src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           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->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 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_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 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
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184         vvdwsum          = _mm_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               x+j_coord_offsetC,x+j_coord_offsetD,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_ps(ix0,jx0);
 207             dy00             = _mm_sub_ps(iy0,jy0);
 208             dz00             = _mm_sub_ps(iz0,jz0);
 209             dx10             = _mm_sub_ps(ix1,jx0);
 210             dy10             = _mm_sub_ps(iy1,jy0);
 211             dz10             = _mm_sub_ps(iz1,jz0);
 212             dx20             = _mm_sub_ps(ix2,jx0);
 213             dy20             = _mm_sub_ps(iy2,jy0);
 214             dz20             = _mm_sub_ps(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224
 225             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 226             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 227             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 228
 229             /* Load parameters for j particles */
 230             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 231                                                               charge+jnrC+0,charge+jnrD+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234             vdwjidx0C        = 2*vdwtype[jnrC+0];
 235             vdwjidx0D        = 2*vdwtype[jnrD+0];
 236
 237             fjx0             = _mm_setzero_ps();
 238             fjy0             = _mm_setzero_ps();
 239             fjz0             = _mm_setzero_ps();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             r00              = _mm_mul_ps(rsq00,rinv00);
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             qq00             = _mm_mul_ps(iq0,jq0);
 249             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,
 251                                          vdwparam+vdwioffset0+vdwjidx0C,
 252                                          vdwparam+vdwioffset0+vdwjidx0D,
 253                                          &c6_00,&c12_00);
 254
 255             /* Calculate table index by multiplying r with table scale and truncate to integer */
 256             rt               = _mm_mul_ps(r00,vftabscale);
 257             vfitab           = _mm_cvttps_epi32(rt);
 258             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 259             vfitab           = _mm_slli_epi32(vfitab,3);
 260
 261             /* COULOMB ELECTROSTATICS */
 262             velec            = _mm_mul_ps(qq00,rinv00);
 263             felec            = _mm_mul_ps(velec,rinvsq00);
 264
 265             /* CUBIC SPLINE TABLE DISPERSION */
 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             velecsum         = _mm_add_ps(velecsum,velec);
 296             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = _mm_add_ps(felec,fvdw);
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_ps(fscal,dx00);
 302             ty               = _mm_mul_ps(fscal,dy00);
 303             tz               = _mm_mul_ps(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm_add_ps(fix0,tx);
 307             fiy0             = _mm_add_ps(fiy0,ty);
 308             fiz0             = _mm_add_ps(fiz0,tz);
 309
 310             fjx0             = _mm_add_ps(fjx0,tx);
 311             fjy0             = _mm_add_ps(fjy0,ty);
 312             fjz0             = _mm_add_ps(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq10             = _mm_mul_ps(iq1,jq0);
 320
 321             /* COULOMB ELECTROSTATICS */
 322             velec            = _mm_mul_ps(qq10,rinv10);
 323             felec            = _mm_mul_ps(velec,rinvsq10);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velecsum         = _mm_add_ps(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_ps(fscal,dx10);
 332             ty               = _mm_mul_ps(fscal,dy10);
 333             tz               = _mm_mul_ps(fscal,dz10);
 334
 335             /* Update vectorial force */
 336             fix1             = _mm_add_ps(fix1,tx);
 337             fiy1             = _mm_add_ps(fiy1,ty);
 338             fiz1             = _mm_add_ps(fiz1,tz);
 339
 340             fjx0             = _mm_add_ps(fjx0,tx);
 341             fjy0             = _mm_add_ps(fjy0,ty);
 342             fjz0             = _mm_add_ps(fjz0,tz);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq20             = _mm_mul_ps(iq2,jq0);
 350
 351             /* COULOMB ELECTROSTATICS */
 352             velec            = _mm_mul_ps(qq20,rinv20);
 353             felec            = _mm_mul_ps(velec,rinvsq20);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm_mul_ps(fscal,dx20);
 362             ty               = _mm_mul_ps(fscal,dy20);
 363             tz               = _mm_mul_ps(fscal,dz20);
 364
 365             /* Update vectorial force */
