src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_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_elec->data;
 127     vftabscale       = _mm_set1_ps(kernel_data->table_elec->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           = sse2_invsqrt_f(rsq00);
 219             rinv10           = sse2_invsqrt_f(rsq10);
 220             rinv20           = sse2_invsqrt_f(rsq20);
 221
 222             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 226                                                               charge+jnrC+0,charge+jnrD+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229             vdwjidx0C        = 2*vdwtype[jnrC+0];
 230             vdwjidx0D        = 2*vdwtype[jnrD+0];
 231
 232             fjx0             = _mm_setzero_ps();
 233             fjy0             = _mm_setzero_ps();
 234             fjz0             = _mm_setzero_ps();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             r00              = _mm_mul_ps(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             qq00             = _mm_mul_ps(iq0,jq0);
 244             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,
 246                                          vdwparam+vdwioffset0+vdwjidx0C,
 247                                          vdwparam+vdwioffset0+vdwjidx0D,
 248                                          &c6_00,&c12_00);
 249
 250             /* Calculate table index by multiplying r with table scale and truncate to integer */
 251             rt               = _mm_mul_ps(r00,vftabscale);
 252             vfitab           = _mm_cvttps_epi32(rt);
 253             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 254             vfitab           = _mm_slli_epi32(vfitab,2);
 255
 256             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 257             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 258             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 259             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 260             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 261             _MM_TRANSPOSE4_PS(Y,F,G,H);
 262             Heps             = _mm_mul_ps(vfeps,H);
 263             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 264             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 265             velec            = _mm_mul_ps(qq00,VV);
 266             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 267             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 268
 269             /* LENNARD-JONES DISPERSION/REPULSION */
 270
 271             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 272             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 273             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 274             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 275             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velecsum         = _mm_add_ps(velecsum,velec);
 279             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_ps(felec,fvdw);
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm_mul_ps(fscal,dx00);
 285             ty               = _mm_mul_ps(fscal,dy00);
 286             tz               = _mm_mul_ps(fscal,dz00);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_add_ps(fix0,tx);
 290             fiy0             = _mm_add_ps(fiy0,ty);
 291             fiz0             = _mm_add_ps(fiz0,tz);
 292
 293             fjx0             = _mm_add_ps(fjx0,tx);
 294             fjy0             = _mm_add_ps(fjy0,ty);
 295             fjz0             = _mm_add_ps(fjz0,tz);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             r10              = _mm_mul_ps(rsq10,rinv10);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_ps(iq1,jq0);
 305
 306             /* Calculate table index by multiplying r with table scale and truncate to integer */
 307             rt               = _mm_mul_ps(r10,vftabscale);
 308             vfitab           = _mm_cvttps_epi32(rt);
 309             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 310             vfitab           = _mm_slli_epi32(vfitab,2);
 311
 312             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 313             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 314             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 315             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 316             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 317             _MM_TRANSPOSE4_PS(Y,F,G,H);
 318             Heps             = _mm_mul_ps(vfeps,H);
 319             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 320             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 321             velec            = _mm_mul_ps(qq10,VV);
 322             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 323             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 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             r20              = _mm_mul_ps(rsq20,rinv20);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_ps(iq2,jq0);
 352
 353             /* Calculate table index by multiplying r with table scale and truncate to integer */
 354             rt               = _mm_mul_ps(r20,vftabscale);
 355             vfitab           = _mm_cvttps_epi32(rt);
 356             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 357             vfitab           = _mm_slli_epi32(vfitab,2);
 358
 359             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 360             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 361             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 362             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 363             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 364             _MM_TRANSPOSE4_PS(Y,F,G,H);
 365             Heps             = _mm_mul_ps(vfeps,H);
 366             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 367             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 368             velec            = _mm_mul_ps(qq20,VV);
 369             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 370             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm_mul_ps(fscal,dx20);
 379             ty               = _mm_mul_ps(fscal,dy20);
 380             tz               = _mm_mul_ps(fscal,dz20);
 381
 382             /* Update vectorial force */
 383             fix2             = _mm_add_ps(fix2,tx);
 384             fiy2             = _mm_add_ps(fiy2,ty);
 385             fiz2             = _mm_add_ps(fiz2,tz);
 386
 387             fjx0             = _mm_add_ps(fjx0,tx);
 388             fjy0             = _mm_add_ps(fjy0,ty);
 389             fjz0             = _mm_add_ps(fjz0,tz);
 390
 391             fjptrA             = f+j_coord_offsetA;
 392             fjptrB             = f+j_coord_offsetB;
 393             fjptrC             = f+j_coord_offsetC;
 394             fjptrD             = f+j_coord_offsetD;
 395
 396             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 397
 398             /* Inner loop uses 142 flops */
 399         }
 400
 401         if(jidx<j_index_end)
 402         {
 403
 404             /* Get j neighbor index, and coordinate index */
 405             jnrlistA         = jjnr[jidx];
 406             jnrlistB         = jjnr[jidx+1];
 407             jnrlistC         = jjnr[jidx+2];
 408             jnrlistD         = jjnr[jidx+3];
 409             /* Sign of each element will be negative for non-real atoms.
