src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse2_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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  78     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  79     __m128i          ewitab;
  80     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  83     real             rswitch_scalar,d_scalar;
  84     __m128           dummy_mask,cutoff_mask;
  85     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  86     __m128           one     = _mm_set1_ps(1.0);
  87     __m128           two     = _mm_set1_ps(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm_set1_ps(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101     nvdwtype         = fr->ntype;
 102     vdwparam         = fr->nbfp;
 103     vdwtype          = mdatoms->typeA;
 104
 105     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 106     ewtab            = fr->ic->tabq_coul_FDV0;
 107     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 108     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 109
 110     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 111     rcutoff_scalar   = fr->rcoulomb;
 112     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 113     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 114
 115     rswitch_scalar   = fr->rcoulomb_switch;
 116     rswitch          = _mm_set1_ps(rswitch_scalar);
 117     /* Setup switch parameters */
 118     d_scalar         = rcutoff_scalar-rswitch_scalar;
 119     d                = _mm_set1_ps(d_scalar);
 120     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 121     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 122     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 123     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 124     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 125     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142         shX              = shiftvec[i_shift_offset+XX];
 143         shY              = shiftvec[i_shift_offset+YY];
 144         shZ              = shiftvec[i_shift_offset+ZZ];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 156         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 157         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162
 163         /* Load parameters for i particles */
 164         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 165         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 166
 167         /* Reset potential sums */
 168         velecsum         = _mm_setzero_ps();
 169         vvdwsum          = _mm_setzero_ps();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               x+j_coord_offsetC,x+j_coord_offsetD,
 189                                               &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm_sub_ps(ix0,jx0);
 193             dy00             = _mm_sub_ps(iy0,jy0);
 194             dz00             = _mm_sub_ps(iz0,jz0);
 195
 196             /* Calculate squared distance and things based on it */
 197             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 198
 199             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 200
 201             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 202
 203             /* Load parameters for j particles */
 204             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 205                                                               charge+jnrC+0,charge+jnrD+0);
 206             vdwjidx0A        = 2*vdwtype[jnrA+0];
 207             vdwjidx0B        = 2*vdwtype[jnrB+0];
 208             vdwjidx0C        = 2*vdwtype[jnrC+0];
 209             vdwjidx0D        = 2*vdwtype[jnrD+0];
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             if (gmx_mm_any_lt(rsq00,rcutoff2))
 216             {
 217
 218             r00              = _mm_mul_ps(rsq00,rinv00);
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             qq00             = _mm_mul_ps(iq0,jq0);
 222             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 223                                          vdwparam+vdwioffset0+vdwjidx0B,
 224                                          vdwparam+vdwioffset0+vdwjidx0C,
 225                                          vdwparam+vdwioffset0+vdwjidx0D,
 226                                          &c6_00,&c12_00);
 227
 228             /* EWALD ELECTROSTATICS */
 229
 230             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 231             ewrt             = _mm_mul_ps(r00,ewtabscale);
 232             ewitab           = _mm_cvttps_epi32(ewrt);
 233             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 234             ewitab           = _mm_slli_epi32(ewitab,2);
 235             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 236             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 237             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 238             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 239             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 240             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 241             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 242             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 243             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 244
 245             /* LENNARD-JONES DISPERSION/REPULSION */
 246
 247             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 248             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 249             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 250             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 251             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 252
 253             d                = _mm_sub_ps(r00,rswitch);
 254             d                = _mm_max_ps(d,_mm_setzero_ps());
 255             d2               = _mm_mul_ps(d,d);
 256             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 257
 258             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 259
 260             /* Evaluate switch function */
 261             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 262             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 263             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 264             velec            = _mm_mul_ps(velec,sw);
 265             vvdw             = _mm_mul_ps(vvdw,sw);
 266             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velec            = _mm_and_ps(velec,cutoff_mask);
 270             velecsum         = _mm_add_ps(velecsum,velec);
 271             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 272             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 273
 274             fscal            = _mm_add_ps(felec,fvdw);
 275
 276             fscal            = _mm_and_ps(fscal,cutoff_mask);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_ps(fscal,dx00);
 280             ty               = _mm_mul_ps(fscal,dy00);
 281             tz               = _mm_mul_ps(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm_add_ps(fix0,tx);
 285             fiy0             = _mm_add_ps(fiy0,ty);
 286             fiz0             = _mm_add_ps(fiz0,tz);
 287
 288             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 289                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 290                                                    tx,ty,tz);
 291
 292             }
 293
 294             /* Inner loop uses 83 flops */
 295         }
 296
 297         if(jidx<j_index_end)
 298         {
 299
 300             /* Get j neighbor index, and coordinate index */
 301             jnrA             = jjnr[jidx];
 302             jnrB             = jjnr[jidx+1];
 303             jnrC             = jjnr[jidx+2];
 304             jnrD             = jjnr[jidx+3];
 305
 306             /* Sign of each element will be negative for non-real atoms.
