src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_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 SSE, 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           tx,ty,tz,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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128i          ewitab;
  93     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 127     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 128     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 129     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm_setzero_ps();
 165         fiy0             = _mm_setzero_ps();
 166         fiz0             = _mm_setzero_ps();
 167         fix1             = _mm_setzero_ps();
 168         fiy1             = _mm_setzero_ps();
 169         fiz1             = _mm_setzero_ps();
 170         fix2             = _mm_setzero_ps();
 171         fiy2             = _mm_setzero_ps();
 172         fiz2             = _mm_setzero_ps();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vvdwsum          = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_ps(ix0,jx0);
 199             dy00             = _mm_sub_ps(iy0,jy0);
 200             dz00             = _mm_sub_ps(iz0,jz0);
 201             dx10             = _mm_sub_ps(ix1,jx0);
 202             dy10             = _mm_sub_ps(iy1,jy0);
 203             dz10             = _mm_sub_ps(iz1,jz0);
 204             dx20             = _mm_sub_ps(ix2,jx0);
 205             dy20             = _mm_sub_ps(iy2,jy0);
 206             dz20             = _mm_sub_ps(iz2,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 212
 213             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 214             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 215             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 216
 217             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 218             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 219             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             fjx0             = _mm_setzero_ps();
 230             fjy0             = _mm_setzero_ps();
 231             fjz0             = _mm_setzero_ps();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _mm_mul_ps(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm_mul_ps(iq0,jq0);
 241             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,
 243                                          vdwparam+vdwioffset0+vdwjidx0C,
 244                                          vdwparam+vdwioffset0+vdwjidx0D,
 245                                          &c6_00,&c12_00);
 246
 247             /* Calculate table index by multiplying r with table scale and truncate to integer */
 248             rt               = _mm_mul_ps(r00,vftabscale);
 249             vfitab           = _mm_cvttps_epi32(rt);
 250             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 251             vfitab           = _mm_slli_epi32(vfitab,3);
 252
 253             /* EWALD ELECTROSTATICS */
 254
 255             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 256             ewrt             = _mm_mul_ps(r00,ewtabscale);
 257             ewitab           = _mm_cvttps_epi32(ewrt);
 258             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 259             ewitab           = _mm_slli_epi32(ewitab,2);
 260             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 261             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 262             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 263             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 264             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 265             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 266             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 267             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 268             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 269
 270             /* CUBIC SPLINE TABLE DISPERSION */
 271             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 272             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 273             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 274             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 275             _MM_TRANSPOSE4_PS(Y,F,G,H);
 276             Heps             = _mm_mul_ps(vfeps,H);
 277             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 278             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 279             vvdw6            = _mm_mul_ps(c6_00,VV);
 280             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 281             fvdw6            = _mm_mul_ps(c6_00,FF);
 282
 283             /* CUBIC SPLINE TABLE REPULSION */
 284             vfitab           = _mm_add_epi32(vfitab,ifour);
 285             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 286             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 287             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 288             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 289             _MM_TRANSPOSE4_PS(Y,F,G,H);
 290             Heps             = _mm_mul_ps(vfeps,H);
 291             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 292             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 293             vvdw12           = _mm_mul_ps(c12_00,VV);
 294             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 295             fvdw12           = _mm_mul_ps(c12_00,FF);
 296             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 297             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 302
 303             fscal            = _mm_add_ps(felec,fvdw);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_ps(fscal,dx00);
 307             ty               = _mm_mul_ps(fscal,dy00);
 308             tz               = _mm_mul_ps(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_add_ps(fix0,tx);
 312             fiy0             = _mm_add_ps(fiy0,ty);
 313             fiz0             = _mm_add_ps(fiz0,tz);
 314
 315             fjx0             = _mm_add_ps(fjx0,tx);
 316             fjy0             = _mm_add_ps(fjy0,ty);
 317             fjz0             = _mm_add_ps(fjz0,tz);
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             r10              = _mm_mul_ps(rsq10,rinv10);
 324
 325             /* Compute parameters for interactions between i and j atoms */
 326             qq10             = _mm_mul_ps(iq1,jq0);
