src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_sse2_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 sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse2_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     __m128i          ewitab;
 107     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     real             *ewtab;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 132
 133     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 134     ewtab            = fr->ic->tabq_coul_FDV0;
 135     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 136     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 141     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 142     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 143     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181         fix1             = _mm_setzero_ps();
 182         fiy1             = _mm_setzero_ps();
 183         fiz1             = _mm_setzero_ps();
 184         fix2             = _mm_setzero_ps();
 185         fiy2             = _mm_setzero_ps();
 186         fiz2             = _mm_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_ps();
 190         vvdwsum          = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx10             = _mm_sub_ps(ix1,jx0);
 216             dy10             = _mm_sub_ps(iy1,jy0);
 217             dz10             = _mm_sub_ps(iz1,jz0);
 218             dx20             = _mm_sub_ps(ix2,jx0);
 219             dy20             = _mm_sub_ps(iy2,jy0);
 220             dz20             = _mm_sub_ps(iz2,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226
 227             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 228             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 229             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 230
 231             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 232             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 234
 235             /* Load parameters for j particles */
 236             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 237                                                               charge+jnrC+0,charge+jnrD+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242
 243             fjx0             = _mm_setzero_ps();
 244             fjy0             = _mm_setzero_ps();
 245             fjz0             = _mm_setzero_ps();
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             r00              = _mm_mul_ps(rsq00,rinv00);
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             qq00             = _mm_mul_ps(iq0,jq0);
 255             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 256                                          vdwparam+vdwioffset0+vdwjidx0B,
 257                                          vdwparam+vdwioffset0+vdwjidx0C,
 258                                          vdwparam+vdwioffset0+vdwjidx0D,
 259                                          &c6_00,&c12_00);
 260
 261             /* Calculate table index by multiplying r with table scale and truncate to integer */
 262             rt               = _mm_mul_ps(r00,vftabscale);
 263             vfitab           = _mm_cvttps_epi32(rt);
 264             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 265             vfitab           = _mm_slli_epi32(vfitab,3);
 266
 267             /* EWALD ELECTROSTATICS */
 268
 269             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 270             ewrt             = _mm_mul_ps(r00,ewtabscale);
 271             ewitab           = _mm_cvttps_epi32(ewrt);
 272             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 273             ewitab           = _mm_slli_epi32(ewitab,2);
 274             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 275             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 276             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 277             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 278             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 279             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 280             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 281             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 282             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 283
 284             /* CUBIC SPLINE TABLE DISPERSION */
 285             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 286             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 287             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 288             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 289             _MM_TRANSPOSE4_PS(Y,F,G,H);
 290             Heps             = _mm_mul_ps(vfeps,H);
 291             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 292             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 293             vvdw6            = _mm_mul_ps(c6_00,VV);
 294             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 295             fvdw6            = _mm_mul_ps(c6_00,FF);
 296
 297             /* CUBIC SPLINE TABLE REPULSION */
 298             vfitab           = _mm_add_epi32(vfitab,ifour);
 299             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 300             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 301             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 302             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 303             _MM_TRANSPOSE4_PS(Y,F,G,H);
 304             Heps             = _mm_mul_ps(vfeps,H);
 305             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 306             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 307             vvdw12           = _mm_mul_ps(c12_00,VV);
 308             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 309             fvdw12           = _mm_mul_ps(c12_00,FF);
 310             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 311             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm_add_ps(velecsum,velec);
 315             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 316
 317             fscal            = _mm_add_ps(felec,fvdw);
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm_mul_ps(fscal,dx00);
 321             ty               = _mm_mul_ps(fscal,dy00);
 322             tz               = _mm_mul_ps(fscal,dz00);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm_add_ps(fix0,tx);
 326             fiy0             = _mm_add_ps(fiy0,ty);
 327             fiz0             = _mm_add_ps(fiz0,tz);
 328
 329             fjx0             = _mm_add_ps(fjx0,tx);
 330             fjy0             = _mm_add_ps(fjy0,ty);