 410              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 411              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 412              */
 413             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 414             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 415             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 416             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 417             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 418             j_coord_offsetA  = DIM*jnrA;
 419             j_coord_offsetB  = DIM*jnrB;
 420             j_coord_offsetC  = DIM*jnrC;
 421             j_coord_offsetD  = DIM*jnrD;
 422
 423             /* load j atom coordinates */
 424             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 425                                               x+j_coord_offsetC,x+j_coord_offsetD,
 426                                               &jx0,&jy0,&jz0);
 427
 428             /* Calculate displacement vector */
 429             dx00             = _mm_sub_ps(ix0,jx0);
 430             dy00             = _mm_sub_ps(iy0,jy0);
 431             dz00             = _mm_sub_ps(iz0,jz0);
 432             dx10             = _mm_sub_ps(ix1,jx0);
 433             dy10             = _mm_sub_ps(iy1,jy0);
 434             dz10             = _mm_sub_ps(iz1,jz0);
 435             dx20             = _mm_sub_ps(ix2,jx0);
 436             dy20             = _mm_sub_ps(iy2,jy0);
 437             dz20             = _mm_sub_ps(iz2,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 441             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 442             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 443
 444             rinv00           = sse2_invsqrt_f(rsq00);
 445             rinv10           = sse2_invsqrt_f(rsq10);
 446             rinv20           = sse2_invsqrt_f(rsq20);
 447
 448             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 449
 450             /* Load parameters for j particles */
 451             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 452                                                               charge+jnrC+0,charge+jnrD+0);
 453             vdwjidx0A        = 2*vdwtype[jnrA+0];
 454             vdwjidx0B        = 2*vdwtype[jnrB+0];
 455             vdwjidx0C        = 2*vdwtype[jnrC+0];
 456             vdwjidx0D        = 2*vdwtype[jnrD+0];
 457
 458             fjx0             = _mm_setzero_ps();
 459             fjy0             = _mm_setzero_ps();
 460             fjz0             = _mm_setzero_ps();
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             r00              = _mm_mul_ps(rsq00,rinv00);
 467             r00              = _mm_andnot_ps(dummy_mask,r00);
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq00             = _mm_mul_ps(iq0,jq0);
 471             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 472                                          vdwparam+vdwioffset0+vdwjidx0B,
 473                                          vdwparam+vdwioffset0+vdwjidx0C,
 474                                          vdwparam+vdwioffset0+vdwjidx0D,
 475                                          &c6_00,&c12_00);
 476
 477             /* Calculate table index by multiplying r with table scale and truncate to integer */
 478             rt               = _mm_mul_ps(r00,vftabscale);
 479             vfitab           = _mm_cvttps_epi32(rt);
 480             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 481             vfitab           = _mm_slli_epi32(vfitab,2);
 482
 483             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 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             velec            = _mm_mul_ps(qq00,VV);
 493             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 494             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 495
 496             /* LENNARD-JONES DISPERSION/REPULSION */
 497
 498             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 499             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 500             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 501             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 502             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm_andnot_ps(dummy_mask,velec);
 506             velecsum         = _mm_add_ps(velecsum,velec);
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = _mm_add_ps(felec,fvdw);
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm_mul_ps(fscal,dx00);
 516             ty               = _mm_mul_ps(fscal,dy00);
 517             tz               = _mm_mul_ps(fscal,dz00);
 518
 519             /* Update vectorial force */
 520             fix0             = _mm_add_ps(fix0,tx);
 521             fiy0             = _mm_add_ps(fiy0,ty);
 522             fiz0             = _mm_add_ps(fiz0,tz);
 523
 524             fjx0             = _mm_add_ps(fjx0,tx);
 525             fjy0             = _mm_add_ps(fjy0,ty);
 526             fjz0             = _mm_add_ps(fjz0,tz);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r10              = _mm_mul_ps(rsq10,rinv10);
 533             r10              = _mm_andnot_ps(dummy_mask,r10);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq10             = _mm_mul_ps(iq1,jq0);
