src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_elec_vdw->data;
 128     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173         fix1             = _mm_setzero_ps();
 174         fiy1             = _mm_setzero_ps();
 175         fiz1             = _mm_setzero_ps();
 176         fix2             = _mm_setzero_ps();
 177         fiy2             = _mm_setzero_ps();
 178         fiz2             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 218
 219             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 220             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 221             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 225                                                               charge+jnrC+0,charge+jnrD+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228             vdwjidx0C        = 2*vdwtype[jnrC+0];
 229             vdwjidx0D        = 2*vdwtype[jnrD+0];
 230
 231             fjx0             = _mm_setzero_ps();
 232             fjy0             = _mm_setzero_ps();
 233             fjz0             = _mm_setzero_ps();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_ps(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm_mul_ps(iq0,jq0);
 243             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,
 245                                          vdwparam+vdwioffset0+vdwjidx0C,
 246                                          vdwparam+vdwioffset0+vdwjidx0D,
 247                                          &c6_00,&c12_00);
 248
 249             /* Calculate table index by multiplying r with table scale and truncate to integer */
 250             rt               = _mm_mul_ps(r00,vftabscale);
 251             vfitab           = _mm_cvttps_epi32(rt);
 252             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 253             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 254
 255             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 256             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 258             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 259             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 260             _MM_TRANSPOSE4_PS(Y,F,G,H);
 261             Heps             = _mm_mul_ps(vfeps,H);
 262             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 263             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 264             velec            = _mm_mul_ps(qq00,VV);
 265             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 266             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 267
 268             /* CUBIC SPLINE TABLE DISPERSION */
 269             vfitab           = _mm_add_epi32(vfitab,ifour);
 270             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 271             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 272             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 273             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 274             _MM_TRANSPOSE4_PS(Y,F,G,H);
 275             Heps             = _mm_mul_ps(vfeps,H);
 276             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 277             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 278             vvdw6            = _mm_mul_ps(c6_00,VV);
 279             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 280             fvdw6            = _mm_mul_ps(c6_00,FF);
 281
 282             /* CUBIC SPLINE TABLE REPULSION */
 283             vfitab           = _mm_add_epi32(vfitab,ifour);
 284             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 285             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 286             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 287             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 288             _MM_TRANSPOSE4_PS(Y,F,G,H);
 289             Heps             = _mm_mul_ps(vfeps,H);
 290             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 291             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 292             vvdw12           = _mm_mul_ps(c12_00,VV);
 293             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 294             fvdw12           = _mm_mul_ps(c12_00,FF);
 295             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 296             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 301
 302             fscal            = _mm_add_ps(felec,fvdw);
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm_mul_ps(fscal,dx00);
 306             ty               = _mm_mul_ps(fscal,dy00);
 307             tz               = _mm_mul_ps(fscal,dz00);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm_add_ps(fix0,tx);
 311             fiy0             = _mm_add_ps(fiy0,ty);
 312             fiz0             = _mm_add_ps(fiz0,tz);
 313
 314             fjx0             = _mm_add_ps(fjx0,tx);
 315             fjy0             = _mm_add_ps(fjy0,ty);
 316             fjz0             = _mm_add_ps(fjz0,tz);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r10              = _mm_mul_ps(rsq10,rinv10);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm_mul_ps(iq1,jq0);
 326
 327             /* Calculate table index by multiplying r with table scale and truncate to integer */
 328             rt               = _mm_mul_ps(r10,vftabscale);
 329             vfitab           = _mm_cvttps_epi32(rt);
 330             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 331             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 332
