_ps(rsq22,rinv22);
2714             r22              = _mm_andnot_ps(dummy_mask,r22);
2715
2716             /* EWALD ELECTROSTATICS */
2717
2718             /* Analytical PME correction */
2719             zeta2            = _mm_mul_ps(beta2,rsq22);
2720             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
2721             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2722             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2723             felec            = _mm_mul_ps(qq22,felec);
2724             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2725             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
2726             velec            = _mm_mul_ps(qq22,velec);
2727
2728             d                = _mm_sub_ps(r22,rswitch);
2729             d                = _mm_max_ps(d,_mm_setzero_ps());
2730             d2               = _mm_mul_ps(d,d);
2731             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2732
2733             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2734
2735             /* Evaluate switch function */
2736             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2737             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
2738             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2739
2740             fscal            = felec;
2741
2742             fscal            = _mm_and_ps(fscal,cutoff_mask);
2743
2744             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2745
2746              /* Update vectorial force */
2747             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2748             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2749             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2750
2751             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2752             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2753             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2754
2755             }
2756
2757             /**************************
2758              * CALCULATE INTERACTIONS *
2759              **************************/
2760
2761             if (gmx_mm_any_lt(rsq23,rcutoff2))
2762             {
2763
2764             r23              = _mm_mul_ps(rsq23,rinv23);
2765             r23              = _mm_andnot_ps(dummy_mask,r23);
2766
2767             /* EWALD ELECTROSTATICS */
2768
2769             /* Analytical PME correction */
2770             zeta2            = _mm_mul_ps(beta2,rsq23);
2771             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
2772             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2773             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2774             felec            = _mm_mul_ps(qq23,felec);
2775             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2776             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv23);
2777             velec            = _mm_mul_ps(qq23,velec);
2778
2779             d                = _mm_sub_ps(r23,rswitch);
2780             d                = _mm_max_ps(d,_mm_setzero_ps());
2781             d2               = _mm_mul_ps(d,d);
2782             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2783
2784             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2785
2786             /* Evaluate switch function */
2787             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2788             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
2789             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
2790
2791             fscal            = felec;
2792
2793             fscal            = _mm_and_ps(fscal,cutoff_mask);
2794
2795             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2796
2797              /* Update vectorial force */
2798             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2799             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2800             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2801
2802             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2803             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2804             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2805
2806             }
2807
2808             /**************************
2809              * CALCULATE INTERACTIONS *
2810              **************************/
2811
2812             if (gmx_mm_any_lt(rsq31,rcutoff2))
2813             {
2814
2815             r31              = _mm_mul_ps(rsq31,rinv31);
2816             r31              = _mm_andnot_ps(dummy_mask,r31);
2817
2818             /* EWALD ELECTROSTATICS */
2819
2820             /* Analytical PME correction */
2821             zeta2            = _mm_mul_ps(beta2,rsq31);
2822             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
2823             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2824             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2825             felec            = _mm_mul_ps(qq31,felec);
2826             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2827             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv31);
2828             velec            = _mm_mul_ps(qq31,velec);
2829
2830             d                = _mm_sub_ps(r31,rswitch);
2831             d                = _mm_max_ps(d,_mm_setzero_ps());
2832             d2               = _mm_mul_ps(d,d);
2833             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2834
2835             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2836
2837             /* Evaluate switch function */
2838             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2839             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
2840             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
2841
2842             fscal            = felec;
2843
2844             fscal            = _mm_and_ps(fscal,cutoff_mask);
2845
2846             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2847
2848              /* Update vectorial force */
2849             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2850             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2851             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2852
2853             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2854             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2855             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2856
2857             }
2858
2859             /**************************
2860              * CALCULATE INTERACTIONS *
2861              **************************/
2862
2863             if (gmx_mm_any_lt(rsq32,rcutoff2))
2864             {
2865
2866             r32              = _mm_mul_ps(rsq32,rinv32);
2867             r32              = _mm_andnot_ps(dummy_mask,r32);
2868
2869             /* EWALD ELECTROSTATICS */
2870
2871             /* Analytical PME correction */
2872             zeta2            = _mm_mul_ps(beta2,rsq32);
2873             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
2874             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2875             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2876             felec            = _mm_mul_ps(qq32,felec);
2877             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2878             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv32);
2879             velec            = _mm_mul_ps(qq32,velec);
2880
2881             d                = _mm_sub_ps(r32,rswitch);
2882             d                = _mm_max_ps(d,_mm_setzero_ps());
2883             d2               = _mm_mul_ps(d,d);
2884             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2885
2886             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2887
2888             /* Evaluate switch function */
2889             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2890             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
2891             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
2892
2893             fscal            = felec;
2894
2895             fscal            = _mm_and_ps(fscal,cutoff_mask);
2896
2897             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2898
2899              /* Update vectorial force */
2900             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2901             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2902             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2903
2904             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2905             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2906             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2907
2908             }
2909
2910             /**************************
2911              * CALCULATE INTERACTIONS *
2912              **************************/
2913
2914             if (gmx_mm_any_lt(rsq33,rcutoff2))
2915             {
2916
2917             r33              = _mm_mul_ps(rsq33,rinv33);
2918             r33              = _mm_andnot_ps(dummy_mask,r33);
2919
2920             /* EWALD ELECTROSTATICS */
2921
2922             /* Analytical PME correction */
2923             zeta2            = _mm_mul_ps(beta2,rsq33);
2924             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
2925             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2926             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2927             felec            = _mm_mul_ps(qq33,felec);
2928             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
2929             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv33);
2930             velec            = _mm_mul_ps(qq33,velec);
2931
2932             d                = _mm_sub_ps(r33,rswitch);
2933             d                = _mm_max_ps(d,_mm_setzero_ps());
2934             d2               = _mm_mul_ps(d,d);
2935             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2936
2937             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2938
2939             /* Evaluate switch function */
2940             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2941             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
2942             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2943
2944             fscal            = felec;
2945
2946             fscal            = _mm_and_ps(fscal,cutoff_mask);
2947
2948             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2949
2950              /* Update vectorial force */
2951             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2952             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2953             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2954
2955             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2956             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2957             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2958
2959             }
2960
2961             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2962             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2963             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2964             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2965
2966             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2967                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2968                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2969
2970             /* Inner loop uses 522 flops */
2971         }
2972
2973         /* End of innermost loop */
2974
2975         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2976                                               f+i_coord_offset,fshift+i_shift_offset);
2977
2978         /* Increment number of inner iterations */
2979         inneriter                  += j_index_end - j_index_start;
2980
2981         /* Outer loop uses 24 flops */
2982     }
2983
2984     /* Increment number of outer iterations */
2985     outeriter        += nri;
2986
2987     /* Update outer/inner flops */
2988
2989     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*522);
2990 }