 327
 328             /* EWALD ELECTROSTATICS */
 329
 330             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 331             ewrt             = _mm_mul_ps(r10,ewtabscale);
 332             ewitab           = _mm_cvttps_epi32(ewrt);
 333             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 334             ewitab           = _mm_slli_epi32(ewitab,2);
 335             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 336             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 337             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 338             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 339             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 340             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 341             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 342             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 343             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velecsum         = _mm_add_ps(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm_mul_ps(fscal,dx10);
 352             ty               = _mm_mul_ps(fscal,dy10);
 353             tz               = _mm_mul_ps(fscal,dz10);
 354
 355             /* Update vectorial force */
 356             fix1             = _mm_add_ps(fix1,tx);
 357             fiy1             = _mm_add_ps(fiy1,ty);
 358             fiz1             = _mm_add_ps(fiz1,tz);
 359
 360             fjx0             = _mm_add_ps(fjx0,tx);
 361             fjy0             = _mm_add_ps(fjy0,ty);
 362             fjz0             = _mm_add_ps(fjz0,tz);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             r20              = _mm_mul_ps(rsq20,rinv20);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq20             = _mm_mul_ps(iq2,jq0);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 376             ewrt             = _mm_mul_ps(r20,ewtabscale);
 377             ewitab           = _mm_cvttps_epi32(ewrt);
 378             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 379             ewitab           = _mm_slli_epi32(ewitab,2);
 380             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 381             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 382             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 383             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 384             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 385             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 386             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 387             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 388             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm_mul_ps(fscal,dx20);
 397             ty               = _mm_mul_ps(fscal,dy20);
 398             tz               = _mm_mul_ps(fscal,dz20);
 399
 400             /* Update vectorial force */
 401             fix2             = _mm_add_ps(fix2,tx);
 402             fiy2             = _mm_add_ps(fiy2,ty);
 403             fiz2             = _mm_add_ps(fiz2,tz);
 404
 405             fjx0             = _mm_add_ps(fjx0,tx);
 406             fjy0             = _mm_add_ps(fjy0,ty);
 407             fjz0             = _mm_add_ps(fjz0,tz);
 408
 409             fjptrA             = f+j_coord_offsetA;
 410             fjptrB             = f+j_coord_offsetB;
 411             fjptrC             = f+j_coord_offsetC;
 412             fjptrD             = f+j_coord_offsetD;
 413
 414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 415
 416             /* Inner loop uses 157 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             /* Sign of each element will be negative for non-real atoms.
 428              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 429              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 430              */
 431             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 432             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 433             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 434             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 435             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                               x+j_coord_offsetC,x+j_coord_offsetD,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_ps(ix0,jx0);
 448             dy00             = _mm_sub_ps(iy0,jy0);
 449             dz00             = _mm_sub_ps(iz0,jz0);
 450             dx10             = _mm_sub_ps(ix1,jx0);
 451             dy10             = _mm_sub_ps(iy1,jy0);
 452             dz10             = _mm_sub_ps(iz1,jz0);
 453             dx20             = _mm_sub_ps(ix2,jx0);
 454             dy20             = _mm_sub_ps(iy2,jy0);
 455             dz20             = _mm_sub_ps(iz2,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 461
 462             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 463             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 464             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 465
 466             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 467             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 468             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 469
 470             /* Load parameters for j particles */
 471             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 472                                                               charge+jnrC+0,charge+jnrD+0);
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474             vdwjidx0B        = 2*vdwtype[jnrB+0];
 475             vdwjidx0C        = 2*vdwtype[jnrC+0];
 476             vdwjidx0D        = 2*vdwtype[jnrD+0];
 477
 478             fjx0             = _mm_setzero_ps();
 479             fjy0             = _mm_setzero_ps();
 480             fjz0             = _mm_setzero_ps();
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             r00              = _mm_mul_ps(rsq00,rinv00);
 487             r00              = _mm_andnot_ps(dummy_mask,r00);
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq00             = _mm_mul_ps(iq0,jq0);
 491             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 492                                          vdwparam+vdwioffset0+vdwjidx0B,
 493                                          vdwparam+vdwioffset0+vdwjidx0C,
 494                                          vdwparam+vdwioffset0+vdwjidx0D,