 333             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 334             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 335             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 336             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 337             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 338             _MM_TRANSPOSE4_PS(Y,F,G,H);
 339             Heps             = _mm_mul_ps(vfeps,H);
 340             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 341             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 342             velec            = _mm_mul_ps(qq10,VV);
 343             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 344             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_ps(fscal,dx10);
 353             ty               = _mm_mul_ps(fscal,dy10);
 354             tz               = _mm_mul_ps(fscal,dz10);
 355
 356             /* Update vectorial force */
 357             fix1             = _mm_add_ps(fix1,tx);
 358             fiy1             = _mm_add_ps(fiy1,ty);
 359             fiz1             = _mm_add_ps(fiz1,tz);
 360
 361             fjx0             = _mm_add_ps(fjx0,tx);
 362             fjy0             = _mm_add_ps(fjy0,ty);
 363             fjz0             = _mm_add_ps(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             r20              = _mm_mul_ps(rsq20,rinv20);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm_mul_ps(iq2,jq0);
 373
 374             /* Calculate table index by multiplying r with table scale and truncate to integer */
 375             rt               = _mm_mul_ps(r20,vftabscale);
 376             vfitab           = _mm_cvttps_epi32(rt);
 377             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 378             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 379
 380             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 381             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 382             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 383             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 384             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 385             _MM_TRANSPOSE4_PS(Y,F,G,H);
 386             Heps             = _mm_mul_ps(vfeps,H);
 387             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 388             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 389             velec            = _mm_mul_ps(qq20,VV);
 390             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 391             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx20);
 400             ty               = _mm_mul_ps(fscal,dy20);
 401             tz               = _mm_mul_ps(fscal,dz20);
 402
 403             /* Update vectorial force */
 404             fix2             = _mm_add_ps(fix2,tx);
 405             fiy2             = _mm_add_ps(fiy2,ty);
 406             fiz2             = _mm_add_ps(fiz2,tz);
 407
 408             fjx0             = _mm_add_ps(fjx0,tx);
 409             fjy0             = _mm_add_ps(fjy0,ty);
 410             fjz0             = _mm_add_ps(fjz0,tz);
 411
 412             fjptrA             = f+j_coord_offsetA;
 413             fjptrB             = f+j_coord_offsetB;
 414             fjptrC             = f+j_coord_offsetC;
 415             fjptrD             = f+j_coord_offsetD;
 416
 417             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 418
 419             /* Inner loop uses 159 flops */
 420         }
 421
 422         if(jidx<j_index_end)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrlistA         = jjnr[jidx];
 427             jnrlistB         = jjnr[jidx+1];
 428             jnrlistC         = jjnr[jidx+2];
 429             jnrlistD         = jjnr[jidx+3];
 430             /* Sign of each element will be negative for non-real atoms.
 431              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 432              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 433              */
 434             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 435             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 436             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 437             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 438             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 439             j_coord_offsetA  = DIM*jnrA;
 440             j_coord_offsetB  = DIM*jnrB;
 441             j_coord_offsetC  = DIM*jnrC;
 442             j_coord_offsetD  = DIM*jnrD;
 443
 444             /* load j atom coordinates */
 445             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 446                                               x+j_coord_offsetC,x+j_coord_offsetD,
 447                                               &jx0,&jy0,&jz0);
 448
 449             /* Calculate displacement vector */
 450             dx00             = _mm_sub_ps(ix0,jx0);
 451             dy00             = _mm_sub_ps(iy0,jy0);
 452             dz00             = _mm_sub_ps(iz0,jz0);
 453             dx10             = _mm_sub_ps(ix1,jx0);
 454             dy10             = _mm_sub_ps(iy1,jy0);
 455             dz10             = _mm_sub_ps(iz1,jz0);
 456             dx20             = _mm_sub_ps(ix2,jx0);
 457             dy20             = _mm_sub_ps(iy2,jy0);
 458             dz20             = _mm_sub_ps(iz2,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 462             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 463             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 464
 465             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 466             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 467             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 468
 469             /* Load parameters for j particles */