 331             fjz0             = _mm_add_ps(fjz0,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             r10              = _mm_mul_ps(rsq10,rinv10);
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq10             = _mm_mul_ps(iq1,jq0);
 341
 342             /* EWALD ELECTROSTATICS */
 343
 344             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 345             ewrt             = _mm_mul_ps(r10,ewtabscale);
 346             ewitab           = _mm_cvttps_epi32(ewrt);
 347             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 348             ewitab           = _mm_slli_epi32(ewitab,2);
 349             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 350             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 351             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 352             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 353             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 354             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 355             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 356             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 357             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_ps(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_ps(fscal,dx10);
 366             ty               = _mm_mul_ps(fscal,dy10);
 367             tz               = _mm_mul_ps(fscal,dz10);
 368
 369             /* Update vectorial force */
 370             fix1             = _mm_add_ps(fix1,tx);
 371             fiy1             = _mm_add_ps(fiy1,ty);
 372             fiz1             = _mm_add_ps(fiz1,tz);
 373
 374             fjx0             = _mm_add_ps(fjx0,tx);
 375             fjy0             = _mm_add_ps(fjy0,ty);
 376             fjz0             = _mm_add_ps(fjz0,tz);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             r20              = _mm_mul_ps(rsq20,rinv20);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq20             = _mm_mul_ps(iq2,jq0);
 386
 387             /* EWALD ELECTROSTATICS */
 388
 389             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 390             ewrt             = _mm_mul_ps(r20,ewtabscale);
 391             ewitab           = _mm_cvttps_epi32(ewrt);
 392             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 393             ewitab           = _mm_slli_epi32(ewitab,2);
 394             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 395             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 396             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 397             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 398             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 399             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 400             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 401             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 402             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velecsum         = _mm_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm_mul_ps(fscal,dx20);
 411             ty               = _mm_mul_ps(fscal,dy20);
 412             tz               = _mm_mul_ps(fscal,dz20);
 413
 414             /* Update vectorial force */
 415             fix2             = _mm_add_ps(fix2,tx);
 416             fiy2             = _mm_add_ps(fiy2,ty);
 417             fiz2             = _mm_add_ps(fiz2,tz);
 418
 419             fjx0             = _mm_add_ps(fjx0,tx);
 420             fjy0             = _mm_add_ps(fjy0,ty);
 421             fjz0             = _mm_add_ps(fjz0,tz);
 422
 423             fjptrA             = f+j_coord_offsetA;
 424             fjptrB             = f+j_coord_offsetB;
 425             fjptrC             = f+j_coord_offsetC;
 426             fjptrD             = f+j_coord_offsetD;
 427
 428             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 429
 430             /* Inner loop uses 157 flops */
 431         }
 432
 433         if(jidx<j_index_end)
 434         {
 435
 436             /* Get j neighbor index, and coordinate index */
 437             jnrlistA         = jjnr[jidx];
 438             jnrlistB         = jjnr[jidx+1];
 439             jnrlistC         = jjnr[jidx+2];
 440             jnrlistD         = jjnr[jidx+3];
 441             /* Sign of each element will be negative for non-real atoms.
 442              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 443              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 444              */
 445             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 446             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 447             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 448             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 449             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 450             j_coord_offsetA  = DIM*jnrA;
 451             j_coord_offsetB  = DIM*jnrB;
 452             j_coord_offsetC  = DIM*jnrC;
 453             j_coord_offsetD  = DIM*jnrD;
 454
 455             /* load j atom coordinates */
 456             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 457                                               x+j_coord_offsetC,x+j_coord_offsetD,
 458                                               &jx0,&jy0,&jz0);
 459
 460             /* Calculate displacement vector */
 461             dx00             = _mm_sub_ps(ix0,jx0);
 462             dy00             = _mm_sub_ps(iy0,jy0);
 463             dz00             = _mm_sub_ps(iz0,jz0);
 464             dx10             = _mm_sub_ps(ix1,jx0);
 465             dy10             = _mm_sub_ps(iy1,jy0);
 466             dz10             = _mm_sub_ps(iz1,jz0);
 467             dx20             = _mm_sub_ps(ix2,jx0);
 468             dy20             = _mm_sub_ps(iy2,jy0);
 469             dz20             = _mm_sub_ps(iz2,jz0);
 470
 471             /* Calculate squared distance and things based on it */
 472             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 473             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 474             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 475
 476             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 477             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 478             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 479
 480             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 481             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 482             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 483
 484             /* Load parameters for j particles */
 485             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 486                                                               charge+jnrC+0,charge+jnrD+0);
