src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_sse4_1_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse4_1_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 135     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 136     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181         fix3             = _mm_setzero_ps();
 182         fiy3             = _mm_setzero_ps();
 183         fiz3             = _mm_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               x+j_coord_offsetC,x+j_coord_offsetD,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_ps(ix0,jx0);
 210             dy00             = _mm_sub_ps(iy0,jy0);
 211             dz00             = _mm_sub_ps(iz0,jz0);
 212             dx10             = _mm_sub_ps(ix1,jx0);
 213             dy10             = _mm_sub_ps(iy1,jy0);
 214             dz10             = _mm_sub_ps(iz1,jz0);
 215             dx20             = _mm_sub_ps(ix2,jx0);
 216             dy20             = _mm_sub_ps(iy2,jy0);
 217             dz20             = _mm_sub_ps(iz2,jz0);
 218             dx30             = _mm_sub_ps(ix3,jx0);
 219             dy30             = _mm_sub_ps(iy3,jy0);
 220             dz30             = _mm_sub_ps(iz3,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 227
 228             rinv00           = sse41_invsqrt_f(rsq00);
 229             rinv10           = sse41_invsqrt_f(rsq10);
 230             rinv20           = sse41_invsqrt_f(rsq20);
 231             rinv30           = sse41_invsqrt_f(rsq30);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             fjx0             = _mm_setzero_ps();
 242             fjy0             = _mm_setzero_ps();
 243             fjz0             = _mm_setzero_ps();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             r00              = _mm_mul_ps(rsq00,rinv00);
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* Calculate table index by multiplying r with table scale and truncate to integer */
 259             rt               = _mm_mul_ps(r00,vftabscale);
 260             vfitab           = _mm_cvttps_epi32(rt);
 261             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 262             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 263
 264             /* CUBIC SPLINE TABLE DISPERSION */
 265             vfitab           = _mm_add_epi32(vfitab,ifour);
 266             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 267             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 268             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 269             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 270             _MM_TRANSPOSE4_PS(Y,F,G,H);
 271             Heps             = _mm_mul_ps(vfeps,H);
 272             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 273             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 274             vvdw6            = _mm_mul_ps(c6_00,VV);
 275             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 276             fvdw6            = _mm_mul_ps(c6_00,FF);
 277
 278             /* CUBIC SPLINE TABLE REPULSION */
 279             vfitab           = _mm_add_epi32(vfitab,ifour);
 280             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 281             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 282             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 283             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 284             _MM_TRANSPOSE4_PS(Y,F,G,H);
 285             Heps             = _mm_mul_ps(vfeps,H);
 286             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 287             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 288             vvdw12           = _mm_mul_ps(c12_00,VV);
 289             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 290             fvdw12           = _mm_mul_ps(c12_00,FF);
 291             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 292             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 296
 297             fscal            = fvdw;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm_mul_ps(fscal,dx00);
 301             ty               = _mm_mul_ps(fscal,dy00);
 302             tz               = _mm_mul_ps(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm_add_ps(fix0,tx);
 306             fiy0             = _mm_add_ps(fiy0,ty);
 307             fiz0             = _mm_add_ps(fiz0,tz);
 308
 309             fjx0             = _mm_add_ps(fjx0,tx);
 310             fjy0             = _mm_add_ps(fjy0,ty);
 311             fjz0             = _mm_add_ps(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             r10              = _mm_mul_ps(rsq10,rinv10);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_ps(iq1,jq0);
 321
 322             /* Calculate table index by multiplying r with table scale and truncate to integer */
 323             rt               = _mm_mul_ps(r10,vftabscale);
 324             vfitab           = _mm_cvttps_epi32(rt);
 325             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 326             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 327
 328             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 329             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 330             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 331             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 332             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 333             _MM_TRANSPOSE4_PS(Y,F,G,H);
 334             Heps             = _mm_mul_ps(vfeps,H);
 335             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 336             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 337             velec            = _mm_mul_ps(qq10,VV);
 338             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 339             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx10);
 348             ty               = _mm_mul_ps(fscal,dy10);
 349             tz               = _mm_mul_ps(fscal,dz10);
 350
 351             /* Update vectorial force */
 352             fix1             = _mm_add_ps(fix1,tx);
 353             fiy1             = _mm_add_ps(fiy1,ty);
 354             fiz1             = _mm_add_ps(fiz1,tz);
 355
 356             fjx0             = _mm_add_ps(fjx0,tx);
 357             fjy0             = _mm_add_ps(fjy0,ty);
 358             fjz0             = _mm_add_ps(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r20              = _mm_mul_ps(rsq20,rinv20);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm_mul_ps(iq2,jq0);
 368
 369             /* Calculate table index by multiplying r with table scale and truncate to integer */
 370             rt               = _mm_mul_ps(r20,vftabscale);
 371             vfitab           = _mm_cvttps_epi32(rt);
 372             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 373             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 374
 375             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 376             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 377             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 378             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 379             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 380             _MM_TRANSPOSE4_PS(Y,F,G,H);
 381             Heps             = _mm_mul_ps(vfeps,H);
 382             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 383             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 384             velec            = _mm_mul_ps(qq20,VV);