 470             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 471                                                               charge+jnrC+0,charge+jnrD+0);
 472             vdwjidx0A        = 2*vdwtype[jnrA+0];
 473             vdwjidx0B        = 2*vdwtype[jnrB+0];
 474             vdwjidx0C        = 2*vdwtype[jnrC+0];
 475             vdwjidx0D        = 2*vdwtype[jnrD+0];
 476
 477             fjx0             = _mm_setzero_ps();
 478             fjy0             = _mm_setzero_ps();
 479             fjz0             = _mm_setzero_ps();
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             r00              = _mm_mul_ps(rsq00,rinv00);
 486             r00              = _mm_andnot_ps(dummy_mask,r00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq00             = _mm_mul_ps(iq0,jq0);
 490             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 491                                          vdwparam+vdwioffset0+vdwjidx0B,
 492                                          vdwparam+vdwioffset0+vdwjidx0C,
 493                                          vdwparam+vdwioffset0+vdwjidx0D,
 494                                          &c6_00,&c12_00);
 495
 496             /* Calculate table index by multiplying r with table scale and truncate to integer */
 497             rt               = _mm_mul_ps(r00,vftabscale);
 498             vfitab           = _mm_cvttps_epi32(rt);
 499             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 500             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 501
 502             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 503             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 504             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 505             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 506             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 507             _MM_TRANSPOSE4_PS(Y,F,G,H);
 508             Heps             = _mm_mul_ps(vfeps,H);
 509             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 510             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 511             velec            = _mm_mul_ps(qq00,VV);
 512             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 513             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 514
 515             /* CUBIC SPLINE TABLE DISPERSION */
 516             vfitab           = _mm_add_epi32(vfitab,ifour);
 517             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 518             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 519             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 520             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 521             _MM_TRANSPOSE4_PS(Y,F,G,H);
 522             Heps             = _mm_mul_ps(vfeps,H);
 523             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 524             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 525             vvdw6            = _mm_mul_ps(c6_00,VV);
 526             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 527             fvdw6            = _mm_mul_ps(c6_00,FF);
 528
 529             /* CUBIC SPLINE TABLE REPULSION */
 530             vfitab           = _mm_add_epi32(vfitab,ifour);
 531             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 532             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 533             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 534             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 535             _MM_TRANSPOSE4_PS(Y,F,G,H);
 536             Heps             = _mm_mul_ps(vfeps,H);
 537             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 538             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 539             vvdw12           = _mm_mul_ps(c12_00,VV);
 540             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 541             fvdw12           = _mm_mul_ps(c12_00,FF);
 542             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 543             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 544
 545             /* Update potential sum for this i atom from the interaction with this j atom. */
 546             velec            = _mm_andnot_ps(dummy_mask,velec);
 547             velecsum         = _mm_add_ps(velecsum,velec);
 548             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 549             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 550
 551             fscal            = _mm_add_ps(felec,fvdw);
 552
 553             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 554
 555             /* Calculate temporary vectorial force */
 556             tx               = _mm_mul_ps(fscal,dx00);
 557             ty               = _mm_mul_ps(fscal,dy00);
 558             tz               = _mm_mul_ps(fscal,dz00);
 559
 560             /* Update vectorial force */
 561             fix0             = _mm_add_ps(fix0,tx);
 562             fiy0             = _mm_add_ps(fiy0,ty);
 563             fiz0             = _mm_add_ps(fiz0,tz);
 564
 565             fjx0             = _mm_add_ps(fjx0,tx);
 566             fjy0             = _mm_add_ps(fjy0,ty);
 567             fjz0             = _mm_add_ps(fjz0,tz);
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             r10              = _mm_mul_ps(rsq10,rinv10);
 574             r10              = _mm_andnot_ps(dummy_mask,r10);
 575
 576             /* Compute parameters for interactions between i and j atoms */
 577             qq10             = _mm_mul_ps(iq1,jq0);
 578
 579             /* Calculate table index by multiplying r with table scale and truncate to integer */