 487             vdwjidx0A        = 2*vdwtype[jnrA+0];
 488             vdwjidx0B        = 2*vdwtype[jnrB+0];
 489             vdwjidx0C        = 2*vdwtype[jnrC+0];
 490             vdwjidx0D        = 2*vdwtype[jnrD+0];
 491
 492             fjx0             = _mm_setzero_ps();
 493             fjy0             = _mm_setzero_ps();
 494             fjz0             = _mm_setzero_ps();
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r00              = _mm_mul_ps(rsq00,rinv00);
 501             r00              = _mm_andnot_ps(dummy_mask,r00);
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             qq00             = _mm_mul_ps(iq0,jq0);
 505             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 506                                          vdwparam+vdwioffset0+vdwjidx0B,
 507                                          vdwparam+vdwioffset0+vdwjidx0C,
 508                                          vdwparam+vdwioffset0+vdwjidx0D,
 509                                          &c6_00,&c12_00);
 510
 511             /* Calculate table index by multiplying r with table scale and truncate to integer */
 512             rt               = _mm_mul_ps(r00,vftabscale);
 513             vfitab           = _mm_cvttps_epi32(rt);
 514             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 515             vfitab           = _mm_slli_epi32(vfitab,3);
 516
 517             /* EWALD ELECTROSTATICS */
 518
 519             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 520             ewrt             = _mm_mul_ps(r00,ewtabscale);
 521             ewitab           = _mm_cvttps_epi32(ewrt);
 522             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 523             ewitab           = _mm_slli_epi32(ewitab,2);
 524             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 525             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 526             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 527             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 528             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 529             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 530             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 531             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 532             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 533
 534             /* CUBIC SPLINE TABLE DISPERSION */
 535             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 536             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 537             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 538             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 539             _MM_TRANSPOSE4_PS(Y,F,G,H);
 540             Heps             = _mm_mul_ps(vfeps,H);
 541             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 542             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 543             vvdw6            = _mm_mul_ps(c6_00,VV);
 544             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 545             fvdw6            = _mm_mul_ps(c6_00,FF);
 546
 547             /* CUBIC SPLINE TABLE REPULSION */
 548             vfitab           = _mm_add_epi32(vfitab,ifour);
 549             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 550             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 551             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 552             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 553             _MM_TRANSPOSE4_PS(Y,F,G,H);
 554             Heps             = _mm_mul_ps(vfeps,H);
 555             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 556             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 557             vvdw12           = _mm_mul_ps(c12_00,VV);
 558             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 559             fvdw12           = _mm_mul_ps(c12_00,FF);
 560             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 561             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _mm_andnot_ps(dummy_mask,velec);
 565             velecsum         = _mm_add_ps(velecsum,velec);
 566             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 567             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 568
 569             fscal            = _mm_add_ps(felec,fvdw);
 570
 571             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm_mul_ps(fscal,dx00);
 575             ty               = _mm_mul_ps(fscal,dy00);
 576             tz               = _mm_mul_ps(fscal,dz00);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_add_ps(fix0,tx);
 580             fiy0             = _mm_add_ps(fiy0,ty);
 581             fiz0             = _mm_add_ps(fiz0,tz);
 582
 583             fjx0             = _mm_add_ps(fjx0,tx);
 584             fjy0             = _mm_add_ps(fjy0,ty);
 585             fjz0             = _mm_add_ps(fjz0,tz);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r10              = _mm_mul_ps(rsq10,rinv10);
 592             r10              = _mm_andnot_ps(dummy_mask,r10);
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq10             = _mm_mul_ps(iq1,jq0);
 596
 597             /* EWALD ELECTROSTATICS */
 598
 599             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 600             ewrt             = _mm_mul_ps(r10,ewtabscale);
 601             ewitab           = _mm_cvttps_epi32(ewrt);
 602             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 603             ewitab           = _mm_slli_epi32(ewitab,2);
 604             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 605             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 606             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 607             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 608             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 609             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 610             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 611             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 612             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm_andnot_ps(dummy_mask,velec);
 616             velecsum         = _mm_add_ps(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm_mul_ps(fscal,dx10);
 624             ty               = _mm_mul_ps(fscal,dy10);
 625             tz               = _mm_mul_ps(fscal,dz10);
 626
 627             /* Update vectorial force */
 628             fix1             = _mm_add_ps(fix1,tx);
 629             fiy1             = _mm_add_ps(fiy1,ty);
 630             fiz1             = _mm_add_ps(fiz1,tz);
 631
 632             fjx0             = _mm_add_ps(fjx0,tx);