 495                                          &c6_00,&c12_00);
 496
 497             /* Calculate table index by multiplying r with table scale and truncate to integer */
 498             rt               = _mm_mul_ps(r00,vftabscale);
 499             vfitab           = _mm_cvttps_epi32(rt);
 500             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 501             vfitab           = _mm_slli_epi32(vfitab,3);
 502
 503             /* EWALD ELECTROSTATICS */
 504
 505             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 506             ewrt             = _mm_mul_ps(r00,ewtabscale);
 507             ewitab           = _mm_cvttps_epi32(ewrt);
 508             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 509             ewitab           = _mm_slli_epi32(ewitab,2);
 510             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 511             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 512             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 513             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 514             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 515             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 516             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 517             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 518             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 519
 520             /* CUBIC SPLINE TABLE DISPERSION */
 521             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 522             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 523             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 524             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 525             _MM_TRANSPOSE4_PS(Y,F,G,H);
 526             Heps             = _mm_mul_ps(vfeps,H);
 527             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 528             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 529             vvdw6            = _mm_mul_ps(c6_00,VV);
 530             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 531             fvdw6            = _mm_mul_ps(c6_00,FF);
 532
 533             /* CUBIC SPLINE TABLE REPULSION */
 534             vfitab           = _mm_add_epi32(vfitab,ifour);
 535             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 536             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 537             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 538             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 539             _MM_TRANSPOSE4_PS(Y,F,G,H);
 540             Heps             = _mm_mul_ps(vfeps,H);
 541             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 542             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 543             vvdw12           = _mm_mul_ps(c12_00,VV);
 544             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 545             fvdw12           = _mm_mul_ps(c12_00,FF);
 546             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 547             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm_andnot_ps(dummy_mask,velec);
 551             velecsum         = _mm_add_ps(velecsum,velec);
 552             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 553             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 554
 555             fscal            = _mm_add_ps(felec,fvdw);
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_ps(fscal,dx00);
 561             ty               = _mm_mul_ps(fscal,dy00);
 562             tz               = _mm_mul_ps(fscal,dz00);
 563
 564             /* Update vectorial force */
 565             fix0             = _mm_add_ps(fix0,tx);
 566             fiy0             = _mm_add_ps(fiy0,ty);
 567             fiz0             = _mm_add_ps(fiz0,tz);
 568
 569             fjx0             = _mm_add_ps(fjx0,tx);
 570             fjy0             = _mm_add_ps(fjy0,ty);
 571             fjz0             = _mm_add_ps(fjz0,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r10              = _mm_mul_ps(rsq10,rinv10);
 578             r10              = _mm_andnot_ps(dummy_mask,r10);
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq10             = _mm_mul_ps(iq1,jq0);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 586             ewrt             = _mm_mul_ps(r10,ewtabscale);
 587             ewitab           = _mm_cvttps_epi32(ewrt);
 588             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 589             ewitab           = _mm_slli_epi32(ewitab,2);
 590             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 591             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 592             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 593             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 594             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 595             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 596             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 597             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 598             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm_andnot_ps(dummy_mask,velec);
 602             velecsum         = _mm_add_ps(velecsum,velec);
 603
 604             fscal            = felec;
 605
 606             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 607
 608             /* Calculate temporary vectorial force */
 609             tx               = _mm_mul_ps(fscal,dx10);
 610             ty               = _mm_mul_ps(fscal,dy10);
 611             tz               = _mm_mul_ps(fscal,dz10);
 612
 613             /* Update vectorial force */
 614             fix1             = _mm_add_ps(fix1,tx);
 615             fiy1             = _mm_add_ps(fiy1,ty);
 616             fiz1             = _mm_add_ps(fiz1,tz);
 617
 618             fjx0             = _mm_add_ps(fjx0,tx);
 619             fjy0             = _mm_add_ps(fjy0,ty);
 620             fjz0             = _mm_add_ps(fjz0,tz);
 621
 622             /**************************
 623              * CALCULATE INTERACTIONS *
 624              **************************/
 625
 626             r20              = _mm_mul_ps(rsq20,rinv20);
 627             r20              = _mm_andnot_ps(dummy_mask,r20);
 628
 629             /* Compute parameters for interactions between i and j atoms */
 630             qq20             = _mm_mul_ps(iq2,jq0);
 631
 632             /* EWALD ELECTROSTATICS */
 633