 580             rt               = _mm_mul_ps(r10,vftabscale);
 581             vfitab           = _mm_cvttps_epi32(rt);
 582             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 583             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 584
 585             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 586             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 587             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 588             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 589             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 590             _MM_TRANSPOSE4_PS(Y,F,G,H);
 591             Heps             = _mm_mul_ps(vfeps,H);
 592             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 593             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 594             velec            = _mm_mul_ps(qq10,VV);
 595             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 596             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velec            = _mm_andnot_ps(dummy_mask,velec);
 600             velecsum         = _mm_add_ps(velecsum,velec);
 601
 602             fscal            = felec;
 603
 604             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 605
 606             /* Calculate temporary vectorial force */
 607             tx               = _mm_mul_ps(fscal,dx10);
 608             ty               = _mm_mul_ps(fscal,dy10);
 609             tz               = _mm_mul_ps(fscal,dz10);
 610
 611             /* Update vectorial force */
 612             fix1             = _mm_add_ps(fix1,tx);
 613             fiy1             = _mm_add_ps(fiy1,ty);
 614             fiz1             = _mm_add_ps(fiz1,tz);
 615
 616             fjx0             = _mm_add_ps(fjx0,tx);
 617             fjy0             = _mm_add_ps(fjy0,ty);
 618             fjz0             = _mm_add_ps(fjz0,tz);
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             r20              = _mm_mul_ps(rsq20,rinv20);
 625             r20              = _mm_andnot_ps(dummy_mask,r20);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq20             = _mm_mul_ps(iq2,jq0);
 629
 630             /* Calculate table index by multiplying r with table scale and truncate to integer */
 631             rt               = _mm_mul_ps(r20,vftabscale);
 632             vfitab           = _mm_cvttps_epi32(rt);
 633             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 634             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 635
 636             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 637             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 638             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 639             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 640             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 641             _MM_TRANSPOSE4_PS(Y,F,G,H);
 642             Heps             = _mm_mul_ps(vfeps,H);
 643             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 644             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 645             velec            = _mm_mul_ps(qq20,VV);
 646             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 647             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 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 162 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*162);
 703 }
 704 /*
 705  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_single
 706  * Electrostatics interaction: CubicSplineTable
 707  * VdW interaction:            CubicSplineTable
 708  * Geometry:                   Water3-Particle
 709  * Calculate force/pot:        Force
 710  */
 711 void
 712 nb_kernel_ElecCSTab_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_unused * 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     __m128           dummy_mask,cutoff_mask;
 761     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 762     __m128           one     = _mm_set1_ps(1.0);
 763     __m128           two     = _mm_set1_ps(2.0);
 764     x                = xx[0];
 765     f                = ff[0];
 766
 767     nri              = nlist->nri;
 768     iinr             = nlist->iinr;
 769     jindex           = nlist->jindex;
 770     jjnr             = nlist->jjnr;
 771     shiftidx         = nlist->shift;
 772     gid              = nlist->gid;
 773     shiftvec         = fr->shift_vec[0];
 774     fshift           = fr->fshift[0];
 775     facel            = _mm_set1_ps(fr->epsfac);
 776     charge           = mdatoms->chargeA;
 777     nvdwtype         = fr->ntype;
 778     vdwparam         = fr->nbfp;
 779     vdwtype          = mdatoms->typeA;
 780
 781     vftab            = kernel_data->table_elec_vdw->data;
 782     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 783
 784     /* Setup water-specific parameters */
 785     inr              = nlist->iinr[0];
 786     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 787     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 788     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 789     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 790
 791     /* Avoid stupid compiler warnings */
 792     jnrA = jnrB = jnrC = jnrD = 0;
 793     j_coord_offsetA = 0;
 794     j_coord_offsetB = 0;
 795     j_coord_offsetC = 0;
 796     j_coord_offsetD = 0;
 797
 798     outeriter        = 0;
 799     inneriter        = 0;
 800
 801     for(iidx=0;iidx<4*DIM;iidx++)
 802     {
 803         scratch[iidx] = 0.0;
 804     }
 805
 806     /* Start outer loop over neighborlists */
 807     for(iidx=0; iidx<nri; iidx++)
