src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_GeomW4P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_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              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m128i          ewitab;
 110     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     real             *ewtab;
 112     __m128           dummy_mask,cutoff_mask;
 113     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 114     __m128           one     = _mm_set1_ps(1.0);
 115     __m128           two     = _mm_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_vdw->data;
 134     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 135
 136     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 144     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 145     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm_setzero_ps();
 182         fiy0             = _mm_setzero_ps();
 183         fiz0             = _mm_setzero_ps();
 184         fix1             = _mm_setzero_ps();
 185         fiy1             = _mm_setzero_ps();
 186         fiz1             = _mm_setzero_ps();
 187         fix2             = _mm_setzero_ps();
 188         fiy2             = _mm_setzero_ps();
 189         fiz2             = _mm_setzero_ps();
 190         fix3             = _mm_setzero_ps();
 191         fiy3             = _mm_setzero_ps();
 192         fiz3             = _mm_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm_setzero_ps();
 196         vvdwsum          = _mm_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_ps(ix0,jx0);
 219             dy00             = _mm_sub_ps(iy0,jy0);
 220             dz00             = _mm_sub_ps(iz0,jz0);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx20             = _mm_sub_ps(ix2,jx0);
 225             dy20             = _mm_sub_ps(iy2,jy0);
 226             dz20             = _mm_sub_ps(iz2,jz0);
 227             dx30             = _mm_sub_ps(ix3,jx0);
 228             dy30             = _mm_sub_ps(iy3,jy0);
 229             dz30             = _mm_sub_ps(iz3,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 233             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 234             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 235             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 236
 237             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 240             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 241
 242             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 244             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 245
 246             /* Load parameters for j particles */
 247             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 248                                                               charge+jnrC+0,charge+jnrD+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251             vdwjidx0C        = 2*vdwtype[jnrC+0];
 252             vdwjidx0D        = 2*vdwtype[jnrD+0];
 253
 254             fjx0             = _mm_setzero_ps();
 255             fjy0             = _mm_setzero_ps();
 256             fjz0             = _mm_setzero_ps();
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             r00              = _mm_mul_ps(rsq00,rinv00);
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 266                                          vdwparam+vdwioffset0+vdwjidx0B,
 267                                          vdwparam+vdwioffset0+vdwjidx0C,
 268                                          vdwparam+vdwioffset0+vdwjidx0D,
 269                                          &c6_00,&c12_00);
 270
 271             /* Calculate table index by multiplying r with table scale and truncate to integer */
 272             rt               = _mm_mul_ps(r00,vftabscale);
 273             vfitab           = _mm_cvttps_epi32(rt);
 274             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 275             vfitab           = _mm_slli_epi32(vfitab,3);
 276
 277             /* CUBIC SPLINE TABLE DISPERSION */
 278             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 280             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 281             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 282             _MM_TRANSPOSE4_PS(Y,F,G,H);
 283             Heps             = _mm_mul_ps(vfeps,H);
 284             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 285             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 286             vvdw6            = _mm_mul_ps(c6_00,VV);
 287             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 288             fvdw6            = _mm_mul_ps(c6_00,FF);
 289
 290             /* CUBIC SPLINE TABLE REPULSION */
 291             vfitab           = _mm_add_epi32(vfitab,ifour);
 292             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 293             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 294             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 295             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 296             _MM_TRANSPOSE4_PS(Y,F,G,H);
 297             Heps             = _mm_mul_ps(vfeps,H);
 298             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 299             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 300             vvdw12           = _mm_mul_ps(c12_00,VV);
 301             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 302             fvdw12           = _mm_mul_ps(c12_00,FF);
 303             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 304             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 308
 309             fscal            = fvdw;
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm_mul_ps(fscal,dx00);
 313             ty               = _mm_mul_ps(fscal,dy00);
 314             tz               = _mm_mul_ps(fscal,dz00);
 315
 316             /* Update vectorial force */
 317             fix0             = _mm_add_ps(fix0,tx);
 318             fiy0             = _mm_add_ps(fiy0,ty);
 319             fiz0             = _mm_add_ps(fiz0,tz);
 320
 321             fjx0             = _mm_add_ps(fjx0,tx);
 322             fjy0             = _mm_add_ps(fjy0,ty);
 323             fjz0             = _mm_add_ps(fjz0,tz);
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             r10              = _mm_mul_ps(rsq10,rinv10);
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq10             = _mm_mul_ps(iq1,jq0);
 333
 334             /* EWALD ELECTROSTATICS */
 335
 336             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 337             ewrt             = _mm_mul_ps(r10,ewtabscale);
 338             ewitab           = _mm_cvttps_epi32(ewrt);
 339             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 340             ewitab           = _mm_slli_epi32(ewitab,2);