 366             fix2             = _mm_add_ps(fix2,tx);
 367             fiy2             = _mm_add_ps(fiy2,ty);
 368             fiz2             = _mm_add_ps(fiz2,tz);
 369
 370             fjx0             = _mm_add_ps(fjx0,tx);
 371             fjy0             = _mm_add_ps(fjy0,ty);
 372             fjz0             = _mm_add_ps(fjz0,tz);
 373
 374             fjptrA             = f+j_coord_offsetA;
 375             fjptrB             = f+j_coord_offsetB;
 376             fjptrC             = f+j_coord_offsetC;
 377             fjptrD             = f+j_coord_offsetD;
 378
 379             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 380
 381             /* Inner loop uses 119 flops */
 382         }
 383
 384         if(jidx<j_index_end)
 385         {
 386
 387             /* Get j neighbor index, and coordinate index */
 388             jnrlistA         = jjnr[jidx];
 389             jnrlistB         = jjnr[jidx+1];
 390             jnrlistC         = jjnr[jidx+2];
 391             jnrlistD         = jjnr[jidx+3];
 392             /* Sign of each element will be negative for non-real atoms.
 393              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 394              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 395              */
 396             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 397             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 398             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 399             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 400             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 401             j_coord_offsetA  = DIM*jnrA;
 402             j_coord_offsetB  = DIM*jnrB;
 403             j_coord_offsetC  = DIM*jnrC;
 404             j_coord_offsetD  = DIM*jnrD;
 405
 406             /* load j atom coordinates */
 407             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 408                                               x+j_coord_offsetC,x+j_coord_offsetD,
 409                                               &jx0,&jy0,&jz0);
 410
 411             /* Calculate displacement vector */
 412             dx00             = _mm_sub_ps(ix0,jx0);
 413             dy00             = _mm_sub_ps(iy0,jy0);
 414             dz00             = _mm_sub_ps(iz0,jz0);
 415             dx10             = _mm_sub_ps(ix1,jx0);
 416             dy10             = _mm_sub_ps(iy1,jy0);
 417             dz10             = _mm_sub_ps(iz1,jz0);
 418             dx20             = _mm_sub_ps(ix2,jx0);
 419             dy20             = _mm_sub_ps(iy2,jy0);
 420             dz20             = _mm_sub_ps(iz2,jz0);
 421
 422             /* Calculate squared distance and things based on it */
 423             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 424             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 425             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 426
 427             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 428             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 429             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 430
 431             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 432             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 433             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 434
 435             /* Load parameters for j particles */
 436             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 437                                                               charge+jnrC+0,charge+jnrD+0);
 438             vdwjidx0A        = 2*vdwtype[jnrA+0];
 439             vdwjidx0B        = 2*vdwtype[jnrB+0];
 440             vdwjidx0C        = 2*vdwtype[jnrC+0];
 441             vdwjidx0D        = 2*vdwtype[jnrD+0];
 442
 443             fjx0             = _mm_setzero_ps();
 444             fjy0             = _mm_setzero_ps();
 445             fjz0             = _mm_setzero_ps();
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             r00              = _mm_mul_ps(rsq00,rinv00);
 452             r00              = _mm_andnot_ps(dummy_mask,r00);
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq00             = _mm_mul_ps(iq0,jq0);
 456             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 457                                          vdwparam+vdwioffset0+vdwjidx0B,
 458                                          vdwparam+vdwioffset0+vdwjidx0C,
 459                                          vdwparam+vdwioffset0+vdwjidx0D,
 460                                          &c6_00,&c12_00);
 461
 462             /* Calculate table index by multiplying r with table scale and truncate to integer */
 463             rt               = _mm_mul_ps(r00,vftabscale);
 464             vfitab           = _mm_cvttps_epi32(rt);
 465             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 466             vfitab           = _mm_slli_epi32(vfitab,3);
 467
 468             /* COULOMB ELECTROSTATICS */
 469             velec            = _mm_mul_ps(qq00,rinv00);
 470             felec            = _mm_mul_ps(velec,rinvsq00);