 634             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 635             ewrt             = _mm_mul_ps(r20,ewtabscale);
 636             ewitab           = _mm_cvttps_epi32(ewrt);
 637             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 638             ewitab           = _mm_slli_epi32(ewitab,2);
 639             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 640             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 641             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 642             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 643             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 644             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 645             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 646             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 647             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 648
 649             /* Update potential sum for this i atom from the interaction with this j atom. */
 650             velec            = _mm_andnot_ps(dummy_mask,velec);
 651             velecsum         = _mm_add_ps(velecsum,velec);
 652
 653             fscal            = felec;
 654
 655             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 656
 657             /* Calculate temporary vectorial force */
 658             tx               = _mm_mul_ps(fscal,dx20);
 659             ty               = _mm_mul_ps(fscal,dy20);
 660             tz               = _mm_mul_ps(fscal,dz20);
 661
 662             /* Update vectorial force */
 663             fix2             = _mm_add_ps(fix2,tx);
 664             fiy2             = _mm_add_ps(fiy2,ty);
 665             fiz2             = _mm_add_ps(fiz2,tz);
 666
 667             fjx0             = _mm_add_ps(fjx0,tx);
 668             fjy0             = _mm_add_ps(fjy0,ty);
 669             fjz0             = _mm_add_ps(fjz0,tz);
 670
 671             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 672             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 673             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 674             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 675
 676             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 677
 678             /* Inner loop uses 160 flops */
 679         }
 680
 681         /* End of innermost loop */
 682
 683         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 684                                               f+i_coord_offset,fshift+i_shift_offset);
 685
 686         ggid                        = gid[iidx];
 687         /* Update potential energies */
 688         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 689         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 690
 691         /* Increment number of inner iterations */
 692         inneriter                  += j_index_end - j_index_start;
 693
 694         /* Outer loop uses 20 flops */
 695     }
 696
 697     /* Increment number of outer iterations */
 698     outeriter        += nri;
 699
 700     /* Update outer/inner flops */
 701
 702     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 703 }
 704 /*
 705  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_single
 706  * Electrostatics interaction: Ewald
 707  * VdW interaction:            CubicSplineTable
 708  * Geometry:                   Water3-Particle
 709  * Calculate force/pot:        Force
 710  */
 711 void
 712 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_single
 713                     (t_nblist * gmx_restrict                nlist,
 714                      rvec * gmx_restrict                    xx,
 715                      rvec * gmx_restrict                    ff,
 716                      t_forcerec * gmx_restrict              fr,
 717                      t_mdatoms * gmx_restrict               mdatoms,
 718                      nb_kernel_data_t * gmx_restrict        kernel_data,
 719                      t_nrnb * gmx_restrict                  nrnb)
 720 {
 721     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 722      * just 0 for non-waters.
 723      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 724      * jnr indices corresponding to data put in the four positions in the SIMD register.
 725      */
 726     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 727     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 728     int              jnrA,jnrB,jnrC,jnrD;
 729     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 730     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 731     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 732     real             rcutoff_scalar;
 733     real             *shiftvec,*fshift,*x,*f;
 734     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 735     real             scratch[4*DIM];
 736     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 737     int              vdwioffset0;
 738     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 739     int              vdwioffset1;
 740     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 741     int              vdwioffset2;
 742     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 743     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 744     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 745     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 746     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 747     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 748     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 749     real             *charge;
 750     int              nvdwtype;
 751     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 752     int              *vdwtype;
 753     real             *vdwparam;
 754     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 755     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 756     __m128i          vfitab;
 757     __m128i          ifour       = _mm_set1_epi32(4);
 758     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 759     real             *vftab;
 760     __m128i          ewitab;
 761     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 762     real             *ewtab;
 763     __m128           dummy_mask,cutoff_mask;
 764     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 765     __m128           one     = _mm_set1_ps(1.0);
 766     __m128           two     = _mm_set1_ps(2.0);
 767     x                = xx[0];
 768     f                = ff[0];
 769