 341             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 342             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 343             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 344             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 345             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 346             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 347             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 348             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 349             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velecsum         = _mm_add_ps(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm_mul_ps(fscal,dx10);
 358             ty               = _mm_mul_ps(fscal,dy10);
 359             tz               = _mm_mul_ps(fscal,dz10);
 360
 361             /* Update vectorial force */
 362             fix1             = _mm_add_ps(fix1,tx);
 363             fiy1             = _mm_add_ps(fiy1,ty);
 364             fiz1             = _mm_add_ps(fiz1,tz);
 365
 366             fjx0             = _mm_add_ps(fjx0,tx);
 367             fjy0             = _mm_add_ps(fjy0,ty);
 368             fjz0             = _mm_add_ps(fjz0,tz);
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             r20              = _mm_mul_ps(rsq20,rinv20);
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq20             = _mm_mul_ps(iq2,jq0);
 378
 379             /* EWALD ELECTROSTATICS */
 380
 381             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 382             ewrt             = _mm_mul_ps(r20,ewtabscale);
 383             ewitab           = _mm_cvttps_epi32(ewrt);
 384             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 385             ewitab           = _mm_slli_epi32(ewitab,2);
 386             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 387             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 388             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 389             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 390             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 391             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 392             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 393             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 394             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             velecsum         = _mm_add_ps(velecsum,velec);
 398
 399             fscal            = felec;
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm_mul_ps(fscal,dx20);
 403             ty               = _mm_mul_ps(fscal,dy20);
 404             tz               = _mm_mul_ps(fscal,dz20);
 405
 406             /* Update vectorial force */
 407             fix2             = _mm_add_ps(fix2,tx);
 408             fiy2             = _mm_add_ps(fiy2,ty);
 409             fiz2             = _mm_add_ps(fiz2,tz);
 410
 411             fjx0             = _mm_add_ps(fjx0,tx);
 412             fjy0             = _mm_add_ps(fjy0,ty);
 413             fjz0             = _mm_add_ps(fjz0,tz);
 414
 415             /**************************
 416              * CALCULATE INTERACTIONS *
 417              **************************/
 418
 419             r30              = _mm_mul_ps(rsq30,rinv30);
 420
 421             /* Compute parameters for interactions between i and j atoms */
 422             qq30             = _mm_mul_ps(iq3,jq0);
 423
 424             /* EWALD ELECTROSTATICS */
 425
 426             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 427             ewrt             = _mm_mul_ps(r30,ewtabscale);
 428             ewitab           = _mm_cvttps_epi32(ewrt);
 429             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 430             ewitab           = _mm_slli_epi32(ewitab,2);
 431             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 432             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 433             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 434             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 435             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 436             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 437             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 438             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 439             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 440
 441             /* Update potential sum for this i atom from the interaction with this j atom. */
 442             velecsum         = _mm_add_ps(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm_mul_ps(fscal,dx30);
 448             ty               = _mm_mul_ps(fscal,dy30);
 449             tz               = _mm_mul_ps(fscal,dz30);
 450
 451             /* Update vectorial force */
 452             fix3             = _mm_add_ps(fix3,tx);
 453             fiy3             = _mm_add_ps(fiy3,ty);
 454             fiz3             = _mm_add_ps(fiz3,tz);
 455
 456             fjx0             = _mm_add_ps(fjx0,tx);
 457             fjy0             = _mm_add_ps(fjy0,ty);
 458             fjz0             = _mm_add_ps(fjz0,tz);
 459
 460             fjptrA             = f+j_coord_offsetA;
 461             fjptrB             = f+j_coord_offsetB;
 462             fjptrC             = f+j_coord_offsetC;
 463             fjptrD             = f+j_coord_offsetD;
 464
 465             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 466
 467             /* Inner loop uses 179 flops */
 468         }
 469
 470         if(jidx<j_index_end)
 471         {
 472
 473             /* Get j neighbor index, and coordinate index */
 474             jnrlistA         = jjnr[jidx];
 475             jnrlistB         = jjnr[jidx+1];
 476             jnrlistC         = jjnr[jidx+2];
 477             jnrlistD         = jjnr[jidx+3];
 478             /* Sign of each element will be negative for non-real atoms.
 479              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 480              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 481              */
 482             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 483             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 484             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 485             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 486             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 487             j_coord_offsetA  = DIM*jnrA;
 488             j_coord_offsetB  = DIM*jnrB;
 489             j_coord_offsetC  = DIM*jnrC;
 490             j_coord_offsetD  = DIM*jnrD;
 491
 492             /* load j atom coordinates */
 493             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 494                                               x+j_coord_offsetC,x+j_coord_offsetD,
 495                                               &jx0,&jy0,&jz0);
 496
 497             /* Calculate displacement vector */
 498             dx00             = _mm_sub_ps(ix0,jx0);
 499             dy00             = _mm_sub_ps(iy0,jy0);
 500             dz00             = _mm_sub_ps(iz0,jz0);
 501             dx10             = _mm_sub_ps(ix1,jx0);
 502             dy10             = _mm_sub_ps(iy1,jy0);
 503             dz10             = _mm_sub_ps(iz1,jz0);
 504             dx20             = _mm_sub_ps(ix2,jx0);
 505             dy20             = _mm_sub_ps(iy2,jy0);