 385             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 386             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm_add_ps(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm_mul_ps(fscal,dx20);
 395             ty               = _mm_mul_ps(fscal,dy20);
 396             tz               = _mm_mul_ps(fscal,dz20);
 397
 398             /* Update vectorial force */
 399             fix2             = _mm_add_ps(fix2,tx);
 400             fiy2             = _mm_add_ps(fiy2,ty);
 401             fiz2             = _mm_add_ps(fiz2,tz);
 402
 403             fjx0             = _mm_add_ps(fjx0,tx);
 404             fjy0             = _mm_add_ps(fjy0,ty);
 405             fjz0             = _mm_add_ps(fjz0,tz);
 406
 407             /**************************
 408              * CALCULATE INTERACTIONS *
 409              **************************/
 410
 411             r30              = _mm_mul_ps(rsq30,rinv30);
 412
 413             /* Compute parameters for interactions between i and j atoms */
 414             qq30             = _mm_mul_ps(iq3,jq0);
 415
 416             /* Calculate table index by multiplying r with table scale and truncate to integer */
 417             rt               = _mm_mul_ps(r30,vftabscale);
 418             vfitab           = _mm_cvttps_epi32(rt);
 419             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 420             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 421
 422             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 423             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 424             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 425             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 426             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 427             _MM_TRANSPOSE4_PS(Y,F,G,H);
 428             Heps             = _mm_mul_ps(vfeps,H);
 429             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 430             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 431             velec            = _mm_mul_ps(qq30,VV);
 432             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 433             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 434
 435             /* Update potential sum for this i atom from the interaction with this j atom. */
 436             velecsum         = _mm_add_ps(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             /* Calculate temporary vectorial force */
 441             tx               = _mm_mul_ps(fscal,dx30);
 442             ty               = _mm_mul_ps(fscal,dy30);
 443             tz               = _mm_mul_ps(fscal,dz30);
 444
 445             /* Update vectorial force */
 446             fix3             = _mm_add_ps(fix3,tx);
 447             fiy3             = _mm_add_ps(fiy3,ty);
 448             fiz3             = _mm_add_ps(fiz3,tz);
 449
 450             fjx0             = _mm_add_ps(fjx0,tx);
 451             fjy0             = _mm_add_ps(fjy0,ty);
 452             fjz0             = _mm_add_ps(fjz0,tz);
 453
 454             fjptrA             = f+j_coord_offsetA;
 455             fjptrB             = f+j_coord_offsetB;
 456             fjptrC             = f+j_coord_offsetC;
 457             fjptrD             = f+j_coord_offsetD;
 458
 459             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 460
 461             /* Inner loop uses 185 flops */
 462         }
 463
 464         if(jidx<j_index_end)
 465         {
 466
 467             /* Get j neighbor index, and coordinate index */
 468             jnrlistA         = jjnr[jidx];
 469             jnrlistB         = jjnr[jidx+1];
 470             jnrlistC         = jjnr[jidx+2];
 471             jnrlistD         = jjnr[jidx+3];
 472             /* Sign of each element will be negative for non-real atoms.
 473              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 474              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 475              */
 476             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 477             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 478             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 479             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 480             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 481             j_coord_offsetA  = DIM*jnrA;
 482             j_coord_offsetB  = DIM*jnrB;
 483             j_coord_offsetC  = DIM*jnrC;
 484             j_coord_offsetD  = DIM*jnrD;
 485
 486             /* load j atom coordinates */
 487             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 488                                               x+j_coord_offsetC,x+j_coord_offsetD,
 489                                               &jx0,&jy0,&jz0);
 490
 491             /* Calculate displacement vector */
 492             dx00             = _mm_sub_ps(ix0,jx0);
 493             dy00             = _mm_sub_ps(iy0,jy0);
 494             dz00             = _mm_sub_ps(iz0,jz0);
 495             dx10             = _mm_sub_ps(ix1,jx0);
 496             dy10             = _mm_sub_ps(iy1,jy0);
 497             dz10             = _mm_sub_ps(iz1,jz0);
 498             dx20             = _mm_sub_ps(ix2,jx0);
 499             dy20             = _mm_sub_ps(iy2,jy0);
 500             dz20             = _mm_sub_ps(iz2,jz0);
 501             dx30             = _mm_sub_ps(ix3,jx0);
 502             dy30             = _mm_sub_ps(iy3,jy0);
 503             dz30             = _mm_sub_ps(iz3,jz0);
 504
 505             /* Calculate squared distance and things based on it */
 506             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 507             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 508             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 509             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 510
 511             rinv00           = sse41_invsqrt_f(rsq00);
 512             rinv10           = sse41_invsqrt_f(rsq10);
 513             rinv20           = sse41_invsqrt_f(rsq20);
 514             rinv30           = sse41_invsqrt_f(rsq30);
 515
 516             /* Load parameters for j particles */
 517             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 518                                                               charge+jnrC+0,charge+jnrD+0);
 519             vdwjidx0A        = 2*vdwtype[jnrA+0];
 520             vdwjidx0B        = 2*vdwtype[jnrB+0];
 521             vdwjidx0C        = 2*vdwtype[jnrC+0];
 522             vdwjidx0D        = 2*vdwtype[jnrD+0];
 523
 524             fjx0             = _mm_setzero_ps();
 525             fjy0             = _mm_setzero_ps();
 526             fjz0             = _mm_setzero_ps();
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r00              = _mm_mul_ps(rsq00,rinv00);
 533             r00              = _mm_andnot_ps(dummy_mask,r00);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 537                                          vdwparam+vdwioffset0+vdwjidx0B,
 538                                          vdwparam+vdwioffset0+vdwjidx0C,
 539                                          vdwparam+vdwioffset0+vdwjidx0D,
 540                                          &c6_00,&c12_00);
 541
 542             /* Calculate table index by multiplying r with table scale and truncate to integer */
 543             rt               = _mm_mul_ps(r00,vftabscale);
