src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_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_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 132     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 133     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 167                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 168
 169         fix0             = _mm_setzero_ps();
 170         fiy0             = _mm_setzero_ps();
 171         fiz0             = _mm_setzero_ps();
 172         fix1             = _mm_setzero_ps();
 173         fiy1             = _mm_setzero_ps();
 174         fiz1             = _mm_setzero_ps();
 175         fix2             = _mm_setzero_ps();
 176         fiy2             = _mm_setzero_ps();
 177         fiz2             = _mm_setzero_ps();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_ps();
 181         vvdwsum          = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206             dx10             = _mm_sub_ps(ix1,jx0);
 207             dy10             = _mm_sub_ps(iy1,jy0);
 208             dz10             = _mm_sub_ps(iz1,jz0);
 209             dx20             = _mm_sub_ps(ix2,jx0);
 210             dy20             = _mm_sub_ps(iy2,jy0);
 211             dz20             = _mm_sub_ps(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 217
 218             rinv00           = sse41_invsqrt_f(rsq00);
 219             rinv10           = sse41_invsqrt_f(rsq10);
 220             rinv20           = sse41_invsqrt_f(rsq20);
 221
 222             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 223             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 228                                                               charge+jnrC+0,charge+jnrD+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231             vdwjidx0C        = 2*vdwtype[jnrC+0];
 232             vdwjidx0D        = 2*vdwtype[jnrD+0];
 233
 234             fjx0             = _mm_setzero_ps();
 235             fjy0             = _mm_setzero_ps();
 236             fjz0             = _mm_setzero_ps();
 237
 238             /**************************
 239              * CALCULATE INTERACTIONS *
 240              **************************/
 241
 242             r00              = _mm_mul_ps(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             qq00             = _mm_mul_ps(iq0,jq0);
 246             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 247                                          vdwparam+vdwioffset0+vdwjidx0B,
 248                                          vdwparam+vdwioffset0+vdwjidx0C,
 249                                          vdwparam+vdwioffset0+vdwjidx0D,
 250                                          &c6_00,&c12_00);
 251
 252             /* Calculate table index by multiplying r with table scale and truncate to integer */
 253             rt               = _mm_mul_ps(r00,vftabscale);
 254             vfitab           = _mm_cvttps_epi32(rt);
 255             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 256             vfitab           = _mm_slli_epi32(vfitab,3);
 257
 258             /* COULOMB ELECTROSTATICS */
 259             velec            = _mm_mul_ps(qq00,rinv00);
 260             felec            = _mm_mul_ps(velec,rinvsq00);
 261
 262             /* CUBIC SPLINE TABLE DISPERSION */
 263             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 264             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 265             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 266             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 267             _MM_TRANSPOSE4_PS(Y,F,G,H);
 268             Heps             = _mm_mul_ps(vfeps,H);
 269             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 270             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 271             vvdw6            = _mm_mul_ps(c6_00,VV);
 272             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 273             fvdw6            = _mm_mul_ps(c6_00,FF);
 274
 275             /* CUBIC SPLINE TABLE REPULSION */
 276             vfitab           = _mm_add_epi32(vfitab,ifour);
 277             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 278             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 279             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 280             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 281             _MM_TRANSPOSE4_PS(Y,F,G,H);
 282             Heps             = _mm_mul_ps(vfeps,H);
 283             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 284             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 285             vvdw12           = _mm_mul_ps(c12_00,VV);
 286             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 287             fvdw12           = _mm_mul_ps(c12_00,FF);
