src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_sse4_1_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 135     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 136     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181         fix3             = _mm_setzero_ps();
 182         fiy3             = _mm_setzero_ps();
 183         fiz3             = _mm_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               x+j_coord_offsetC,x+j_coord_offsetD,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_ps(ix0,jx0);
 210             dy00             = _mm_sub_ps(iy0,jy0);
 211             dz00             = _mm_sub_ps(iz0,jz0);
 212             dx10             = _mm_sub_ps(ix1,jx0);
 213             dy10             = _mm_sub_ps(iy1,jy0);
 214             dz10             = _mm_sub_ps(iz1,jz0);
 215             dx20             = _mm_sub_ps(ix2,jx0);
 216             dy20             = _mm_sub_ps(iy2,jy0);
 217             dz20             = _mm_sub_ps(iz2,jz0);
 218             dx30             = _mm_sub_ps(ix3,jx0);
 219             dy30             = _mm_sub_ps(iy3,jy0);
 220             dz30             = _mm_sub_ps(iz3,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 227
 228             rinv00           = sse41_invsqrt_f(rsq00);
 229             rinv10           = sse41_invsqrt_f(rsq10);
 230             rinv20           = sse41_invsqrt_f(rsq20);
 231             rinv30           = sse41_invsqrt_f(rsq30);
 232
 233             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 234             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 235             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                               charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm_setzero_ps();
 246             fjy0             = _mm_setzero_ps();
 247             fjz0             = _mm_setzero_ps();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             r00              = _mm_mul_ps(rsq00,rinv00);
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 257                                          vdwparam+vdwioffset0+vdwjidx0B,
 258                                          vdwparam+vdwioffset0+vdwjidx0C,
 259                                          vdwparam+vdwioffset0+vdwjidx0D,
 260                                          &c6_00,&c12_00);
 261
 262             /* Calculate table index by multiplying r with table scale and truncate to integer */
 263             rt               = _mm_mul_ps(r00,vftabscale);
 264             vfitab           = _mm_cvttps_epi32(rt);
 265             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 266             vfitab           = _mm_slli_epi32(vfitab,3);
 267
 268             /* CUBIC SPLINE TABLE DISPERSION */
 269             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Heps             = _mm_mul_ps(vfeps,H);
 275             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 276             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 277             vvdw6            = _mm_mul_ps(c6_00,VV);
 278             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 279             fvdw6            = _mm_mul_ps(c6_00,FF);
 280
 281             /* CUBIC SPLINE TABLE REPULSION */
 282             vfitab           = _mm_add_epi32(vfitab,ifour);
 283             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 284             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 285             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 286             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 287             _MM_TRANSPOSE4_PS(Y,F,G,H);
 288             Heps             = _mm_mul_ps(vfeps,H);
 289             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 290             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 291             vvdw12           = _mm_mul_ps(c12_00,VV);
 292             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 293             fvdw12           = _mm_mul_ps(c12_00,FF);
 294             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 295             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 299
 300             fscal            = fvdw;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_ps(fscal,dx00);
 304             ty               = _mm_mul_ps(fscal,dy00);
 305             tz               = _mm_mul_ps(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm_add_ps(fix0,tx);
 309             fiy0             = _mm_add_ps(fiy0,ty);
 310             fiz0             = _mm_add_ps(fiz0,tz);
 311
 312             fjx0             = _mm_add_ps(fjx0,tx);
 313             fjy0             = _mm_add_ps(fjy0,ty);
 314             fjz0             = _mm_add_ps(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_ps(iq1,jq0);
 322
 323             /* COULOMB ELECTROSTATICS */
 324             velec            = _mm_mul_ps(qq10,rinv10);
 325             felec            = _mm_mul_ps(velec,rinvsq10);
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm_add_ps(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_ps(fscal,dx10);
 334             ty               = _mm_mul_ps(fscal,dy10);
 335             tz               = _mm_mul_ps(fscal,dz10);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm_add_ps(fix1,tx);
 339             fiy1             = _mm_add_ps(fiy1,ty);
 340             fiz1             = _mm_add_ps(fiz1,tz);
 341
 342             fjx0             = _mm_add_ps(fjx0,tx);
 343             fjy0             = _mm_add_ps(fjy0,ty);
 344             fjz0             = _mm_add_ps(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_ps(iq2,jq0);
 352
 353             /* COULOMB ELECTROSTATICS */
 354             velec            = _mm_mul_ps(qq20,rinv20);