 544             vfitab           = _mm_cvttps_epi32(rt);
 545             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 546             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 547
 548             /* CUBIC SPLINE TABLE DISPERSION */
 549             vfitab           = _mm_add_epi32(vfitab,ifour);
 550             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 551             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 552             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 553             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 554             _MM_TRANSPOSE4_PS(Y,F,G,H);
 555             Heps             = _mm_mul_ps(vfeps,H);
 556             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 557             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 558             vvdw6            = _mm_mul_ps(c6_00,VV);
 559             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 560             fvdw6            = _mm_mul_ps(c6_00,FF);
 561
 562             /* CUBIC SPLINE TABLE REPULSION */
 563             vfitab           = _mm_add_epi32(vfitab,ifour);
 564             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 565             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 566             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 567             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 568             _MM_TRANSPOSE4_PS(Y,F,G,H);
 569             Heps             = _mm_mul_ps(vfeps,H);
 570             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 571             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 572             vvdw12           = _mm_mul_ps(c12_00,VV);
 573             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 574             fvdw12           = _mm_mul_ps(c12_00,FF);
 575             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 576             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 580             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 581
 582             fscal            = fvdw;
 583
 584             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 585
 586             /* Calculate temporary vectorial force */
 587             tx               = _mm_mul_ps(fscal,dx00);
 588             ty               = _mm_mul_ps(fscal,dy00);
 589             tz               = _mm_mul_ps(fscal,dz00);
 590
 591             /* Update vectorial force */
 592             fix0             = _mm_add_ps(fix0,tx);
 593             fiy0             = _mm_add_ps(fiy0,ty);
 594             fiz0             = _mm_add_ps(fiz0,tz);
 595
 596             fjx0             = _mm_add_ps(fjx0,tx);
 597             fjy0             = _mm_add_ps(fjy0,ty);
 598             fjz0             = _mm_add_ps(fjz0,tz);
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             r10              = _mm_mul_ps(rsq10,rinv10);
 605             r10              = _mm_andnot_ps(dummy_mask,r10);
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq10             = _mm_mul_ps(iq1,jq0);
 609
 610             /* Calculate table index by multiplying r with table scale and truncate to integer */
 611             rt               = _mm_mul_ps(r10,vftabscale);
 612             vfitab           = _mm_cvttps_epi32(rt);
 613             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 614             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 615
 616             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 617             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 618             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 619             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 620             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 621             _MM_TRANSPOSE4_PS(Y,F,G,H);
 622             Heps             = _mm_mul_ps(vfeps,H);
 623             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 624             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 625             velec            = _mm_mul_ps(qq10,VV);
 626             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 627             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             velec            = _mm_andnot_ps(dummy_mask,velec);
 631             velecsum         = _mm_add_ps(velecsum,velec);
 632
 633             fscal            = felec;
 634
 635             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 636
 637             /* Calculate temporary vectorial force */
 638             tx               = _mm_mul_ps(fscal,dx10);
 639             ty               = _mm_mul_ps(fscal,dy10);
 640             tz               = _mm_mul_ps(fscal,dz10);
 641
 642             /* Update vectorial force */
 643             fix1             = _mm_add_ps(fix1,tx);
 644             fiy1             = _mm_add_ps(fiy1,ty);
 645             fiz1             = _mm_add_ps(fiz1,tz);
 646
 647             fjx0             = _mm_add_ps(fjx0,tx);
 648             fjy0             = _mm_add_ps(fjy0,ty);
 649             fjz0             = _mm_add_ps(fjz0,tz);
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             r20              = _mm_mul_ps(rsq20,rinv20);
 656             r20              = _mm_andnot_ps(dummy_mask,r20);
 657
 658             /* Compute parameters for interactions between i and j atoms */
 659             qq20             = _mm_mul_ps(iq2,jq0);
 660
 661             /* Calculate table index by multiplying r with table scale and truncate to integer */
 662             rt               = _mm_mul_ps(r20,vftabscale);
 663             vfitab           = _mm_cvttps_epi32(rt);
 664             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 665             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 666
 667             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 668             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 669             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 670             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 671             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 672             _MM_TRANSPOSE4_PS(Y,F,G,H);
 673             Heps             = _mm_mul_ps(vfeps,H);
 674             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 675             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 676             velec            = _mm_mul_ps(qq20,VV);
 677             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 678             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 679
 680             /* Update potential sum for this i atom from the interaction with this j atom. */
 681             velec            = _mm_andnot_ps(dummy_mask,velec);
 682             velecsum         = _mm_add_ps(velecsum,velec);
 683
 684             fscal            = felec;
 685
 686             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 687
 688             /* Calculate temporary vectorial force */
 689             tx               = _mm_mul_ps(fscal,dx20);
 690             ty               = _mm_mul_ps(fscal,dy20);
 691             tz               = _mm_mul_ps(fscal,dz20);
 692
 693             /* Update vectorial force */
 694             fix2             = _mm_add_ps(fix2,tx);
 695             fiy2             = _mm_add_ps(fiy2,ty);
 696             fiz2             = _mm_add_ps(fiz2,tz);
 697
 698             fjx0             = _mm_add_ps(fjx0,tx);
 699             fjy0             = _mm_add_ps(fjy0,ty);