 288             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 289             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             velecsum         = _mm_add_ps(velecsum,velec);
 293             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 294
 295             fscal            = _mm_add_ps(felec,fvdw);
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_ps(fscal,dx00);
 299             ty               = _mm_mul_ps(fscal,dy00);
 300             tz               = _mm_mul_ps(fscal,dz00);
 301
 302             /* Update vectorial force */
 303             fix0             = _mm_add_ps(fix0,tx);
 304             fiy0             = _mm_add_ps(fiy0,ty);
 305             fiz0             = _mm_add_ps(fiz0,tz);
 306
 307             fjx0             = _mm_add_ps(fjx0,tx);
 308             fjy0             = _mm_add_ps(fjy0,ty);
 309             fjz0             = _mm_add_ps(fjz0,tz);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq10             = _mm_mul_ps(iq1,jq0);
 317
 318             /* COULOMB ELECTROSTATICS */
 319             velec            = _mm_mul_ps(qq10,rinv10);
 320             felec            = _mm_mul_ps(velec,rinvsq10);
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velecsum         = _mm_add_ps(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm_mul_ps(fscal,dx10);
 329             ty               = _mm_mul_ps(fscal,dy10);
 330             tz               = _mm_mul_ps(fscal,dz10);
 331
 332             /* Update vectorial force */
 333             fix1             = _mm_add_ps(fix1,tx);
 334             fiy1             = _mm_add_ps(fiy1,ty);
 335             fiz1             = _mm_add_ps(fiz1,tz);
 336
 337             fjx0             = _mm_add_ps(fjx0,tx);
 338             fjy0             = _mm_add_ps(fjy0,ty);
 339             fjz0             = _mm_add_ps(fjz0,tz);
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq20             = _mm_mul_ps(iq2,jq0);
 347
 348             /* COULOMB ELECTROSTATICS */
 349             velec            = _mm_mul_ps(qq20,rinv20);
 350             felec            = _mm_mul_ps(velec,rinvsq20);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velecsum         = _mm_add_ps(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm_mul_ps(fscal,dx20);
 359             ty               = _mm_mul_ps(fscal,dy20);
 360             tz               = _mm_mul_ps(fscal,dz20);
 361
 362             /* Update vectorial force */
 363             fix2             = _mm_add_ps(fix2,tx);
 364             fiy2             = _mm_add_ps(fiy2,ty);
 365             fiz2             = _mm_add_ps(fiz2,tz);
 366
 367             fjx0             = _mm_add_ps(fjx0,tx);
 368             fjy0             = _mm_add_ps(fjy0,ty);
 369             fjz0             = _mm_add_ps(fjz0,tz);
 370
 371             fjptrA             = f+j_coord_offsetA;
 372             fjptrB             = f+j_coord_offsetB;
 373             fjptrC             = f+j_coord_offsetC;
 374             fjptrD             = f+j_coord_offsetD;
 375
 376             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 377
 378             /* Inner loop uses 119 flops */
 379         }
 380
 381         if(jidx<j_index_end)
 382         {
 383
 384             /* Get j neighbor index, and coordinate index */
 385             jnrlistA         = jjnr[jidx];
 386             jnrlistB         = jjnr[jidx+1];
 387             jnrlistC         = jjnr[jidx+2];
 388             jnrlistD         = jjnr[jidx+3];
 389             /* Sign of each element will be negative for non-real atoms.
 390              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 391              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 392              */
 393             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 394             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 395             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 396             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 397             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 398             j_coord_offsetA  = DIM*jnrA;
 399             j_coord_offsetB  = DIM*jnrB;
 400             j_coord_offsetC  = DIM*jnrC;
 401             j_coord_offsetD  = DIM*jnrD;
 402
 403             /* load j atom coordinates */
 404             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 405                                               x+j_coord_offsetC,x+j_coord_offsetD,
 406                                               &jx0,&jy0,&jz0);
 407
 408             /* Calculate displacement vector */
 409             dx00             = _mm_sub_ps(ix0,jx0);
 410             dy00             = _mm_sub_ps(iy0,jy0);
 411             dz00             = _mm_sub_ps(iz0,jz0);
 412             dx10             = _mm_sub_ps(ix1,jx0);
 413             dy10             = _mm_sub_ps(iy1,jy0);
 414             dz10             = _mm_sub_ps(iz1,jz0);
 415             dx20             = _mm_sub_ps(ix2,jx0);
