src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse2_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          vfitab;
 108     __m128i          ifour       = _mm_set1_epi32(4);
 109     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 110     real             *vftab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_elec_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 138     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 139     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm_setzero_ps();
 176         fiy0             = _mm_setzero_ps();
 177         fiz0             = _mm_setzero_ps();
 178         fix1             = _mm_setzero_ps();
 179         fiy1             = _mm_setzero_ps();
 180         fiz1             = _mm_setzero_ps();
 181         fix2             = _mm_setzero_ps();
 182         fiy2             = _mm_setzero_ps();
 183         fiz2             = _mm_setzero_ps();
 184         fix3             = _mm_setzero_ps();
 185         fiy3             = _mm_setzero_ps();
 186         fiz3             = _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             dx30             = _mm_sub_ps(ix3,jx0);
 222             dy30             = _mm_sub_ps(iy3,jy0);
 223             dz30             = _mm_sub_ps(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 229             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 230
 231             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 232             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 234             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 238                                                               charge+jnrC+0,charge+jnrD+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243
 244             fjx0             = _mm_setzero_ps();
 245             fjy0             = _mm_setzero_ps();
 246             fjz0             = _mm_setzero_ps();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm_mul_ps(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 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(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 266
 267             /* CUBIC SPLINE TABLE DISPERSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Heps             = _mm_mul_ps(vfeps,H);
 275             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 276             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 277             vvdw6            = _mm_mul_ps(c6_00,VV);
 278             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 279             fvdw6            = _mm_mul_ps(c6_00,FF);
 280
 281             /* CUBIC SPLINE TABLE REPULSION */
 282             vfitab           = _mm_add_epi32(vfitab,ifour);
 283             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 284             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 285             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 286             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 287             _MM_TRANSPOSE4_PS(Y,F,G,H);
 288             Heps             = _mm_mul_ps(vfeps,H);
 289             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 290             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 291             vvdw12           = _mm_mul_ps(c12_00,VV);
 292             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 293             fvdw12           = _mm_mul_ps(c12_00,FF);
 294             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 295             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 299
 300             fscal            = fvdw;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_ps(fscal,dx00);
 304             ty               = _mm_mul_ps(fscal,dy00);
 305             tz               = _mm_mul_ps(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm_add_ps(fix0,tx);
 309             fiy0             = _mm_add_ps(fiy0,ty);
 310             fiz0             = _mm_add_ps(fiz0,tz);
 311
 312             fjx0             = _mm_add_ps(fjx0,tx);
 313             fjy0             = _mm_add_ps(fjy0,ty);
 314             fjz0             = _mm_add_ps(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r10              = _mm_mul_ps(rsq10,rinv10);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_ps(iq1,jq0);
 324
 325             /* Calculate table index by multiplying r with table scale and truncate to integer */
 326             rt               = _mm_mul_ps(r10,vftabscale);
 327             vfitab           = _mm_cvttps_epi32(rt);
 328             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 329             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 330
 331             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 332             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 333             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 334             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 335             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 336             _MM_TRANSPOSE4_PS(Y,F,G,H);
 337             Heps             = _mm_mul_ps(vfeps,H);
 338             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 339             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 340             velec            = _mm_mul_ps(qq10,VV);
 341             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 342             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_ps(fscal,dx10);
 351             ty               = _mm_mul_ps(fscal,dy10);
 352             tz               = _mm_mul_ps(fscal,dz10);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm_add_ps(fix1,tx);
 356             fiy1             = _mm_add_ps(fiy1,ty);
 357             fiz1             = _mm_add_ps(fiz1,tz);
 358
 359             fjx0             = _mm_add_ps(fjx0,tx);
 360             fjy0             = _mm_add_ps(fjy0,ty);
 361             fjz0             = _mm_add_ps(fjz0,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r20              = _mm_mul_ps(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_ps(iq2,jq0);
 371
 372             /* Calculate table index by multiplying r with table scale and truncate to integer */
 373             rt               = _mm_mul_ps(r20,vftabscale);
 374             vfitab           = _mm_cvttps_epi32(rt);
 375             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 376             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 377