 700             fjz0             = _mm_add_ps(fjz0,tz);
 701
 702             /**************************
 703              * CALCULATE INTERACTIONS *
 704              **************************/
 705
 706             r30              = _mm_mul_ps(rsq30,rinv30);
 707             r30              = _mm_andnot_ps(dummy_mask,r30);
 708
 709             /* Compute parameters for interactions between i and j atoms */
 710             qq30             = _mm_mul_ps(iq3,jq0);
 711
 712             /* Calculate table index by multiplying r with table scale and truncate to integer */
 713             rt               = _mm_mul_ps(r30,vftabscale);
 714             vfitab           = _mm_cvttps_epi32(rt);
 715             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 716             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 717
 718             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 719             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 720             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 721             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 722             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 723             _MM_TRANSPOSE4_PS(Y,F,G,H);
 724             Heps             = _mm_mul_ps(vfeps,H);
 725             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 726             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 727             velec            = _mm_mul_ps(qq30,VV);
 728             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 729             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 730
 731             /* Update potential sum for this i atom from the interaction with this j atom. */
 732             velec            = _mm_andnot_ps(dummy_mask,velec);
 733             velecsum         = _mm_add_ps(velecsum,velec);
 734
 735             fscal            = felec;
 736
 737             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 738
 739             /* Calculate temporary vectorial force */
 740             tx               = _mm_mul_ps(fscal,dx30);
 741             ty               = _mm_mul_ps(fscal,dy30);
 742             tz               = _mm_mul_ps(fscal,dz30);
 743
 744             /* Update vectorial force */
 745             fix3             = _mm_add_ps(fix3,tx);
 746             fiy3             = _mm_add_ps(fiy3,ty);
 747             fiz3             = _mm_add_ps(fiz3,tz);
 748
 749             fjx0             = _mm_add_ps(fjx0,tx);
 750             fjy0             = _mm_add_ps(fjy0,ty);
 751             fjz0             = _mm_add_ps(fjz0,tz);
 752
 753             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 754             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 755             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 756             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 757
 758             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 759
 760             /* Inner loop uses 189 flops */
 761         }
 762
 763         /* End of innermost loop */
 764
 765         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 766                                               f+i_coord_offset,fshift+i_shift_offset);
 767
 768         ggid                        = gid[iidx];
 769         /* Update potential energies */
 770         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 771         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 772
 773         /* Increment number of inner iterations */
 774         inneriter                  += j_index_end - j_index_start;
 775
 776         /* Outer loop uses 26 flops */
 777     }
 778
 779     /* Increment number of outer iterations */
 780     outeriter        += nri;
 781
 782     /* Update outer/inner flops */
 783
 784     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*189);
 785 }
 786 /*
 787  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse4_1_single
 788  * Electrostatics interaction: CubicSplineTable
 789  * VdW interaction:            CubicSplineTable
 790  * Geometry:                   Water4-Particle
 791  * Calculate force/pot:        Force
 792  */
 793 void
 794 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse4_1_single
 795                     (t_nblist                    * gmx_restrict       nlist,
 796                      rvec                        * gmx_restrict          xx,
 797                      rvec                        * gmx_restrict          ff,
 798                      struct t_forcerec           * gmx_restrict          fr,
 799                      t_mdatoms                   * gmx_restrict     mdatoms,
 800                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 801                      t_nrnb                      * gmx_restrict        nrnb)
 802 {
 803     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 804      * just 0 for non-waters.
 805      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 806      * jnr indices corresponding to data put in the four positions in the SIMD register.
 807      */
 808     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 809     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 810     int              jnrA,jnrB,jnrC,jnrD;
 811     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 812     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 813     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 814     real             rcutoff_scalar;
 815     real             *shiftvec,*fshift,*x,*f;
 816     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 817     real             scratch[4*DIM];
 818     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 819     int              vdwioffset0;
 820     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 821     int              vdwioffset1;
 822     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 823     int              vdwioffset2;
 824     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 825     int              vdwioffset3;
 826     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 827     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 828     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 829     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 830     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 831     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 832     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 833     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 834     real             *charge;
 835     int              nvdwtype;
 836     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 837     int              *vdwtype;
 838     real             *vdwparam;
 839     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 840     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 841     __m128i          vfitab;
 842     __m128i          ifour       = _mm_set1_epi32(4);
 843     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 844     real             *vftab;
 845     __m128           dummy_mask,cutoff_mask;
 846     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 847     __m128           one     = _mm_set1_ps(1.0);
 848     __m128           two     = _mm_set1_ps(2.0);
 849     x                = xx[0];
 850     f                = ff[0];
 851
 852     nri              = nlist->nri;
 853     iinr             = nlist->iinr;
 854     jindex           = nlist->jindex;
 855     jjnr             = nlist->jjnr;