 307              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 308              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 309              */
 310             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 311             jnrA       = (jnrA>=0) ? jnrA : 0;
 312             jnrB       = (jnrB>=0) ? jnrB : 0;
 313             jnrC       = (jnrC>=0) ? jnrC : 0;
 314             jnrD       = (jnrD>=0) ? jnrD : 0;
 315
 316             j_coord_offsetA  = DIM*jnrA;
 317             j_coord_offsetB  = DIM*jnrB;
 318             j_coord_offsetC  = DIM*jnrC;
 319             j_coord_offsetD  = DIM*jnrD;
 320
 321             /* load j atom coordinates */
 322             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 323                                               x+j_coord_offsetC,x+j_coord_offsetD,
 324                                               &jx0,&jy0,&jz0);
 325
 326             /* Calculate displacement vector */
 327             dx00             = _mm_sub_ps(ix0,jx0);
 328             dy00             = _mm_sub_ps(iy0,jy0);
 329             dz00             = _mm_sub_ps(iz0,jz0);
 330
 331             /* Calculate squared distance and things based on it */
 332             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 333
 334             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 335
 336             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 337
 338             /* Load parameters for j particles */
 339             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 340                                                               charge+jnrC+0,charge+jnrD+0);
 341             vdwjidx0A        = 2*vdwtype[jnrA+0];
 342             vdwjidx0B        = 2*vdwtype[jnrB+0];
 343             vdwjidx0C        = 2*vdwtype[jnrC+0];
 344             vdwjidx0D        = 2*vdwtype[jnrD+0];
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             if (gmx_mm_any_lt(rsq00,rcutoff2))
 351             {
 352
 353             r00              = _mm_mul_ps(rsq00,rinv00);
 354             r00              = _mm_andnot_ps(dummy_mask,r00);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq00             = _mm_mul_ps(iq0,jq0);
 358             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 359                                          vdwparam+vdwioffset0+vdwjidx0B,
 360                                          vdwparam+vdwioffset0+vdwjidx0C,
 361                                          vdwparam+vdwioffset0+vdwjidx0D,
 362                                          &c6_00,&c12_00);
 363
 364             /* EWALD ELECTROSTATICS */
 365
 366             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 367             ewrt             = _mm_mul_ps(r00,ewtabscale);
 368             ewitab           = _mm_cvttps_epi32(ewrt);
 369             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 370             ewitab           = _mm_slli_epi32(ewitab,2);
 371             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 372             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 373             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 374             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 375             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 376             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 377             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 378             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 379             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 380
 381             /* LENNARD-JONES DISPERSION/REPULSION */
 382
 383             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 384             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 385             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 386             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 387             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 388
 389             d                = _mm_sub_ps(r00,rswitch);
 390             d                = _mm_max_ps(d,_mm_setzero_ps());
 391             d2               = _mm_mul_ps(d,d);
 392             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 393
 394             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 395
 396             /* Evaluate switch function */
 397             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 398             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 399             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 400             velec            = _mm_mul_ps(velec,sw);
 401             vvdw             = _mm_mul_ps(vvdw,sw);
 402             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_and_ps(velec,cutoff_mask);
 406             velec            = _mm_andnot_ps(dummy_mask,velec);
 407             velecsum         = _mm_add_ps(velecsum,velec);
 408             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 409             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 410             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 411
 412             fscal            = _mm_add_ps(felec,fvdw);
 413
 414             fscal            = _mm_and_ps(fscal,cutoff_mask);
 415
 416             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 417
 418             /* Calculate temporary vectorial force */
 419             tx               = _mm_mul_ps(fscal,dx00);
 420             ty               = _mm_mul_ps(fscal,dy00);
 421             tz               = _mm_mul_ps(fscal,dz00);
 422
 423             /* Update vectorial force */
 424             fix0             = _mm_add_ps(fix0,tx);
 425             fiy0             = _mm_add_ps(fiy0,ty);
 426             fiz0             = _mm_add_ps(fiz0,tz);
 427
 428             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 429                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 430                                                    tx,ty,tz);
 431
 432             }
 433
 434             /* Inner loop uses 84 flops */
 435         }
 436
 437         /* End of innermost loop */
 438
 439         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 440                                               f+i_coord_offset,fshift+i_shift_offset);