 537
 538             /* Calculate table index by multiplying r with table scale and truncate to integer */
 539             rt               = _mm_mul_ps(r10,vftabscale);
 540             vfitab           = _mm_cvttps_epi32(rt);
 541             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 542             vfitab           = _mm_slli_epi32(vfitab,2);
 543
 544             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 545             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 546             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 547             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 548             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 549             _MM_TRANSPOSE4_PS(Y,F,G,H);
 550             Heps             = _mm_mul_ps(vfeps,H);
 551             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 552             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 553             velec            = _mm_mul_ps(qq10,VV);
 554             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 555             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm_andnot_ps(dummy_mask,velec);
 559             velecsum         = _mm_add_ps(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_ps(fscal,dx10);
 567             ty               = _mm_mul_ps(fscal,dy10);
 568             tz               = _mm_mul_ps(fscal,dz10);
 569
 570             /* Update vectorial force */
 571             fix1             = _mm_add_ps(fix1,tx);
 572             fiy1             = _mm_add_ps(fiy1,ty);
 573             fiz1             = _mm_add_ps(fiz1,tz);
 574
 575             fjx0             = _mm_add_ps(fjx0,tx);
 576             fjy0             = _mm_add_ps(fjy0,ty);
 577             fjz0             = _mm_add_ps(fjz0,tz);
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             r20              = _mm_mul_ps(rsq20,rinv20);
 584             r20              = _mm_andnot_ps(dummy_mask,r20);
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq20             = _mm_mul_ps(iq2,jq0);
 588
 589             /* Calculate table index by multiplying r with table scale and truncate to integer */
 590             rt               = _mm_mul_ps(r20,vftabscale);
 591             vfitab           = _mm_cvttps_epi32(rt);
 592             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 593             vfitab           = _mm_slli_epi32(vfitab,2);
 594
 595             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 596             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 597             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 598             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 599             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 600             _MM_TRANSPOSE4_PS(Y,F,G,H);
 601             Heps             = _mm_mul_ps(vfeps,H);
 602             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 603             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 604             velec            = _mm_mul_ps(qq20,VV);
 605             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 606             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velec            = _mm_andnot_ps(dummy_mask,velec);
 610             velecsum         = _mm_add_ps(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 615
 616             /* Calculate temporary vectorial force */
 617             tx               = _mm_mul_ps(fscal,dx20);
 618             ty               = _mm_mul_ps(fscal,dy20);
 619             tz               = _mm_mul_ps(fscal,dz20);
 620
 621             /* Update vectorial force */
 622             fix2             = _mm_add_ps(fix2,tx);
 623             fiy2             = _mm_add_ps(fiy2,ty);
 624             fiz2             = _mm_add_ps(fiz2,tz);
 625
 626             fjx0             = _mm_add_ps(fjx0,tx);
 627             fjy0             = _mm_add_ps(fjy0,ty);
 628             fjz0             = _mm_add_ps(fjz0,tz);
 629
 630             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 631             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 632             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 633             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 634
 635             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 636
 637             /* Inner loop uses 145 flops */
 638         }
 639
 640         /* End of innermost loop */
 641
 642         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 643                                               f+i_coord_offset,fshift+i_shift_offset);
 644
 645         ggid                        = gid[iidx];
 646         /* Update potential energies */
 647         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 648         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 649
 650         /* Increment number of inner iterations */
 651         inneriter                  += j_index_end - j_index_start;
 652
 653         /* Outer loop uses 20 flops */
 654     }
 655
 656     /* Increment number of outer iterations */
 657     outeriter        += nri;
 658
 659     /* Update outer/inner flops */