 506             dz20             = _mm_sub_ps(iz2,jz0);
 507             dx30             = _mm_sub_ps(ix3,jx0);
 508             dy30             = _mm_sub_ps(iy3,jy0);
 509             dz30             = _mm_sub_ps(iz3,jz0);
 510
 511             /* Calculate squared distance and things based on it */
 512             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 513             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 514             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 515             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 516
 517             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 518             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 519             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 520             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 521
 522             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 523             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 524             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 525
 526             /* Load parameters for j particles */
 527             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 528                                                               charge+jnrC+0,charge+jnrD+0);
 529             vdwjidx0A        = 2*vdwtype[jnrA+0];
 530             vdwjidx0B        = 2*vdwtype[jnrB+0];
 531             vdwjidx0C        = 2*vdwtype[jnrC+0];
 532             vdwjidx0D        = 2*vdwtype[jnrD+0];
 533
 534             fjx0             = _mm_setzero_ps();
 535             fjy0             = _mm_setzero_ps();
 536             fjz0             = _mm_setzero_ps();
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             r00              = _mm_mul_ps(rsq00,rinv00);
 543             r00              = _mm_andnot_ps(dummy_mask,r00);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 547                                          vdwparam+vdwioffset0+vdwjidx0B,
 548                                          vdwparam+vdwioffset0+vdwjidx0C,
 549                                          vdwparam+vdwioffset0+vdwjidx0D,
 550                                          &c6_00,&c12_00);
 551
 552             /* Calculate table index by multiplying r with table scale and truncate to integer */
 553             rt               = _mm_mul_ps(r00,vftabscale);
 554             vfitab           = _mm_cvttps_epi32(rt);
 555             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 556             vfitab           = _mm_slli_epi32(vfitab,3);
 557
 558             /* CUBIC SPLINE TABLE DISPERSION */
 559             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 560             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 561             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 562             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 563             _MM_TRANSPOSE4_PS(Y,F,G,H);
 564             Heps             = _mm_mul_ps(vfeps,H);
 565             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 566             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 567             vvdw6            = _mm_mul_ps(c6_00,VV);
 568             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 569             fvdw6            = _mm_mul_ps(c6_00,FF);
 570
 571             /* CUBIC SPLINE TABLE REPULSION */
 572             vfitab           = _mm_add_epi32(vfitab,ifour);
 573             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 574             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 575             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 576             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 577             _MM_TRANSPOSE4_PS(Y,F,G,H);
 578             Heps             = _mm_mul_ps(vfeps,H);
 579             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 580             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 581             vvdw12           = _mm_mul_ps(c12_00,VV);
 582             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 583             fvdw12           = _mm_mul_ps(c12_00,FF);
 584             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 585             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 589             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 590
 591             fscal            = fvdw;
 592
 593             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_ps(fscal,dx00);
 597             ty               = _mm_mul_ps(fscal,dy00);
 598             tz               = _mm_mul_ps(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm_add_ps(fix0,tx);
 602             fiy0             = _mm_add_ps(fiy0,ty);
 603             fiz0             = _mm_add_ps(fiz0,tz);
 604
 605             fjx0             = _mm_add_ps(fjx0,tx);
 606             fjy0             = _mm_add_ps(fjy0,ty);
 607             fjz0             = _mm_add_ps(fjz0,tz);
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             r10              = _mm_mul_ps(rsq10,rinv10);
 614             r10              = _mm_andnot_ps(dummy_mask,r10);
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq10             = _mm_mul_ps(iq1,jq0);
 618
 619             /* EWALD ELECTROSTATICS */
 620
 621             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 622             ewrt             = _mm_mul_ps(r10,ewtabscale);
 623             ewitab           = _mm_cvttps_epi32(ewrt);
 624             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 625             ewitab           = _mm_slli_epi32(ewitab,2);
 626             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 627             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 628             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 629             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 630             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 631             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 632             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 633             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 634             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 635
 636             /* Update potential sum for this i atom from the interaction with this j atom. */
 637             velec            = _mm_andnot_ps(dummy_mask,velec);
 638             velecsum         = _mm_add_ps(velecsum,velec);
 639
 640             fscal            = felec;
 641
 642             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 643
 644             /* Calculate temporary vectorial force */
 645             tx               = _mm_mul_ps(fscal,dx10);
 646             ty               = _mm_mul_ps(fscal,dy10);
 647             tz               = _mm_mul_ps(fscal,dz10);
 648
 649             /* Update vectorial force */
 650             fix1             = _mm_add_ps(fix1,tx);
 651             fiy1             = _mm_add_ps(fiy1,ty);
 652             fiz1             = _mm_add_ps(fiz1,tz);
 653
 654             fjx0             = _mm_add_ps(fjx0,tx);