 471
 472             /* CUBIC SPLINE TABLE DISPERSION */
 473             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 474             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 475             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 476             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 477             _MM_TRANSPOSE4_PS(Y,F,G,H);
 478             Heps             = _mm_mul_ps(vfeps,H);
 479             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 480             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 481             vvdw6            = _mm_mul_ps(c6_00,VV);
 482             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 483             fvdw6            = _mm_mul_ps(c6_00,FF);
 484
 485             /* CUBIC SPLINE TABLE REPULSION */
 486             vfitab           = _mm_add_epi32(vfitab,ifour);
 487             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 488             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 489             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 490             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 491             _MM_TRANSPOSE4_PS(Y,F,G,H);
 492             Heps             = _mm_mul_ps(vfeps,H);
 493             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 494             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 495             vvdw12           = _mm_mul_ps(c12_00,VV);
 496             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 497             fvdw12           = _mm_mul_ps(c12_00,FF);
 498             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 499             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velec            = _mm_andnot_ps(dummy_mask,velec);
 503             velecsum         = _mm_add_ps(velecsum,velec);
 504             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 505             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 506
 507             fscal            = _mm_add_ps(felec,fvdw);
 508
 509             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm_mul_ps(fscal,dx00);
 513             ty               = _mm_mul_ps(fscal,dy00);
 514             tz               = _mm_mul_ps(fscal,dz00);
 515
 516             /* Update vectorial force */
 517             fix0             = _mm_add_ps(fix0,tx);
 518             fiy0             = _mm_add_ps(fiy0,ty);
 519             fiz0             = _mm_add_ps(fiz0,tz);
 520
 521             fjx0             = _mm_add_ps(fjx0,tx);
 522             fjy0             = _mm_add_ps(fjy0,ty);
 523             fjz0             = _mm_add_ps(fjz0,tz);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq10             = _mm_mul_ps(iq1,jq0);
 531
 532             /* COULOMB ELECTROSTATICS */
 533             velec            = _mm_mul_ps(qq10,rinv10);
 534             felec            = _mm_mul_ps(velec,rinvsq10);
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             velec            = _mm_andnot_ps(dummy_mask,velec);
 538             velecsum         = _mm_add_ps(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm_mul_ps(fscal,dx10);
 546             ty               = _mm_mul_ps(fscal,dy10);
 547             tz               = _mm_mul_ps(fscal,dz10);
 548
 549             /* Update vectorial force */
 550             fix1             = _mm_add_ps(fix1,tx);
 551             fiy1             = _mm_add_ps(fiy1,ty);
 552             fiz1             = _mm_add_ps(fiz1,tz);
 553
 554             fjx0             = _mm_add_ps(fjx0,tx);
 555             fjy0             = _mm_add_ps(fjy0,ty);
 556             fjz0             = _mm_add_ps(fjz0,tz);
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq20             = _mm_mul_ps(iq2,jq0);
 564
 565             /* COULOMB ELECTROSTATICS */
 566             velec            = _mm_mul_ps(qq20,rinv20);
 567             felec            = _mm_mul_ps(velec,rinvsq20);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 576
 577             /* Calculate temporary vectorial force */
 578             tx               = _mm_mul_ps(fscal,dx20);
 579             ty               = _mm_mul_ps(fscal,dy20);
 580             tz               = _mm_mul_ps(fscal,dz20);
 581
 582             /* Update vectorial force */
 583             fix2             = _mm_add_ps(fix2,tx);
 584             fiy2             = _mm_add_ps(fiy2,ty);
 585             fiz2             = _mm_add_ps(fiz2,tz);
 586
 587             fjx0             = _mm_add_ps(fjx0,tx);
 588             fjy0             = _mm_add_ps(fjy0,ty);
 589             fjz0             = _mm_add_ps(fjz0,tz);
 590
 591             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 592             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 593             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 594             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 595
 596             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 597
 598             /* Inner loop uses 120 flops */
 599         }
 600
 601         /* End of innermost loop */
 602