 633             fjy0             = _mm_add_ps(fjy0,ty);
 634             fjz0             = _mm_add_ps(fjz0,tz);
 635
 636             /**************************
 637              * CALCULATE INTERACTIONS *
 638              **************************/
 639
 640             r20              = _mm_mul_ps(rsq20,rinv20);
 641             r20              = _mm_andnot_ps(dummy_mask,r20);
 642
 643             /* Compute parameters for interactions between i and j atoms */
 644             qq20             = _mm_mul_ps(iq2,jq0);
 645
 646             /* EWALD ELECTROSTATICS */
 647
 648             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 649             ewrt             = _mm_mul_ps(r20,ewtabscale);
 650             ewitab           = _mm_cvttps_epi32(ewrt);
 651             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 652             ewitab           = _mm_slli_epi32(ewitab,2);
 653             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 654             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 655             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 656             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 657             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 658             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 659             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 660             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 661             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 662
 663             /* Update potential sum for this i atom from the interaction with this j atom. */
 664             velec            = _mm_andnot_ps(dummy_mask,velec);
 665             velecsum         = _mm_add_ps(velecsum,velec);
 666
 667             fscal            = felec;
 668
 669             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 670
 671             /* Calculate temporary vectorial force */
 672             tx               = _mm_mul_ps(fscal,dx20);
 673             ty               = _mm_mul_ps(fscal,dy20);
 674             tz               = _mm_mul_ps(fscal,dz20);
 675
 676             /* Update vectorial force */
 677             fix2             = _mm_add_ps(fix2,tx);
 678             fiy2             = _mm_add_ps(fiy2,ty);
 679             fiz2             = _mm_add_ps(fiz2,tz);
 680
 681             fjx0             = _mm_add_ps(fjx0,tx);
 682             fjy0             = _mm_add_ps(fjy0,ty);
 683             fjz0             = _mm_add_ps(fjz0,tz);
 684
 685             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 686             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 687             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 688             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 689
 690             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 691
 692             /* Inner loop uses 160 flops */
 693         }
 694
 695         /* End of innermost loop */
 696
 697         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 698                                               f+i_coord_offset,fshift+i_shift_offset);
 699
 700         ggid                        = gid[iidx];
 701         /* Update potential energies */
 702         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 703         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 704
 705         /* Increment number of inner iterations */
 706         inneriter                  += j_index_end - j_index_start;
 707
 708         /* Outer loop uses 20 flops */
 709     }
 710
 711     /* Increment number of outer iterations */
 712     outeriter        += nri;
 713
 714     /* Update outer/inner flops */
 715
 716     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 717 }
 718 /*
 719  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_single
 720  * Electrostatics interaction: Ewald
 721  * VdW interaction:            CubicSplineTable
 722  * Geometry:                   Water3-Particle
 723  * Calculate force/pot:        Force
 724  */
 725 void
 726 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_single
 727                     (t_nblist                    * gmx_restrict       nlist,
 728                      rvec                        * gmx_restrict          xx,
 729                      rvec                        * gmx_restrict          ff,
 730                      t_forcerec                  * gmx_restrict          fr,
 731                      t_mdatoms                   * gmx_restrict     mdatoms,
 732                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 733                      t_nrnb                      * gmx_restrict        nrnb)
 734 {
 735     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 736      * just 0 for non-waters.
 737      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 738      * jnr indices corresponding to data put in the four positions in the SIMD register.
 739      */
 740     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 741     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 742     int              jnrA,jnrB,jnrC,jnrD;
 743     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 744     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 745     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 746     real             rcutoff_scalar;
 747     real             *shiftvec,*fshift,*x,*f;
 748     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 749     real             scratch[4*DIM];
 750     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 751     int              vdwioffset0;
 752     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 753     int              vdwioffset1;
 754     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 755     int              vdwioffset2;
 756     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 757     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 758     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 759     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 760     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 761     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 762     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 763     real             *charge;
 764     int              nvdwtype;
 765     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 766     int              *vdwtype;
 767     real             *vdwparam;
 768     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 769     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 770     __m128i          vfitab;
 771     __m128i          ifour       = _mm_set1_epi32(4);
 772     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 773     real             *vftab;
 774     __m128i          ewitab;