 856     shiftidx         = nlist->shift;
 857     gid              = nlist->gid;
 858     shiftvec         = fr->shift_vec[0];
 859     fshift           = fr->fshift[0];
 860     facel            = _mm_set1_ps(fr->ic->epsfac);
 861     charge           = mdatoms->chargeA;
 862     nvdwtype         = fr->ntype;
 863     vdwparam         = fr->nbfp;
 864     vdwtype          = mdatoms->typeA;
 865
 866     vftab            = kernel_data->table_elec_vdw->data;
 867     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 868
 869     /* Setup water-specific parameters */
 870     inr              = nlist->iinr[0];
 871     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 872     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 873     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 874     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 875
 876     /* Avoid stupid compiler warnings */
 877     jnrA = jnrB = jnrC = jnrD = 0;
 878     j_coord_offsetA = 0;
 879     j_coord_offsetB = 0;
 880     j_coord_offsetC = 0;
 881     j_coord_offsetD = 0;
 882
 883     outeriter        = 0;
 884     inneriter        = 0;
 885
 886     for(iidx=0;iidx<4*DIM;iidx++)
 887     {
 888         scratch[iidx] = 0.0;
 889     }
 890
 891     /* Start outer loop over neighborlists */
 892     for(iidx=0; iidx<nri; iidx++)
 893     {
 894         /* Load shift vector for this list */
 895         i_shift_offset   = DIM*shiftidx[iidx];
 896
 897         /* Load limits for loop over neighbors */
 898         j_index_start    = jindex[iidx];
 899         j_index_end      = jindex[iidx+1];
 900
 901         /* Get outer coordinate index */
 902         inr              = iinr[iidx];
 903         i_coord_offset   = DIM*inr;
 904
 905         /* Load i particle coords and add shift vector */
 906         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 907                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 908
 909         fix0             = _mm_setzero_ps();
 910         fiy0             = _mm_setzero_ps();
 911         fiz0             = _mm_setzero_ps();
 912         fix1             = _mm_setzero_ps();
 913         fiy1             = _mm_setzero_ps();
 914         fiz1             = _mm_setzero_ps();
 915         fix2             = _mm_setzero_ps();
 916         fiy2             = _mm_setzero_ps();
 917         fiz2             = _mm_setzero_ps();
 918         fix3             = _mm_setzero_ps();
 919         fiy3             = _mm_setzero_ps();
 920         fiz3             = _mm_setzero_ps();
 921
 922         /* Start inner kernel loop */
 923         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 924         {
 925
 926             /* Get j neighbor index, and coordinate index */
 927             jnrA             = jjnr[jidx];
 928             jnrB             = jjnr[jidx+1];
 929             jnrC             = jjnr[jidx+2];
 930             jnrD             = jjnr[jidx+3];
 931             j_coord_offsetA  = DIM*jnrA;
 932             j_coord_offsetB  = DIM*jnrB;
 933             j_coord_offsetC  = DIM*jnrC;
 934             j_coord_offsetD  = DIM*jnrD;
 935
 936             /* load j atom coordinates */
 937             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 938                                               x+j_coord_offsetC,x+j_coord_offsetD,
 939                                               &jx0,&jy0,&jz0);
 940
 941             /* Calculate displacement vector */
 942             dx00             = _mm_sub_ps(ix0,jx0);
 943             dy00             = _mm_sub_ps(iy0,jy0);
 944             dz00             = _mm_sub_ps(iz0,jz0);
 945             dx10             = _mm_sub_ps(ix1,jx0);
 946             dy10             = _mm_sub_ps(iy1,jy0);
 947             dz10             = _mm_sub_ps(iz1,jz0);
 948             dx20             = _mm_sub_ps(ix2,jx0);
 949             dy20             = _mm_sub_ps(iy2,jy0);
 950             dz20             = _mm_sub_ps(iz2,jz0);
 951             dx30             = _mm_sub_ps(ix3,jx0);
 952             dy30             = _mm_sub_ps(iy3,jy0);
 953             dz30             = _mm_sub_ps(iz3,jz0);
 954
 955             /* Calculate squared distance and things based on it */
 956             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 957             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 958             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 959             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 960
 961             rinv00           = sse41_invsqrt_f(rsq00);
 962             rinv10           = sse41_invsqrt_f(rsq10);
 963             rinv20           = sse41_invsqrt_f(rsq20);
 964             rinv30           = sse41_invsqrt_f(rsq30);
 965
 966             /* Load parameters for j particles */
 967             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 968                                                               charge+jnrC+0,charge+jnrD+0);
 969             vdwjidx0A        = 2*vdwtype[jnrA+0];
 970             vdwjidx0B        = 2*vdwtype[jnrB+0];
 971             vdwjidx0C        = 2*vdwtype[jnrC+0];
 972             vdwjidx0D        = 2*vdwtype[jnrD+0];
 973
 974             fjx0             = _mm_setzero_ps();
 975             fjy0             = _mm_setzero_ps();
 976             fjz0             = _mm_setzero_ps();
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             r00              = _mm_mul_ps(rsq00,rinv00);
 983
 984             /* Compute parameters for interactions between i and j atoms */
 985             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 986                                          vdwparam+vdwioffset0+vdwjidx0B,
 987                                          vdwparam+vdwioffset0+vdwjidx0C,
 988                                          vdwparam+vdwioffset0+vdwjidx0D,
 989                                          &c6_00,&c12_00);
 990
 991             /* Calculate table index by multiplying r with table scale and truncate to integer */
 992             rt               = _mm_mul_ps(r00,vftabscale);
 993             vfitab           = _mm_cvttps_epi32(rt);
 994             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 995             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 996
 997             /* CUBIC SPLINE TABLE DISPERSION */
 998             vfitab           = _mm_add_epi32(vfitab,ifour);
 999             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1000             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1001             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1002             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1003             _MM_TRANSPOSE4_PS(Y,F,G,H);
1004             Heps             = _mm_mul_ps(vfeps,H);
1005             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1006             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1007             fvdw6            = _mm_mul_ps(c6_00,FF);
1008
1009             /* CUBIC SPLINE TABLE REPULSION */
1010             vfitab           = _mm_add_epi32(vfitab,ifour);
1011             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1012             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1013             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1014             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1015             _MM_TRANSPOSE4_PS(Y,F,G,H);
1016             Heps             = _mm_mul_ps(vfeps,H);