 603         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 604                                               f+i_coord_offset,fshift+i_shift_offset);
 605
 606         ggid                        = gid[iidx];
 607         /* Update potential energies */
 608         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 609         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 610
 611         /* Increment number of inner iterations */
 612         inneriter                  += j_index_end - j_index_start;
 613
 614         /* Outer loop uses 20 flops */
 615     }
 616
 617     /* Increment number of outer iterations */
 618     outeriter        += nri;
 619
 620     /* Update outer/inner flops */
 621
 622     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 623 }
 624 /*
 625  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_single
 626  * Electrostatics interaction: Coulomb
 627  * VdW interaction:            CubicSplineTable
 628  * Geometry:                   Water3-Particle
 629  * Calculate force/pot:        Force
 630  */
 631 void
 632 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_single
 633                     (t_nblist                    * gmx_restrict       nlist,
 634                      rvec                        * gmx_restrict          xx,
 635                      rvec                        * gmx_restrict          ff,
 636                      t_forcerec                  * gmx_restrict          fr,
 637                      t_mdatoms                   * gmx_restrict     mdatoms,
 638                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 639                      t_nrnb                      * gmx_restrict        nrnb)
 640 {
 641     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 642      * just 0 for non-waters.
 643      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 644      * jnr indices corresponding to data put in the four positions in the SIMD register.
 645      */
 646     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 647     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 648     int              jnrA,jnrB,jnrC,jnrD;
 649     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 650     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 651     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 652     real             rcutoff_scalar;
 653     real             *shiftvec,*fshift,*x,*f;
 654     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 655     real             scratch[4*DIM];
 656     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 657     int              vdwioffset0;
 658     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 659     int              vdwioffset1;
 660     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 661     int              vdwioffset2;
 662     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 663     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 664     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 665     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 666     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 667     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 668     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 669     real             *charge;
 670     int              nvdwtype;
 671     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 672     int              *vdwtype;
 673     real             *vdwparam;
 674     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 675     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 676     __m128i          vfitab;
 677     __m128i          ifour       = _mm_set1_epi32(4);
 678     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 679     real             *vftab;
 680     __m128           dummy_mask,cutoff_mask;
 681     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 682     __m128           one     = _mm_set1_ps(1.0);
 683     __m128           two     = _mm_set1_ps(2.0);
 684     x                = xx[0];
 685     f                = ff[0];
 686
 687     nri              = nlist->nri;
 688     iinr             = nlist->iinr;
 689     jindex           = nlist->jindex;
 690     jjnr             = nlist->jjnr;
 691     shiftidx         = nlist->shift;
 692     gid              = nlist->gid;
 693     shiftvec         = fr->shift_vec[0];
 694     fshift           = fr->fshift[0];
 695     facel            = _mm_set1_ps(fr->epsfac);
 696     charge           = mdatoms->chargeA;
 697     nvdwtype         = fr->ntype;
 698     vdwparam         = fr->nbfp;
 699     vdwtype          = mdatoms->typeA;
 700
 701     vftab            = kernel_data->table_vdw->data;
 702     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 703
 704     /* Setup water-specific parameters */
 705     inr              = nlist->iinr[0];
 706     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 707     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 708     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 709     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 710
 711     /* Avoid stupid compiler warnings */