 378             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 379             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 380             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 381             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 382             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 383             _MM_TRANSPOSE4_PS(Y,F,G,H);
 384             Heps             = _mm_mul_ps(vfeps,H);
 385             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 386             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 387             velec            = _mm_mul_ps(qq20,VV);
 388             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 389             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_ps(fscal,dx20);
 398             ty               = _mm_mul_ps(fscal,dy20);
 399             tz               = _mm_mul_ps(fscal,dz20);
 400
 401             /* Update vectorial force */
 402             fix2             = _mm_add_ps(fix2,tx);
 403             fiy2             = _mm_add_ps(fiy2,ty);
 404             fiz2             = _mm_add_ps(fiz2,tz);
 405
 406             fjx0             = _mm_add_ps(fjx0,tx);
 407             fjy0             = _mm_add_ps(fjy0,ty);
 408             fjz0             = _mm_add_ps(fjz0,tz);
 409
 410             /**************************
 411              * CALCULATE INTERACTIONS *
 412              **************************/
 413
 414             r30              = _mm_mul_ps(rsq30,rinv30);
 415
 416             /* Compute parameters for interactions between i and j atoms */
 417             qq30             = _mm_mul_ps(iq3,jq0);
 418
 419             /* Calculate table index by multiplying r with table scale and truncate to integer */
 420             rt               = _mm_mul_ps(r30,vftabscale);
 421             vfitab           = _mm_cvttps_epi32(rt);
 422             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 423             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 424
 425             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 426             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 427             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 428             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 429             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 430             _MM_TRANSPOSE4_PS(Y,F,G,H);
 431             Heps             = _mm_mul_ps(vfeps,H);
 432             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 433             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 434             velec            = _mm_mul_ps(qq30,VV);
 435             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 436             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velecsum         = _mm_add_ps(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             /* Calculate temporary vectorial force */
 444             tx               = _mm_mul_ps(fscal,dx30);
 445             ty               = _mm_mul_ps(fscal,dy30);
 446             tz               = _mm_mul_ps(fscal,dz30);
 447
 448             /* Update vectorial force */
 449             fix3             = _mm_add_ps(fix3,tx);
 450             fiy3             = _mm_add_ps(fiy3,ty);
 451             fiz3             = _mm_add_ps(fiz3,tz);
 452
 453             fjx0             = _mm_add_ps(fjx0,tx);
 454             fjy0             = _mm_add_ps(fjy0,ty);
 455             fjz0             = _mm_add_ps(fjz0,tz);
 456
 457             fjptrA             = f+j_coord_offsetA;
 458             fjptrB             = f+j_coord_offsetB;
 459             fjptrC             = f+j_coord_offsetC;
 460             fjptrD             = f+j_coord_offsetD;
 461
 462             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 463
 464             /* Inner loop uses 185 flops */
 465         }
 466
 467         if(jidx<j_index_end)
 468         {
 469
 470             /* Get j neighbor index, and coordinate index */
 471             jnrlistA         = jjnr[jidx];
 472             jnrlistB         = jjnr[jidx+1];
 473             jnrlistC         = jjnr[jidx+2];
 474             jnrlistD         = jjnr[jidx+3];
 475             /* Sign of each element will be negative for non-real atoms.
 476              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 477              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 478              */
 479             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 480             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 481             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 482             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 483             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 484             j_coord_offsetA  = DIM*jnrA;
 485             j_coord_offsetB  = DIM*jnrB;
 486             j_coord_offsetC  = DIM*jnrC;
 487             j_coord_offsetD  = DIM*jnrD;
 488
 489             /* load j atom coordinates */
 490             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 491                                               x+j_coord_offsetC,x+j_coord_offsetD,
 492                                               &jx0,&jy0,&jz0);
 493
 494             /* Calculate displacement vector */
 495             dx00             = _mm_sub_ps(ix0,jx0);
 496             dy00             = _mm_sub_ps(iy0,jy0);
 497             dz00             = _mm_sub_ps(iz0,jz0);
 498             dx10             = _mm_sub_ps(ix1,jx0);
 499             dy10             = _mm_sub_ps(iy1,jy0);
 500             dz10             = _mm_sub_ps(iz1,jz0);
 501             dx20             = _mm_sub_ps(ix2,jx0);
 502             dy20             = _mm_sub_ps(iy2,jy0);
 503             dz20             = _mm_sub_ps(iz2,jz0);
 504             dx30             = _mm_sub_ps(ix3,jx0);
 505             dy30             = _mm_sub_ps(iy3,jy0);
 506             dz30             = _mm_sub_ps(iz3,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 510             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 511             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 512             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 513
 514             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 515             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 516             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 517             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 521                                                               charge+jnrC+0,charge+jnrD+0);