 770     nri              = nlist->nri;
 771     iinr             = nlist->iinr;
 772     jindex           = nlist->jindex;
 773     jjnr             = nlist->jjnr;
 774     shiftidx         = nlist->shift;
 775     gid              = nlist->gid;
 776     shiftvec         = fr->shift_vec[0];
 777     fshift           = fr->fshift[0];
 778     facel            = _mm_set1_ps(fr->epsfac);
 779     charge           = mdatoms->chargeA;
 780     nvdwtype         = fr->ntype;
 781     vdwparam         = fr->nbfp;
 782     vdwtype          = mdatoms->typeA;
 783
 784     vftab            = kernel_data->table_vdw->data;
 785     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 786
 787     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 788     ewtab            = fr->ic->tabq_coul_F;
 789     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 790     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 791
 792     /* Setup water-specific parameters */
 793     inr              = nlist->iinr[0];
 794     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 795     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 796     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 797     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 798
 799     /* Avoid stupid compiler warnings */
 800     jnrA = jnrB = jnrC = jnrD = 0;
 801     j_coord_offsetA = 0;
 802     j_coord_offsetB = 0;
 803     j_coord_offsetC = 0;
 804     j_coord_offsetD = 0;
 805
 806     outeriter        = 0;
 807     inneriter        = 0;
 808
 809     for(iidx=0;iidx<4*DIM;iidx++)
 810     {
 811         scratch[iidx] = 0.0;
 812     }
 813
 814     /* Start outer loop over neighborlists */
 815     for(iidx=0; iidx<nri; iidx++)
 816     {
 817         /* Load shift vector for this list */
 818         i_shift_offset   = DIM*shiftidx[iidx];
 819
 820         /* Load limits for loop over neighbors */
 821         j_index_start    = jindex[iidx];
 822         j_index_end      = jindex[iidx+1];
 823
 824         /* Get outer coordinate index */
 825         inr              = iinr[iidx];
 826         i_coord_offset   = DIM*inr;
 827
 828         /* Load i particle coords and add shift vector */
 829         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 830                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 831
 832         fix0             = _mm_setzero_ps();
 833         fiy0             = _mm_setzero_ps();
 834         fiz0             = _mm_setzero_ps();
 835         fix1             = _mm_setzero_ps();
 836         fiy1             = _mm_setzero_ps();
 837         fiz1             = _mm_setzero_ps();
 838         fix2             = _mm_setzero_ps();
 839         fiy2             = _mm_setzero_ps();
 840         fiz2             = _mm_setzero_ps();
 841
 842         /* Start inner kernel loop */
 843         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 844         {
 845
 846             /* Get j neighbor index, and coordinate index */
 847             jnrA             = jjnr[jidx];
 848             jnrB             = jjnr[jidx+1];
 849             jnrC             = jjnr[jidx+2];
 850             jnrD             = jjnr[jidx+3];
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853             j_coord_offsetC  = DIM*jnrC;
 854             j_coord_offsetD  = DIM*jnrD;
 855
 856             /* load j atom coordinates */
 857             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 858                                               x+j_coord_offsetC,x+j_coord_offsetD,
 859                                               &jx0,&jy0,&jz0);
 860
 861             /* Calculate displacement vector */
 862             dx00             = _mm_sub_ps(ix0,jx0);
 863             dy00             = _mm_sub_ps(iy0,jy0);
 864             dz00             = _mm_sub_ps(iz0,jz0);
 865             dx10             = _mm_sub_ps(ix1,jx0);
 866             dy10             = _mm_sub_ps(iy1,jy0);
 867             dz10             = _mm_sub_ps(iz1,jz0);
 868             dx20             = _mm_sub_ps(ix2,jx0);
 869             dy20             = _mm_sub_ps(iy2,jy0);
 870             dz20             = _mm_sub_ps(iz2,jz0);
 871
 872             /* Calculate squared distance and things based on it */
 873             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 874             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 875             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 876
 877             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 878             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 879             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 880
 881             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 882             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 883             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 884
 885             /* Load parameters for j particles */
 886             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 887                                                               charge+jnrC+0,charge+jnrD+0);
 888             vdwjidx0A        = 2*vdwtype[jnrA+0];
 889             vdwjidx0B        = 2*vdwtype[jnrB+0];
 890             vdwjidx0C        = 2*vdwtype[jnrC+0];
 891             vdwjidx0D        = 2*vdwtype[jnrD+0];
 892
 893             fjx0             = _mm_setzero_ps();
 894             fjy0             = _mm_setzero_ps();
 895             fjz0             = _mm_setzero_ps();
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             r00              = _mm_mul_ps(rsq00,rinv00);
 902
 903             /* Compute parameters for interactions between i and j atoms */
 904             qq00             = _mm_mul_ps(iq0,jq0);
 905             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 906                                          vdwparam+vdwioffset0+vdwjidx0B,
 907                                          vdwparam+vdwioffset0+vdwjidx0C,
 908                                          vdwparam+vdwioffset0+vdwjidx0D,
 909                                          &c6_00,&c12_00);
 910
 911             /* Calculate table index by multiplying r with table scale and truncate to integer */
 912             rt               = _mm_mul_ps(r00,vftabscale);
 913             vfitab           = _mm_cvttps_epi32(rt);
 914             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 915             vfitab           = _mm_slli_epi32(vfitab,3);
 916
 917             /* EWALD ELECTROSTATICS */
 918