 712     jnrA = jnrB = jnrC = jnrD = 0;
 713     j_coord_offsetA = 0;
 714     j_coord_offsetB = 0;
 715     j_coord_offsetC = 0;
 716     j_coord_offsetD = 0;
 717
 718     outeriter        = 0;
 719     inneriter        = 0;
 720
 721     for(iidx=0;iidx<4*DIM;iidx++)
 722     {
 723         scratch[iidx] = 0.0;
 724     }
 725
 726     /* Start outer loop over neighborlists */
 727     for(iidx=0; iidx<nri; iidx++)
 728     {
 729         /* Load shift vector for this list */
 730         i_shift_offset   = DIM*shiftidx[iidx];
 731
 732         /* Load limits for loop over neighbors */
 733         j_index_start    = jindex[iidx];
 734         j_index_end      = jindex[iidx+1];
 735
 736         /* Get outer coordinate index */
 737         inr              = iinr[iidx];
 738         i_coord_offset   = DIM*inr;
 739
 740         /* Load i particle coords and add shift vector */
 741         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 742                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 743
 744         fix0             = _mm_setzero_ps();
 745         fiy0             = _mm_setzero_ps();
 746         fiz0             = _mm_setzero_ps();
 747         fix1             = _mm_setzero_ps();
 748         fiy1             = _mm_setzero_ps();
 749         fiz1             = _mm_setzero_ps();
 750         fix2             = _mm_setzero_ps();
 751         fiy2             = _mm_setzero_ps();
 752         fiz2             = _mm_setzero_ps();
 753
 754         /* Start inner kernel loop */
 755         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 756         {
 757
 758             /* Get j neighbor index, and coordinate index */
 759             jnrA             = jjnr[jidx];
 760             jnrB             = jjnr[jidx+1];
 761             jnrC             = jjnr[jidx+2];
 762             jnrD             = jjnr[jidx+3];
 763             j_coord_offsetA  = DIM*jnrA;
 764             j_coord_offsetB  = DIM*jnrB;
 765             j_coord_offsetC  = DIM*jnrC;
 766             j_coord_offsetD  = DIM*jnrD;
 767
 768             /* load j atom coordinates */
 769             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 770                                               x+j_coord_offsetC,x+j_coord_offsetD,
 771                                               &jx0,&jy0,&jz0);
 772
 773             /* Calculate displacement vector */
 774             dx00             = _mm_sub_ps(ix0,jx0);
 775             dy00             = _mm_sub_ps(iy0,jy0);
 776             dz00             = _mm_sub_ps(iz0,jz0);
 777             dx10             = _mm_sub_ps(ix1,jx0);
 778             dy10             = _mm_sub_ps(iy1,jy0);
 779             dz10             = _mm_sub_ps(iz1,jz0);
 780             dx20             = _mm_sub_ps(ix2,jx0);
 781             dy20             = _mm_sub_ps(iy2,jy0);
 782             dz20             = _mm_sub_ps(iz2,jz0);
 783
 784             /* Calculate squared distance and things based on it */
 785             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 786             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 787             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 788
 789             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 790             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 791             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 792
 793             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 794             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 795             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 796
 797             /* Load parameters for j particles */
 798             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 799                                                               charge+jnrC+0,charge+jnrD+0);
 800             vdwjidx0A        = 2*vdwtype[jnrA+0];
 801             vdwjidx0B        = 2*vdwtype[jnrB+0];
 802             vdwjidx0C        = 2*vdwtype[jnrC+0];
 803             vdwjidx0D        = 2*vdwtype[jnrD+0];
 804
 805             fjx0             = _mm_setzero_ps();
 806             fjy0             = _mm_setzero_ps();
 807             fjz0             = _mm_setzero_ps();
 808
 809             /**************************
 810              * CALCULATE INTERACTIONS *
 811              **************************/
 812
 813             r00              = _mm_mul_ps(rsq00,rinv00);
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             qq00             = _mm_mul_ps(iq0,jq0);
 817             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 818                                          vdwparam+vdwioffset0+vdwjidx0B,
 819                                          vdwparam+vdwioffset0+vdwjidx0C,
 820                                          vdwparam+vdwioffset0+vdwjidx0D,
 821                                          &c6_00,&c12_00);
 822
 823             /* Calculate table index by multiplying r with table scale and truncate to integer */
 824             rt               = _mm_mul_ps(r00,vftabscale);
 825             vfitab           = _mm_cvttps_epi32(rt);