 416             dy20             = _mm_sub_ps(iy2,jy0);
 417             dz20             = _mm_sub_ps(iz2,jz0);
 418
 419             /* Calculate squared distance and things based on it */
 420             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 421             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 422             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 423
 424             rinv00           = sse41_invsqrt_f(rsq00);
 425             rinv10           = sse41_invsqrt_f(rsq10);
 426             rinv20           = sse41_invsqrt_f(rsq20);
 427
 428             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 429             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 430             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 431
 432             /* Load parameters for j particles */
 433             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 434                                                               charge+jnrC+0,charge+jnrD+0);
 435             vdwjidx0A        = 2*vdwtype[jnrA+0];
 436             vdwjidx0B        = 2*vdwtype[jnrB+0];
 437             vdwjidx0C        = 2*vdwtype[jnrC+0];
 438             vdwjidx0D        = 2*vdwtype[jnrD+0];
 439
 440             fjx0             = _mm_setzero_ps();
 441             fjy0             = _mm_setzero_ps();
 442             fjz0             = _mm_setzero_ps();
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             r00              = _mm_mul_ps(rsq00,rinv00);
 449             r00              = _mm_andnot_ps(dummy_mask,r00);
 450
 451             /* Compute parameters for interactions between i and j atoms */
 452             qq00             = _mm_mul_ps(iq0,jq0);
 453             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 454                                          vdwparam+vdwioffset0+vdwjidx0B,
 455                                          vdwparam+vdwioffset0+vdwjidx0C,
 456                                          vdwparam+vdwioffset0+vdwjidx0D,
 457                                          &c6_00,&c12_00);
 458
 459             /* Calculate table index by multiplying r with table scale and truncate to integer */
 460             rt               = _mm_mul_ps(r00,vftabscale);
 461             vfitab           = _mm_cvttps_epi32(rt);
 462             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 463             vfitab           = _mm_slli_epi32(vfitab,3);
 464
 465             /* COULOMB ELECTROSTATICS */
 466             velec            = _mm_mul_ps(qq00,rinv00);
 467             felec            = _mm_mul_ps(velec,rinvsq00);
 468
 469             /* CUBIC SPLINE TABLE DISPERSION */
 470             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 471             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 472             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 473             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 474             _MM_TRANSPOSE4_PS(Y,F,G,H);
 475             Heps             = _mm_mul_ps(vfeps,H);
 476             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 477             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 478             vvdw6            = _mm_mul_ps(c6_00,VV);
 479             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 480             fvdw6            = _mm_mul_ps(c6_00,FF);
 481
 482             /* CUBIC SPLINE TABLE REPULSION */
 483             vfitab           = _mm_add_epi32(vfitab,ifour);
 484             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 485             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 486             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 487             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 488             _MM_TRANSPOSE4_PS(Y,F,G,H);
 489             Heps             = _mm_mul_ps(vfeps,H);
 490             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 491             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 492             vvdw12           = _mm_mul_ps(c12_00,VV);
 493             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 494             fvdw12           = _mm_mul_ps(c12_00,FF);
 495             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 496             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             velec            = _mm_andnot_ps(dummy_mask,velec);
 500             velecsum         = _mm_add_ps(velecsum,velec);
 501             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 502             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 503
 504             fscal            = _mm_add_ps(felec,fvdw);
 505
 506             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm_mul_ps(fscal,dx00);
 510             ty               = _mm_mul_ps(fscal,dy00);
 511             tz               = _mm_mul_ps(fscal,dz00);
 512