 355             felec            = _mm_mul_ps(velec,rinvsq20);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velecsum         = _mm_add_ps(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm_mul_ps(fscal,dx20);
 364             ty               = _mm_mul_ps(fscal,dy20);
 365             tz               = _mm_mul_ps(fscal,dz20);
 366
 367             /* Update vectorial force */
 368             fix2             = _mm_add_ps(fix2,tx);
 369             fiy2             = _mm_add_ps(fiy2,ty);
 370             fiz2             = _mm_add_ps(fiz2,tz);
 371
 372             fjx0             = _mm_add_ps(fjx0,tx);
 373             fjy0             = _mm_add_ps(fjy0,ty);
 374             fjz0             = _mm_add_ps(fjz0,tz);
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             /* Compute parameters for interactions between i and j atoms */
 381             qq30             = _mm_mul_ps(iq3,jq0);
 382
 383             /* COULOMB ELECTROSTATICS */
 384             velec            = _mm_mul_ps(qq30,rinv30);
 385             felec            = _mm_mul_ps(velec,rinvsq30);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm_mul_ps(fscal,dx30);
 394             ty               = _mm_mul_ps(fscal,dy30);
 395             tz               = _mm_mul_ps(fscal,dz30);
 396
 397             /* Update vectorial force */
 398             fix3             = _mm_add_ps(fix3,tx);
 399             fiy3             = _mm_add_ps(fiy3,ty);
 400             fiz3             = _mm_add_ps(fiz3,tz);
 401
 402             fjx0             = _mm_add_ps(fjx0,tx);
 403             fjy0             = _mm_add_ps(fjy0,ty);
 404             fjz0             = _mm_add_ps(fjz0,tz);
 405
 406             fjptrA             = f+j_coord_offsetA;
 407             fjptrB             = f+j_coord_offsetB;
 408             fjptrC             = f+j_coord_offsetC;
 409             fjptrD             = f+j_coord_offsetD;
 410
 411             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 412
 413             /* Inner loop uses 140 flops */
 414         }
 415
 416         if(jidx<j_index_end)
 417         {
 418
 419             /* Get j neighbor index, and coordinate index */
 420             jnrlistA         = jjnr[jidx];
 421             jnrlistB         = jjnr[jidx+1];
 422             jnrlistC         = jjnr[jidx+2];
 423             jnrlistD         = jjnr[jidx+3];
 424             /* Sign of each element will be negative for non-real atoms.
 425              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 426              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 427              */
 428             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 429             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 430             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 431             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 432             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 433             j_coord_offsetA  = DIM*jnrA;
 434             j_coord_offsetB  = DIM*jnrB;
 435             j_coord_offsetC  = DIM*jnrC;
 436             j_coord_offsetD  = DIM*jnrD;
 437
 438             /* load j atom coordinates */
 439             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 440                                               x+j_coord_offsetC,x+j_coord_offsetD,
 441                                               &jx0,&jy0,&jz0);
 442
 443             /* Calculate displacement vector */
 444             dx00             = _mm_sub_ps(ix0,jx0);
 445             dy00             = _mm_sub_ps(iy0,jy0);
 446             dz00             = _mm_sub_ps(iz0,jz0);
 447             dx10             = _mm_sub_ps(ix1,jx0);
 448             dy10             = _mm_sub_ps(iy1,jy0);
 449             dz10             = _mm_sub_ps(iz1,jz0);
 450             dx20             = _mm_sub_ps(ix2,jx0);
 451             dy20             = _mm_sub_ps(iy2,jy0);
 452             dz20             = _mm_sub_ps(iz2,jz0);
 453             dx30             = _mm_sub_ps(ix3,jx0);
 454             dy30             = _mm_sub_ps(iy3,jy0);
 455             dz30             = _mm_sub_ps(iz3,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 461             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 462
 463             rinv00           = sse41_invsqrt_f(rsq00);
 464             rinv10           = sse41_invsqrt_f(rsq10);
 465             rinv20           = sse41_invsqrt_f(rsq20);
 466             rinv30           = sse41_invsqrt_f(rsq30);
 467
 468             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 470             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 471
 472             /* Load parameters for j particles */
 473             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 474                                                               charge+jnrC+0,charge+jnrD+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479
 480             fjx0             = _mm_setzero_ps();
 481             fjy0             = _mm_setzero_ps();
 482             fjz0             = _mm_setzero_ps();
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             r00              = _mm_mul_ps(rsq00,rinv00);
 489             r00              = _mm_andnot_ps(dummy_mask,r00);
 490
 491             /* Compute parameters for interactions between i and j atoms */
 492             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 493                                          vdwparam+vdwioffset0+vdwjidx0B,
 494                                          vdwparam+vdwioffset0+vdwjidx0C,
 495                                          vdwparam+vdwioffset0+vdwjidx0D,
 496                                          &c6_00,&c12_00);
 497
 498             /* Calculate table index by multiplying r with table scale and truncate to integer */