 826             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 827             vfitab           = _mm_slli_epi32(vfitab,3);
 828
 829             /* COULOMB ELECTROSTATICS */
 830             velec            = _mm_mul_ps(qq00,rinv00);
 831             felec            = _mm_mul_ps(velec,rinvsq00);
 832
 833             /* CUBIC SPLINE TABLE DISPERSION */
 834             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 835             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 836             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 837             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 838             _MM_TRANSPOSE4_PS(Y,F,G,H);
 839             Heps             = _mm_mul_ps(vfeps,H);
 840             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 841             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 842             fvdw6            = _mm_mul_ps(c6_00,FF);
 843
 844             /* CUBIC SPLINE TABLE REPULSION */
 845             vfitab           = _mm_add_epi32(vfitab,ifour);
 846             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 847             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 848             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 849             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 850             _MM_TRANSPOSE4_PS(Y,F,G,H);
 851             Heps             = _mm_mul_ps(vfeps,H);
 852             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 853             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 854             fvdw12           = _mm_mul_ps(c12_00,FF);
 855             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 856
 857             fscal            = _mm_add_ps(felec,fvdw);
 858
 859             /* Calculate temporary vectorial force */
 860             tx               = _mm_mul_ps(fscal,dx00);
 861             ty               = _mm_mul_ps(fscal,dy00);
 862             tz               = _mm_mul_ps(fscal,dz00);
 863
 864             /* Update vectorial force */
 865             fix0             = _mm_add_ps(fix0,tx);
 866             fiy0             = _mm_add_ps(fiy0,ty);
 867             fiz0             = _mm_add_ps(fiz0,tz);
 868
 869             fjx0             = _mm_add_ps(fjx0,tx);
 870             fjy0             = _mm_add_ps(fjy0,ty);
 871             fjz0             = _mm_add_ps(fjz0,tz);
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             /* Compute parameters for interactions between i and j atoms */
 878             qq10             = _mm_mul_ps(iq1,jq0);
 879
 880             /* COULOMB ELECTROSTATICS */
 881             velec            = _mm_mul_ps(qq10,rinv10);
 882             felec            = _mm_mul_ps(velec,rinvsq10);
 883
 884             fscal            = felec;
 885
 886             /* Calculate temporary vectorial force */
 887             tx               = _mm_mul_ps(fscal,dx10);
 888             ty               = _mm_mul_ps(fscal,dy10);
 889             tz               = _mm_mul_ps(fscal,dz10);
 890
 891             /* Update vectorial force */
 892             fix1             = _mm_add_ps(fix1,tx);
 893             fiy1             = _mm_add_ps(fiy1,ty);
 894             fiz1             = _mm_add_ps(fiz1,tz);
 895
 896             fjx0             = _mm_add_ps(fjx0,tx);
 897             fjy0             = _mm_add_ps(fjy0,ty);
 898             fjz0             = _mm_add_ps(fjz0,tz);
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq20             = _mm_mul_ps(iq2,jq0);
 906
 907             /* COULOMB ELECTROSTATICS */
 908             velec            = _mm_mul_ps(qq20,rinv20);
 909             felec            = _mm_mul_ps(velec,rinvsq20);
 910
 911             fscal            = felec;
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_ps(fscal,dx20);
 915             ty               = _mm_mul_ps(fscal,dy20);
 916             tz               = _mm_mul_ps(fscal,dz20);
 917
 918             /* Update vectorial force */
 919             fix2             = _mm_add_ps(fix2,tx);
 920             fiy2             = _mm_add_ps(fiy2,ty);
 921             fiz2             = _mm_add_ps(fiz2,tz);
 922
 923             fjx0             = _mm_add_ps(fjx0,tx);
 924             fjy0             = _mm_add_ps(fjy0,ty);
 925             fjz0             = _mm_add_ps(fjz0,tz);
 926
 927             fjptrA             = f+j_coord_offsetA;
 928             fjptrB             = f+j_coord_offsetB;
 929             fjptrC             = f+j_coord_offsetC;
 930             fjptrD             = f+j_coord_offsetD;
 931
 932             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 933
 934             /* Inner loop uses 108 flops */
 935         }
 936
 937         if(jidx<j_index_end)
 938         {
 939
 940             /* Get j neighbor index, and coordinate index */
 941             jnrlistA         = jjnr[jidx];
 942             jnrlistB         = jjnr[jidx+1];
 943             jnrlistC         = jjnr[jidx+2];
 944             jnrlistD         = jjnr[jidx+3];
 945             /* Sign of each element will be negative for non-real atoms.