 919             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 920             ewrt             = _mm_mul_ps(r00,ewtabscale);
 921             ewitab           = _mm_cvttps_epi32(ewrt);
 922             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 923             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 924                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 925                                          &ewtabF,&ewtabFn);
 926             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 927             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 928
 929             /* CUBIC SPLINE TABLE DISPERSION */
 930             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 931             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 932             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 933             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 934             _MM_TRANSPOSE4_PS(Y,F,G,H);
 935             Heps             = _mm_mul_ps(vfeps,H);
 936             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 937             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 938             fvdw6            = _mm_mul_ps(c6_00,FF);
 939
 940             /* CUBIC SPLINE TABLE REPULSION */
 941             vfitab           = _mm_add_epi32(vfitab,ifour);
 942             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 943             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 944             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 945             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 946             _MM_TRANSPOSE4_PS(Y,F,G,H);
 947             Heps             = _mm_mul_ps(vfeps,H);
 948             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 949             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 950             fvdw12           = _mm_mul_ps(c12_00,FF);
 951             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 952
 953             fscal            = _mm_add_ps(felec,fvdw);
 954
 955             /* Calculate temporary vectorial force */
 956             tx               = _mm_mul_ps(fscal,dx00);
 957             ty               = _mm_mul_ps(fscal,dy00);
 958             tz               = _mm_mul_ps(fscal,dz00);
 959
 960             /* Update vectorial force */
 961             fix0             = _mm_add_ps(fix0,tx);
 962             fiy0             = _mm_add_ps(fiy0,ty);
 963             fiz0             = _mm_add_ps(fiz0,tz);
 964
 965             fjx0             = _mm_add_ps(fjx0,tx);
 966             fjy0             = _mm_add_ps(fjy0,ty);
 967             fjz0             = _mm_add_ps(fjz0,tz);
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r10              = _mm_mul_ps(rsq10,rinv10);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq10             = _mm_mul_ps(iq1,jq0);
 977
 978             /* EWALD ELECTROSTATICS */
 979
 980             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 981             ewrt             = _mm_mul_ps(r10,ewtabscale);
 982             ewitab           = _mm_cvttps_epi32(ewrt);
 983             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 984             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 985                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 986                                          &ewtabF,&ewtabFn);
 987             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 988             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 989
 990             fscal            = felec;
 991
 992             /* Calculate temporary vectorial force */
 993             tx               = _mm_mul_ps(fscal,dx10);
 994             ty               = _mm_mul_ps(fscal,dy10);
 995             tz               = _mm_mul_ps(fscal,dz10);
 996
 997             /* Update vectorial force */
 998             fix1             = _mm_add_ps(fix1,tx);
 999             fiy1             = _mm_add_ps(fiy1,ty);
1000             fiz1             = _mm_add_ps(fiz1,tz);
1001
1002             fjx0             = _mm_add_ps(fjx0,tx);
1003             fjy0             = _mm_add_ps(fjy0,ty);
1004             fjz0             = _mm_add_ps(fjz0,tz);
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             r20              = _mm_mul_ps(rsq20,rinv20);
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             qq20             = _mm_mul_ps(iq2,jq0);
1014
1015             /* EWALD ELECTROSTATICS */
1016
1017             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1018             ewrt             = _mm_mul_ps(r20,ewtabscale);
1019             ewitab           = _mm_cvttps_epi32(ewrt);
1020             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1021             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1022                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1023                                          &ewtabF,&ewtabFn);
1024             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1025             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1026
1027             fscal            = felec;
1028
1029             /* Calculate temporary vectorial force */
1030             tx               = _mm_mul_ps(fscal,dx20);
1031             ty               = _mm_mul_ps(fscal,dy20);
1032             tz               = _mm_mul_ps(fscal,dz20);
1033
1034             /* Update vectorial force */
1035             fix2             = _mm_add_ps(fix2,tx);
1036             fiy2             = _mm_add_ps(fiy2,ty);
1037             fiz2             = _mm_add_ps(fiz2,tz);
1038
1039             fjx0             = _mm_add_ps(fjx0,tx);
1040             fjy0             = _mm_add_ps(fjy0,ty);
1041             fjz0             = _mm_add_ps(fjz0,tz);
1042
1043             fjptrA             = f+j_coord_offsetA;
1044             fjptrB             = f+j_coord_offsetB;
1045             fjptrC             = f+j_coord_offsetC;
1046             fjptrD             = f+j_coord_offsetD;
1047
1048             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1049
1050             /* Inner loop uses 134 flops */
1051         }
1052
1053         if(jidx<j_index_end)
1054         {
1055
1056             /* Get j neighbor index, and coordinate index */
1057             jnrlistA         = jjnr[jidx];
1058             jnrlistB         = jjnr[jidx+1];
1059             jnrlistC         = jjnr[jidx+2];
1060             jnrlistD         = jjnr[jidx+3];
1061             /* Sign of each element will be negative for non-real atoms.