 499             rt               = _mm_mul_ps(r00,vftabscale);
 500             vfitab           = _mm_cvttps_epi32(rt);
 501             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 502             vfitab           = _mm_slli_epi32(vfitab,3);
 503
 504             /* CUBIC SPLINE TABLE DISPERSION */
 505             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 506             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 507             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 508             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 509             _MM_TRANSPOSE4_PS(Y,F,G,H);
 510             Heps             = _mm_mul_ps(vfeps,H);
 511             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 512             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 513             vvdw6            = _mm_mul_ps(c6_00,VV);
 514             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 515             fvdw6            = _mm_mul_ps(c6_00,FF);
 516
 517             /* CUBIC SPLINE TABLE REPULSION */
 518             vfitab           = _mm_add_epi32(vfitab,ifour);
 519             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 520             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 521             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 522             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 523             _MM_TRANSPOSE4_PS(Y,F,G,H);
 524             Heps             = _mm_mul_ps(vfeps,H);
 525             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 526             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 527             vvdw12           = _mm_mul_ps(c12_00,VV);
 528             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 529             fvdw12           = _mm_mul_ps(c12_00,FF);
 530             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 531             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 535             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 536
 537             fscal            = fvdw;
 538
 539             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_ps(fscal,dx00);
 543             ty               = _mm_mul_ps(fscal,dy00);
 544             tz               = _mm_mul_ps(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm_add_ps(fix0,tx);
 548             fiy0             = _mm_add_ps(fiy0,ty);
 549             fiz0             = _mm_add_ps(fiz0,tz);
 550
 551             fjx0             = _mm_add_ps(fjx0,tx);
 552             fjy0             = _mm_add_ps(fjy0,ty);
 553             fjz0             = _mm_add_ps(fjz0,tz);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq10             = _mm_mul_ps(iq1,jq0);
 561
 562             /* COULOMB ELECTROSTATICS */
 563             velec            = _mm_mul_ps(qq10,rinv10);
 564             felec            = _mm_mul_ps(velec,rinvsq10);
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velec            = _mm_andnot_ps(dummy_mask,velec);
 568             velecsum         = _mm_add_ps(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm_mul_ps(fscal,dx10);
 576             ty               = _mm_mul_ps(fscal,dy10);
 577             tz               = _mm_mul_ps(fscal,dz10);
 578
 579             /* Update vectorial force */
 580             fix1             = _mm_add_ps(fix1,tx);
 581             fiy1             = _mm_add_ps(fiy1,ty);
 582             fiz1             = _mm_add_ps(fiz1,tz);
 583
 584             fjx0             = _mm_add_ps(fjx0,tx);
 585             fjy0             = _mm_add_ps(fjy0,ty);
 586             fjz0             = _mm_add_ps(fjz0,tz);
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             /* Compute parameters for interactions between i and j atoms */
 593             qq20             = _mm_mul_ps(iq2,jq0);
 594
 595             /* COULOMB ELECTROSTATICS */
 596             velec            = _mm_mul_ps(qq20,rinv20);
 597             felec            = _mm_mul_ps(velec,rinvsq20);
 598
 599             /* Update potential sum for this i atom from the interaction with this j atom. */
 600             velec            = _mm_andnot_ps(dummy_mask,velec);
 601             velecsum         = _mm_add_ps(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm_mul_ps(fscal,dx20);
 609             ty               = _mm_mul_ps(fscal,dy20);
 610             tz               = _mm_mul_ps(fscal,dz20);
 611
 612             /* Update vectorial force */
 613             fix2             = _mm_add_ps(fix2,tx);
 614             fiy2             = _mm_add_ps(fiy2,ty);
 615             fiz2             = _mm_add_ps(fiz2,tz);
 616
 617             fjx0             = _mm_add_ps(fjx0,tx);
 618             fjy0             = _mm_add_ps(fjy0,ty);
 619             fjz0             = _mm_add_ps(fjz0,tz);
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq30             = _mm_mul_ps(iq3,jq0);
 627
 628             /* COULOMB ELECTROSTATICS */
 629             velec            = _mm_mul_ps(qq30,rinv30);
 630             felec            = _mm_mul_ps(velec,rinvsq30);
 631
 632             /* Update potential sum for this i atom from the interaction with this j atom. */
 633             velec            = _mm_andnot_ps(dummy_mask,velec);
 634             velecsum         = _mm_add_ps(velecsum,velec);
 635
 636             fscal            = felec;
 637
 638             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 639
 640             /* Calculate temporary vectorial force */
 641             tx               = _mm_mul_ps(fscal,dx30);
 642             ty               = _mm_mul_ps(fscal,dy30);
 643             tz               = _mm_mul_ps(fscal,dz30);
 644
 645             /* Update vectorial force */
 646             fix3             = _mm_add_ps(fix3,tx);
 647             fiy3             = _mm_add_ps(fiy3,ty);