 522             vdwjidx0A        = 2*vdwtype[jnrA+0];
 523             vdwjidx0B        = 2*vdwtype[jnrB+0];
 524             vdwjidx0C        = 2*vdwtype[jnrC+0];
 525             vdwjidx0D        = 2*vdwtype[jnrD+0];
 526
 527             fjx0             = _mm_setzero_ps();
 528             fjy0             = _mm_setzero_ps();
 529             fjz0             = _mm_setzero_ps();
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r00              = _mm_mul_ps(rsq00,rinv00);
 536             r00              = _mm_andnot_ps(dummy_mask,r00);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 540                                          vdwparam+vdwioffset0+vdwjidx0B,
 541                                          vdwparam+vdwioffset0+vdwjidx0C,
 542                                          vdwparam+vdwioffset0+vdwjidx0D,
 543                                          &c6_00,&c12_00);
 544
 545             /* Calculate table index by multiplying r with table scale and truncate to integer */
 546             rt               = _mm_mul_ps(r00,vftabscale);
 547             vfitab           = _mm_cvttps_epi32(rt);
 548             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 549             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 550
 551             /* CUBIC SPLINE TABLE DISPERSION */
 552             vfitab           = _mm_add_epi32(vfitab,ifour);
 553             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 554             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 555             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 556             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 557             _MM_TRANSPOSE4_PS(Y,F,G,H);
 558             Heps             = _mm_mul_ps(vfeps,H);
 559             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 560             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 561             vvdw6            = _mm_mul_ps(c6_00,VV);
 562             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 563             fvdw6            = _mm_mul_ps(c6_00,FF);
 564
 565             /* CUBIC SPLINE TABLE REPULSION */
 566             vfitab           = _mm_add_epi32(vfitab,ifour);
 567             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 568             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 569             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 570             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 571             _MM_TRANSPOSE4_PS(Y,F,G,H);
 572             Heps             = _mm_mul_ps(vfeps,H);
 573             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 574             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 575             vvdw12           = _mm_mul_ps(c12_00,VV);
 576             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 577             fvdw12           = _mm_mul_ps(c12_00,FF);
 578             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 579             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 583             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 584
 585             fscal            = fvdw;
 586
 587             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 588
 589             /* Calculate temporary vectorial force */
 590             tx               = _mm_mul_ps(fscal,dx00);
 591             ty               = _mm_mul_ps(fscal,dy00);
 592             tz               = _mm_mul_ps(fscal,dz00);
 593
 594             /* Update vectorial force */
 595             fix0             = _mm_add_ps(fix0,tx);
 596             fiy0             = _mm_add_ps(fiy0,ty);
 597             fiz0             = _mm_add_ps(fiz0,tz);
 598
 599             fjx0             = _mm_add_ps(fjx0,tx);
 600             fjy0             = _mm_add_ps(fjy0,ty);
 601             fjz0             = _mm_add_ps(fjz0,tz);
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             r10              = _mm_mul_ps(rsq10,rinv10);
 608             r10              = _mm_andnot_ps(dummy_mask,r10);
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq10             = _mm_mul_ps(iq1,jq0);
 612
 613             /* Calculate table index by multiplying r with table scale and truncate to integer */
 614             rt               = _mm_mul_ps(r10,vftabscale);
 615             vfitab           = _mm_cvttps_epi32(rt);
 616             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 617             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 618
 619             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 620             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 621             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 622             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 623             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 624             _MM_TRANSPOSE4_PS(Y,F,G,H);
 625             Heps             = _mm_mul_ps(vfeps,H);
 626             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 627             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 628             velec            = _mm_mul_ps(qq10,VV);
 629             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 630             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 631
 632             /* Update potential sum for this i atom from the interaction with this j atom. */
 633             velec            = _mm_andnot_ps(dummy_mask,velec);
 634             velecsum         = _mm_add_ps(velecsum,velec);
 635
 636             fscal            = felec;
 637
 638             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 639
 640             /* Calculate temporary vectorial force */
 641             tx               = _mm_mul_ps(fscal,dx10);
 642             ty               = _mm_mul_ps(fscal,dy10);
 643             tz               = _mm_mul_ps(fscal,dz10);
 644
 645             /* Update vectorial force */
 646             fix1             = _mm_add_ps(fix1,tx);
 647             fiy1             = _mm_add_ps(fiy1,ty);
 648             fiz1             = _mm_add_ps(fiz1,tz);
 649
 650             fjx0             = _mm_add_ps(fjx0,tx);
 651             fjy0             = _mm_add_ps(fjy0,ty);
 652             fjz0             = _mm_add_ps(fjz0,tz);
 653
 654             /**************************
 655              * CALCULATE INTERACTIONS *
 656              **************************/
 657
 658             r20              = _mm_mul_ps(rsq20,rinv20);