1062              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1063              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1064              */
1065             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1066             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1067             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1068             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1069             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1070             j_coord_offsetA  = DIM*jnrA;
1071             j_coord_offsetB  = DIM*jnrB;
1072             j_coord_offsetC  = DIM*jnrC;
1073             j_coord_offsetD  = DIM*jnrD;
1074
1075             /* load j atom coordinates */
1076             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1077                                               x+j_coord_offsetC,x+j_coord_offsetD,
1078                                               &jx0,&jy0,&jz0);
1079
1080             /* Calculate displacement vector */
1081             dx00             = _mm_sub_ps(ix0,jx0);
1082             dy00             = _mm_sub_ps(iy0,jy0);
1083             dz00             = _mm_sub_ps(iz0,jz0);
1084             dx10             = _mm_sub_ps(ix1,jx0);
1085             dy10             = _mm_sub_ps(iy1,jy0);
1086             dz10             = _mm_sub_ps(iz1,jz0);
1087             dx20             = _mm_sub_ps(ix2,jx0);
1088             dy20             = _mm_sub_ps(iy2,jy0);
1089             dz20             = _mm_sub_ps(iz2,jz0);
1090
1091             /* Calculate squared distance and things based on it */
1092             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1093             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1094             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1095
1096             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1097             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1098             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1099
1100             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1101             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1102             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1103
1104             /* Load parameters for j particles */
1105             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1106                                                               charge+jnrC+0,charge+jnrD+0);
1107             vdwjidx0A        = 2*vdwtype[jnrA+0];
1108             vdwjidx0B        = 2*vdwtype[jnrB+0];
1109             vdwjidx0C        = 2*vdwtype[jnrC+0];
1110             vdwjidx0D        = 2*vdwtype[jnrD+0];
1111
1112             fjx0             = _mm_setzero_ps();
1113             fjy0             = _mm_setzero_ps();
1114             fjz0             = _mm_setzero_ps();
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             r00              = _mm_mul_ps(rsq00,rinv00);
1121             r00              = _mm_andnot_ps(dummy_mask,r00);
1122
1123             /* Compute parameters for interactions between i and j atoms */
1124             qq00             = _mm_mul_ps(iq0,jq0);
1125             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1126                                          vdwparam+vdwioffset0+vdwjidx0B,
1127                                          vdwparam+vdwioffset0+vdwjidx0C,
1128                                          vdwparam+vdwioffset0+vdwjidx0D,
1129                                          &c6_00,&c12_00);
1130
1131             /* Calculate table index by multiplying r with table scale and truncate to integer */
1132             rt               = _mm_mul_ps(r00,vftabscale);
1133             vfitab           = _mm_cvttps_epi32(rt);
1134             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1135             vfitab           = _mm_slli_epi32(vfitab,3);
1136
1137             /* EWALD ELECTROSTATICS */
1138
1139             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1140             ewrt             = _mm_mul_ps(r00,ewtabscale);
1141             ewitab           = _mm_cvttps_epi32(ewrt);
1142             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1143             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1144                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1145                                          &ewtabF,&ewtabFn);
1146             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1147             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1148
1149             /* CUBIC SPLINE TABLE DISPERSION */
1150             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1151             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1152             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1153             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1154             _MM_TRANSPOSE4_PS(Y,F,G,H);
1155             Heps             = _mm_mul_ps(vfeps,H);
1156             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1157             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1158             fvdw6            = _mm_mul_ps(c6_00,FF);
1159
1160             /* CUBIC SPLINE TABLE REPULSION */
1161             vfitab           = _mm_add_epi32(vfitab,ifour);
1162             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1163             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1164             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1165             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1166             _MM_TRANSPOSE4_PS(Y,F,G,H);
1167             Heps             = _mm_mul_ps(vfeps,H);