 648             fiz3             = _mm_add_ps(fiz3,tz);
 649
 650             fjx0             = _mm_add_ps(fjx0,tx);
 651             fjy0             = _mm_add_ps(fjy0,ty);
 652             fjz0             = _mm_add_ps(fjz0,tz);
 653
 654             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 655             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 656             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 657             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 658
 659             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 660
 661             /* Inner loop uses 141 flops */
 662         }
 663
 664         /* End of innermost loop */
 665
 666         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 667                                               f+i_coord_offset,fshift+i_shift_offset);
 668
 669         ggid                        = gid[iidx];
 670         /* Update potential energies */
 671         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 672         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 673
 674         /* Increment number of inner iterations */
 675         inneriter                  += j_index_end - j_index_start;
 676
 677         /* Outer loop uses 26 flops */
 678     }
 679
 680     /* Increment number of outer iterations */
 681     outeriter        += nri;
 682
 683     /* Update outer/inner flops */
 684
 685     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 686 }
 687 /*
 688  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_single
 689  * Electrostatics interaction: Coulomb
 690  * VdW interaction:            CubicSplineTable
 691  * Geometry:                   Water4-Particle
 692  * Calculate force/pot:        Force
 693  */
 694 void
 695 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_single
 696                     (t_nblist                    * gmx_restrict       nlist,
 697                      rvec                        * gmx_restrict          xx,
 698                      rvec                        * gmx_restrict          ff,
 699                      struct t_forcerec           * gmx_restrict          fr,
 700                      t_mdatoms                   * gmx_restrict     mdatoms,
 701                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 702                      t_nrnb                      * gmx_restrict        nrnb)
 703 {
 704     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 705      * just 0 for non-waters.
 706      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 707      * jnr indices corresponding to data put in the four positions in the SIMD register.
 708      */
 709     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 710     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 711     int              jnrA,jnrB,jnrC,jnrD;
 712     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 713     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 714     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 715     real             rcutoff_scalar;
 716     real             *shiftvec,*fshift,*x,*f;
 717     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 718     real             scratch[4*DIM];
 719     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 720     int              vdwioffset0;
 721     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 722     int              vdwioffset1;
 723     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 724     int              vdwioffset2;
 725     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 726     int              vdwioffset3;
 727     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 728     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 729     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 730     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 731     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 732     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 733     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 734     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 735     real             *charge;
 736     int              nvdwtype;
 737     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 738     int              *vdwtype;
 739     real             *vdwparam;
 740     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 741     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 742     __m128i          vfitab;
 743     __m128i          ifour       = _mm_set1_epi32(4);
 744     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 745     real             *vftab;
 746     __m128           dummy_mask,cutoff_mask;
 747     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 748     __m128           one     = _mm_set1_ps(1.0);
 749     __m128           two     = _mm_set1_ps(2.0);
 750     x                = xx[0];
 751     f                = ff[0];
 752
 753     nri              = nlist->nri;
 754     iinr             = nlist->iinr;
 755     jindex           = nlist->jindex;
 756     jjnr             = nlist->jjnr;
 757     shiftidx         = nlist->shift;
 758     gid              = nlist->gid;
 759     shiftvec         = fr->shift_vec[0];
 760     fshift           = fr->fshift[0];
 761     facel            = _mm_set1_ps(fr->ic->epsfac);
 762     charge           = mdatoms->chargeA;
 763     nvdwtype         = fr->ntype;
 764     vdwparam         = fr->nbfp;
 765     vdwtype          = mdatoms->typeA;
 766
 767     vftab            = kernel_data->table_vdw->data;
 768     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 769
 770     /* Setup water-specific parameters */
 771     inr              = nlist->iinr[0];