 659             r20              = _mm_andnot_ps(dummy_mask,r20);
 660
 661             /* Compute parameters for interactions between i and j atoms */
 662             qq20             = _mm_mul_ps(iq2,jq0);
 663
 664             /* Calculate table index by multiplying r with table scale and truncate to integer */
 665             rt               = _mm_mul_ps(r20,vftabscale);
 666             vfitab           = _mm_cvttps_epi32(rt);
 667             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 668             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 669
 670             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 671             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 672             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 673             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 674             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 675             _MM_TRANSPOSE4_PS(Y,F,G,H);
 676             Heps             = _mm_mul_ps(vfeps,H);
 677             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 678             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 679             velec            = _mm_mul_ps(qq20,VV);
 680             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 681             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 682
 683             /* Update potential sum for this i atom from the interaction with this j atom. */
 684             velec            = _mm_andnot_ps(dummy_mask,velec);
 685             velecsum         = _mm_add_ps(velecsum,velec);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm_mul_ps(fscal,dx20);
 693             ty               = _mm_mul_ps(fscal,dy20);
 694             tz               = _mm_mul_ps(fscal,dz20);
 695
 696             /* Update vectorial force */
 697             fix2             = _mm_add_ps(fix2,tx);
 698             fiy2             = _mm_add_ps(fiy2,ty);
 699             fiz2             = _mm_add_ps(fiz2,tz);
 700
 701             fjx0             = _mm_add_ps(fjx0,tx);
 702             fjy0             = _mm_add_ps(fjy0,ty);
 703             fjz0             = _mm_add_ps(fjz0,tz);
 704
 705             /**************************
 706              * CALCULATE INTERACTIONS *
 707              **************************/
 708
 709             r30              = _mm_mul_ps(rsq30,rinv30);
 710             r30              = _mm_andnot_ps(dummy_mask,r30);
 711
 712             /* Compute parameters for interactions between i and j atoms */
 713             qq30             = _mm_mul_ps(iq3,jq0);
 714
 715             /* Calculate table index by multiplying r with table scale and truncate to integer */
 716             rt               = _mm_mul_ps(r30,vftabscale);
 717             vfitab           = _mm_cvttps_epi32(rt);
 718             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 719             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 720
 721             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 722             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 723             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 724             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 725             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 726             _MM_TRANSPOSE4_PS(Y,F,G,H);
 727             Heps             = _mm_mul_ps(vfeps,H);
 728             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 729             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 730             velec            = _mm_mul_ps(qq30,VV);
 731             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 732             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 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 189 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*189);
 788 }
 789 /*
 790  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse2_single
 791  * Electrostatics interaction: CubicSplineTable
 792  * VdW interaction:            CubicSplineTable
 793  * Geometry:                   Water4-Particle
 794  * Calculate force/pot:        Force
 795  */
 796 void
 797 nb_kernel_ElecCSTab_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     __m128           dummy_mask,cutoff_mask;
 849     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 850     __m128           one     = _mm_set1_ps(1.0);
 851     __m128           two     = _mm_set1_ps(2.0);
 852     x                = xx[0];
 853     f                = ff[0];
 854
 855     nri              = nlist->nri;
 856     iinr             = nlist->iinr;
 857     jindex           = nlist->jindex;
 858     jjnr             = nlist->jjnr;
 859     shiftidx         = nlist->shift;
 860     gid              = nlist->gid;
 861     shiftvec         = fr->shift_vec[0];
 862     fshift           = fr->fshift[0];
 863     facel            = _mm_set1_ps(fr->epsfac);
 864     charge           = mdatoms->chargeA;
 865     nvdwtype         = fr->ntype;
 866     vdwparam         = fr->nbfp;
 867     vdwtype          = mdatoms->typeA;
 868
 869     vftab            = kernel_data->table_elec_vdw->data;
 870     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 871
 872     /* Setup water-specific parameters */
 873     inr              = nlist->iinr[0];
 874     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 875     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 876     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 877     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 878
 879     /* Avoid stupid compiler warnings */
 880     jnrA = jnrB = jnrC = jnrD = 0;
 881     j_coord_offsetA = 0;
 882     j_coord_offsetB = 0;
 883     j_coord_offsetC = 0;
 884     j_coord_offsetD = 0;
 885
 886     outeriter        = 0;
 887     inneriter        = 0;
 888
 889     for(iidx=0;iidx<4*DIM;iidx++)
 890     {
 891         scratch[iidx] = 0.0;
 892     }
 893
 894     /* Start outer loop over neighborlists */
 895     for(iidx=0; iidx<nri; iidx++)
 896     {
 897         /* Load shift vector for this list */
 898         i_shift_offset   = DIM*shiftidx[iidx];
 899
 900         /* Load limits for loop over neighbors */
 901         j_index_start    = jindex[iidx];
 902         j_index_end      = jindex[iidx+1];
 903
 904         /* Get outer coordinate index */
 905         inr              = iinr[iidx];
 906         i_coord_offset   = DIM*inr;
 907
 908         /* Load i particle coords and add shift vector */