 513             /* Update vectorial force */
 514             fix0             = _mm_add_ps(fix0,tx);
 515             fiy0             = _mm_add_ps(fiy0,ty);
 516             fiz0             = _mm_add_ps(fiz0,tz);
 517
 518             fjx0             = _mm_add_ps(fjx0,tx);
 519             fjy0             = _mm_add_ps(fjy0,ty);
 520             fjz0             = _mm_add_ps(fjz0,tz);
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq10             = _mm_mul_ps(iq1,jq0);
 528
 529             /* COULOMB ELECTROSTATICS */
 530             velec            = _mm_mul_ps(qq10,rinv10);
 531             felec            = _mm_mul_ps(velec,rinvsq10);
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             velec            = _mm_andnot_ps(dummy_mask,velec);
 535             velecsum         = _mm_add_ps(velecsum,velec);
 536
 537             fscal            = felec;
 538
 539             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_ps(fscal,dx10);
 543             ty               = _mm_mul_ps(fscal,dy10);
 544             tz               = _mm_mul_ps(fscal,dz10);
 545
 546             /* Update vectorial force */
 547             fix1             = _mm_add_ps(fix1,tx);
 548             fiy1             = _mm_add_ps(fiy1,ty);
 549             fiz1             = _mm_add_ps(fiz1,tz);
 550
 551             fjx0             = _mm_add_ps(fjx0,tx);
 552             fjy0             = _mm_add_ps(fjy0,ty);
 553             fjz0             = _mm_add_ps(fjz0,tz);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq20             = _mm_mul_ps(iq2,jq0);
 561
 562             /* COULOMB ELECTROSTATICS */
 563             velec            = _mm_mul_ps(qq20,rinv20);
 564             felec            = _mm_mul_ps(velec,rinvsq20);
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velec            = _mm_andnot_ps(dummy_mask,velec);
 568             velecsum         = _mm_add_ps(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm_mul_ps(fscal,dx20);
 576             ty               = _mm_mul_ps(fscal,dy20);
 577             tz               = _mm_mul_ps(fscal,dz20);
 578
 579             /* Update vectorial force */
 580             fix2             = _mm_add_ps(fix2,tx);
 581             fiy2             = _mm_add_ps(fiy2,ty);
 582             fiz2             = _mm_add_ps(fiz2,tz);
 583
 584             fjx0             = _mm_add_ps(fjx0,tx);
 585             fjy0             = _mm_add_ps(fjy0,ty);
 586             fjz0             = _mm_add_ps(fjz0,tz);
 587
 588             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 589             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 590             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 591             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 592
 593             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 594
 595             /* Inner loop uses 120 flops */
 596         }
 597
 598         /* End of innermost loop */
 599
 600         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 601                                               f+i_coord_offset,fshift+i_shift_offset);
 602
 603         ggid                        = gid[iidx];
 604         /* Update potential energies */
 605         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 606         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 607
 608         /* Increment number of inner iterations */
 609         inneriter                  += j_index_end - j_index_start;
 610
 611         /* Outer loop uses 20 flops */
 612     }
 613
 614     /* Increment number of outer iterations */
 615     outeriter        += nri;
 616
 617     /* Update outer/inner flops */
 618
 619     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 620 }
 621 /*
 622  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_single
 623  * Electrostatics interaction: Coulomb
 624  * VdW interaction:            CubicSplineTable
 625  * Geometry:                   Water3-Particle
 626  * Calculate force/pot:        Force
 627  */
 628 void
 629 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_single
 630                     (t_nblist                    * gmx_restrict       nlist,
 631                      rvec                        * gmx_restrict          xx,
 632                      rvec                        * gmx_restrict          ff,
 633                      struct t_forcerec           * gmx_restrict          fr,
 634                      t_mdatoms                   * gmx_restrict     mdatoms,
 635                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 636                      t_nrnb                      * gmx_restrict        nrnb)
 637 {
 638     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 639      * just 0 for non-waters.
 640      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 641      * jnr indices corresponding to data put in the four positions in the SIMD register.