 441
 442         ggid                        = gid[iidx];
 443         /* Update potential energies */
 444         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 445         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 446
 447         /* Increment number of inner iterations */
 448         inneriter                  += j_index_end - j_index_start;
 449
 450         /* Outer loop uses 12 flops */
 451     }
 452
 453     /* Increment number of outer iterations */
 454     outeriter        += nri;
 455
 456     /* Update outer/inner flops */
 457
 458     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*84);
 459 }
 460 /*
 461  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse2_single
 462  * Electrostatics interaction: Ewald
 463  * VdW interaction:            LennardJones
 464  * Geometry:                   Particle-Particle
 465  * Calculate force/pot:        Force
 466  */
 467 void
 468 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse2_single
 469                     (t_nblist * gmx_restrict                nlist,
 470                      rvec * gmx_restrict                    xx,
 471                      rvec * gmx_restrict                    ff,
 472                      t_forcerec * gmx_restrict              fr,
 473                      t_mdatoms * gmx_restrict               mdatoms,
 474                      nb_kernel_data_t * gmx_restrict        kernel_data,
 475                      t_nrnb * gmx_restrict                  nrnb)
 476 {
 477     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 478      * just 0 for non-waters.
 479      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 480      * jnr indices corresponding to data put in the four positions in the SIMD register.
 481      */
 482     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 483     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 484     int              jnrA,jnrB,jnrC,jnrD;
 485     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 486     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 487     real             shX,shY,shZ,rcutoff_scalar;
 488     real             *shiftvec,*fshift,*x,*f;
 489     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 490     int              vdwioffset0;
 491     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 492     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 493     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 494     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 495     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 496     real             *charge;
 497     int              nvdwtype;
 498     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 499     int              *vdwtype;
 500     real             *vdwparam;
 501     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 502     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 503     __m128i          ewitab;
 504     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 505     real             *ewtab;
 506     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 507     real             rswitch_scalar,d_scalar;
 508     __m128           dummy_mask,cutoff_mask;
 509     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 510     __m128           one     = _mm_set1_ps(1.0);
 511     __m128           two     = _mm_set1_ps(2.0);
 512     x                = xx[0];
 513     f                = ff[0];
 514
 515     nri              = nlist->nri;
 516     iinr             = nlist->iinr;
 517     jindex           = nlist->jindex;
 518     jjnr             = nlist->jjnr;
 519     shiftidx         = nlist->shift;
 520     gid              = nlist->gid;
 521     shiftvec         = fr->shift_vec[0];
 522     fshift           = fr->fshift[0];
 523     facel            = _mm_set1_ps(fr->epsfac);
 524     charge           = mdatoms->chargeA;
 525     nvdwtype         = fr->ntype;
 526     vdwparam         = fr->nbfp;
 527     vdwtype          = mdatoms->typeA;
 528
 529     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 530     ewtab            = fr->ic->tabq_coul_FDV0;
 531     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 532     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 533
 534     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 535     rcutoff_scalar   = fr->rcoulomb;
 536     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 537     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 538
 539     rswitch_scalar   = fr->rcoulomb_switch;
 540     rswitch          = _mm_set1_ps(rswitch_scalar);
 541     /* Setup switch parameters */
 542     d_scalar         = rcutoff_scalar-rswitch_scalar;
 543     d                = _mm_set1_ps(d_scalar);
 544     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 545     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 546     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 547     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 548     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 549     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 550
 551     /* Avoid stupid compiler warnings */
 552     jnrA = jnrB = jnrC = jnrD = 0;
 553     j_coord_offsetA = 0;
 554     j_coord_offsetB = 0;
 555     j_coord_offsetC = 0;
 556     j_coord_offsetD = 0;
 557
 558     outeriter        = 0;
 559     inneriter        = 0;
 560
 561     /* Start outer loop over neighborlists */
 562     for(iidx=0; iidx<nri; iidx++)
 563     {
 564         /* Load shift vector for this list */
 565         i_shift_offset   = DIM*shiftidx[iidx];
 566         shX              = shiftvec[i_shift_offset+XX];
 567         shY              = shiftvec[i_shift_offset+YY];
 568         shZ              = shiftvec[i_shift_offset+ZZ];
 569
 570         /* Load limits for loop over neighbors */
 571         j_index_start    = jindex[iidx];
 572         j_index_end      = jindex[iidx+1];
 573
 574         /* Get outer coordinate index */
 575         inr              = iinr[iidx];
 576         i_coord_offset   = DIM*inr;
 577
 578         /* Load i particle coords and add shift vector */
 579         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 580         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 581         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 582