 909         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 910                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 911
 912         fix0             = _mm_setzero_ps();
 913         fiy0             = _mm_setzero_ps();
 914         fiz0             = _mm_setzero_ps();
 915         fix1             = _mm_setzero_ps();
 916         fiy1             = _mm_setzero_ps();
 917         fiz1             = _mm_setzero_ps();
 918         fix2             = _mm_setzero_ps();
 919         fiy2             = _mm_setzero_ps();
 920         fiz2             = _mm_setzero_ps();
 921         fix3             = _mm_setzero_ps();
 922         fiy3             = _mm_setzero_ps();
 923         fiz3             = _mm_setzero_ps();
 924
 925         /* Start inner kernel loop */
 926         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 927         {
 928
 929             /* Get j neighbor index, and coordinate index */
 930             jnrA             = jjnr[jidx];
 931             jnrB             = jjnr[jidx+1];
 932             jnrC             = jjnr[jidx+2];
 933             jnrD             = jjnr[jidx+3];
 934             j_coord_offsetA  = DIM*jnrA;
 935             j_coord_offsetB  = DIM*jnrB;
 936             j_coord_offsetC  = DIM*jnrC;
 937             j_coord_offsetD  = DIM*jnrD;
 938
 939             /* load j atom coordinates */
 940             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 941                                               x+j_coord_offsetC,x+j_coord_offsetD,
 942                                               &jx0,&jy0,&jz0);
 943
 944             /* Calculate displacement vector */
 945             dx00             = _mm_sub_ps(ix0,jx0);
 946             dy00             = _mm_sub_ps(iy0,jy0);
 947             dz00             = _mm_sub_ps(iz0,jz0);
 948             dx10             = _mm_sub_ps(ix1,jx0);
 949             dy10             = _mm_sub_ps(iy1,jy0);
 950             dz10             = _mm_sub_ps(iz1,jz0);
 951             dx20             = _mm_sub_ps(ix2,jx0);
 952             dy20             = _mm_sub_ps(iy2,jy0);
 953             dz20             = _mm_sub_ps(iz2,jz0);
 954             dx30             = _mm_sub_ps(ix3,jx0);
 955             dy30             = _mm_sub_ps(iy3,jy0);
 956             dz30             = _mm_sub_ps(iz3,jz0);
 957
 958             /* Calculate squared distance and things based on it */
 959             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 960             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 961             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 962             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 963
 964             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 965             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 966             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 967             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 968
 969             /* Load parameters for j particles */
 970             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 971                                                               charge+jnrC+0,charge+jnrD+0);
 972             vdwjidx0A        = 2*vdwtype[jnrA+0];
 973             vdwjidx0B        = 2*vdwtype[jnrB+0];
 974             vdwjidx0C        = 2*vdwtype[jnrC+0];
 975             vdwjidx0D        = 2*vdwtype[jnrD+0];
 976
 977             fjx0             = _mm_setzero_ps();
 978             fjy0             = _mm_setzero_ps();
 979             fjz0             = _mm_setzero_ps();
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             r00              = _mm_mul_ps(rsq00,rinv00);
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 989                                          vdwparam+vdwioffset0+vdwjidx0B,
 990                                          vdwparam+vdwioffset0+vdwjidx0C,
 991                                          vdwparam+vdwioffset0+vdwjidx0D,
 992                                          &c6_00,&c12_00);
 993
 994             /* Calculate table index by multiplying r with table scale and truncate to integer */
 995             rt               = _mm_mul_ps(r00,vftabscale);
 996             vfitab           = _mm_cvttps_epi32(rt);
 997             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 998             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 999
1000             /* CUBIC SPLINE TABLE DISPERSION */
1001             vfitab           = _mm_add_epi32(vfitab,ifour);
1002             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1003             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1004             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1005             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1006             _MM_TRANSPOSE4_PS(Y,F,G,H);
1007             Heps             = _mm_mul_ps(vfeps,H);
1008             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1009             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1010             fvdw6            = _mm_mul_ps(c6_00,FF);
1011
1012             /* CUBIC SPLINE TABLE REPULSION */
1013             vfitab           = _mm_add_epi32(vfitab,ifour);
1014             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1015             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1016             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1017             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1018             _MM_TRANSPOSE4_PS(Y,F,G,H);
1019             Heps             = _mm_mul_ps(vfeps,H);
1020             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1021             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1022             fvdw12           = _mm_mul_ps(c12_00,FF);
1023             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1024
1025             fscal            = fvdw;
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm_mul_ps(fscal,dx00);
1029             ty               = _mm_mul_ps(fscal,dy00);
1030             tz               = _mm_mul_ps(fscal,dz00);
1031
1032             /* Update vectorial force */
1033             fix0             = _mm_add_ps(fix0,tx);
1034             fiy0             = _mm_add_ps(fiy0,ty);
1035             fiz0             = _mm_add_ps(fiz0,tz);
1036
1037             fjx0             = _mm_add_ps(fjx0,tx);
1038             fjy0             = _mm_add_ps(fjy0,ty);
1039             fjz0             = _mm_add_ps(fjz0,tz);