 660
 661     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 662 }
 663 /*
 664  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
 665  * Electrostatics interaction: CubicSplineTable
 666  * VdW interaction:            LennardJones
 667  * Geometry:                   Water3-Particle
 668  * Calculate force/pot:        Force
 669  */
 670 void
 671 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
 672                     (t_nblist                    * gmx_restrict       nlist,
 673                      rvec                        * gmx_restrict          xx,
 674                      rvec                        * gmx_restrict          ff,
 675                      struct t_forcerec           * gmx_restrict          fr,
 676                      t_mdatoms                   * gmx_restrict     mdatoms,
 677                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 678                      t_nrnb                      * gmx_restrict        nrnb)
 679 {
 680     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 681      * just 0 for non-waters.
 682      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 683      * jnr indices corresponding to data put in the four positions in the SIMD register.
 684      */
 685     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 686     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 687     int              jnrA,jnrB,jnrC,jnrD;
 688     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 689     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 690     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 691     real             rcutoff_scalar;
 692     real             *shiftvec,*fshift,*x,*f;
 693     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 694     real             scratch[4*DIM];
 695     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 696     int              vdwioffset0;
 697     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 698     int              vdwioffset1;
 699     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 700     int              vdwioffset2;
 701     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 702     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 703     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 704     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 705     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 706     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 707     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 708     real             *charge;
 709     int              nvdwtype;
 710     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 711     int              *vdwtype;
 712     real             *vdwparam;
 713     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 714     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 715     __m128i          vfitab;
 716     __m128i          ifour       = _mm_set1_epi32(4);
 717     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 718     real             *vftab;
 719     __m128           dummy_mask,cutoff_mask;
 720     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 721     __m128           one     = _mm_set1_ps(1.0);
 722     __m128           two     = _mm_set1_ps(2.0);
 723     x                = xx[0];
 724     f                = ff[0];
 725
 726     nri              = nlist->nri;
 727     iinr             = nlist->iinr;
 728     jindex           = nlist->jindex;
 729     jjnr             = nlist->jjnr;
 730     shiftidx         = nlist->shift;
 731     gid              = nlist->gid;
 732     shiftvec         = fr->shift_vec[0];
 733     fshift           = fr->fshift[0];
 734     facel            = _mm_set1_ps(fr->ic->epsfac);
 735     charge           = mdatoms->chargeA;
 736     nvdwtype         = fr->ntype;
 737     vdwparam         = fr->nbfp;
 738     vdwtype          = mdatoms->typeA;
 739
 740     vftab            = kernel_data->table_elec->data;
 741     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 742
 743     /* Setup water-specific parameters */
 744     inr              = nlist->iinr[0];
 745     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 746     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 747     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 748     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 749
 750     /* Avoid stupid compiler warnings */
 751     jnrA = jnrB = jnrC = jnrD = 0;
 752     j_coord_offsetA = 0;
 753     j_coord_offsetB = 0;
 754     j_coord_offsetC = 0;
 755     j_coord_offsetD = 0;
 756
 757     outeriter        = 0;
 758     inneriter        = 0;
 759
 760     for(iidx=0;iidx<4*DIM;iidx++)
 761     {
 762         scratch[iidx] = 0.0;
 763     }
 764
 765     /* Start outer loop over neighborlists */
 766     for(iidx=0; iidx<nri; iidx++)
 767     {
 768         /* Load shift vector for this list */
 769         i_shift_offset   = DIM*shiftidx[iidx];
 770
 771         /* Load limits for loop over neighbors */
 772         j_index_start    = jindex[iidx];
 773         j_index_end      = jindex[iidx+1];
 774
 775         /* Get outer coordinate index */
 776         inr              = iinr[iidx];
 777         i_coord_offset   = DIM*inr;
 778
 779         /* Load i particle coords and add shift vector */