 772     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 773     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 774     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 775     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 776
 777     /* Avoid stupid compiler warnings */
 778     jnrA = jnrB = jnrC = jnrD = 0;
 779     j_coord_offsetA = 0;
 780     j_coord_offsetB = 0;
 781     j_coord_offsetC = 0;
 782     j_coord_offsetD = 0;
 783
 784     outeriter        = 0;
 785     inneriter        = 0;
 786
 787     for(iidx=0;iidx<4*DIM;iidx++)
 788     {
 789         scratch[iidx] = 0.0;
 790     }
 791
 792     /* Start outer loop over neighborlists */
 793     for(iidx=0; iidx<nri; iidx++)
 794     {
 795         /* Load shift vector for this list */
 796         i_shift_offset   = DIM*shiftidx[iidx];
 797
 798         /* Load limits for loop over neighbors */
 799         j_index_start    = jindex[iidx];
 800         j_index_end      = jindex[iidx+1];
 801
 802         /* Get outer coordinate index */
 803         inr              = iinr[iidx];
 804         i_coord_offset   = DIM*inr;
 805
 806         /* Load i particle coords and add shift vector */
 807         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 808                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 809
 810         fix0             = _mm_setzero_ps();
 811         fiy0             = _mm_setzero_ps();
 812         fiz0             = _mm_setzero_ps();
 813         fix1             = _mm_setzero_ps();
 814         fiy1             = _mm_setzero_ps();
 815         fiz1             = _mm_setzero_ps();
 816         fix2             = _mm_setzero_ps();
 817         fiy2             = _mm_setzero_ps();
 818         fiz2             = _mm_setzero_ps();
 819         fix3             = _mm_setzero_ps();
 820         fiy3             = _mm_setzero_ps();
 821         fiz3             = _mm_setzero_ps();
 822
 823         /* Start inner kernel loop */
 824         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 825         {
 826
 827             /* Get j neighbor index, and coordinate index */
 828             jnrA             = jjnr[jidx];
 829             jnrB             = jjnr[jidx+1];
 830             jnrC             = jjnr[jidx+2];
 831             jnrD             = jjnr[jidx+3];
 832             j_coord_offsetA  = DIM*jnrA;
 833             j_coord_offsetB  = DIM*jnrB;
 834             j_coord_offsetC  = DIM*jnrC;
 835             j_coord_offsetD  = DIM*jnrD;
 836
 837             /* load j atom coordinates */
 838             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 839                                               x+j_coord_offsetC,x+j_coord_offsetD,
 840                                               &jx0,&jy0,&jz0);
 841
 842             /* Calculate displacement vector */
 843             dx00             = _mm_sub_ps(ix0,jx0);
 844             dy00             = _mm_sub_ps(iy0,jy0);
 845             dz00             = _mm_sub_ps(iz0,jz0);
 846             dx10             = _mm_sub_ps(ix1,jx0);
 847             dy10             = _mm_sub_ps(iy1,jy0);
 848             dz10             = _mm_sub_ps(iz1,jz0);
 849             dx20             = _mm_sub_ps(ix2,jx0);
 850             dy20             = _mm_sub_ps(iy2,jy0);
 851             dz20             = _mm_sub_ps(iz2,jz0);
 852             dx30             = _mm_sub_ps(ix3,jx0);
 853             dy30             = _mm_sub_ps(iy3,jy0);
 854             dz30             = _mm_sub_ps(iz3,jz0);
 855
 856             /* Calculate squared distance and things based on it */
 857             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 858             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 859             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 860             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 861
 862             rinv00           = sse41_invsqrt_f(rsq00);
 863             rinv10           = sse41_invsqrt_f(rsq10);
 864             rinv20           = sse41_invsqrt_f(rsq20);
 865             rinv30           = sse41_invsqrt_f(rsq30);
 866
 867             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 868             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 869             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 870
 871             /* Load parameters for j particles */
 872             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 873                                                               charge+jnrC+0,charge+jnrD+0);
 874             vdwjidx0A        = 2*vdwtype[jnrA+0];
 875             vdwjidx0B        = 2*vdwtype[jnrB+0];
 876             vdwjidx0C        = 2*vdwtype[jnrC+0];
 877             vdwjidx0D        = 2*vdwtype[jnrD+0];
 878
 879             fjx0             = _mm_setzero_ps();
 880             fjy0             = _mm_setzero_ps();
 881             fjz0             = _mm_setzero_ps();
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             r00              = _mm_mul_ps(rsq00,rinv00);
 888
 889             /* Compute parameters for interactions between i and j atoms */
 890             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 891                                          vdwparam+vdwioffset0+vdwjidx0B,
 892                                          vdwparam+vdwioffset0+vdwjidx0C,
 893                                          vdwparam+vdwioffset0+vdwjidx0D,
 894                                          &c6_00,&c12_00);
 895
 896             /* Calculate table index by multiplying r with table scale and truncate to integer */
 897             rt               = _mm_mul_ps(r00,vftabscale);
 898             vfitab           = _mm_cvttps_epi32(rt);
 899             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 900             vfitab           = _mm_slli_epi32(vfitab,3);
 901