 583         fix0             = _mm_setzero_ps();
 584         fiy0             = _mm_setzero_ps();
 585         fiz0             = _mm_setzero_ps();
 586
 587         /* Load parameters for i particles */
 588         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 589         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 590
 591         /* Start inner kernel loop */
 592         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 593         {
 594
 595             /* Get j neighbor index, and coordinate index */
 596             jnrA             = jjnr[jidx];
 597             jnrB             = jjnr[jidx+1];
 598             jnrC             = jjnr[jidx+2];
 599             jnrD             = jjnr[jidx+3];
 600
 601             j_coord_offsetA  = DIM*jnrA;
 602             j_coord_offsetB  = DIM*jnrB;
 603             j_coord_offsetC  = DIM*jnrC;
 604             j_coord_offsetD  = DIM*jnrD;
 605
 606             /* load j atom coordinates */
 607             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 608                                               x+j_coord_offsetC,x+j_coord_offsetD,
 609                                               &jx0,&jy0,&jz0);
 610
 611             /* Calculate displacement vector */
 612             dx00             = _mm_sub_ps(ix0,jx0);
 613             dy00             = _mm_sub_ps(iy0,jy0);
 614             dz00             = _mm_sub_ps(iz0,jz0);
 615
 616             /* Calculate squared distance and things based on it */
 617             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 618
 619             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 620
 621             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 622
 623             /* Load parameters for j particles */
 624             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 625                                                               charge+jnrC+0,charge+jnrD+0);
 626             vdwjidx0A        = 2*vdwtype[jnrA+0];
 627             vdwjidx0B        = 2*vdwtype[jnrB+0];
 628             vdwjidx0C        = 2*vdwtype[jnrC+0];
 629             vdwjidx0D        = 2*vdwtype[jnrD+0];
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             if (gmx_mm_any_lt(rsq00,rcutoff2))
 636             {
 637
 638             r00              = _mm_mul_ps(rsq00,rinv00);
 639
 640             /* Compute parameters for interactions between i and j atoms */
 641             qq00             = _mm_mul_ps(iq0,jq0);
 642             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 643                                          vdwparam+vdwioffset0+vdwjidx0B,
 644                                          vdwparam+vdwioffset0+vdwjidx0C,
 645                                          vdwparam+vdwioffset0+vdwjidx0D,
 646                                          &c6_00,&c12_00);
 647
 648             /* EWALD ELECTROSTATICS */
 649
 650             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 651             ewrt             = _mm_mul_ps(r00,ewtabscale);
 652             ewitab           = _mm_cvttps_epi32(ewrt);
 653             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 654             ewitab           = _mm_slli_epi32(ewitab,2);
 655             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 656             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 657             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 658             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 659             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 660             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 661             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 662             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 663             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 664
 665             /* LENNARD-JONES DISPERSION/REPULSION */
 666
 667             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 668             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 669             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 670             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 671             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 672
 673             d                = _mm_sub_ps(r00,rswitch);
 674             d                = _mm_max_ps(d,_mm_setzero_ps());
 675             d2               = _mm_mul_ps(d,d);
 676             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 677
 678             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 679
 680             /* Evaluate switch function */
 681             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 682             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 683             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 684             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 685
 686             fscal            = _mm_add_ps(felec,fvdw);
 687
 688             fscal            = _mm_and_ps(fscal,cutoff_mask);
 689
 690             /* Calculate temporary vectorial force */
 691             tx               = _mm_mul_ps(fscal,dx00);
 692             ty               = _mm_mul_ps(fscal,dy00);
 693             tz               = _mm_mul_ps(fscal,dz00);
 694
 695             /* Update vectorial force */
 696             fix0             = _mm_add_ps(fix0,tx);
 697             fiy0             = _mm_add_ps(fiy0,ty);
 698             fiz0             = _mm_add_ps(fiz0,tz);
 699
 700             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 701                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 702                                                    tx,ty,tz);
 703
 704             }
 705
 706             /* Inner loop uses 77 flops */
 707         }
 708
 709         if(jidx<j_index_end)
 710         {
 711
 712             /* Get j neighbor index, and coordinate index */
 713             jnrA             = jjnr[jidx];
 714             jnrB             = jjnr[jidx+1];
 715             jnrC             = jjnr[jidx+2];
 716             jnrD             = jjnr[jidx+3];
 717
 718             /* Sign of each element will be negative for non-real atoms.