 642      */
 643     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 644     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 645     int              jnrA,jnrB,jnrC,jnrD;
 646     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 647     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 648     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 649     real             rcutoff_scalar;
 650     real             *shiftvec,*fshift,*x,*f;
 651     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 652     real             scratch[4*DIM];
 653     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 654     int              vdwioffset0;
 655     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 656     int              vdwioffset1;
 657     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 658     int              vdwioffset2;
 659     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 660     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 661     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 662     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 663     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 664     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 665     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 666     real             *charge;
 667     int              nvdwtype;
 668     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 669     int              *vdwtype;
 670     real             *vdwparam;
 671     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 672     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 673     __m128i          vfitab;
 674     __m128i          ifour       = _mm_set1_epi32(4);
 675     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 676     real             *vftab;
 677     __m128           dummy_mask,cutoff_mask;
 678     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 679     __m128           one     = _mm_set1_ps(1.0);
 680     __m128           two     = _mm_set1_ps(2.0);
 681     x                = xx[0];
 682     f                = ff[0];
 683
 684     nri              = nlist->nri;
 685     iinr             = nlist->iinr;
 686     jindex           = nlist->jindex;
 687     jjnr             = nlist->jjnr;
 688     shiftidx         = nlist->shift;
 689     gid              = nlist->gid;
 690     shiftvec         = fr->shift_vec[0];
 691     fshift           = fr->fshift[0];
 692     facel            = _mm_set1_ps(fr->ic->epsfac);
 693     charge           = mdatoms->chargeA;
 694     nvdwtype         = fr->ntype;
 695     vdwparam         = fr->nbfp;
 696     vdwtype          = mdatoms->typeA;
 697
 698     vftab            = kernel_data->table_vdw->data;
 699     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 700
 701     /* Setup water-specific parameters */
 702     inr              = nlist->iinr[0];
 703     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 704     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 705     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 706     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 707
 708     /* Avoid stupid compiler warnings */
 709     jnrA = jnrB = jnrC = jnrD = 0;
 710     j_coord_offsetA = 0;
 711     j_coord_offsetB = 0;
 712     j_coord_offsetC = 0;
 713     j_coord_offsetD = 0;
 714
 715     outeriter        = 0;
 716     inneriter        = 0;
 717
 718     for(iidx=0;iidx<4*DIM;iidx++)
 719     {
 720         scratch[iidx] = 0.0;
 721     }
 722
 723     /* Start outer loop over neighborlists */
 724     for(iidx=0; iidx<nri; iidx++)
 725     {
 726         /* Load shift vector for this list */
 727         i_shift_offset   = DIM*shiftidx[iidx];
 728
 729         /* Load limits for loop over neighbors */
 730         j_index_start    = jindex[iidx];
 731         j_index_end      = jindex[iidx+1];
 732
 733         /* Get outer coordinate index */
 734         inr              = iinr[iidx];
 735         i_coord_offset   = DIM*inr;
 736
 737         /* Load i particle coords and add shift vector */
 738         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 739                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 740
 741         fix0             = _mm_setzero_ps();
 742         fiy0             = _mm_setzero_ps();
 743         fiz0             = _mm_setzero_ps();
 744         fix1             = _mm_setzero_ps();
 745         fiy1             = _mm_setzero_ps();
 746         fiz1             = _mm_setzero_ps();
 747         fix2             = _mm_setzero_ps();
 748         fiy2             = _mm_setzero_ps();
 749         fiz2             = _mm_setzero_ps();
 750
 751         /* Start inner kernel loop */
 752         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 753         {
 754
 755             /* Get j neighbor index, and coordinate index */
 756             jnrA             = jjnr[jidx];
 757             jnrB             = jjnr[jidx+1];
 758             jnrC             = jjnr[jidx+2];
 759             jnrD             = jjnr[jidx+3];
 760             j_coord_offsetA  = DIM*jnrA;
 761             j_coord_offsetB  = DIM*jnrB;
 762             j_coord_offsetC  = DIM*jnrC;
 763             j_coord_offsetD  = DIM*jnrD;
 764
 765             /* load j atom coordinates */
 766             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 767                                               x+j_coord_offsetC,x+j_coord_offsetD,
 768                                               &jx0,&jy0,&jz0);