1017             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1018             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1019             fvdw12           = _mm_mul_ps(c12_00,FF);
1020             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1021
1022             fscal            = fvdw;
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm_mul_ps(fscal,dx00);
1026             ty               = _mm_mul_ps(fscal,dy00);
1027             tz               = _mm_mul_ps(fscal,dz00);
1028
1029             /* Update vectorial force */
1030             fix0             = _mm_add_ps(fix0,tx);
1031             fiy0             = _mm_add_ps(fiy0,ty);
1032             fiz0             = _mm_add_ps(fiz0,tz);
1033
1034             fjx0             = _mm_add_ps(fjx0,tx);
1035             fjy0             = _mm_add_ps(fjy0,ty);
1036             fjz0             = _mm_add_ps(fjz0,tz);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r10              = _mm_mul_ps(rsq10,rinv10);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq10             = _mm_mul_ps(iq1,jq0);
1046
1047             /* Calculate table index by multiplying r with table scale and truncate to integer */
1048             rt               = _mm_mul_ps(r10,vftabscale);
1049             vfitab           = _mm_cvttps_epi32(rt);
1050             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1051             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1052
1053             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1054             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1055             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1056             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1057             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1058             _MM_TRANSPOSE4_PS(Y,F,G,H);
1059             Heps             = _mm_mul_ps(vfeps,H);
1060             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1061             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1062             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1063
1064             fscal            = felec;
1065
1066             /* Calculate temporary vectorial force */
1067             tx               = _mm_mul_ps(fscal,dx10);
1068             ty               = _mm_mul_ps(fscal,dy10);
1069             tz               = _mm_mul_ps(fscal,dz10);
1070
1071             /* Update vectorial force */
1072             fix1             = _mm_add_ps(fix1,tx);
1073             fiy1             = _mm_add_ps(fiy1,ty);
1074             fiz1             = _mm_add_ps(fiz1,tz);
1075
1076             fjx0             = _mm_add_ps(fjx0,tx);
1077             fjy0             = _mm_add_ps(fjy0,ty);
1078             fjz0             = _mm_add_ps(fjz0,tz);
1079
1080             /**************************
1081              * CALCULATE INTERACTIONS *
1082              **************************/
1083
1084             r20              = _mm_mul_ps(rsq20,rinv20);
1085
1086             /* Compute parameters for interactions between i and j atoms */
1087             qq20             = _mm_mul_ps(iq2,jq0);
1088
1089             /* Calculate table index by multiplying r with table scale and truncate to integer */
1090             rt               = _mm_mul_ps(r20,vftabscale);
1091             vfitab           = _mm_cvttps_epi32(rt);
1092             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1093             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1094
1095             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1096             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1097             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1098             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1099             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1100             _MM_TRANSPOSE4_PS(Y,F,G,H);
1101             Heps             = _mm_mul_ps(vfeps,H);
1102             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1103             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1104             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1105
1106             fscal            = felec;
1107
1108             /* Calculate temporary vectorial force */
1109             tx               = _mm_mul_ps(fscal,dx20);
1110             ty               = _mm_mul_ps(fscal,dy20);
1111             tz               = _mm_mul_ps(fscal,dz20);
1112
1113             /* Update vectorial force */
1114             fix2             = _mm_add_ps(fix2,tx);
1115             fiy2             = _mm_add_ps(fiy2,ty);
1116             fiz2             = _mm_add_ps(fiz2,tz);
1117
1118             fjx0             = _mm_add_ps(fjx0,tx);
1119             fjy0             = _mm_add_ps(fjy0,ty);
1120             fjz0             = _mm_add_ps(fjz0,tz);
1121
1122             /**************************
1123              * CALCULATE INTERACTIONS *
1124              **************************/
1125
1126             r30              = _mm_mul_ps(rsq30,rinv30);
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             qq30             = _mm_mul_ps(iq3,jq0);
1130
1131             /* Calculate table index by multiplying r with table scale and truncate to integer */
1132             rt               = _mm_mul_ps(r30,vftabscale);
1133             vfitab           = _mm_cvttps_epi32(rt);
1134             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1135             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1136
1137             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1138             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1139             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1140             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1141             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1142             _MM_TRANSPOSE4_PS(Y,F,G,H);
1143             Heps             = _mm_mul_ps(vfeps,H);
1144             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1145             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1146             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1147
1148             fscal            = felec;
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm_mul_ps(fscal,dx30);
1152             ty               = _mm_mul_ps(fscal,dy30);
1153             tz               = _mm_mul_ps(fscal,dz30);
1154
1155             /* Update vectorial force */
1156             fix3             = _mm_add_ps(fix3,tx);
1157             fiy3             = _mm_add_ps(fiy3,ty);
1158             fiz3             = _mm_add_ps(fiz3,tz);
1159
1160             fjx0             = _mm_add_ps(fjx0,tx);
1161             fjy0             = _mm_add_ps(fjy0,ty);
1162             fjz0             = _mm_add_ps(fjz0,tz);
1163
1164             fjptrA             = f+j_coord_offsetA;
1165             fjptrB             = f+j_coord_offsetB;
1166             fjptrC             = f+j_coord_offsetC;
1167             fjptrD             = f+j_coord_offsetD;
1168
1169             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1170
1171             /* Inner loop uses 165 flops */
1172         }
1173
1174         if(jidx<j_index_end)
1175         {
1176
1177             /* Get j neighbor index, and coordinate index */
1178             jnrlistA         = jjnr[jidx];
1179             jnrlistB         = jjnr[jidx+1];
1180             jnrlistC         = jjnr[jidx+2];
1181             jnrlistD         = jjnr[jidx+3];
1182             /* Sign of each element will be negative for non-real atoms.