 655             fjy0             = _mm_add_ps(fjy0,ty);
 656             fjz0             = _mm_add_ps(fjz0,tz);
 657
 658             /**************************
 659              * CALCULATE INTERACTIONS *
 660              **************************/
 661
 662             r20              = _mm_mul_ps(rsq20,rinv20);
 663             r20              = _mm_andnot_ps(dummy_mask,r20);
 664
 665             /* Compute parameters for interactions between i and j atoms */
 666             qq20             = _mm_mul_ps(iq2,jq0);
 667
 668             /* EWALD ELECTROSTATICS */
 669
 670             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 671             ewrt             = _mm_mul_ps(r20,ewtabscale);
 672             ewitab           = _mm_cvttps_epi32(ewrt);
 673             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 674             ewitab           = _mm_slli_epi32(ewitab,2);
 675             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 676             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 677             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 678             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 679             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 680             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 681             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 682             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 683             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 684
 685             /* Update potential sum for this i atom from the interaction with this j atom. */
 686             velec            = _mm_andnot_ps(dummy_mask,velec);
 687             velecsum         = _mm_add_ps(velecsum,velec);
 688
 689             fscal            = felec;
 690
 691             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm_mul_ps(fscal,dx20);
 695             ty               = _mm_mul_ps(fscal,dy20);
 696             tz               = _mm_mul_ps(fscal,dz20);
 697
 698             /* Update vectorial force */
 699             fix2             = _mm_add_ps(fix2,tx);
 700             fiy2             = _mm_add_ps(fiy2,ty);
 701             fiz2             = _mm_add_ps(fiz2,tz);
 702
 703             fjx0             = _mm_add_ps(fjx0,tx);
 704             fjy0             = _mm_add_ps(fjy0,ty);
 705             fjz0             = _mm_add_ps(fjz0,tz);
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             r30              = _mm_mul_ps(rsq30,rinv30);
 712             r30              = _mm_andnot_ps(dummy_mask,r30);
 713
 714             /* Compute parameters for interactions between i and j atoms */
 715             qq30             = _mm_mul_ps(iq3,jq0);
 716
 717             /* EWALD ELECTROSTATICS */
 718
 719             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 720             ewrt             = _mm_mul_ps(r30,ewtabscale);
 721             ewitab           = _mm_cvttps_epi32(ewrt);
 722             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 723             ewitab           = _mm_slli_epi32(ewitab,2);
 724             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 725             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 726             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 727             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 728             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 729             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 730             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 731             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 732             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 733
 734             /* Update potential sum for this i atom from the interaction with this j atom. */
 735             velec            = _mm_andnot_ps(dummy_mask,velec);
 736             velecsum         = _mm_add_ps(velecsum,velec);
 737
 738             fscal            = felec;
 739
 740             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 741
 742             /* Calculate temporary vectorial force */
 743             tx               = _mm_mul_ps(fscal,dx30);
 744             ty               = _mm_mul_ps(fscal,dy30);
 745             tz               = _mm_mul_ps(fscal,dz30);
 746
 747             /* Update vectorial force */
 748             fix3             = _mm_add_ps(fix3,tx);
 749             fiy3             = _mm_add_ps(fiy3,ty);
 750             fiz3             = _mm_add_ps(fiz3,tz);
 751
 752             fjx0             = _mm_add_ps(fjx0,tx);
 753             fjy0             = _mm_add_ps(fjy0,ty);
 754             fjz0             = _mm_add_ps(fjz0,tz);
 755
 756             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 757             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 758             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 759             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 760
 761             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 762
 763             /* Inner loop uses 183 flops */
 764         }
 765
 766         /* End of innermost loop */
 767
 768         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 769                                               f+i_coord_offset,fshift+i_shift_offset);
 770
 771         ggid                        = gid[iidx];
 772         /* Update potential energies */
 773         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 774         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 775
 776         /* Increment number of inner iterations */
 777         inneriter                  += j_index_end - j_index_start;
 778
 779         /* Outer loop uses 26 flops */
 780     }
 781
 782     /* Increment number of outer iterations */
 783     outeriter        += nri;
 784
 785     /* Update outer/inner flops */
 786
 787     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*183);
 788 }
 789 /*
 790  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse2_single
 791  * Electrostatics interaction: Ewald
 792  * VdW interaction:            CubicSplineTable
 793  * Geometry:                   Water4-Particle
 794  * Calculate force/pot:        Force
 795  */
 796 void
 797 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse2_single
 798                     (t_nblist                    * gmx_restrict       nlist,
 799                      rvec                        * gmx_restrict          xx,
 800                      rvec                        * gmx_restrict          ff,
 801                      t_forcerec                  * gmx_restrict          fr,
 802                      t_mdatoms                   * gmx_restrict     mdatoms,
 803                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 804                      t_nrnb                      * gmx_restrict        nrnb)
 805 {
 806     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 807      * just 0 for non-waters.
 808      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 809      * jnr indices corresponding to data put in the four positions in the SIMD register.