 769
 770             /* Calculate displacement vector */
 771             dx00             = _mm_sub_ps(ix0,jx0);
 772             dy00             = _mm_sub_ps(iy0,jy0);
 773             dz00             = _mm_sub_ps(iz0,jz0);
 774             dx10             = _mm_sub_ps(ix1,jx0);
 775             dy10             = _mm_sub_ps(iy1,jy0);
 776             dz10             = _mm_sub_ps(iz1,jz0);
 777             dx20             = _mm_sub_ps(ix2,jx0);
 778             dy20             = _mm_sub_ps(iy2,jy0);
 779             dz20             = _mm_sub_ps(iz2,jz0);
 780
 781             /* Calculate squared distance and things based on it */
 782             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 783             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 784             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 785
 786             rinv00           = sse41_invsqrt_f(rsq00);
 787             rinv10           = sse41_invsqrt_f(rsq10);
 788             rinv20           = sse41_invsqrt_f(rsq20);
 789
 790             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 791             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 792             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 793
 794             /* Load parameters for j particles */
 795             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 796                                                               charge+jnrC+0,charge+jnrD+0);
 797             vdwjidx0A        = 2*vdwtype[jnrA+0];
 798             vdwjidx0B        = 2*vdwtype[jnrB+0];
 799             vdwjidx0C        = 2*vdwtype[jnrC+0];
 800             vdwjidx0D        = 2*vdwtype[jnrD+0];
 801
 802             fjx0             = _mm_setzero_ps();
 803             fjy0             = _mm_setzero_ps();
 804             fjz0             = _mm_setzero_ps();
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             r00              = _mm_mul_ps(rsq00,rinv00);
 811
 812             /* Compute parameters for interactions between i and j atoms */
 813             qq00             = _mm_mul_ps(iq0,jq0);
 814             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 815                                          vdwparam+vdwioffset0+vdwjidx0B,
 816                                          vdwparam+vdwioffset0+vdwjidx0C,
 817                                          vdwparam+vdwioffset0+vdwjidx0D,
 818                                          &c6_00,&c12_00);
 819
 820             /* Calculate table index by multiplying r with table scale and truncate to integer */
 821             rt               = _mm_mul_ps(r00,vftabscale);
 822             vfitab           = _mm_cvttps_epi32(rt);
 823             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 824             vfitab           = _mm_slli_epi32(vfitab,3);
 825
 826             /* COULOMB ELECTROSTATICS */
 827             velec            = _mm_mul_ps(qq00,rinv00);
 828             felec            = _mm_mul_ps(velec,rinvsq00);
 829
 830             /* CUBIC SPLINE TABLE DISPERSION */
 831             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 832             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 833             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 834             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 835             _MM_TRANSPOSE4_PS(Y,F,G,H);
 836             Heps             = _mm_mul_ps(vfeps,H);
 837             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 838             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 839             fvdw6            = _mm_mul_ps(c6_00,FF);
 840
 841             /* CUBIC SPLINE TABLE REPULSION */
 842             vfitab           = _mm_add_epi32(vfitab,ifour);
 843             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 844             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 845             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 846             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 847             _MM_TRANSPOSE4_PS(Y,F,G,H);
 848             Heps             = _mm_mul_ps(vfeps,H);
 849             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 850             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 851             fvdw12           = _mm_mul_ps(c12_00,FF);
 852             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 853
 854             fscal            = _mm_add_ps(felec,fvdw);
 855
 856             /* Calculate temporary vectorial force */
 857             tx               = _mm_mul_ps(fscal,dx00);
 858             ty               = _mm_mul_ps(fscal,dy00);
 859             tz               = _mm_mul_ps(fscal,dz00);
 860
 861             /* Update vectorial force */
 862             fix0             = _mm_add_ps(fix0,tx);
 863             fiy0             = _mm_add_ps(fiy0,ty);
 864             fiz0             = _mm_add_ps(fiz0,tz);
 865
 866             fjx0             = _mm_add_ps(fjx0,tx);
 867             fjy0             = _mm_add_ps(fjy0,ty);
 868             fjz0             = _mm_add_ps(fjz0,tz);
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq10             = _mm_mul_ps(iq1,jq0);
 876
 877             /* COULOMB ELECTROSTATICS */
 878             velec            = _mm_mul_ps(qq10,rinv10);
 879             felec            = _mm_mul_ps(velec,rinvsq10);
 880
 881             fscal            = felec;
 882
 883             /* Calculate temporary vectorial force */