 808     {
 809         /* Load shift vector for this list */
 810         i_shift_offset   = DIM*shiftidx[iidx];
 811
 812         /* Load limits for loop over neighbors */
 813         j_index_start    = jindex[iidx];
 814         j_index_end      = jindex[iidx+1];
 815
 816         /* Get outer coordinate index */
 817         inr              = iinr[iidx];
 818         i_coord_offset   = DIM*inr;
 819
 820         /* Load i particle coords and add shift vector */
 821         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 822                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 823
 824         fix0             = _mm_setzero_ps();
 825         fiy0             = _mm_setzero_ps();
 826         fiz0             = _mm_setzero_ps();
 827         fix1             = _mm_setzero_ps();
 828         fiy1             = _mm_setzero_ps();
 829         fiz1             = _mm_setzero_ps();
 830         fix2             = _mm_setzero_ps();
 831         fiy2             = _mm_setzero_ps();
 832         fiz2             = _mm_setzero_ps();
 833
 834         /* Start inner kernel loop */
 835         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 836         {
 837
 838             /* Get j neighbor index, and coordinate index */
 839             jnrA             = jjnr[jidx];
 840             jnrB             = jjnr[jidx+1];
 841             jnrC             = jjnr[jidx+2];
 842             jnrD             = jjnr[jidx+3];
 843             j_coord_offsetA  = DIM*jnrA;
 844             j_coord_offsetB  = DIM*jnrB;
 845             j_coord_offsetC  = DIM*jnrC;
 846             j_coord_offsetD  = DIM*jnrD;
 847
 848             /* load j atom coordinates */
 849             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 850                                               x+j_coord_offsetC,x+j_coord_offsetD,
 851                                               &jx0,&jy0,&jz0);
 852
 853             /* Calculate displacement vector */
 854             dx00             = _mm_sub_ps(ix0,jx0);
 855             dy00             = _mm_sub_ps(iy0,jy0);
 856             dz00             = _mm_sub_ps(iz0,jz0);
 857             dx10             = _mm_sub_ps(ix1,jx0);
 858             dy10             = _mm_sub_ps(iy1,jy0);
 859             dz10             = _mm_sub_ps(iz1,jz0);
 860             dx20             = _mm_sub_ps(ix2,jx0);
 861             dy20             = _mm_sub_ps(iy2,jy0);
 862             dz20             = _mm_sub_ps(iz2,jz0);
 863
 864             /* Calculate squared distance and things based on it */
 865             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 866             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 867             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 868
 869             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 870             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 871             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 872
 873             /* Load parameters for j particles */
 874             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 875                                                               charge+jnrC+0,charge+jnrD+0);
 876             vdwjidx0A        = 2*vdwtype[jnrA+0];
 877             vdwjidx0B        = 2*vdwtype[jnrB+0];
 878             vdwjidx0C        = 2*vdwtype[jnrC+0];
 879             vdwjidx0D        = 2*vdwtype[jnrD+0];
 880
 881             fjx0             = _mm_setzero_ps();
 882             fjy0             = _mm_setzero_ps();
 883             fjz0             = _mm_setzero_ps();
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             r00              = _mm_mul_ps(rsq00,rinv00);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq00             = _mm_mul_ps(iq0,jq0);
 893             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 894                                          vdwparam+vdwioffset0+vdwjidx0B,
 895                                          vdwparam+vdwioffset0+vdwjidx0C,
 896                                          vdwparam+vdwioffset0+vdwjidx0D,
 897                                          &c6_00,&c12_00);
 898
 899             /* Calculate table index by multiplying r with table scale and truncate to integer */
 900             rt               = _mm_mul_ps(r00,vftabscale);
 901             vfitab           = _mm_cvttps_epi32(rt);
 902             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 903             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 904
 905             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 906             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 907             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 908             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 909             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 910             _MM_TRANSPOSE4_PS(Y,F,G,H);
 911             Heps             = _mm_mul_ps(vfeps,H);
 912             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 913             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 914             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 915
 916             /* CUBIC SPLINE TABLE DISPERSION */
 917             vfitab           = _mm_add_epi32(vfitab,ifour);
 918             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 919             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 920             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 921             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 922             _MM_TRANSPOSE4_PS(Y,F,G,H);
 923             Heps             = _mm_mul_ps(vfeps,H);