 775     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 776     real             *ewtab;
 777     __m128           dummy_mask,cutoff_mask;
 778     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 779     __m128           one     = _mm_set1_ps(1.0);
 780     __m128           two     = _mm_set1_ps(2.0);
 781     x                = xx[0];
 782     f                = ff[0];
 783
 784     nri              = nlist->nri;
 785     iinr             = nlist->iinr;
 786     jindex           = nlist->jindex;
 787     jjnr             = nlist->jjnr;
 788     shiftidx         = nlist->shift;
 789     gid              = nlist->gid;
 790     shiftvec         = fr->shift_vec[0];
 791     fshift           = fr->fshift[0];
 792     facel            = _mm_set1_ps(fr->epsfac);
 793     charge           = mdatoms->chargeA;
 794     nvdwtype         = fr->ntype;
 795     vdwparam         = fr->nbfp;
 796     vdwtype          = mdatoms->typeA;
 797
 798     vftab            = kernel_data->table_vdw->data;
 799     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 800
 801     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 802     ewtab            = fr->ic->tabq_coul_F;
 803     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 804     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 805
 806     /* Setup water-specific parameters */
 807     inr              = nlist->iinr[0];
 808     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 809     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 810     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 811     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 812
 813     /* Avoid stupid compiler warnings */
 814     jnrA = jnrB = jnrC = jnrD = 0;
 815     j_coord_offsetA = 0;
 816     j_coord_offsetB = 0;
 817     j_coord_offsetC = 0;
 818     j_coord_offsetD = 0;
 819
 820     outeriter        = 0;
 821     inneriter        = 0;
 822
 823     for(iidx=0;iidx<4*DIM;iidx++)
 824     {
 825         scratch[iidx] = 0.0;
 826     }
 827
 828     /* Start outer loop over neighborlists */
 829     for(iidx=0; iidx<nri; iidx++)
 830     {
 831         /* Load shift vector for this list */
 832         i_shift_offset   = DIM*shiftidx[iidx];
 833
 834         /* Load limits for loop over neighbors */
 835         j_index_start    = jindex[iidx];
 836         j_index_end      = jindex[iidx+1];
 837
 838         /* Get outer coordinate index */
 839         inr              = iinr[iidx];
 840         i_coord_offset   = DIM*inr;
 841
 842         /* Load i particle coords and add shift vector */
 843         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 844                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 845
 846         fix0             = _mm_setzero_ps();
 847         fiy0             = _mm_setzero_ps();
 848         fiz0             = _mm_setzero_ps();
 849         fix1             = _mm_setzero_ps();
 850         fiy1             = _mm_setzero_ps();
 851         fiz1             = _mm_setzero_ps();
 852         fix2             = _mm_setzero_ps();
 853         fiy2             = _mm_setzero_ps();
 854         fiz2             = _mm_setzero_ps();
 855
 856         /* Start inner kernel loop */
 857         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 858         {
 859
 860             /* Get j neighbor index, and coordinate index */
 861             jnrA             = jjnr[jidx];
 862             jnrB             = jjnr[jidx+1];
 863             jnrC             = jjnr[jidx+2];
 864             jnrD             = jjnr[jidx+3];
 865             j_coord_offsetA  = DIM*jnrA;
 866             j_coord_offsetB  = DIM*jnrB;
 867             j_coord_offsetC  = DIM*jnrC;
 868             j_coord_offsetD  = DIM*jnrD;
 869
 870             /* load j atom coordinates */
 871             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 872                                               x+j_coord_offsetC,x+j_coord_offsetD,
 873                                               &jx0,&jy0,&jz0);
 874
 875             /* Calculate displacement vector */
 876             dx00             = _mm_sub_ps(ix0,jx0);
 877             dy00             = _mm_sub_ps(iy0,jy0);
 878             dz00             = _mm_sub_ps(iz0,jz0);
 879             dx10             = _mm_sub_ps(ix1,jx0);
 880             dy10             = _mm_sub_ps(iy1,jy0);
 881             dz10             = _mm_sub_ps(iz1,jz0);
 882             dx20             = _mm_sub_ps(ix2,jx0);
 883             dy20             = _mm_sub_ps(iy2,jy0);
 884             dz20             = _mm_sub_ps(iz2,jz0);
 885
 886             /* Calculate squared distance and things based on it */
 887             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 888             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 889             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 890
 891             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 892             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 893             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 894
 895             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 896             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 897             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 898
 899             /* Load parameters for j particles */
 900             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 901                                                               charge+jnrC+0,charge+jnrD+0);
 902             vdwjidx0A        = 2*vdwtype[jnrA+0];
 903             vdwjidx0B        = 2*vdwtype[jnrB+0];
 904             vdwjidx0C        = 2*vdwtype[jnrC+0];
 905             vdwjidx0D        = 2*vdwtype[jnrD+0];
 906
 907             fjx0             = _mm_setzero_ps();
 908             fjy0             = _mm_setzero_ps();
 909             fjz0             = _mm_setzero_ps();
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             r00              = _mm_mul_ps(rsq00,rinv00);
 916
 917             /* Compute parameters for interactions between i and j atoms */
 918             qq00             = _mm_mul_ps(iq0,jq0);
 919             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 920                                          vdwparam+vdwioffset0+vdwjidx0B,
 921                                          vdwparam+vdwioffset0+vdwjidx0C,
 922                                          vdwparam+vdwioffset0+vdwjidx0D,
 923                                          &c6_00,&c12_00);
 924
 925             /* Calculate table index by multiplying r with table scale and truncate to integer */
 926             rt               = _mm_mul_ps(r00,vftabscale);
 927             vfitab           = _mm_cvttps_epi32(rt);