 902             /* CUBIC SPLINE TABLE DISPERSION */
 903             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 904             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 905             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 906             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 907             _MM_TRANSPOSE4_PS(Y,F,G,H);
 908             Heps             = _mm_mul_ps(vfeps,H);
 909             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 910             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 911             fvdw6            = _mm_mul_ps(c6_00,FF);
 912
 913             /* CUBIC SPLINE TABLE REPULSION */
 914             vfitab           = _mm_add_epi32(vfitab,ifour);
 915             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 916             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 917             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 918             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 919             _MM_TRANSPOSE4_PS(Y,F,G,H);
 920             Heps             = _mm_mul_ps(vfeps,H);
 921             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 922             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 923             fvdw12           = _mm_mul_ps(c12_00,FF);
 924             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 925
 926             fscal            = fvdw;
 927
 928             /* Calculate temporary vectorial force */
 929             tx               = _mm_mul_ps(fscal,dx00);
 930             ty               = _mm_mul_ps(fscal,dy00);
 931             tz               = _mm_mul_ps(fscal,dz00);
 932
 933             /* Update vectorial force */
 934             fix0             = _mm_add_ps(fix0,tx);
 935             fiy0             = _mm_add_ps(fiy0,ty);
 936             fiz0             = _mm_add_ps(fiz0,tz);
 937
 938             fjx0             = _mm_add_ps(fjx0,tx);
 939             fjy0             = _mm_add_ps(fjy0,ty);
 940             fjz0             = _mm_add_ps(fjz0,tz);
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             qq10             = _mm_mul_ps(iq1,jq0);
 948
 949             /* COULOMB ELECTROSTATICS */
 950             velec            = _mm_mul_ps(qq10,rinv10);
 951             felec            = _mm_mul_ps(velec,rinvsq10);
 952
 953             fscal            = felec;
 954
 955             /* Calculate temporary vectorial force */
 956             tx               = _mm_mul_ps(fscal,dx10);
 957             ty               = _mm_mul_ps(fscal,dy10);
 958             tz               = _mm_mul_ps(fscal,dz10);
 959
 960             /* Update vectorial force */
 961             fix1             = _mm_add_ps(fix1,tx);
 962             fiy1             = _mm_add_ps(fiy1,ty);
 963             fiz1             = _mm_add_ps(fiz1,tz);
 964
 965             fjx0             = _mm_add_ps(fjx0,tx);
 966             fjy0             = _mm_add_ps(fjy0,ty);
 967             fjz0             = _mm_add_ps(fjz0,tz);
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             /* Compute parameters for interactions between i and j atoms */
 974             qq20             = _mm_mul_ps(iq2,jq0);
 975
 976             /* COULOMB ELECTROSTATICS */
 977             velec            = _mm_mul_ps(qq20,rinv20);
 978             felec            = _mm_mul_ps(velec,rinvsq20);
 979
 980             fscal            = felec;
 981
 982             /* Calculate temporary vectorial force */
 983             tx               = _mm_mul_ps(fscal,dx20);
 984             ty               = _mm_mul_ps(fscal,dy20);
 985             tz               = _mm_mul_ps(fscal,dz20);
 986
 987             /* Update vectorial force */
 988             fix2             = _mm_add_ps(fix2,tx);
 989             fiy2             = _mm_add_ps(fiy2,ty);
 990             fiz2             = _mm_add_ps(fiz2,tz);
 991
 992             fjx0             = _mm_add_ps(fjx0,tx);
 993             fjy0             = _mm_add_ps(fjy0,ty);
 994             fjz0             = _mm_add_ps(fjz0,tz);
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq30             = _mm_mul_ps(iq3,jq0);
1002
1003             /* COULOMB ELECTROSTATICS */
1004             velec            = _mm_mul_ps(qq30,rinv30);
1005             felec            = _mm_mul_ps(velec,rinvsq30);
1006
1007             fscal            = felec;
1008
1009             /* Calculate temporary vectorial force */
1010             tx               = _mm_mul_ps(fscal,dx30);
1011             ty               = _mm_mul_ps(fscal,dy30);
1012             tz               = _mm_mul_ps(fscal,dz30);
1013
1014             /* Update vectorial force */
1015             fix3             = _mm_add_ps(fix3,tx);
1016             fiy3             = _mm_add_ps(fiy3,ty);
1017             fiz3             = _mm_add_ps(fiz3,tz);
1018
1019             fjx0             = _mm_add_ps(fjx0,tx);
1020             fjy0             = _mm_add_ps(fjy0,ty);
1021             fjz0             = _mm_add_ps(fjz0,tz);
1022
1023             fjptrA             = f+j_coord_offsetA;
1024             fjptrB             = f+j_coord_offsetB;
1025             fjptrC             = f+j_coord_offsetC;
1026             fjptrD             = f+j_coord_offsetD;
1027
1028             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1029
1030             /* Inner loop uses 129 flops */
1031         }
1032
1033         if(jidx<j_index_end)
1034         {
1035
1036             /* Get j neighbor index, and coordinate index */
1037             jnrlistA         = jjnr[jidx];
1038             jnrlistB         = jjnr[jidx+1];
1039             jnrlistC         = jjnr[jidx+2];
1040             jnrlistD         = jjnr[jidx+3];
1041             /* Sign of each element will be negative for non-real atoms.