 924             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 925             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 926             fvdw6            = _mm_mul_ps(c6_00,FF);
 927
 928             /* CUBIC SPLINE TABLE REPULSION */
 929             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             fvdw12           = _mm_mul_ps(c12_00,FF);
 939             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 940
 941             fscal            = _mm_add_ps(felec,fvdw);
 942
 943             /* Calculate temporary vectorial force */
 944             tx               = _mm_mul_ps(fscal,dx00);
 945             ty               = _mm_mul_ps(fscal,dy00);
 946             tz               = _mm_mul_ps(fscal,dz00);
 947
 948             /* Update vectorial force */
 949             fix0             = _mm_add_ps(fix0,tx);
 950             fiy0             = _mm_add_ps(fiy0,ty);
 951             fiz0             = _mm_add_ps(fiz0,tz);
 952
 953             fjx0             = _mm_add_ps(fjx0,tx);
 954             fjy0             = _mm_add_ps(fjy0,ty);
 955             fjz0             = _mm_add_ps(fjz0,tz);
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r10              = _mm_mul_ps(rsq10,rinv10);
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq10             = _mm_mul_ps(iq1,jq0);
 965
 966             /* Calculate table index by multiplying r with table scale and truncate to integer */
 967             rt               = _mm_mul_ps(r10,vftabscale);
 968             vfitab           = _mm_cvttps_epi32(rt);
 969             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 970             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 971
 972             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 973             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 974             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 975             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 976             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 977             _MM_TRANSPOSE4_PS(Y,F,G,H);
 978             Heps             = _mm_mul_ps(vfeps,H);
 979             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 980             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 981             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 982
 983             fscal            = felec;
 984
 985             /* Calculate temporary vectorial force */
 986             tx               = _mm_mul_ps(fscal,dx10);
 987             ty               = _mm_mul_ps(fscal,dy10);
 988             tz               = _mm_mul_ps(fscal,dz10);
 989
 990             /* Update vectorial force */
 991             fix1             = _mm_add_ps(fix1,tx);
 992             fiy1             = _mm_add_ps(fiy1,ty);
 993             fiz1             = _mm_add_ps(fiz1,tz);
 994
 995             fjx0             = _mm_add_ps(fjx0,tx);
 996             fjy0             = _mm_add_ps(fjy0,ty);
 997             fjz0             = _mm_add_ps(fjz0,tz);
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             r20              = _mm_mul_ps(rsq20,rinv20);
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             qq20             = _mm_mul_ps(iq2,jq0);
1007
1008             /* Calculate table index by multiplying r with table scale and truncate to integer */
1009             rt               = _mm_mul_ps(r20,vftabscale);
1010             vfitab           = _mm_cvttps_epi32(rt);
1011             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1012             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1013
1014             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1015             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1016             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1017             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1018             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1019             _MM_TRANSPOSE4_PS(Y,F,G,H);
1020             Heps             = _mm_mul_ps(vfeps,H);
1021             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1022             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1023             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1024
1025             fscal            = felec;
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm_mul_ps(fscal,dx20);
1029             ty               = _mm_mul_ps(fscal,dy20);
1030             tz               = _mm_mul_ps(fscal,dz20);
1031
1032             /* Update vectorial force */
1033             fix2             = _mm_add_ps(fix2,tx);
1034             fiy2             = _mm_add_ps(fiy2,ty);
1035             fiz2             = _mm_add_ps(fiz2,tz);
1036
1037             fjx0             = _mm_add_ps(fjx0,tx);
1038             fjy0             = _mm_add_ps(fjy0,ty);
1039             fjz0             = _mm_add_ps(fjz0,tz);
1040
1041             fjptrA             = f+j_coord_offsetA;
1042             fjptrB             = f+j_coord_offsetB;
1043             fjptrC             = f+j_coord_offsetC;
1044             fjptrD             = f+j_coord_offsetD;
1045
1046             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1047
1048             /* Inner loop uses 139 flops */
1049         }
1050
1051         if(jidx<j_index_end)
1052         {
1053
1054             /* Get j neighbor index, and coordinate index */
1055             jnrlistA         = jjnr[jidx];
1056             jnrlistB         = jjnr[jidx+1];
1057             jnrlistC         = jjnr[jidx+2];
1058             jnrlistD         = jjnr[jidx+3];
1059             /* Sign of each element will be negative for non-real atoms.