 810      */
 811     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 812     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 813     int              jnrA,jnrB,jnrC,jnrD;
 814     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 815     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 816     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 817     real             rcutoff_scalar;
 818     real             *shiftvec,*fshift,*x,*f;
 819     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 820     real             scratch[4*DIM];
 821     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 822     int              vdwioffset0;
 823     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 824     int              vdwioffset1;
 825     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 826     int              vdwioffset2;
 827     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 828     int              vdwioffset3;
 829     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 830     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 831     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 832     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 833     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 834     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 835     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 836     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 837     real             *charge;
 838     int              nvdwtype;
 839     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 840     int              *vdwtype;
 841     real             *vdwparam;
 842     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 843     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 844     __m128i          vfitab;
 845     __m128i          ifour       = _mm_set1_epi32(4);
 846     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 847     real             *vftab;
 848     __m128i          ewitab;
 849     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 850     real             *ewtab;
 851     __m128           dummy_mask,cutoff_mask;
 852     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 853     __m128           one     = _mm_set1_ps(1.0);
 854     __m128           two     = _mm_set1_ps(2.0);
 855     x                = xx[0];
 856     f                = ff[0];
 857
 858     nri              = nlist->nri;
 859     iinr             = nlist->iinr;
 860     jindex           = nlist->jindex;
 861     jjnr             = nlist->jjnr;
 862     shiftidx         = nlist->shift;
 863     gid              = nlist->gid;
 864     shiftvec         = fr->shift_vec[0];
 865     fshift           = fr->fshift[0];
 866     facel            = _mm_set1_ps(fr->epsfac);
 867     charge           = mdatoms->chargeA;
 868     nvdwtype         = fr->ntype;
 869     vdwparam         = fr->nbfp;
 870     vdwtype          = mdatoms->typeA;
 871
 872     vftab            = kernel_data->table_vdw->data;
 873     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 874
 875     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 876     ewtab            = fr->ic->tabq_coul_F;
 877     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 878     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 879
 880     /* Setup water-specific parameters */
 881     inr              = nlist->iinr[0];
 882     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 883     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 884     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 885     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 886
 887     /* Avoid stupid compiler warnings */
 888     jnrA = jnrB = jnrC = jnrD = 0;
 889     j_coord_offsetA = 0;
 890     j_coord_offsetB = 0;
 891     j_coord_offsetC = 0;
 892     j_coord_offsetD = 0;
 893
 894     outeriter        = 0;
 895     inneriter        = 0;
 896
 897     for(iidx=0;iidx<4*DIM;iidx++)
 898     {
 899         scratch[iidx] = 0.0;
 900     }
 901
 902     /* Start outer loop over neighborlists */
 903     for(iidx=0; iidx<nri; iidx++)
 904     {
 905         /* Load shift vector for this list */
 906         i_shift_offset   = DIM*shiftidx[iidx];
 907
 908         /* Load limits for loop over neighbors */
 909         j_index_start    = jindex[iidx];
 910         j_index_end      = jindex[iidx+1];
 911
 912         /* Get outer coordinate index */
 913         inr              = iinr[iidx];
 914         i_coord_offset   = DIM*inr;
 915
 916         /* Load i particle coords and add shift vector */
 917         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 918                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 919
 920         fix0             = _mm_setzero_ps();
 921         fiy0             = _mm_setzero_ps();
 922         fiz0             = _mm_setzero_ps();
 923         fix1             = _mm_setzero_ps();
 924         fiy1             = _mm_setzero_ps();
 925         fiz1             = _mm_setzero_ps();
 926         fix2             = _mm_setzero_ps();
 927         fiy2             = _mm_setzero_ps();
 928         fiz2             = _mm_setzero_ps();
 929         fix3             = _mm_setzero_ps();
 930         fiy3             = _mm_setzero_ps();
 931         fiz3             = _mm_setzero_ps();
 932
 933         /* Start inner kernel loop */
 934         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 935         {
 936
 937             /* Get j neighbor index, and coordinate index */
 938             jnrA             = jjnr[jidx];
 939             jnrB             = jjnr[jidx+1];
 940             jnrC             = jjnr[jidx+2];
 941             jnrD             = jjnr[jidx+3];
 942             j_coord_offsetA  = DIM*jnrA;
 943             j_coord_offsetB  = DIM*jnrB;
 944             j_coord_offsetC  = DIM*jnrC;
 945             j_coord_offsetD  = DIM*jnrD;
 946
 947             /* load j atom coordinates */
 948             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 949                                               x+j_coord_offsetC,x+j_coord_offsetD,
 950                                               &jx0,&jy0,&jz0);
 951
 952             /* Calculate displacement vector */
 953             dx00             = _mm_sub_ps(ix0,jx0);
 954             dy00             = _mm_sub_ps(iy0,jy0);
 955             dz00             = _mm_sub_ps(iz0,jz0);
 956             dx10             = _mm_sub_ps(ix1,jx0);
 957             dy10             = _mm_sub_ps(iy1,jy0);
 958             dz10             = _mm_sub_ps(iz1,jz0);
 959             dx20             = _mm_sub_ps(ix2,jx0);
 960             dy20             = _mm_sub_ps(iy2,jy0);
 961             dz20             = _mm_sub_ps(iz2,jz0);
 962             dx30             = _mm_sub_ps(ix3,jx0);
 963             dy30             = _mm_sub_ps(iy3,jy0);
 964             dz30             = _mm_sub_ps(iz3,jz0);
 965
 966             /* Calculate squared distance and things based on it */
 967             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 968             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 969             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 970             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 971
 972             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 973             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 974             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 975             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 976
 977             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 978             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 979             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 980
 981             /* Load parameters for j particles */
 982             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 983                                                               charge+jnrC+0,charge+jnrD+0);
 984             vdwjidx0A        = 2*vdwtype[jnrA+0];
 985             vdwjidx0B        = 2*vdwtype[jnrB+0];
 986             vdwjidx0C        = 2*vdwtype[jnrC+0];
 987             vdwjidx0D        = 2*vdwtype[jnrD+0];
 988
 989             fjx0             = _mm_setzero_ps();
 990             fjy0             = _mm_setzero_ps();
 991             fjz0             = _mm_setzero_ps();