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             r10              = _mm_mul_ps(rsq10,rinv10);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq10             = _mm_mul_ps(iq1,jq0);
1049
1050             /* Calculate table index by multiplying r with table scale and truncate to integer */
1051             rt               = _mm_mul_ps(r10,vftabscale);
1052             vfitab           = _mm_cvttps_epi32(rt);
1053             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1054             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1055
1056             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1057             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1058             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1059             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1060             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1061             _MM_TRANSPOSE4_PS(Y,F,G,H);
1062             Heps             = _mm_mul_ps(vfeps,H);
1063             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1064             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1065             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1066
1067             fscal            = felec;
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm_mul_ps(fscal,dx10);
1071             ty               = _mm_mul_ps(fscal,dy10);
1072             tz               = _mm_mul_ps(fscal,dz10);
1073
1074             /* Update vectorial force */
1075             fix1             = _mm_add_ps(fix1,tx);
1076             fiy1             = _mm_add_ps(fiy1,ty);
1077             fiz1             = _mm_add_ps(fiz1,tz);
1078
1079             fjx0             = _mm_add_ps(fjx0,tx);
1080             fjy0             = _mm_add_ps(fjy0,ty);
1081             fjz0             = _mm_add_ps(fjz0,tz);
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             r20              = _mm_mul_ps(rsq20,rinv20);
1088
1089             /* Compute parameters for interactions between i and j atoms */
1090             qq20             = _mm_mul_ps(iq2,jq0);
1091
1092             /* Calculate table index by multiplying r with table scale and truncate to integer */
1093             rt               = _mm_mul_ps(r20,vftabscale);
1094             vfitab           = _mm_cvttps_epi32(rt);
1095             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1096             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1097
1098             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1099             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1100             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1101             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1102             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1103             _MM_TRANSPOSE4_PS(Y,F,G,H);
1104             Heps             = _mm_mul_ps(vfeps,H);
1105             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1106             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1107             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1108
1109             fscal            = felec;
1110
1111             /* Calculate temporary vectorial force */
1112             tx               = _mm_mul_ps(fscal,dx20);
1113             ty               = _mm_mul_ps(fscal,dy20);
1114             tz               = _mm_mul_ps(fscal,dz20);
1115
1116             /* Update vectorial force */
1117             fix2             = _mm_add_ps(fix2,tx);
1118             fiy2             = _mm_add_ps(fiy2,ty);
1119             fiz2             = _mm_add_ps(fiz2,tz);
1120
1121             fjx0             = _mm_add_ps(fjx0,tx);
1122             fjy0             = _mm_add_ps(fjy0,ty);
1123             fjz0             = _mm_add_ps(fjz0,tz);
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             r30              = _mm_mul_ps(rsq30,rinv30);
1130
1131             /* Compute parameters for interactions between i and j atoms */
1132             qq30             = _mm_mul_ps(iq3,jq0);
1133
1134             /* Calculate table index by multiplying r with table scale and truncate to integer */
1135             rt               = _mm_mul_ps(r30,vftabscale);
1136             vfitab           = _mm_cvttps_epi32(rt);
1137             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1138             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1139
1140             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1141             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1142             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1143             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1144             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1145             _MM_TRANSPOSE4_PS(Y,F,G,H);
1146             Heps             = _mm_mul_ps(vfeps,H);
1147             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1148             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1149             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1150
1151             fscal            = felec;
1152
1153             /* Calculate temporary vectorial force */
1154             tx               = _mm_mul_ps(fscal,dx30);
1155             ty               = _mm_mul_ps(fscal,dy30);
1156             tz               = _mm_mul_ps(fscal,dz30);
1157
1158             /* Update vectorial force */
1159             fix3             = _mm_add_ps(fix3,tx);
1160             fiy3             = _mm_add_ps(fiy3,ty);
1161             fiz3             = _mm_add_ps(fiz3,tz);
1162
1163             fjx0             = _mm_add_ps(fjx0,tx);
1164             fjy0             = _mm_add_ps(fjy0,ty);
1165             fjz0             = _mm_add_ps(fjz0,tz);
1166
1167             fjptrA             = f+j_coord_offsetA;
1168             fjptrB             = f+j_coord_offsetB;
1169             fjptrC             = f+j_coord_offsetC;
1170             fjptrD             = f+j_coord_offsetD;
1171
1172             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1173
1174             /* Inner loop uses 165 flops */
1175         }
1176
1177         if(jidx<j_index_end)
1178         {
1179
1180             /* Get j neighbor index, and coordinate index */
1181             jnrlistA         = jjnr[jidx];
1182             jnrlistB         = jjnr[jidx+1];
1183             jnrlistC         = jjnr[jidx+2];
1184             jnrlistD         = jjnr[jidx+3];
1185             /* Sign of each element will be negative for non-real atoms.