1183              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1184              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1185              */
1186             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1187             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1188             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1189             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1190             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1191             j_coord_offsetA  = DIM*jnrA;
1192             j_coord_offsetB  = DIM*jnrB;
1193             j_coord_offsetC  = DIM*jnrC;
1194             j_coord_offsetD  = DIM*jnrD;
1195
1196             /* load j atom coordinates */
1197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1198                                               x+j_coord_offsetC,x+j_coord_offsetD,
1199                                               &jx0,&jy0,&jz0);
1200
1201             /* Calculate displacement vector */
1202             dx00             = _mm_sub_ps(ix0,jx0);
1203             dy00             = _mm_sub_ps(iy0,jy0);
1204             dz00             = _mm_sub_ps(iz0,jz0);
1205             dx10             = _mm_sub_ps(ix1,jx0);
1206             dy10             = _mm_sub_ps(iy1,jy0);
1207             dz10             = _mm_sub_ps(iz1,jz0);
1208             dx20             = _mm_sub_ps(ix2,jx0);
1209             dy20             = _mm_sub_ps(iy2,jy0);
1210             dz20             = _mm_sub_ps(iz2,jz0);
1211             dx30             = _mm_sub_ps(ix3,jx0);
1212             dy30             = _mm_sub_ps(iy3,jy0);
1213             dz30             = _mm_sub_ps(iz3,jz0);
1214
1215             /* Calculate squared distance and things based on it */
1216             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1217             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1218             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1219             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1220
1221             rinv00           = sse41_invsqrt_f(rsq00);
1222             rinv10           = sse41_invsqrt_f(rsq10);
1223             rinv20           = sse41_invsqrt_f(rsq20);
1224             rinv30           = sse41_invsqrt_f(rsq30);
1225
1226             /* Load parameters for j particles */
1227             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1228                                                               charge+jnrC+0,charge+jnrD+0);
1229             vdwjidx0A        = 2*vdwtype[jnrA+0];
1230             vdwjidx0B        = 2*vdwtype[jnrB+0];
1231             vdwjidx0C        = 2*vdwtype[jnrC+0];
1232             vdwjidx0D        = 2*vdwtype[jnrD+0];
1233
1234             fjx0             = _mm_setzero_ps();
1235             fjy0             = _mm_setzero_ps();
1236             fjz0             = _mm_setzero_ps();
1237
1238             /**************************
1239              * CALCULATE INTERACTIONS *
1240              **************************/
1241
1242             r00              = _mm_mul_ps(rsq00,rinv00);
1243             r00              = _mm_andnot_ps(dummy_mask,r00);
1244
1245             /* Compute parameters for interactions between i and j atoms */
1246             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1247                                          vdwparam+vdwioffset0+vdwjidx0B,
1248                                          vdwparam+vdwioffset0+vdwjidx0C,
1249                                          vdwparam+vdwioffset0+vdwjidx0D,
1250                                          &c6_00,&c12_00);
1251
1252             /* Calculate table index by multiplying r with table scale and truncate to integer */
1253             rt               = _mm_mul_ps(r00,vftabscale);
1254             vfitab           = _mm_cvttps_epi32(rt);
1255             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1256             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1257
1258             /* CUBIC SPLINE TABLE DISPERSION */
1259             vfitab           = _mm_add_epi32(vfitab,ifour);
1260             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1261             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1262             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1263             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1264             _MM_TRANSPOSE4_PS(Y,F,G,H);
1265             Heps             = _mm_mul_ps(vfeps,H);
1266             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1267             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1268             fvdw6            = _mm_mul_ps(c6_00,FF);
1269
1270             /* CUBIC SPLINE TABLE REPULSION */
1271             vfitab           = _mm_add_epi32(vfitab,ifour);
1272             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1273             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1274             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1275             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1276             _MM_TRANSPOSE4_PS(Y,F,G,H);
1277             Heps             = _mm_mul_ps(vfeps,H);
1278             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1279             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1280             fvdw12           = _mm_mul_ps(c12_00,FF);
1281             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1282
1283             fscal            = fvdw;
1284
1285             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1286
1287             /* Calculate temporary vectorial force */
1288             tx               = _mm_mul_ps(fscal,dx00);
1289             ty               = _mm_mul_ps(fscal,dy00);
1290             tz               = _mm_mul_ps(fscal,dz00);
1291
1292             /* Update vectorial force */
1293             fix0             = _mm_add_ps(fix0,tx);
1294             fiy0             = _mm_add_ps(fiy0,ty);
1295             fiz0             = _mm_add_ps(fiz0,tz);
1296
1297             fjx0             = _mm_add_ps(fjx0,tx);
1298             fjy0             = _mm_add_ps(fjy0,ty);
1299             fjz0             = _mm_add_ps(fjz0,tz);
1300
1301             /**************************
1302              * CALCULATE INTERACTIONS *
1303              **************************/
1304
1305             r10              = _mm_mul_ps(rsq10,rinv10);
1306             r10              = _mm_andnot_ps(dummy_mask,r10);
1307
1308             /* Compute parameters for interactions between i and j atoms */
1309             qq10             = _mm_mul_ps(iq1,jq0);
1310
1311             /* Calculate table index by multiplying r with table scale and truncate to integer */
1312             rt               = _mm_mul_ps(r10,vftabscale);
1313             vfitab           = _mm_cvttps_epi32(rt);
1314             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1315             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1316