 928             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 929             vfitab           = _mm_slli_epi32(vfitab,3);
 930
 931             /* EWALD ELECTROSTATICS */
 932
 933             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 934             ewrt             = _mm_mul_ps(r00,ewtabscale);
 935             ewitab           = _mm_cvttps_epi32(ewrt);
 936             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 937             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 938                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 939                                          &ewtabF,&ewtabFn);
 940             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 941             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 942
 943             /* CUBIC SPLINE TABLE DISPERSION */
 944             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 945             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 946             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 947             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 948             _MM_TRANSPOSE4_PS(Y,F,G,H);
 949             Heps             = _mm_mul_ps(vfeps,H);
 950             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 951             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 952             fvdw6            = _mm_mul_ps(c6_00,FF);
 953
 954             /* CUBIC SPLINE TABLE REPULSION */
 955             vfitab           = _mm_add_epi32(vfitab,ifour);
 956             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 957             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 958             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 959             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 960             _MM_TRANSPOSE4_PS(Y,F,G,H);
 961             Heps             = _mm_mul_ps(vfeps,H);
 962             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 963             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 964             fvdw12           = _mm_mul_ps(c12_00,FF);
 965             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 966
 967             fscal            = _mm_add_ps(felec,fvdw);
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm_mul_ps(fscal,dx00);
 971             ty               = _mm_mul_ps(fscal,dy00);
 972             tz               = _mm_mul_ps(fscal,dz00);
 973
 974             /* Update vectorial force */
 975             fix0             = _mm_add_ps(fix0,tx);
 976             fiy0             = _mm_add_ps(fiy0,ty);
 977             fiz0             = _mm_add_ps(fiz0,tz);
 978
 979             fjx0             = _mm_add_ps(fjx0,tx);
 980             fjy0             = _mm_add_ps(fjy0,ty);
 981             fjz0             = _mm_add_ps(fjz0,tz);
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             r10              = _mm_mul_ps(rsq10,rinv10);
 988
 989             /* Compute parameters for interactions between i and j atoms */
 990             qq10             = _mm_mul_ps(iq1,jq0);
 991
 992             /* EWALD ELECTROSTATICS */
 993
 994             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 995             ewrt             = _mm_mul_ps(r10,ewtabscale);
 996             ewitab           = _mm_cvttps_epi32(ewrt);
 997             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 998             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 999                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1000                                          &ewtabF,&ewtabFn);
1001             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1002             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1003
1004             fscal            = felec;
1005
1006             /* Calculate temporary vectorial force */
1007             tx               = _mm_mul_ps(fscal,dx10);
1008             ty               = _mm_mul_ps(fscal,dy10);
1009             tz               = _mm_mul_ps(fscal,dz10);
1010
1011             /* Update vectorial force */
1012             fix1             = _mm_add_ps(fix1,tx);
1013             fiy1             = _mm_add_ps(fiy1,ty);
1014             fiz1             = _mm_add_ps(fiz1,tz);
1015
1016             fjx0             = _mm_add_ps(fjx0,tx);
1017             fjy0             = _mm_add_ps(fjy0,ty);
1018             fjz0             = _mm_add_ps(fjz0,tz);
1019
1020             /**************************
1021              * CALCULATE INTERACTIONS *
1022              **************************/
1023
1024             r20              = _mm_mul_ps(rsq20,rinv20);
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq20             = _mm_mul_ps(iq2,jq0);
1028
1029             /* EWALD ELECTROSTATICS */
1030
1031             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1032             ewrt             = _mm_mul_ps(r20,ewtabscale);
1033             ewitab           = _mm_cvttps_epi32(ewrt);
1034             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1035             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1036                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1037                                          &ewtabF,&ewtabFn);
1038             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1039             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1040
1041             fscal            = felec;
1042
1043             /* Calculate temporary vectorial force */
1044             tx               = _mm_mul_ps(fscal,dx20);
1045             ty               = _mm_mul_ps(fscal,dy20);
1046             tz               = _mm_mul_ps(fscal,dz20);
1047
1048             /* Update vectorial force */
1049             fix2             = _mm_add_ps(fix2,tx);
1050             fiy2             = _mm_add_ps(fiy2,ty);
1051             fiz2             = _mm_add_ps(fiz2,tz);
1052
1053             fjx0             = _mm_add_ps(fjx0,tx);
1054             fjy0             = _mm_add_ps(fjy0,ty);
1055             fjz0             = _mm_add_ps(fjz0,tz);
1056
1057             fjptrA             = f+j_coord_offsetA;
1058             fjptrB             = f+j_coord_offsetB;
1059             fjptrC             = f+j_coord_offsetC;
1060             fjptrD             = f+j_coord_offsetD;
1061
1062             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1063
1064             /* Inner loop uses 134 flops */
1065         }
1066
1067         if(jidx<j_index_end)
1068         {
1069
1070             /* Get j neighbor index, and coordinate index */
1071             jnrlistA         = jjnr[jidx];
1072             jnrlistB         = jjnr[jidx+1];
1073             jnrlistC         = jjnr[jidx+2];
1074             jnrlistD         = jjnr[jidx+3];
1075             /* Sign of each element will be negative for non-real atoms.