 780         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 781                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 782
 783         fix0             = _mm_setzero_ps();
 784         fiy0             = _mm_setzero_ps();
 785         fiz0             = _mm_setzero_ps();
 786         fix1             = _mm_setzero_ps();
 787         fiy1             = _mm_setzero_ps();
 788         fiz1             = _mm_setzero_ps();
 789         fix2             = _mm_setzero_ps();
 790         fiy2             = _mm_setzero_ps();
 791         fiz2             = _mm_setzero_ps();
 792
 793         /* Start inner kernel loop */
 794         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 795         {
 796
 797             /* Get j neighbor index, and coordinate index */
 798             jnrA             = jjnr[jidx];
 799             jnrB             = jjnr[jidx+1];
 800             jnrC             = jjnr[jidx+2];
 801             jnrD             = jjnr[jidx+3];
 802             j_coord_offsetA  = DIM*jnrA;
 803             j_coord_offsetB  = DIM*jnrB;
 804             j_coord_offsetC  = DIM*jnrC;
 805             j_coord_offsetD  = DIM*jnrD;
 806
 807             /* load j atom coordinates */
 808             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 809                                               x+j_coord_offsetC,x+j_coord_offsetD,
 810                                               &jx0,&jy0,&jz0);
 811
 812             /* Calculate displacement vector */
 813             dx00             = _mm_sub_ps(ix0,jx0);
 814             dy00             = _mm_sub_ps(iy0,jy0);
 815             dz00             = _mm_sub_ps(iz0,jz0);
 816             dx10             = _mm_sub_ps(ix1,jx0);
 817             dy10             = _mm_sub_ps(iy1,jy0);
 818             dz10             = _mm_sub_ps(iz1,jz0);
 819             dx20             = _mm_sub_ps(ix2,jx0);
 820             dy20             = _mm_sub_ps(iy2,jy0);
 821             dz20             = _mm_sub_ps(iz2,jz0);
 822
 823             /* Calculate squared distance and things based on it */
 824             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 825             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 826             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 827
 828             rinv00           = sse2_invsqrt_f(rsq00);
 829             rinv10           = sse2_invsqrt_f(rsq10);
 830             rinv20           = sse2_invsqrt_f(rsq20);
 831
 832             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 833
 834             /* Load parameters for j particles */
 835             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 836                                                               charge+jnrC+0,charge+jnrD+0);
 837             vdwjidx0A        = 2*vdwtype[jnrA+0];
 838             vdwjidx0B        = 2*vdwtype[jnrB+0];
 839             vdwjidx0C        = 2*vdwtype[jnrC+0];
 840             vdwjidx0D        = 2*vdwtype[jnrD+0];
 841
 842             fjx0             = _mm_setzero_ps();
 843             fjy0             = _mm_setzero_ps();
 844             fjz0             = _mm_setzero_ps();
 845
 846             /**************************
 847              * CALCULATE INTERACTIONS *
 848              **************************/
 849
 850             r00              = _mm_mul_ps(rsq00,rinv00);
 851
 852             /* Compute parameters for interactions between i and j atoms */
 853             qq00             = _mm_mul_ps(iq0,jq0);
 854             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 855                                          vdwparam+vdwioffset0+vdwjidx0B,
 856                                          vdwparam+vdwioffset0+vdwjidx0C,
 857                                          vdwparam+vdwioffset0+vdwjidx0D,
 858                                          &c6_00,&c12_00);
 859
 860             /* Calculate table index by multiplying r with table scale and truncate to integer */
 861             rt               = _mm_mul_ps(r00,vftabscale);
 862             vfitab           = _mm_cvttps_epi32(rt);
 863             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 864             vfitab           = _mm_slli_epi32(vfitab,2);
 865
 866             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 867             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 868             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 869             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 870             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 871             _MM_TRANSPOSE4_PS(Y,F,G,H);
 872             Heps             = _mm_mul_ps(vfeps,H);
 873             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 874             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 875             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 876
 877             /* LENNARD-JONES DISPERSION/REPULSION */
 878
 879             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 880             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 881
 882             fscal            = _mm_add_ps(felec,fvdw);
 883
 884             /* Calculate temporary vectorial force */