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             r00              = _mm_mul_ps(rsq00,rinv00);
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1001                                          vdwparam+vdwioffset0+vdwjidx0B,
1002                                          vdwparam+vdwioffset0+vdwjidx0C,
1003                                          vdwparam+vdwioffset0+vdwjidx0D,
1004                                          &c6_00,&c12_00);
1005
1006             /* Calculate table index by multiplying r with table scale and truncate to integer */
1007             rt               = _mm_mul_ps(r00,vftabscale);
1008             vfitab           = _mm_cvttps_epi32(rt);
1009             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1010             vfitab           = _mm_slli_epi32(vfitab,3);
1011
1012             /* CUBIC SPLINE TABLE DISPERSION */
1013             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1014             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1015             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1016             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1017             _MM_TRANSPOSE4_PS(Y,F,G,H);
1018             Heps             = _mm_mul_ps(vfeps,H);
1019             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1020             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1021             fvdw6            = _mm_mul_ps(c6_00,FF);
1022
1023             /* CUBIC SPLINE TABLE REPULSION */
1024             vfitab           = _mm_add_epi32(vfitab,ifour);
1025             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1026             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1027             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1028             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1029             _MM_TRANSPOSE4_PS(Y,F,G,H);
1030             Heps             = _mm_mul_ps(vfeps,H);
1031             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1032             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1033             fvdw12           = _mm_mul_ps(c12_00,FF);
1034             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1035
1036             fscal            = fvdw;
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm_mul_ps(fscal,dx00);
1040             ty               = _mm_mul_ps(fscal,dy00);
1041             tz               = _mm_mul_ps(fscal,dz00);
1042
1043             /* Update vectorial force */
1044             fix0             = _mm_add_ps(fix0,tx);
1045             fiy0             = _mm_add_ps(fiy0,ty);
1046             fiz0             = _mm_add_ps(fiz0,tz);
1047
1048             fjx0             = _mm_add_ps(fjx0,tx);
1049             fjy0             = _mm_add_ps(fjy0,ty);
1050             fjz0             = _mm_add_ps(fjz0,tz);
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             r10              = _mm_mul_ps(rsq10,rinv10);
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq10             = _mm_mul_ps(iq1,jq0);
1060
1061             /* EWALD ELECTROSTATICS */
1062
1063             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1064             ewrt             = _mm_mul_ps(r10,ewtabscale);
1065             ewitab           = _mm_cvttps_epi32(ewrt);
1066             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1067             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1068                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1069                                          &ewtabF,&ewtabFn);
1070             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1071             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1072
1073             fscal            = felec;
1074
1075             /* Calculate temporary vectorial force */
1076             tx               = _mm_mul_ps(fscal,dx10);
1077             ty               = _mm_mul_ps(fscal,dy10);
1078             tz               = _mm_mul_ps(fscal,dz10);
1079
1080             /* Update vectorial force */
1081             fix1             = _mm_add_ps(fix1,tx);
1082             fiy1             = _mm_add_ps(fiy1,ty);
1083             fiz1             = _mm_add_ps(fiz1,tz);
1084
1085             fjx0             = _mm_add_ps(fjx0,tx);
1086             fjy0             = _mm_add_ps(fjy0,ty);
1087             fjz0             = _mm_add_ps(fjz0,tz);
1088
1089             /**************************
1090              * CALCULATE INTERACTIONS *
1091              **************************/
1092
1093             r20              = _mm_mul_ps(rsq20,rinv20);
1094
1095             /* Compute parameters for interactions between i and j atoms */
1096             qq20             = _mm_mul_ps(iq2,jq0);
1097
1098             /* EWALD ELECTROSTATICS */
1099
1100             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1101             ewrt             = _mm_mul_ps(r20,ewtabscale);
1102             ewitab           = _mm_cvttps_epi32(ewrt);
1103             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1104             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1105                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1106                                          &ewtabF,&ewtabFn);
1107             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1108             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1109
1110             fscal            = felec;
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm_mul_ps(fscal,dx20);
1114             ty               = _mm_mul_ps(fscal,dy20);
1115             tz               = _mm_mul_ps(fscal,dz20);
1116
1117             /* Update vectorial force */
1118             fix2             = _mm_add_ps(fix2,tx);
1119             fiy2             = _mm_add_ps(fiy2,ty);
1120             fiz2             = _mm_add_ps(fiz2,tz);
1121
1122             fjx0             = _mm_add_ps(fjx0,tx);
1123             fjy0             = _mm_add_ps(fjy0,ty);
1124             fjz0             = _mm_add_ps(fjz0,tz);
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             r30              = _mm_mul_ps(rsq30,rinv30);
1131
1132             /* Compute parameters for interactions between i and j atoms */
1133             qq30             = _mm_mul_ps(iq3,jq0);
1134
1135             /* EWALD ELECTROSTATICS */
1136
1137             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1138             ewrt             = _mm_mul_ps(r30,ewtabscale);
1139             ewitab           = _mm_cvttps_epi32(ewrt);
1140             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1141             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1142                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1143                                          &ewtabF,&ewtabFn);
1144             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1145             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1146
1147             fscal            = felec;
1148
1149             /* Calculate temporary vectorial force */
1150             tx               = _mm_mul_ps(fscal,dx30);
1151             ty               = _mm_mul_ps(fscal,dy30);
1152             tz               = _mm_mul_ps(fscal,dz30);
1153
1154             /* Update vectorial force */
1155             fix3             = _mm_add_ps(fix3,tx);
1156             fiy3             = _mm_add_ps(fiy3,ty);
1157             fiz3             = _mm_add_ps(fiz3,tz);
1158
1159             fjx0             = _mm_add_ps(fjx0,tx);
1160             fjy0             = _mm_add_ps(fjy0,ty);
1161             fjz0             = _mm_add_ps(fjz0,tz);
1162
1163             fjptrA             = f+j_coord_offsetA;
1164             fjptrB             = f+j_coord_offsetB;
1165             fjptrC             = f+j_coord_offsetC;
1166             fjptrD             = f+j_coord_offsetD;
1167
1168             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1169
1170             /* Inner loop uses 156 flops */
1171         }
1172
1173         if(jidx<j_index_end)
1174         {
1175
1176             /* Get j neighbor index, and coordinate index */
1177             jnrlistA         = jjnr[jidx];
1178             jnrlistB         = jjnr[jidx+1];
1179             jnrlistC         = jjnr[jidx+2];
1180             jnrlistD         = jjnr[jidx+3];
1181             /* Sign of each element will be negative for non-real atoms.