1317             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1318             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1319             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1320             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1321             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1322             _MM_TRANSPOSE4_PS(Y,F,G,H);
1323             Heps             = _mm_mul_ps(vfeps,H);
1324             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1325             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1326             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1327
1328             fscal            = felec;
1329
1330             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1331
1332             /* Calculate temporary vectorial force */
1333             tx               = _mm_mul_ps(fscal,dx10);
1334             ty               = _mm_mul_ps(fscal,dy10);
1335             tz               = _mm_mul_ps(fscal,dz10);
1336
1337             /* Update vectorial force */
1338             fix1             = _mm_add_ps(fix1,tx);
1339             fiy1             = _mm_add_ps(fiy1,ty);
1340             fiz1             = _mm_add_ps(fiz1,tz);
1341
1342             fjx0             = _mm_add_ps(fjx0,tx);
1343             fjy0             = _mm_add_ps(fjy0,ty);
1344             fjz0             = _mm_add_ps(fjz0,tz);
1345
1346             /**************************
1347              * CALCULATE INTERACTIONS *
1348              **************************/
1349
1350             r20              = _mm_mul_ps(rsq20,rinv20);
1351             r20              = _mm_andnot_ps(dummy_mask,r20);
1352
1353             /* Compute parameters for interactions between i and j atoms */
1354             qq20             = _mm_mul_ps(iq2,jq0);
1355
1356             /* Calculate table index by multiplying r with table scale and truncate to integer */
1357             rt               = _mm_mul_ps(r20,vftabscale);
1358             vfitab           = _mm_cvttps_epi32(rt);
1359             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1360             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1361
1362             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1363             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1364             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1365             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1366             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1367             _MM_TRANSPOSE4_PS(Y,F,G,H);
1368             Heps             = _mm_mul_ps(vfeps,H);
1369             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1370             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1371             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1372
1373             fscal            = felec;
1374
1375             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1376
1377             /* Calculate temporary vectorial force */
1378             tx               = _mm_mul_ps(fscal,dx20);
1379             ty               = _mm_mul_ps(fscal,dy20);
1380             tz               = _mm_mul_ps(fscal,dz20);
1381
1382             /* Update vectorial force */
1383             fix2             = _mm_add_ps(fix2,tx);
1384             fiy2             = _mm_add_ps(fiy2,ty);
1385             fiz2             = _mm_add_ps(fiz2,tz);
1386
1387             fjx0             = _mm_add_ps(fjx0,tx);
1388             fjy0             = _mm_add_ps(fjy0,ty);
1389             fjz0             = _mm_add_ps(fjz0,tz);
1390
1391             /**************************
1392              * CALCULATE INTERACTIONS *
1393              **************************/
1394
1395             r30              = _mm_mul_ps(rsq30,rinv30);
1396             r30              = _mm_andnot_ps(dummy_mask,r30);
1397
1398             /* Compute parameters for interactions between i and j atoms */
1399             qq30             = _mm_mul_ps(iq3,jq0);
1400
1401             /* Calculate table index by multiplying r with table scale and truncate to integer */
1402             rt               = _mm_mul_ps(r30,vftabscale);
1403             vfitab           = _mm_cvttps_epi32(rt);
1404             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1405             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1406
1407             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1408             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1409             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1410             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1411             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1412             _MM_TRANSPOSE4_PS(Y,F,G,H);
1413             Heps             = _mm_mul_ps(vfeps,H);
1414             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1415             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1416             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1417
1418             fscal            = felec;
1419
1420             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1421
1422             /* Calculate temporary vectorial force */
1423             tx               = _mm_mul_ps(fscal,dx30);
1424             ty               = _mm_mul_ps(fscal,dy30);
1425             tz               = _mm_mul_ps(fscal,dz30);
1426
1427             /* Update vectorial force */
1428             fix3             = _mm_add_ps(fix3,tx);
1429             fiy3             = _mm_add_ps(fiy3,ty);
1430             fiz3             = _mm_add_ps(fiz3,tz);
1431
1432             fjx0             = _mm_add_ps(fjx0,tx);
1433             fjy0             = _mm_add_ps(fjy0,ty);
1434             fjz0             = _mm_add_ps(fjz0,tz);
1435
1436             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1437             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1438             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1439             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1440
1441             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1442
1443             /* Inner loop uses 169 flops */
1444         }
1445
1446         /* End of innermost loop */
1447
1448         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1449                                               f+i_coord_offset,fshift+i_shift_offset);
1450
1451         /* Increment number of inner iterations */
1452         inneriter                  += j_index_end - j_index_start;
1453
1454         /* Outer loop uses 24 flops */
1455     }
1456
1457     /* Increment number of outer iterations */
1458     outeriter        += nri;
1459
1460     /* Update outer/inner flops */
1461
1462     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*169);
1463 }