 885             tx               = _mm_mul_ps(fscal,dx00);
 886             ty               = _mm_mul_ps(fscal,dy00);
 887             tz               = _mm_mul_ps(fscal,dz00);
 888
 889             /* Update vectorial force */
 890             fix0             = _mm_add_ps(fix0,tx);
 891             fiy0             = _mm_add_ps(fiy0,ty);
 892             fiz0             = _mm_add_ps(fiz0,tz);
 893
 894             fjx0             = _mm_add_ps(fjx0,tx);
 895             fjy0             = _mm_add_ps(fjy0,ty);
 896             fjz0             = _mm_add_ps(fjz0,tz);
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             r10              = _mm_mul_ps(rsq10,rinv10);
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq10             = _mm_mul_ps(iq1,jq0);
 906
 907             /* Calculate table index by multiplying r with table scale and truncate to integer */
 908             rt               = _mm_mul_ps(r10,vftabscale);
 909             vfitab           = _mm_cvttps_epi32(rt);
 910             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 911             vfitab           = _mm_slli_epi32(vfitab,2);
 912
 913             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 914             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 915             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 916             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 917             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 918             _MM_TRANSPOSE4_PS(Y,F,G,H);
 919             Heps             = _mm_mul_ps(vfeps,H);
 920             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 921             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 922             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 923
 924             fscal            = felec;
 925
 926             /* Calculate temporary vectorial force */
 927             tx               = _mm_mul_ps(fscal,dx10);
 928             ty               = _mm_mul_ps(fscal,dy10);
 929             tz               = _mm_mul_ps(fscal,dz10);
 930
 931             /* Update vectorial force */
 932             fix1             = _mm_add_ps(fix1,tx);
 933             fiy1             = _mm_add_ps(fiy1,ty);
 934             fiz1             = _mm_add_ps(fiz1,tz);
 935
 936             fjx0             = _mm_add_ps(fjx0,tx);
 937             fjy0             = _mm_add_ps(fjy0,ty);
 938             fjz0             = _mm_add_ps(fjz0,tz);
 939
 940             /**************************
 941              * CALCULATE INTERACTIONS *
 942              **************************/
 943
 944             r20              = _mm_mul_ps(rsq20,rinv20);
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             qq20             = _mm_mul_ps(iq2,jq0);
 948
 949             /* Calculate table index by multiplying r with table scale and truncate to integer */
 950             rt               = _mm_mul_ps(r20,vftabscale);
 951             vfitab           = _mm_cvttps_epi32(rt);
 952             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 953             vfitab           = _mm_slli_epi32(vfitab,2);
 954
 955             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 956             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 957             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 958             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 959             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 960             _MM_TRANSPOSE4_PS(Y,F,G,H);
 961             Heps             = _mm_mul_ps(vfeps,H);
 962             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 963             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 964             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 965
 966             fscal            = felec;
 967
 968             /* Calculate temporary vectorial force */
 969             tx               = _mm_mul_ps(fscal,dx20);
 970             ty               = _mm_mul_ps(fscal,dy20);
 971             tz               = _mm_mul_ps(fscal,dz20);
 972
 973             /* Update vectorial force */
 974             fix2             = _mm_add_ps(fix2,tx);
 975             fiy2             = _mm_add_ps(fiy2,ty);
 976             fiz2             = _mm_add_ps(fiz2,tz);
 977
 978             fjx0             = _mm_add_ps(fjx0,tx);
 979             fjy0             = _mm_add_ps(fjy0,ty);
 980             fjz0             = _mm_add_ps(fjz0,tz);
 981
 982             fjptrA             = f+j_coord_offsetA;
 983             fjptrB             = f+j_coord_offsetB;
 984             fjptrC             = f+j_coord_offsetC;
 985             fjptrD             = f+j_coord_offsetD;
 986
 987             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 988
 989             /* Inner loop uses 125 flops */
 990         }
 991
 992         if(jidx<j_index_end)
 993         {
 994
 995             /* Get j neighbor index, and coordinate index */
 996             jnrlistA         = jjnr[jidx];
 997             jnrlistB         = jjnr[jidx+1];
 998             jnrlistC         = jjnr[jidx+2];
 999             jnrlistD         = jjnr[jidx+3];
1000             /* Sign of each element will be negative for non-real atoms.