1076              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1077              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1078              */
1079             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1080             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1081             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1082             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1083             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1084             j_coord_offsetA  = DIM*jnrA;
1085             j_coord_offsetB  = DIM*jnrB;
1086             j_coord_offsetC  = DIM*jnrC;
1087             j_coord_offsetD  = DIM*jnrD;
1088
1089             /* load j atom coordinates */
1090             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1091                                               x+j_coord_offsetC,x+j_coord_offsetD,
1092                                               &jx0,&jy0,&jz0);
1093
1094             /* Calculate displacement vector */
1095             dx00             = _mm_sub_ps(ix0,jx0);
1096             dy00             = _mm_sub_ps(iy0,jy0);
1097             dz00             = _mm_sub_ps(iz0,jz0);
1098             dx10             = _mm_sub_ps(ix1,jx0);
1099             dy10             = _mm_sub_ps(iy1,jy0);
1100             dz10             = _mm_sub_ps(iz1,jz0);
1101             dx20             = _mm_sub_ps(ix2,jx0);
1102             dy20             = _mm_sub_ps(iy2,jy0);
1103             dz20             = _mm_sub_ps(iz2,jz0);
1104
1105             /* Calculate squared distance and things based on it */
1106             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1107             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1108             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1109
1110             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1111             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1112             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1113
1114             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1115             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1116             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1117
1118             /* Load parameters for j particles */
1119             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1120                                                               charge+jnrC+0,charge+jnrD+0);
1121             vdwjidx0A        = 2*vdwtype[jnrA+0];
1122             vdwjidx0B        = 2*vdwtype[jnrB+0];
1123             vdwjidx0C        = 2*vdwtype[jnrC+0];
1124             vdwjidx0D        = 2*vdwtype[jnrD+0];
1125
1126             fjx0             = _mm_setzero_ps();
1127             fjy0             = _mm_setzero_ps();
1128             fjz0             = _mm_setzero_ps();
1129
1130             /**************************
1131              * CALCULATE INTERACTIONS *
1132              **************************/
1133
1134             r00              = _mm_mul_ps(rsq00,rinv00);
1135             r00              = _mm_andnot_ps(dummy_mask,r00);
1136
1137             /* Compute parameters for interactions between i and j atoms */
1138             qq00             = _mm_mul_ps(iq0,jq0);
1139             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1140                                          vdwparam+vdwioffset0+vdwjidx0B,
1141                                          vdwparam+vdwioffset0+vdwjidx0C,
1142                                          vdwparam+vdwioffset0+vdwjidx0D,
1143                                          &c6_00,&c12_00);
1144
1145             /* Calculate table index by multiplying r with table scale and truncate to integer */
1146             rt               = _mm_mul_ps(r00,vftabscale);
1147             vfitab           = _mm_cvttps_epi32(rt);
1148             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1149             vfitab           = _mm_slli_epi32(vfitab,3);
1150
1151             /* EWALD ELECTROSTATICS */
1152
1153             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1154             ewrt             = _mm_mul_ps(r00,ewtabscale);
1155             ewitab           = _mm_cvttps_epi32(ewrt);
1156             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1157             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1158                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1159                                          &ewtabF,&ewtabFn);
1160             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1161             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1162
1163             /* CUBIC SPLINE TABLE DISPERSION */
1164             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1165             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1166             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1167             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1168             _MM_TRANSPOSE4_PS(Y,F,G,H);
1169             Heps             = _mm_mul_ps(vfeps,H);
1170             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1171             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1172             fvdw6            = _mm_mul_ps(c6_00,FF);
1173
1174             /* CUBIC SPLINE TABLE REPULSION */
1175             vfitab           = _mm_add_epi32(vfitab,ifour);
1176             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1177             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1178             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1179             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1180             _MM_TRANSPOSE4_PS(Y,F,G,H);