 884             tx               = _mm_mul_ps(fscal,dx10);
 885             ty               = _mm_mul_ps(fscal,dy10);
 886             tz               = _mm_mul_ps(fscal,dz10);
 887
 888             /* Update vectorial force */
 889             fix1             = _mm_add_ps(fix1,tx);
 890             fiy1             = _mm_add_ps(fiy1,ty);
 891             fiz1             = _mm_add_ps(fiz1,tz);
 892
 893             fjx0             = _mm_add_ps(fjx0,tx);
 894             fjy0             = _mm_add_ps(fjy0,ty);
 895             fjz0             = _mm_add_ps(fjz0,tz);
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq20             = _mm_mul_ps(iq2,jq0);
 903
 904             /* COULOMB ELECTROSTATICS */
 905             velec            = _mm_mul_ps(qq20,rinv20);
 906             felec            = _mm_mul_ps(velec,rinvsq20);
 907
 908             fscal            = felec;
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm_mul_ps(fscal,dx20);
 912             ty               = _mm_mul_ps(fscal,dy20);
 913             tz               = _mm_mul_ps(fscal,dz20);
 914
 915             /* Update vectorial force */
 916             fix2             = _mm_add_ps(fix2,tx);
 917             fiy2             = _mm_add_ps(fiy2,ty);
 918             fiz2             = _mm_add_ps(fiz2,tz);
 919
 920             fjx0             = _mm_add_ps(fjx0,tx);
 921             fjy0             = _mm_add_ps(fjy0,ty);
 922             fjz0             = _mm_add_ps(fjz0,tz);
 923
 924             fjptrA             = f+j_coord_offsetA;
 925             fjptrB             = f+j_coord_offsetB;
 926             fjptrC             = f+j_coord_offsetC;
 927             fjptrD             = f+j_coord_offsetD;
 928
 929             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 930
 931             /* Inner loop uses 108 flops */
 932         }
 933
 934         if(jidx<j_index_end)
 935         {
 936
 937             /* Get j neighbor index, and coordinate index */
 938             jnrlistA         = jjnr[jidx];
 939             jnrlistB         = jjnr[jidx+1];
 940             jnrlistC         = jjnr[jidx+2];
 941             jnrlistD         = jjnr[jidx+3];
 942             /* Sign of each element will be negative for non-real atoms.
 943              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 944              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 945              */
 946             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 947             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 948             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 949             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 950             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 951             j_coord_offsetA  = DIM*jnrA;
 952             j_coord_offsetB  = DIM*jnrB;
 953             j_coord_offsetC  = DIM*jnrC;
 954             j_coord_offsetD  = DIM*jnrD;
 955
 956             /* load j atom coordinates */
 957             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 958                                               x+j_coord_offsetC,x+j_coord_offsetD,
 959                                               &jx0,&jy0,&jz0);
 960
 961             /* Calculate displacement vector */
 962             dx00             = _mm_sub_ps(ix0,jx0);
 963             dy00             = _mm_sub_ps(iy0,jy0);
 964             dz00             = _mm_sub_ps(iz0,jz0);
 965             dx10             = _mm_sub_ps(ix1,jx0);
 966             dy10             = _mm_sub_ps(iy1,jy0);
 967             dz10             = _mm_sub_ps(iz1,jz0);
 968             dx20             = _mm_sub_ps(ix2,jx0);
 969             dy20             = _mm_sub_ps(iy2,jy0);
 970             dz20             = _mm_sub_ps(iz2,jz0);
 971
 972             /* Calculate squared distance and things based on it */
 973             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 974             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 975             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 976
 977             rinv00           = sse41_invsqrt_f(rsq00);
 978             rinv10           = sse41_invsqrt_f(rsq10);
 979             rinv20           = sse41_invsqrt_f(rsq20);
 980
 981             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 982             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 983             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 984
 985             /* Load parameters for j particles */
 986             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 987                                                               charge+jnrC+0,charge+jnrD+0);
 988             vdwjidx0A        = 2*vdwtype[jnrA+0];
 989             vdwjidx0B        = 2*vdwtype[jnrB+0];
 990             vdwjidx0C        = 2*vdwtype[jnrC+0];
 991             vdwjidx0D        = 2*vdwtype[jnrD+0];
 992
 993             fjx0             = _mm_setzero_ps();
 994             fjy0             = _mm_setzero_ps();
 995             fjz0             = _mm_setzero_ps();
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             r00              = _mm_mul_ps(rsq00,rinv00);
1002             r00              = _mm_andnot_ps(dummy_mask,r00);
1003
1004             /* Compute parameters for interactions between i and j atoms */
1005             qq00             = _mm_mul_ps(iq0,jq0);
1006             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1007                                          vdwparam+vdwioffset0+vdwjidx0B,