1042              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1043              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1044              */
1045             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1046             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1047             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1048             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1049             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1050             j_coord_offsetA  = DIM*jnrA;
1051             j_coord_offsetB  = DIM*jnrB;
1052             j_coord_offsetC  = DIM*jnrC;
1053             j_coord_offsetD  = DIM*jnrD;
1054
1055             /* load j atom coordinates */
1056             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1057                                               x+j_coord_offsetC,x+j_coord_offsetD,
1058                                               &jx0,&jy0,&jz0);
1059
1060             /* Calculate displacement vector */
1061             dx00             = _mm_sub_ps(ix0,jx0);
1062             dy00             = _mm_sub_ps(iy0,jy0);
1063             dz00             = _mm_sub_ps(iz0,jz0);
1064             dx10             = _mm_sub_ps(ix1,jx0);
1065             dy10             = _mm_sub_ps(iy1,jy0);
1066             dz10             = _mm_sub_ps(iz1,jz0);
1067             dx20             = _mm_sub_ps(ix2,jx0);
1068             dy20             = _mm_sub_ps(iy2,jy0);
1069             dz20             = _mm_sub_ps(iz2,jz0);
1070             dx30             = _mm_sub_ps(ix3,jx0);
1071             dy30             = _mm_sub_ps(iy3,jy0);
1072             dz30             = _mm_sub_ps(iz3,jz0);
1073
1074             /* Calculate squared distance and things based on it */
1075             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1076             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1077             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1078             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1079
1080             rinv00           = sse41_invsqrt_f(rsq00);
1081             rinv10           = sse41_invsqrt_f(rsq10);
1082             rinv20           = sse41_invsqrt_f(rsq20);
1083             rinv30           = sse41_invsqrt_f(rsq30);
1084
1085             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1086             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1087             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1088
1089             /* Load parameters for j particles */
1090             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1091                                                               charge+jnrC+0,charge+jnrD+0);
1092             vdwjidx0A        = 2*vdwtype[jnrA+0];
1093             vdwjidx0B        = 2*vdwtype[jnrB+0];
1094             vdwjidx0C        = 2*vdwtype[jnrC+0];
1095             vdwjidx0D        = 2*vdwtype[jnrD+0];
1096
1097             fjx0             = _mm_setzero_ps();
1098             fjy0             = _mm_setzero_ps();
1099             fjz0             = _mm_setzero_ps();
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r00              = _mm_mul_ps(rsq00,rinv00);
1106             r00              = _mm_andnot_ps(dummy_mask,r00);
1107
1108             /* Compute parameters for interactions between i and j atoms */
1109             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1110                                          vdwparam+vdwioffset0+vdwjidx0B,
1111                                          vdwparam+vdwioffset0+vdwjidx0C,
1112                                          vdwparam+vdwioffset0+vdwjidx0D,
1113                                          &c6_00,&c12_00);
1114
1115             /* Calculate table index by multiplying r with table scale and truncate to integer */
1116             rt               = _mm_mul_ps(r00,vftabscale);
1117             vfitab           = _mm_cvttps_epi32(rt);
1118             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1119             vfitab           = _mm_slli_epi32(vfitab,3);
1120
1121             /* CUBIC SPLINE TABLE DISPERSION */
1122             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1123             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1124             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1125             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1126             _MM_TRANSPOSE4_PS(Y,F,G,H);
1127             Heps             = _mm_mul_ps(vfeps,H);
1128             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1129             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1130             fvdw6            = _mm_mul_ps(c6_00,FF);
1131
1132             /* CUBIC SPLINE TABLE REPULSION */
1133             vfitab           = _mm_add_epi32(vfitab,ifour);
1134             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1135             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1136             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1137             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1138             _MM_TRANSPOSE4_PS(Y,F,G,H);
1139             Heps             = _mm_mul_ps(vfeps,H);
1140             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1141             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1142             fvdw12           = _mm_mul_ps(c12_00,FF);