1001              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1002              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1003              */
1004             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1005             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1006             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1007             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1008             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1009             j_coord_offsetA  = DIM*jnrA;
1010             j_coord_offsetB  = DIM*jnrB;
1011             j_coord_offsetC  = DIM*jnrC;
1012             j_coord_offsetD  = DIM*jnrD;
1013
1014             /* load j atom coordinates */
1015             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1016                                               x+j_coord_offsetC,x+j_coord_offsetD,
1017                                               &jx0,&jy0,&jz0);
1018
1019             /* Calculate displacement vector */
1020             dx00             = _mm_sub_ps(ix0,jx0);
1021             dy00             = _mm_sub_ps(iy0,jy0);
1022             dz00             = _mm_sub_ps(iz0,jz0);
1023             dx10             = _mm_sub_ps(ix1,jx0);
1024             dy10             = _mm_sub_ps(iy1,jy0);
1025             dz10             = _mm_sub_ps(iz1,jz0);
1026             dx20             = _mm_sub_ps(ix2,jx0);
1027             dy20             = _mm_sub_ps(iy2,jy0);
1028             dz20             = _mm_sub_ps(iz2,jz0);
1029
1030             /* Calculate squared distance and things based on it */
1031             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1032             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1033             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1034
1035             rinv00           = sse2_invsqrt_f(rsq00);
1036             rinv10           = sse2_invsqrt_f(rsq10);
1037             rinv20           = sse2_invsqrt_f(rsq20);
1038
1039             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1040
1041             /* Load parameters for j particles */
1042             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1043                                                               charge+jnrC+0,charge+jnrD+0);
1044             vdwjidx0A        = 2*vdwtype[jnrA+0];
1045             vdwjidx0B        = 2*vdwtype[jnrB+0];
1046             vdwjidx0C        = 2*vdwtype[jnrC+0];
1047             vdwjidx0D        = 2*vdwtype[jnrD+0];
1048
1049             fjx0             = _mm_setzero_ps();
1050             fjy0             = _mm_setzero_ps();
1051             fjz0             = _mm_setzero_ps();
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             r00              = _mm_mul_ps(rsq00,rinv00);
1058             r00              = _mm_andnot_ps(dummy_mask,r00);
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             qq00             = _mm_mul_ps(iq0,jq0);
1062             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1063                                          vdwparam+vdwioffset0+vdwjidx0B,
1064                                          vdwparam+vdwioffset0+vdwjidx0C,
1065                                          vdwparam+vdwioffset0+vdwjidx0D,
1066                                          &c6_00,&c12_00);
1067
1068             /* Calculate table index by multiplying r with table scale and truncate to integer */
1069             rt               = _mm_mul_ps(r00,vftabscale);
1070             vfitab           = _mm_cvttps_epi32(rt);
1071             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1072             vfitab           = _mm_slli_epi32(vfitab,2);
1073
1074             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1075             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1076             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1077             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1078             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1079             _MM_TRANSPOSE4_PS(Y,F,G,H);
1080             Heps             = _mm_mul_ps(vfeps,H);
1081             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1082             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1083             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1084
1085             /* LENNARD-JONES DISPERSION/REPULSION */
1086
1087             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1088             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1089
1090             fscal            = _mm_add_ps(felec,fvdw);
1091
1092             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm_mul_ps(fscal,dx00);
1096             ty               = _mm_mul_ps(fscal,dy00);
1097             tz               = _mm_mul_ps(fscal,dz00);
1098
1099             /* Update vectorial force */
1100             fix0             = _mm_add_ps(fix0,tx);