1168             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1169             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1170             fvdw12           = _mm_mul_ps(c12_00,FF);
1171             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1172
1173             fscal            = _mm_add_ps(felec,fvdw);
1174
1175             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1176
1177             /* Calculate temporary vectorial force */
1178             tx               = _mm_mul_ps(fscal,dx00);
1179             ty               = _mm_mul_ps(fscal,dy00);
1180             tz               = _mm_mul_ps(fscal,dz00);
1181
1182             /* Update vectorial force */
1183             fix0             = _mm_add_ps(fix0,tx);
1184             fiy0             = _mm_add_ps(fiy0,ty);
1185             fiz0             = _mm_add_ps(fiz0,tz);
1186
1187             fjx0             = _mm_add_ps(fjx0,tx);
1188             fjy0             = _mm_add_ps(fjy0,ty);
1189             fjz0             = _mm_add_ps(fjz0,tz);
1190
1191             /**************************
1192              * CALCULATE INTERACTIONS *
1193              **************************/
1194
1195             r10              = _mm_mul_ps(rsq10,rinv10);
1196             r10              = _mm_andnot_ps(dummy_mask,r10);
1197
1198             /* Compute parameters for interactions between i and j atoms */
1199             qq10             = _mm_mul_ps(iq1,jq0);
1200
1201             /* EWALD ELECTROSTATICS */
1202
1203             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1204             ewrt             = _mm_mul_ps(r10,ewtabscale);
1205             ewitab           = _mm_cvttps_epi32(ewrt);
1206             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1207             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1208                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1209                                          &ewtabF,&ewtabFn);
1210             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1211             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1212
1213             fscal            = felec;
1214
1215             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1216
1217             /* Calculate temporary vectorial force */
1218             tx               = _mm_mul_ps(fscal,dx10);
1219             ty               = _mm_mul_ps(fscal,dy10);
1220             tz               = _mm_mul_ps(fscal,dz10);
1221
1222             /* Update vectorial force */
1223             fix1             = _mm_add_ps(fix1,tx);
1224             fiy1             = _mm_add_ps(fiy1,ty);
1225             fiz1             = _mm_add_ps(fiz1,tz);
1226
1227             fjx0             = _mm_add_ps(fjx0,tx);
1228             fjy0             = _mm_add_ps(fjy0,ty);
1229             fjz0             = _mm_add_ps(fjz0,tz);
1230
1231             /**************************
1232              * CALCULATE INTERACTIONS *
1233              **************************/
1234
1235             r20              = _mm_mul_ps(rsq20,rinv20);
1236             r20              = _mm_andnot_ps(dummy_mask,r20);
1237
1238             /* Compute parameters for interactions between i and j atoms */
1239             qq20             = _mm_mul_ps(iq2,jq0);
1240
1241             /* EWALD ELECTROSTATICS */
1242
1243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1244             ewrt             = _mm_mul_ps(r20,ewtabscale);
1245             ewitab           = _mm_cvttps_epi32(ewrt);
1246             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1247             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1248                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1249                                          &ewtabF,&ewtabFn);
1250             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1251             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1252
1253             fscal            = felec;
1254
1255             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1256
1257             /* Calculate temporary vectorial force */
1258             tx               = _mm_mul_ps(fscal,dx20);
1259             ty               = _mm_mul_ps(fscal,dy20);
1260             tz               = _mm_mul_ps(fscal,dz20);
1261
1262             /* Update vectorial force */
1263             fix2             = _mm_add_ps(fix2,tx);
1264             fiy2             = _mm_add_ps(fiy2,ty);
1265             fiz2             = _mm_add_ps(fiz2,tz);
1266
1267             fjx0             = _mm_add_ps(fjx0,tx);
1268             fjy0             = _mm_add_ps(fjy0,ty);
1269             fjz0             = _mm_add_ps(fjz0,tz);
1270
1271             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1272             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1273             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1274             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1275
1276             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1277
1278             /* Inner loop uses 137 flops */
1279         }
1280
1281         /* End of innermost loop */
1282
1283         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1284                                               f+i_coord_offset,fshift+i_shift_offset);
1285
1286         /* Increment number of inner iterations */
1287         inneriter                  += j_index_end - j_index_start;
1288
1289         /* Outer loop uses 18 flops */
1290     }
1291
1292     /* Increment number of outer iterations */
1293     outeriter        += nri;
1294
1295     /* Update outer/inner flops */
1296
1297     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1298 }