1186              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1187              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1188              */
1189             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1190             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1191             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1192             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1193             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1194             j_coord_offsetA  = DIM*jnrA;
1195             j_coord_offsetB  = DIM*jnrB;
1196             j_coord_offsetC  = DIM*jnrC;
1197             j_coord_offsetD  = DIM*jnrD;
1198
1199             /* load j atom coordinates */
1200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1201                                               x+j_coord_offsetC,x+j_coord_offsetD,
1202                                               &jx0,&jy0,&jz0);
1203
1204             /* Calculate displacement vector */
1205             dx00             = _mm_sub_ps(ix0,jx0);
1206             dy00             = _mm_sub_ps(iy0,jy0);
1207             dz00             = _mm_sub_ps(iz0,jz0);
1208             dx10             = _mm_sub_ps(ix1,jx0);
1209             dy10             = _mm_sub_ps(iy1,jy0);
1210             dz10             = _mm_sub_ps(iz1,jz0);
1211             dx20             = _mm_sub_ps(ix2,jx0);
1212             dy20             = _mm_sub_ps(iy2,jy0);
1213             dz20             = _mm_sub_ps(iz2,jz0);
1214             dx30             = _mm_sub_ps(ix3,jx0);
1215             dy30             = _mm_sub_ps(iy3,jy0);
1216             dz30             = _mm_sub_ps(iz3,jz0);
1217
1218             /* Calculate squared distance and things based on it */
1219             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1220             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1221             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1222             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1223
1224             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1225             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1226             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1227             rinv30           = gmx_mm_invsqrt_ps(rsq30);
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(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1260
1261             /* CUBIC SPLINE TABLE DISPERSION */
1262             vfitab           = _mm_add_epi32(vfitab,ifour);
1263             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1264             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1265             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1266             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1267             _MM_TRANSPOSE4_PS(Y,F,G,H);
1268             Heps             = _mm_mul_ps(vfeps,H);
1269             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1270             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1271             fvdw6            = _mm_mul_ps(c6_00,FF);
1272
1273             /* CUBIC SPLINE TABLE REPULSION */
1274             vfitab           = _mm_add_epi32(vfitab,ifour);
1275             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1276             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1277             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1278             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1279             _MM_TRANSPOSE4_PS(Y,F,G,H);
1280             Heps             = _mm_mul_ps(vfeps,H);
1281             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1282             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1283             fvdw12           = _mm_mul_ps(c12_00,FF);
1284             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1285
1286             fscal            = fvdw;
1287
1288             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1289
1290             /* Calculate temporary vectorial force */
1291             tx               = _mm_mul_ps(fscal,dx00);
1292             ty               = _mm_mul_ps(fscal,dy00);
1293             tz               = _mm_mul_ps(fscal,dz00);
1294
1295             /* Update vectorial force */
1296             fix0             = _mm_add_ps(fix0,tx);
1297             fiy0             = _mm_add_ps(fiy0,ty);
1298             fiz0             = _mm_add_ps(fiz0,tz);
1299
1300             fjx0             = _mm_add_ps(fjx0,tx);
1301             fjy0             = _mm_add_ps(fjy0,ty);
1302             fjz0             = _mm_add_ps(fjz0,tz);
1303
1304             /**************************
1305              * CALCULATE INTERACTIONS *
1306              **************************/
1307
1308             r10              = _mm_mul_ps(rsq10,rinv10);
1309             r10              = _mm_andnot_ps(dummy_mask,r10);
1310
1311             /* Compute parameters for interactions between i and j atoms */
1312             qq10             = _mm_mul_ps(iq1,jq0);
1313
1314             /* Calculate table index by multiplying r with table scale and truncate to integer */
1315             rt               = _mm_mul_ps(r10,vftabscale);
1316             vfitab           = _mm_cvttps_epi32(rt);
1317             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1318             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1319
1320             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1321             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1322             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1323             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1324             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1325             _MM_TRANSPOSE4_PS(Y,F,G,H);
1326             Heps             = _mm_mul_ps(vfeps,H);
1327             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1328             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1329             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1330
1331             fscal            = felec;
1332