1060              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1061              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1062              */
1063             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1064             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1065             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1066             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1067             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1068             j_coord_offsetA  = DIM*jnrA;
1069             j_coord_offsetB  = DIM*jnrB;
1070             j_coord_offsetC  = DIM*jnrC;
1071             j_coord_offsetD  = DIM*jnrD;
1072
1073             /* load j atom coordinates */
1074             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1075                                               x+j_coord_offsetC,x+j_coord_offsetD,
1076                                               &jx0,&jy0,&jz0);
1077
1078             /* Calculate displacement vector */
1079             dx00             = _mm_sub_ps(ix0,jx0);
1080             dy00             = _mm_sub_ps(iy0,jy0);
1081             dz00             = _mm_sub_ps(iz0,jz0);
1082             dx10             = _mm_sub_ps(ix1,jx0);
1083             dy10             = _mm_sub_ps(iy1,jy0);
1084             dz10             = _mm_sub_ps(iz1,jz0);
1085             dx20             = _mm_sub_ps(ix2,jx0);
1086             dy20             = _mm_sub_ps(iy2,jy0);
1087             dz20             = _mm_sub_ps(iz2,jz0);
1088
1089             /* Calculate squared distance and things based on it */
1090             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1091             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1092             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1093
1094             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1095             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1096             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1097
1098             /* Load parameters for j particles */
1099             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1100                                                               charge+jnrC+0,charge+jnrD+0);
1101             vdwjidx0A        = 2*vdwtype[jnrA+0];
1102             vdwjidx0B        = 2*vdwtype[jnrB+0];
1103             vdwjidx0C        = 2*vdwtype[jnrC+0];
1104             vdwjidx0D        = 2*vdwtype[jnrD+0];
1105
1106             fjx0             = _mm_setzero_ps();
1107             fjy0             = _mm_setzero_ps();
1108             fjz0             = _mm_setzero_ps();
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             r00              = _mm_mul_ps(rsq00,rinv00);
1115             r00              = _mm_andnot_ps(dummy_mask,r00);
1116
1117             /* Compute parameters for interactions between i and j atoms */
1118             qq00             = _mm_mul_ps(iq0,jq0);
1119             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1120                                          vdwparam+vdwioffset0+vdwjidx0B,
1121                                          vdwparam+vdwioffset0+vdwjidx0C,
1122                                          vdwparam+vdwioffset0+vdwjidx0D,
1123                                          &c6_00,&c12_00);
1124
1125             /* Calculate table index by multiplying r with table scale and truncate to integer */
1126             rt               = _mm_mul_ps(r00,vftabscale);
1127             vfitab           = _mm_cvttps_epi32(rt);
1128             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1129             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1130
1131             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1132             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1133             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1134             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1135             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1136             _MM_TRANSPOSE4_PS(Y,F,G,H);
1137             Heps             = _mm_mul_ps(vfeps,H);
1138             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1139             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1140             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1141
1142             /* CUBIC SPLINE TABLE DISPERSION */
1143             vfitab           = _mm_add_epi32(vfitab,ifour);
1144             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1145             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1146             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1147             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1148             _MM_TRANSPOSE4_PS(Y,F,G,H);
1149             Heps             = _mm_mul_ps(vfeps,H);
1150             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1151             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1152             fvdw6            = _mm_mul_ps(c6_00,FF);
1153
1154             /* CUBIC SPLINE TABLE REPULSION */
1155             vfitab           = _mm_add_epi32(vfitab,ifour);
1156             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1157             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1158             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1159             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1160             _MM_TRANSPOSE4_PS(Y,F,G,H);
1161             Heps             = _mm_mul_ps(vfeps,H);
1162             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1163             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1164             fvdw12           = _mm_mul_ps(c12_00,FF);
1165             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1166
1167             fscal            = _mm_add_ps(felec,fvdw);
1168
1169             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1170
1171             /* Calculate temporary vectorial force */