 946              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 947              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 948              */
 949             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 950             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 951             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 952             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 953             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 954             j_coord_offsetA  = DIM*jnrA;
 955             j_coord_offsetB  = DIM*jnrB;
 956             j_coord_offsetC  = DIM*jnrC;
 957             j_coord_offsetD  = DIM*jnrD;
 958
 959             /* load j atom coordinates */
 960             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 961                                               x+j_coord_offsetC,x+j_coord_offsetD,
 962                                               &jx0,&jy0,&jz0);
 963
 964             /* Calculate displacement vector */
 965             dx00             = _mm_sub_ps(ix0,jx0);
 966             dy00             = _mm_sub_ps(iy0,jy0);
 967             dz00             = _mm_sub_ps(iz0,jz0);
 968             dx10             = _mm_sub_ps(ix1,jx0);
 969             dy10             = _mm_sub_ps(iy1,jy0);
 970             dz10             = _mm_sub_ps(iz1,jz0);
 971             dx20             = _mm_sub_ps(ix2,jx0);
 972             dy20             = _mm_sub_ps(iy2,jy0);
 973             dz20             = _mm_sub_ps(iz2,jz0);
 974
 975             /* Calculate squared distance and things based on it */
 976             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 977             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 978             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 979
 980             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 981             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 982             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 983
 984             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 985             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 986             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 987
 988             /* Load parameters for j particles */
 989             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 990                                                               charge+jnrC+0,charge+jnrD+0);
 991             vdwjidx0A        = 2*vdwtype[jnrA+0];
 992             vdwjidx0B        = 2*vdwtype[jnrB+0];
 993             vdwjidx0C        = 2*vdwtype[jnrC+0];
 994             vdwjidx0D        = 2*vdwtype[jnrD+0];
 995
 996             fjx0             = _mm_setzero_ps();
 997             fjy0             = _mm_setzero_ps();
 998             fjz0             = _mm_setzero_ps();
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r00              = _mm_mul_ps(rsq00,rinv00);
1005             r00              = _mm_andnot_ps(dummy_mask,r00);
1006
1007             /* Compute parameters for interactions between i and j atoms */
1008             qq00             = _mm_mul_ps(iq0,jq0);
1009             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1010                                          vdwparam+vdwioffset0+vdwjidx0B,
1011                                          vdwparam+vdwioffset0+vdwjidx0C,
1012                                          vdwparam+vdwioffset0+vdwjidx0D,
1013                                          &c6_00,&c12_00);
1014
1015             /* Calculate table index by multiplying r with table scale and truncate to integer */
1016             rt               = _mm_mul_ps(r00,vftabscale);
1017             vfitab           = _mm_cvttps_epi32(rt);
1018             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1019             vfitab           = _mm_slli_epi32(vfitab,3);
1020
1021             /* COULOMB ELECTROSTATICS */
1022             velec            = _mm_mul_ps(qq00,rinv00);
1023             felec            = _mm_mul_ps(velec,rinvsq00);
1024
1025             /* CUBIC SPLINE TABLE DISPERSION */
1026             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1027             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1028             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1029             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1030             _MM_TRANSPOSE4_PS(Y,F,G,H);
1031             Heps             = _mm_mul_ps(vfeps,H);