1333             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1334
1335             /* Calculate temporary vectorial force */
1336             tx               = _mm_mul_ps(fscal,dx10);
1337             ty               = _mm_mul_ps(fscal,dy10);
1338             tz               = _mm_mul_ps(fscal,dz10);
1339
1340             /* Update vectorial force */
1341             fix1             = _mm_add_ps(fix1,tx);
1342             fiy1             = _mm_add_ps(fiy1,ty);
1343             fiz1             = _mm_add_ps(fiz1,tz);
1344
1345             fjx0             = _mm_add_ps(fjx0,tx);
1346             fjy0             = _mm_add_ps(fjy0,ty);
1347             fjz0             = _mm_add_ps(fjz0,tz);
1348
1349             /**************************
1350              * CALCULATE INTERACTIONS *
1351              **************************/
1352
1353             r20              = _mm_mul_ps(rsq20,rinv20);
1354             r20              = _mm_andnot_ps(dummy_mask,r20);
1355
1356             /* Compute parameters for interactions between i and j atoms */
1357             qq20             = _mm_mul_ps(iq2,jq0);
1358
1359             /* Calculate table index by multiplying r with table scale and truncate to integer */
1360             rt               = _mm_mul_ps(r20,vftabscale);
1361             vfitab           = _mm_cvttps_epi32(rt);
1362             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1363             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1364
1365             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1366             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1367             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1368             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1369             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1370             _MM_TRANSPOSE4_PS(Y,F,G,H);
1371             Heps             = _mm_mul_ps(vfeps,H);
1372             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1373             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1374             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1375
1376             fscal            = felec;
1377
1378             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1379
1380             /* Calculate temporary vectorial force */
1381             tx               = _mm_mul_ps(fscal,dx20);
1382             ty               = _mm_mul_ps(fscal,dy20);
1383             tz               = _mm_mul_ps(fscal,dz20);
1384
1385             /* Update vectorial force */
1386             fix2             = _mm_add_ps(fix2,tx);
1387             fiy2             = _mm_add_ps(fiy2,ty);
1388             fiz2             = _mm_add_ps(fiz2,tz);
1389
1390             fjx0             = _mm_add_ps(fjx0,tx);
1391             fjy0             = _mm_add_ps(fjy0,ty);
1392             fjz0             = _mm_add_ps(fjz0,tz);
1393
1394             /**************************
1395              * CALCULATE INTERACTIONS *
1396              **************************/
1397
1398             r30              = _mm_mul_ps(rsq30,rinv30);
1399             r30              = _mm_andnot_ps(dummy_mask,r30);
1400
1401             /* Compute parameters for interactions between i and j atoms */
1402             qq30             = _mm_mul_ps(iq3,jq0);
1403
1404             /* Calculate table index by multiplying r with table scale and truncate to integer */
1405             rt               = _mm_mul_ps(r30,vftabscale);
1406             vfitab           = _mm_cvttps_epi32(rt);
1407             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1408             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1409
1410             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1411             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1412             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1413             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1414             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1415             _MM_TRANSPOSE4_PS(Y,F,G,H);
1416             Heps             = _mm_mul_ps(vfeps,H);
1417             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1418             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1419             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1420
1421             fscal            = felec;
1422
1423             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1424
1425             /* Calculate temporary vectorial force */
1426             tx               = _mm_mul_ps(fscal,dx30);
1427             ty               = _mm_mul_ps(fscal,dy30);
1428             tz               = _mm_mul_ps(fscal,dz30);
1429
1430             /* Update vectorial force */
1431             fix3             = _mm_add_ps(fix3,tx);
1432             fiy3             = _mm_add_ps(fiy3,ty);
1433             fiz3             = _mm_add_ps(fiz3,tz);
1434
1435             fjx0             = _mm_add_ps(fjx0,tx);
1436             fjy0             = _mm_add_ps(fjy0,ty);
1437             fjz0             = _mm_add_ps(fjz0,tz);
1438
1439             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1440             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1441             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1442             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1443
1444             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1445
1446             /* Inner loop uses 169 flops */
1447         }
1448
1449         /* End of innermost loop */
1450
1451         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1452                                               f+i_coord_offset,fshift+i_shift_offset);
1453
1454         /* Increment number of inner iterations */
1455         inneriter                  += j_index_end - j_index_start;
1456
1457         /* Outer loop uses 24 flops */
1458     }
1459
1460     /* Increment number of outer iterations */
1461     outeriter        += nri;
1462
1463     /* Update outer/inner flops */
1464
1465     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*169);
1466 }