1182              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1183              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1184              */
1185             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1186             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1187             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1188             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1189             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1190             j_coord_offsetA  = DIM*jnrA;
1191             j_coord_offsetB  = DIM*jnrB;
1192             j_coord_offsetC  = DIM*jnrC;
1193             j_coord_offsetD  = DIM*jnrD;
1194
1195             /* load j atom coordinates */
1196             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1197                                               x+j_coord_offsetC,x+j_coord_offsetD,
1198                                               &jx0,&jy0,&jz0);
1199
1200             /* Calculate displacement vector */
1201             dx00             = _mm_sub_ps(ix0,jx0);
1202             dy00             = _mm_sub_ps(iy0,jy0);
1203             dz00             = _mm_sub_ps(iz0,jz0);
1204             dx10             = _mm_sub_ps(ix1,jx0);
1205             dy10             = _mm_sub_ps(iy1,jy0);
1206             dz10             = _mm_sub_ps(iz1,jz0);
1207             dx20             = _mm_sub_ps(ix2,jx0);
1208             dy20             = _mm_sub_ps(iy2,jy0);
1209             dz20             = _mm_sub_ps(iz2,jz0);
1210             dx30             = _mm_sub_ps(ix3,jx0);
1211             dy30             = _mm_sub_ps(iy3,jy0);
1212             dz30             = _mm_sub_ps(iz3,jz0);
1213
1214             /* Calculate squared distance and things based on it */
1215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1216             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1217             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1218             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1219
1220             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1221             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1222             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1223             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1224
1225             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1226             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1227             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1228
1229             /* Load parameters for j particles */
1230             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1231                                                               charge+jnrC+0,charge+jnrD+0);
1232             vdwjidx0A        = 2*vdwtype[jnrA+0];
1233             vdwjidx0B        = 2*vdwtype[jnrB+0];
1234             vdwjidx0C        = 2*vdwtype[jnrC+0];
1235             vdwjidx0D        = 2*vdwtype[jnrD+0];
1236
1237             fjx0             = _mm_setzero_ps();
1238             fjy0             = _mm_setzero_ps();
1239             fjz0             = _mm_setzero_ps();
1240
1241             /**************************
1242              * CALCULATE INTERACTIONS *
1243              **************************/
1244
1245             r00              = _mm_mul_ps(rsq00,rinv00);
1246             r00              = _mm_andnot_ps(dummy_mask,r00);
1247
1248             /* Compute parameters for interactions between i and j atoms */
1249             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1250                                          vdwparam+vdwioffset0+vdwjidx0B,
1251                                          vdwparam+vdwioffset0+vdwjidx0C,
1252                                          vdwparam+vdwioffset0+vdwjidx0D,
1253                                          &c6_00,&c12_00);
1254
1255             /* Calculate table index by multiplying r with table scale and truncate to integer */
1256             rt               = _mm_mul_ps(r00,vftabscale);
1257             vfitab           = _mm_cvttps_epi32(rt);
1258             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1259             vfitab           = _mm_slli_epi32(vfitab,3);
1260
1261             /* CUBIC SPLINE TABLE DISPERSION */
1262             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1263             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1264             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1265             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1266             _MM_TRANSPOSE4_PS(Y,F,G,H);
1267             Heps             = _mm_mul_ps(vfeps,H);
1268             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1269             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1270             fvdw6            = _mm_mul_ps(c6_00,FF);
1271
1272             /* CUBIC SPLINE TABLE REPULSION */
1273             vfitab           = _mm_add_epi32(vfitab,ifour);
1274             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1275             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1276             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1277             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1278             _MM_TRANSPOSE4_PS(Y,F,G,H);
1279             Heps             = _mm_mul_ps(vfeps,H);
1280             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1281             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1282             fvdw12           = _mm_mul_ps(c12_00,FF);
1283             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1284
1285             fscal            = fvdw;
1286
1287             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1288
1289             /* Calculate temporary vectorial force */
1290             tx               = _mm_mul_ps(fscal,dx00);
1291             ty               = _mm_mul_ps(fscal,dy00);
1292             tz               = _mm_mul_ps(fscal,dz00);
1293
1294             /* Update vectorial force */
1295             fix0             = _mm_add_ps(fix0,tx);
1296             fiy0             = _mm_add_ps(fiy0,ty);
1297             fiz0             = _mm_add_ps(fiz0,tz);
1298
1299             fjx0             = _mm_add_ps(fjx0,tx);
1300             fjy0             = _mm_add_ps(fjy0,ty);
1301             fjz0             = _mm_add_ps(fjz0,tz);
1302
1303             /**************************
1304              * CALCULATE INTERACTIONS *