1143             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1144
1145             fscal            = fvdw;
1146
1147             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1148
1149             /* Calculate temporary vectorial force */
1150             tx               = _mm_mul_ps(fscal,dx00);
1151             ty               = _mm_mul_ps(fscal,dy00);
1152             tz               = _mm_mul_ps(fscal,dz00);
1153
1154             /* Update vectorial force */
1155             fix0             = _mm_add_ps(fix0,tx);
1156             fiy0             = _mm_add_ps(fiy0,ty);
1157             fiz0             = _mm_add_ps(fiz0,tz);
1158
1159             fjx0             = _mm_add_ps(fjx0,tx);
1160             fjy0             = _mm_add_ps(fjy0,ty);
1161             fjz0             = _mm_add_ps(fjz0,tz);
1162
1163             /**************************
1164              * CALCULATE INTERACTIONS *
1165              **************************/
1166
1167             /* Compute parameters for interactions between i and j atoms */
1168             qq10             = _mm_mul_ps(iq1,jq0);
1169
1170             /* COULOMB ELECTROSTATICS */
1171             velec            = _mm_mul_ps(qq10,rinv10);
1172             felec            = _mm_mul_ps(velec,rinvsq10);
1173
1174             fscal            = felec;
1175
1176             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1177
1178             /* Calculate temporary vectorial force */
1179             tx               = _mm_mul_ps(fscal,dx10);
1180             ty               = _mm_mul_ps(fscal,dy10);
1181             tz               = _mm_mul_ps(fscal,dz10);
1182
1183             /* Update vectorial force */
1184             fix1             = _mm_add_ps(fix1,tx);
1185             fiy1             = _mm_add_ps(fiy1,ty);
1186             fiz1             = _mm_add_ps(fiz1,tz);
1187
1188             fjx0             = _mm_add_ps(fjx0,tx);
1189             fjy0             = _mm_add_ps(fjy0,ty);
1190             fjz0             = _mm_add_ps(fjz0,tz);
1191
1192             /**************************
1193              * CALCULATE INTERACTIONS *
1194              **************************/
1195
1196             /* Compute parameters for interactions between i and j atoms */
1197             qq20             = _mm_mul_ps(iq2,jq0);
1198
1199             /* COULOMB ELECTROSTATICS */
1200             velec            = _mm_mul_ps(qq20,rinv20);
1201             felec            = _mm_mul_ps(velec,rinvsq20);
1202
1203             fscal            = felec;
1204
1205             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm_mul_ps(fscal,dx20);
1209             ty               = _mm_mul_ps(fscal,dy20);
1210             tz               = _mm_mul_ps(fscal,dz20);
1211
1212             /* Update vectorial force */
1213             fix2             = _mm_add_ps(fix2,tx);
1214             fiy2             = _mm_add_ps(fiy2,ty);
1215             fiz2             = _mm_add_ps(fiz2,tz);
1216
1217             fjx0             = _mm_add_ps(fjx0,tx);
1218             fjy0             = _mm_add_ps(fjy0,ty);
1219             fjz0             = _mm_add_ps(fjz0,tz);
1220
1221             /**************************
1222              * CALCULATE INTERACTIONS *
1223              **************************/
1224
1225             /* Compute parameters for interactions between i and j atoms */
1226             qq30             = _mm_mul_ps(iq3,jq0);
1227
1228             /* COULOMB ELECTROSTATICS */
1229             velec            = _mm_mul_ps(qq30,rinv30);
1230             felec            = _mm_mul_ps(velec,rinvsq30);
1231
1232             fscal            = felec;
1233
1234             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1235
1236             /* Calculate temporary vectorial force */
1237             tx               = _mm_mul_ps(fscal,dx30);
1238             ty               = _mm_mul_ps(fscal,dy30);
1239             tz               = _mm_mul_ps(fscal,dz30);
1240
1241             /* Update vectorial force */
1242             fix3             = _mm_add_ps(fix3,tx);
1243             fiy3             = _mm_add_ps(fiy3,ty);
1244             fiz3             = _mm_add_ps(fiz3,tz);
1245
1246             fjx0             = _mm_add_ps(fjx0,tx);
1247             fjy0             = _mm_add_ps(fjy0,ty);
1248             fjz0             = _mm_add_ps(fjz0,tz);
1249
1250             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1251             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1252             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1253             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1254
1255             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1256
1257             /* Inner loop uses 130 flops */
1258         }
1259
1260         /* End of innermost loop */
1261
1262         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1263                                               f+i_coord_offset,fshift+i_shift_offset);
1264
1265         /* Increment number of inner iterations */
1266         inneriter                  += j_index_end - j_index_start;
1267
1268         /* Outer loop uses 24 flops */
1269     }
1270
1271     /* Increment number of outer iterations */
1272     outeriter        += nri;
1273
1274     /* Update outer/inner flops */
1275
1276     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1277 }