1008                                          vdwparam+vdwioffset0+vdwjidx0C,
1009                                          vdwparam+vdwioffset0+vdwjidx0D,
1010                                          &c6_00,&c12_00);
1011
1012             /* Calculate table index by multiplying r with table scale and truncate to integer */
1013             rt               = _mm_mul_ps(r00,vftabscale);
1014             vfitab           = _mm_cvttps_epi32(rt);
1015             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1016             vfitab           = _mm_slli_epi32(vfitab,3);
1017
1018             /* COULOMB ELECTROSTATICS */
1019             velec            = _mm_mul_ps(qq00,rinv00);
1020             felec            = _mm_mul_ps(velec,rinvsq00);
1021
1022             /* CUBIC SPLINE TABLE DISPERSION */
1023             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1024             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1025             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1026             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1027             _MM_TRANSPOSE4_PS(Y,F,G,H);
1028             Heps             = _mm_mul_ps(vfeps,H);
1029             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1030             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1031             fvdw6            = _mm_mul_ps(c6_00,FF);
1032
1033             /* CUBIC SPLINE TABLE REPULSION */
1034             vfitab           = _mm_add_epi32(vfitab,ifour);
1035             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1036             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1037             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1038             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1039             _MM_TRANSPOSE4_PS(Y,F,G,H);
1040             Heps             = _mm_mul_ps(vfeps,H);
1041             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1042             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1043             fvdw12           = _mm_mul_ps(c12_00,FF);
1044             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1045
1046             fscal            = _mm_add_ps(felec,fvdw);
1047
1048             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm_mul_ps(fscal,dx00);
1052             ty               = _mm_mul_ps(fscal,dy00);
1053             tz               = _mm_mul_ps(fscal,dz00);
1054
1055             /* Update vectorial force */
1056             fix0             = _mm_add_ps(fix0,tx);
1057             fiy0             = _mm_add_ps(fiy0,ty);
1058             fiz0             = _mm_add_ps(fiz0,tz);
1059
1060             fjx0             = _mm_add_ps(fjx0,tx);
1061             fjy0             = _mm_add_ps(fjy0,ty);
1062             fjz0             = _mm_add_ps(fjz0,tz);
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq10             = _mm_mul_ps(iq1,jq0);
1070
1071             /* COULOMB ELECTROSTATICS */
1072             velec            = _mm_mul_ps(qq10,rinv10);
1073             felec            = _mm_mul_ps(velec,rinvsq10);
1074
1075             fscal            = felec;
1076
1077             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1078
1079             /* Calculate temporary vectorial force */
1080             tx               = _mm_mul_ps(fscal,dx10);
1081             ty               = _mm_mul_ps(fscal,dy10);
1082             tz               = _mm_mul_ps(fscal,dz10);
1083
1084             /* Update vectorial force */
1085             fix1             = _mm_add_ps(fix1,tx);
1086             fiy1             = _mm_add_ps(fiy1,ty);
1087             fiz1             = _mm_add_ps(fiz1,tz);
1088
1089             fjx0             = _mm_add_ps(fjx0,tx);
1090             fjy0             = _mm_add_ps(fjy0,ty);
1091             fjz0             = _mm_add_ps(fjz0,tz);
1092
1093             /**************************
1094              * CALCULATE INTERACTIONS *
1095              **************************/
1096
1097             /* Compute parameters for interactions between i and j atoms */
1098             qq20             = _mm_mul_ps(iq2,jq0);
1099
1100             /* COULOMB ELECTROSTATICS */
1101             velec            = _mm_mul_ps(qq20,rinv20);
1102             felec            = _mm_mul_ps(velec,rinvsq20);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm_andnot_ps(dummy_mask,fscal);
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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1123             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1124             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1125             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1126
1127             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1128
1129             /* Inner loop uses 109 flops */
1130         }
1131
1132         /* End of innermost loop */
1133
1134         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1135                                               f+i_coord_offset,fshift+i_shift_offset);
1136
1137         /* Increment number of inner iterations */
1138         inneriter                  += j_index_end - j_index_start;
1139
1140         /* Outer loop uses 18 flops */
1141     }
1142
1143     /* Increment number of outer iterations */
1144     outeriter        += nri;
1145
1146     /* Update outer/inner flops */
1147
1148     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1149 }