 719              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 720              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 721              */
 722             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 723             jnrA       = (jnrA>=0) ? jnrA : 0;
 724             jnrB       = (jnrB>=0) ? jnrB : 0;
 725             jnrC       = (jnrC>=0) ? jnrC : 0;
 726             jnrD       = (jnrD>=0) ? jnrD : 0;
 727
 728             j_coord_offsetA  = DIM*jnrA;
 729             j_coord_offsetB  = DIM*jnrB;
 730             j_coord_offsetC  = DIM*jnrC;
 731             j_coord_offsetD  = DIM*jnrD;
 732
 733             /* load j atom coordinates */
 734             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 735                                               x+j_coord_offsetC,x+j_coord_offsetD,
 736                                               &jx0,&jy0,&jz0);
 737
 738             /* Calculate displacement vector */
 739             dx00             = _mm_sub_ps(ix0,jx0);
 740             dy00             = _mm_sub_ps(iy0,jy0);
 741             dz00             = _mm_sub_ps(iz0,jz0);
 742
 743             /* Calculate squared distance and things based on it */
 744             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 745
 746             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 747
 748             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 749
 750             /* Load parameters for j particles */
 751             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 752                                                               charge+jnrC+0,charge+jnrD+0);
 753             vdwjidx0A        = 2*vdwtype[jnrA+0];
 754             vdwjidx0B        = 2*vdwtype[jnrB+0];
 755             vdwjidx0C        = 2*vdwtype[jnrC+0];
 756             vdwjidx0D        = 2*vdwtype[jnrD+0];
 757
 758             /**************************
 759              * CALCULATE INTERACTIONS *
 760              **************************/
 761
 762             if (gmx_mm_any_lt(rsq00,rcutoff2))
 763             {
 764
 765             r00              = _mm_mul_ps(rsq00,rinv00);
 766             r00              = _mm_andnot_ps(dummy_mask,r00);
 767
 768             /* Compute parameters for interactions between i and j atoms */
 769             qq00             = _mm_mul_ps(iq0,jq0);
 770             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 771                                          vdwparam+vdwioffset0+vdwjidx0B,
 772                                          vdwparam+vdwioffset0+vdwjidx0C,
 773                                          vdwparam+vdwioffset0+vdwjidx0D,
 774                                          &c6_00,&c12_00);
 775
 776             /* EWALD ELECTROSTATICS */
 777
 778             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 779             ewrt             = _mm_mul_ps(r00,ewtabscale);
 780             ewitab           = _mm_cvttps_epi32(ewrt);
 781             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 782             ewitab           = _mm_slli_epi32(ewitab,2);
 783             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 784             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 785             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 786             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 787             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 788             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 789             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 790             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 791             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 792
 793             /* LENNARD-JONES DISPERSION/REPULSION */
 794
 795             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 796             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 797             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 798             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 799             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 800
 801             d                = _mm_sub_ps(r00,rswitch);
 802             d                = _mm_max_ps(d,_mm_setzero_ps());
 803             d2               = _mm_mul_ps(d,d);
 804             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 805
 806             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 807
 808             /* Evaluate switch function */
 809             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 810             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 811             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 812             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 813
 814             fscal            = _mm_add_ps(felec,fvdw);
 815
 816             fscal            = _mm_and_ps(fscal,cutoff_mask);
 817
 818             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 819
 820             /* Calculate temporary vectorial force */
 821             tx               = _mm_mul_ps(fscal,dx00);
 822             ty               = _mm_mul_ps(fscal,dy00);
 823             tz               = _mm_mul_ps(fscal,dz00);
 824
 825             /* Update vectorial force */
 826             fix0             = _mm_add_ps(fix0,tx);
 827             fiy0             = _mm_add_ps(fiy0,ty);
 828             fiz0             = _mm_add_ps(fiz0,tz);
 829
 830             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 831                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 832                                                    tx,ty,tz);
 833
 834             }
 835
 836             /* Inner loop uses 78 flops */
 837         }
 838
 839         /* End of innermost loop */
 840
 841         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 842                                               f+i_coord_offset,fshift+i_shift_offset);
 843
 844         /* Increment number of inner iterations */
 845         inneriter                  += j_index_end - j_index_start;
 846
 847         /* Outer loop uses 10 flops */
 848     }
 849
 850     /* Increment number of outer iterations */
 851     outeriter        += nri;
 852
 853     /* Update outer/inner flops */
 854
 855     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*78);
 856 }