1181             Heps             = _mm_mul_ps(vfeps,H);
1182             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1183             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1184             fvdw12           = _mm_mul_ps(c12_00,FF);
1185             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1186
1187             fscal            = _mm_add_ps(felec,fvdw);
1188
1189             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1190
1191             /* Calculate temporary vectorial force */
1192             tx               = _mm_mul_ps(fscal,dx00);
1193             ty               = _mm_mul_ps(fscal,dy00);
1194             tz               = _mm_mul_ps(fscal,dz00);
1195
1196             /* Update vectorial force */
1197             fix0             = _mm_add_ps(fix0,tx);
1198             fiy0             = _mm_add_ps(fiy0,ty);
1199             fiz0             = _mm_add_ps(fiz0,tz);
1200
1201             fjx0             = _mm_add_ps(fjx0,tx);
1202             fjy0             = _mm_add_ps(fjy0,ty);
1203             fjz0             = _mm_add_ps(fjz0,tz);
1204
1205             /**************************
1206              * CALCULATE INTERACTIONS *
1207              **************************/
1208
1209             r10              = _mm_mul_ps(rsq10,rinv10);
1210             r10              = _mm_andnot_ps(dummy_mask,r10);
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq10             = _mm_mul_ps(iq1,jq0);
1214
1215             /* EWALD ELECTROSTATICS */
1216
1217             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1218             ewrt             = _mm_mul_ps(r10,ewtabscale);
1219             ewitab           = _mm_cvttps_epi32(ewrt);
1220             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1221             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1222                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1223                                          &ewtabF,&ewtabFn);
1224             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1225             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1226
1227             fscal            = felec;
1228
1229             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1230
1231             /* Calculate temporary vectorial force */
1232             tx               = _mm_mul_ps(fscal,dx10);
1233             ty               = _mm_mul_ps(fscal,dy10);
1234             tz               = _mm_mul_ps(fscal,dz10);
1235
1236             /* Update vectorial force */
1237             fix1             = _mm_add_ps(fix1,tx);
1238             fiy1             = _mm_add_ps(fiy1,ty);
1239             fiz1             = _mm_add_ps(fiz1,tz);
1240
1241             fjx0             = _mm_add_ps(fjx0,tx);
1242             fjy0             = _mm_add_ps(fjy0,ty);
1243             fjz0             = _mm_add_ps(fjz0,tz);
1244
1245             /**************************
1246              * CALCULATE INTERACTIONS *
1247              **************************/
1248
1249             r20              = _mm_mul_ps(rsq20,rinv20);
1250             r20              = _mm_andnot_ps(dummy_mask,r20);
1251
1252             /* Compute parameters for interactions between i and j atoms */
1253             qq20             = _mm_mul_ps(iq2,jq0);
1254
1255             /* EWALD ELECTROSTATICS */
1256
1257             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1258             ewrt             = _mm_mul_ps(r20,ewtabscale);
1259             ewitab           = _mm_cvttps_epi32(ewrt);
1260             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1261             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1262                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1263                                          &ewtabF,&ewtabFn);
1264             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1265             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1266
1267             fscal            = felec;
1268
1269             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1270
1271             /* Calculate temporary vectorial force */
1272             tx               = _mm_mul_ps(fscal,dx20);
1273             ty               = _mm_mul_ps(fscal,dy20);
1274             tz               = _mm_mul_ps(fscal,dz20);
1275
1276             /* Update vectorial force */
1277             fix2             = _mm_add_ps(fix2,tx);
1278             fiy2             = _mm_add_ps(fiy2,ty);
1279             fiz2             = _mm_add_ps(fiz2,tz);
1280
1281             fjx0             = _mm_add_ps(fjx0,tx);
1282             fjy0             = _mm_add_ps(fjy0,ty);
1283             fjz0             = _mm_add_ps(fjz0,tz);
1284
1285             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1286             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1287             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1288             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1289
1290             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1291
1292             /* Inner loop uses 137 flops */
1293         }
1294
1295         /* End of innermost loop */
1296
1297         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1298                                               f+i_coord_offset,fshift+i_shift_offset);
1299
1300         /* Increment number of inner iterations */
1301         inneriter                  += j_index_end - j_index_start;
1302
1303         /* Outer loop uses 18 flops */
1304     }
1305
1306     /* Increment number of outer iterations */
1307     outeriter        += nri;
1308
1309     /* Update outer/inner flops */
1310
1311     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1312 }