1172             tx               = _mm_mul_ps(fscal,dx00);
1173             ty               = _mm_mul_ps(fscal,dy00);
1174             tz               = _mm_mul_ps(fscal,dz00);
1175
1176             /* Update vectorial force */
1177             fix0             = _mm_add_ps(fix0,tx);
1178             fiy0             = _mm_add_ps(fiy0,ty);
1179             fiz0             = _mm_add_ps(fiz0,tz);
1180
1181             fjx0             = _mm_add_ps(fjx0,tx);
1182             fjy0             = _mm_add_ps(fjy0,ty);
1183             fjz0             = _mm_add_ps(fjz0,tz);
1184
1185             /**************************
1186              * CALCULATE INTERACTIONS *
1187              **************************/
1188
1189             r10              = _mm_mul_ps(rsq10,rinv10);
1190             r10              = _mm_andnot_ps(dummy_mask,r10);
1191
1192             /* Compute parameters for interactions between i and j atoms */
1193             qq10             = _mm_mul_ps(iq1,jq0);
1194
1195             /* Calculate table index by multiplying r with table scale and truncate to integer */
1196             rt               = _mm_mul_ps(r10,vftabscale);
1197             vfitab           = _mm_cvttps_epi32(rt);
1198             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1199             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1200
1201             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1202             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1203             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1204             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1205             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1206             _MM_TRANSPOSE4_PS(Y,F,G,H);
1207             Heps             = _mm_mul_ps(vfeps,H);
1208             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1209             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1210             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1211
1212             fscal            = felec;
1213
1214             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1215
1216             /* Calculate temporary vectorial force */
1217             tx               = _mm_mul_ps(fscal,dx10);
1218             ty               = _mm_mul_ps(fscal,dy10);
1219             tz               = _mm_mul_ps(fscal,dz10);
1220
1221             /* Update vectorial force */
1222             fix1             = _mm_add_ps(fix1,tx);
1223             fiy1             = _mm_add_ps(fiy1,ty);
1224             fiz1             = _mm_add_ps(fiz1,tz);
1225
1226             fjx0             = _mm_add_ps(fjx0,tx);
1227             fjy0             = _mm_add_ps(fjy0,ty);
1228             fjz0             = _mm_add_ps(fjz0,tz);
1229
1230             /**************************
1231              * CALCULATE INTERACTIONS *
1232              **************************/
1233
1234             r20              = _mm_mul_ps(rsq20,rinv20);
1235             r20              = _mm_andnot_ps(dummy_mask,r20);
1236
1237             /* Compute parameters for interactions between i and j atoms */
1238             qq20             = _mm_mul_ps(iq2,jq0);
1239
1240             /* Calculate table index by multiplying r with table scale and truncate to integer */
1241             rt               = _mm_mul_ps(r20,vftabscale);
1242             vfitab           = _mm_cvttps_epi32(rt);
1243             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1244             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1245
1246             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1247             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1248             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1249             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1250             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1251             _MM_TRANSPOSE4_PS(Y,F,G,H);
1252             Heps             = _mm_mul_ps(vfeps,H);
1253             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1254             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1255             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1256
1257             fscal            = felec;
1258
1259             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1260
1261             /* Calculate temporary vectorial force */
1262             tx               = _mm_mul_ps(fscal,dx20);
1263             ty               = _mm_mul_ps(fscal,dy20);
1264             tz               = _mm_mul_ps(fscal,dz20);
1265
1266             /* Update vectorial force */
1267             fix2             = _mm_add_ps(fix2,tx);
1268             fiy2             = _mm_add_ps(fiy2,ty);
1269             fiz2             = _mm_add_ps(fiz2,tz);
1270
1271             fjx0             = _mm_add_ps(fjx0,tx);
1272             fjy0             = _mm_add_ps(fjy0,ty);
1273             fjz0             = _mm_add_ps(fjz0,tz);
1274
1275             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1276             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1277             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1278             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1279
1280             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1281
1282             /* Inner loop uses 142 flops */
1283         }
1284
1285         /* End of innermost loop */
1286
1287         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1288                                               f+i_coord_offset,fshift+i_shift_offset);
1289
1290         /* Increment number of inner iterations */
1291         inneriter                  += j_index_end - j_index_start;
1292
1293         /* Outer loop uses 18 flops */
1294     }
1295
1296     /* Increment number of outer iterations */
1297     outeriter        += nri;
1298
1299     /* Update outer/inner flops */
1300
1301     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1302 }