1305              **************************/
1306
1307             r10              = _mm_mul_ps(rsq10,rinv10);
1308             r10              = _mm_andnot_ps(dummy_mask,r10);
1309
1310             /* Compute parameters for interactions between i and j atoms */
1311             qq10             = _mm_mul_ps(iq1,jq0);
1312
1313             /* EWALD ELECTROSTATICS */
1314
1315             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1316             ewrt             = _mm_mul_ps(r10,ewtabscale);
1317             ewitab           = _mm_cvttps_epi32(ewrt);
1318             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1319             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1320                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1321                                          &ewtabF,&ewtabFn);
1322             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1323             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1324
1325             fscal            = felec;
1326
1327             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1328
1329             /* Calculate temporary vectorial force */
1330             tx               = _mm_mul_ps(fscal,dx10);
1331             ty               = _mm_mul_ps(fscal,dy10);
1332             tz               = _mm_mul_ps(fscal,dz10);
1333
1334             /* Update vectorial force */
1335             fix1             = _mm_add_ps(fix1,tx);
1336             fiy1             = _mm_add_ps(fiy1,ty);
1337             fiz1             = _mm_add_ps(fiz1,tz);
1338
1339             fjx0             = _mm_add_ps(fjx0,tx);
1340             fjy0             = _mm_add_ps(fjy0,ty);
1341             fjz0             = _mm_add_ps(fjz0,tz);
1342
1343             /**************************
1344              * CALCULATE INTERACTIONS *
1345              **************************/
1346
1347             r20              = _mm_mul_ps(rsq20,rinv20);
1348             r20              = _mm_andnot_ps(dummy_mask,r20);
1349
1350             /* Compute parameters for interactions between i and j atoms */
1351             qq20             = _mm_mul_ps(iq2,jq0);
1352
1353             /* EWALD ELECTROSTATICS */
1354
1355             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1356             ewrt             = _mm_mul_ps(r20,ewtabscale);
1357             ewitab           = _mm_cvttps_epi32(ewrt);
1358             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1359             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1360                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1361                                          &ewtabF,&ewtabFn);
1362             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1363             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1364
1365             fscal            = felec;
1366
1367             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1368
1369             /* Calculate temporary vectorial force */
1370             tx               = _mm_mul_ps(fscal,dx20);
1371             ty               = _mm_mul_ps(fscal,dy20);
1372             tz               = _mm_mul_ps(fscal,dz20);
1373
1374             /* Update vectorial force */
1375             fix2             = _mm_add_ps(fix2,tx);
1376             fiy2             = _mm_add_ps(fiy2,ty);
1377             fiz2             = _mm_add_ps(fiz2,tz);
1378
1379             fjx0             = _mm_add_ps(fjx0,tx);
1380             fjy0             = _mm_add_ps(fjy0,ty);
1381             fjz0             = _mm_add_ps(fjz0,tz);
1382
1383             /**************************
1384              * CALCULATE INTERACTIONS *
1385              **************************/
1386
1387             r30              = _mm_mul_ps(rsq30,rinv30);
1388             r30              = _mm_andnot_ps(dummy_mask,r30);
1389
1390             /* Compute parameters for interactions between i and j atoms */
1391             qq30             = _mm_mul_ps(iq3,jq0);
1392
1393             /* EWALD ELECTROSTATICS */
1394
1395             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1396             ewrt             = _mm_mul_ps(r30,ewtabscale);
1397             ewitab           = _mm_cvttps_epi32(ewrt);
1398             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1399             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1400                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1401                                          &ewtabF,&ewtabFn);
1402             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1403             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1404
1405             fscal            = felec;
1406
1407             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1408
1409             /* Calculate temporary vectorial force */
1410             tx               = _mm_mul_ps(fscal,dx30);
1411             ty               = _mm_mul_ps(fscal,dy30);
1412             tz               = _mm_mul_ps(fscal,dz30);
1413
1414             /* Update vectorial force */
1415             fix3             = _mm_add_ps(fix3,tx);
1416             fiy3             = _mm_add_ps(fiy3,ty);
1417             fiz3             = _mm_add_ps(fiz3,tz);
1418
1419             fjx0             = _mm_add_ps(fjx0,tx);
1420             fjy0             = _mm_add_ps(fjy0,ty);
1421             fjz0             = _mm_add_ps(fjz0,tz);
1422
1423             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1424             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1425             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1426             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1427
1428             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1429
1430             /* Inner loop uses 160 flops */
1431         }
1432
1433         /* End of innermost loop */
1434
1435         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1436                                               f+i_coord_offset,fshift+i_shift_offset);
1437
1438         /* Increment number of inner iterations */
1439         inneriter                  += j_index_end - j_index_start;
1440
1441         /* Outer loop uses 24 flops */
1442     }
1443
1444     /* Increment number of outer iterations */
1445     outeriter        += nri;
1446
1447     /* Update outer/inner flops */
1448
1449     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*160);
1450 }