1032             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1033             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1034             fvdw6            = _mm_mul_ps(c6_00,FF);
1035
1036             /* CUBIC SPLINE TABLE REPULSION */
1037             vfitab           = _mm_add_epi32(vfitab,ifour);
1038             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1039             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1040             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1041             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1042             _MM_TRANSPOSE4_PS(Y,F,G,H);
1043             Heps             = _mm_mul_ps(vfeps,H);
1044             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1045             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1046             fvdw12           = _mm_mul_ps(c12_00,FF);
1047             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1048
1049             fscal            = _mm_add_ps(felec,fvdw);
1050
1051             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm_mul_ps(fscal,dx00);
1055             ty               = _mm_mul_ps(fscal,dy00);
1056             tz               = _mm_mul_ps(fscal,dz00);
1057
1058             /* Update vectorial force */
1059             fix0             = _mm_add_ps(fix0,tx);
1060             fiy0             = _mm_add_ps(fiy0,ty);
1061             fiz0             = _mm_add_ps(fiz0,tz);
1062
1063             fjx0             = _mm_add_ps(fjx0,tx);
1064             fjy0             = _mm_add_ps(fjy0,ty);
1065             fjz0             = _mm_add_ps(fjz0,tz);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq10             = _mm_mul_ps(iq1,jq0);
1073
1074             /* COULOMB ELECTROSTATICS */
1075             velec            = _mm_mul_ps(qq10,rinv10);
1076             felec            = _mm_mul_ps(velec,rinvsq10);
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm_mul_ps(fscal,dx10);
1084             ty               = _mm_mul_ps(fscal,dy10);
1085             tz               = _mm_mul_ps(fscal,dz10);
1086
1087             /* Update vectorial force */
1088             fix1             = _mm_add_ps(fix1,tx);
1089             fiy1             = _mm_add_ps(fiy1,ty);
1090             fiz1             = _mm_add_ps(fiz1,tz);
1091
1092             fjx0             = _mm_add_ps(fjx0,tx);
1093             fjy0             = _mm_add_ps(fjy0,ty);
1094             fjz0             = _mm_add_ps(fjz0,tz);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq20             = _mm_mul_ps(iq2,jq0);
1102
1103             /* COULOMB ELECTROSTATICS */
1104             velec            = _mm_mul_ps(qq20,rinv20);
1105             felec            = _mm_mul_ps(velec,rinvsq20);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1110
1111             /* Calculate temporary vectorial force */
1112             tx               = _mm_mul_ps(fscal,dx20);
1113             ty               = _mm_mul_ps(fscal,dy20);
1114             tz               = _mm_mul_ps(fscal,dz20);
1115
1116             /* Update vectorial force */
1117             fix2             = _mm_add_ps(fix2,tx);
1118             fiy2             = _mm_add_ps(fiy2,ty);
1119             fiz2             = _mm_add_ps(fiz2,tz);
1120
1121             fjx0             = _mm_add_ps(fjx0,tx);
1122             fjy0             = _mm_add_ps(fjy0,ty);
1123             fjz0             = _mm_add_ps(fjz0,tz);
1124
1125             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1126             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1127             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1128             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1129
1130             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1131
1132             /* Inner loop uses 109 flops */
1133         }
1134
1135         /* End of innermost loop */
1136
1137         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1138                                               f+i_coord_offset,fshift+i_shift_offset);
1139
1140         /* Increment number of inner iterations */
1141         inneriter                  += j_index_end - j_index_start;
1142
1143         /* Outer loop uses 18 flops */
1144     }
1145
1146     /* Increment number of outer iterations */
1147     outeriter        += nri;
1148
1149     /* Update outer/inner flops */
1150
1151     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1152 }