1101             fiy0             = _mm_add_ps(fiy0,ty);
1102             fiz0             = _mm_add_ps(fiz0,tz);
1103
1104             fjx0             = _mm_add_ps(fjx0,tx);
1105             fjy0             = _mm_add_ps(fjy0,ty);
1106             fjz0             = _mm_add_ps(fjz0,tz);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             r10              = _mm_mul_ps(rsq10,rinv10);
1113             r10              = _mm_andnot_ps(dummy_mask,r10);
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq10             = _mm_mul_ps(iq1,jq0);
1117
1118             /* Calculate table index by multiplying r with table scale and truncate to integer */
1119             rt               = _mm_mul_ps(r10,vftabscale);
1120             vfitab           = _mm_cvttps_epi32(rt);
1121             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1122             vfitab           = _mm_slli_epi32(vfitab,2);
1123
1124             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1125             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1126             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1127             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1128             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1129             _MM_TRANSPOSE4_PS(Y,F,G,H);
1130             Heps             = _mm_mul_ps(vfeps,H);
1131             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1132             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1133             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm_mul_ps(fscal,dx10);
1141             ty               = _mm_mul_ps(fscal,dy10);
1142             tz               = _mm_mul_ps(fscal,dz10);
1143
1144             /* Update vectorial force */
1145             fix1             = _mm_add_ps(fix1,tx);
1146             fiy1             = _mm_add_ps(fiy1,ty);
1147             fiz1             = _mm_add_ps(fiz1,tz);
1148
1149             fjx0             = _mm_add_ps(fjx0,tx);
1150             fjy0             = _mm_add_ps(fjy0,ty);
1151             fjz0             = _mm_add_ps(fjz0,tz);
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             r20              = _mm_mul_ps(rsq20,rinv20);
1158             r20              = _mm_andnot_ps(dummy_mask,r20);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq20             = _mm_mul_ps(iq2,jq0);
1162
1163             /* Calculate table index by multiplying r with table scale and truncate to integer */
1164             rt               = _mm_mul_ps(r20,vftabscale);
1165             vfitab           = _mm_cvttps_epi32(rt);
1166             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1167             vfitab           = _mm_slli_epi32(vfitab,2);
1168
1169             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1170             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1171             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1172             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1173             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1174             _MM_TRANSPOSE4_PS(Y,F,G,H);
1175             Heps             = _mm_mul_ps(vfeps,H);
1176             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1177             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1178             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm_mul_ps(fscal,dx20);
1186             ty               = _mm_mul_ps(fscal,dy20);
1187             tz               = _mm_mul_ps(fscal,dz20);
1188
1189             /* Update vectorial force */
1190             fix2             = _mm_add_ps(fix2,tx);
1191             fiy2             = _mm_add_ps(fiy2,ty);
1192             fiz2             = _mm_add_ps(fiz2,tz);
1193
1194             fjx0             = _mm_add_ps(fjx0,tx);
1195             fjy0             = _mm_add_ps(fjy0,ty);
1196             fjz0             = _mm_add_ps(fjz0,tz);
1197
1198             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1199             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1200             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1201             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1202
1203             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1204
1205             /* Inner loop uses 128 flops */
1206         }
1207
1208         /* End of innermost loop */
1209
1210         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1211                                               f+i_coord_offset,fshift+i_shift_offset);
1212
1213         /* Increment number of inner iterations */
1214         inneriter                  += j_index_end - j_index_start;
1215
1216         /* Outer loop uses 18 flops */
1217     }
1218
1219     /* Increment number of outer iterations */
1220     outeriter        += nri;
1221
1222     /* Update outer/inner flops */
1223
1224     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1225 }