src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_sse2_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 131     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 132     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 167
 168         fix0             = _mm_setzero_ps();
 169         fiy0             = _mm_setzero_ps();
 170         fiz0             = _mm_setzero_ps();
 171         fix1             = _mm_setzero_ps();
 172         fiy1             = _mm_setzero_ps();
 173         fiz1             = _mm_setzero_ps();
 174         fix2             = _mm_setzero_ps();
 175         fiy2             = _mm_setzero_ps();
 176         fiz2             = _mm_setzero_ps();
 177         fix3             = _mm_setzero_ps();
 178         fiy3             = _mm_setzero_ps();
 179         fiz3             = _mm_setzero_ps();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_ps();
 183         vvdwsum          = _mm_setzero_ps();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208             dx10             = _mm_sub_ps(ix1,jx0);
 209             dy10             = _mm_sub_ps(iy1,jy0);
 210             dz10             = _mm_sub_ps(iz1,jz0);
 211             dx20             = _mm_sub_ps(ix2,jx0);
 212             dy20             = _mm_sub_ps(iy2,jy0);
 213             dz20             = _mm_sub_ps(iz2,jz0);
 214             dx30             = _mm_sub_ps(ix3,jx0);
 215             dy30             = _mm_sub_ps(iy3,jy0);
 216             dz30             = _mm_sub_ps(iz3,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 222             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 223
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 227
 228             rinvsq00         = gmx_mm_inv_ps(rsq00);
 229             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 231             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             fjx0             = _mm_setzero_ps();
 242             fjy0             = _mm_setzero_ps();
 243             fjz0             = _mm_setzero_ps();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 251                                          vdwparam+vdwioffset0+vdwjidx0B,
 252                                          vdwparam+vdwioffset0+vdwjidx0C,
 253                                          vdwparam+vdwioffset0+vdwjidx0D,
 254                                          &c6_00,&c12_00);
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 260             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 261             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 262             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 266
 267             fscal            = fvdw;
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm_mul_ps(fscal,dx00);
 271             ty               = _mm_mul_ps(fscal,dy00);
 272             tz               = _mm_mul_ps(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm_add_ps(fix0,tx);
 276             fiy0             = _mm_add_ps(fiy0,ty);
 277             fiz0             = _mm_add_ps(fiz0,tz);
 278
 279             fjx0             = _mm_add_ps(fjx0,tx);
 280             fjy0             = _mm_add_ps(fjy0,ty);
 281             fjz0             = _mm_add_ps(fjz0,tz);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq10             = _mm_mul_ps(iq1,jq0);
 289
 290             /* COULOMB ELECTROSTATICS */
 291             velec            = _mm_mul_ps(qq10,rinv10);
 292             felec            = _mm_mul_ps(velec,rinvsq10);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm_add_ps(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm_mul_ps(fscal,dx10);
 301             ty               = _mm_mul_ps(fscal,dy10);
 302             tz               = _mm_mul_ps(fscal,dz10);
 303
 304             /* Update vectorial force */
 305             fix1             = _mm_add_ps(fix1,tx);
 306             fiy1             = _mm_add_ps(fiy1,ty);
 307             fiz1             = _mm_add_ps(fiz1,tz);
 308
 309             fjx0             = _mm_add_ps(fjx0,tx);
 310             fjy0             = _mm_add_ps(fjy0,ty);
 311             fjz0             = _mm_add_ps(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq20             = _mm_mul_ps(iq2,jq0);
 319
 320             /* COULOMB ELECTROSTATICS */
 321             velec            = _mm_mul_ps(qq20,rinv20);
 322             felec            = _mm_mul_ps(velec,rinvsq20);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm_add_ps(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm_mul_ps(fscal,dx20);
 331             ty               = _mm_mul_ps(fscal,dy20);
 332             tz               = _mm_mul_ps(fscal,dz20);
 333
 334             /* Update vectorial force */
 335             fix2             = _mm_add_ps(fix2,tx);
 336             fiy2             = _mm_add_ps(fiy2,ty);
 337             fiz2             = _mm_add_ps(fiz2,tz);
 338
 339             fjx0             = _mm_add_ps(fjx0,tx);
 340             fjy0             = _mm_add_ps(fjy0,ty);
 341             fjz0             = _mm_add_ps(fjz0,tz);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq30             = _mm_mul_ps(iq3,jq0);
 349
 350             /* COULOMB ELECTROSTATICS */
 351             velec            = _mm_mul_ps(qq30,rinv30);
 352             felec            = _mm_mul_ps(velec,rinvsq30);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velecsum         = _mm_add_ps(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm_mul_ps(fscal,dx30);
 361             ty               = _mm_mul_ps(fscal,dy30);
 362             tz               = _mm_mul_ps(fscal,dz30);
 363
 364             /* Update vectorial force */
 365             fix3             = _mm_add_ps(fix3,tx);
 366             fiy3             = _mm_add_ps(fiy3,ty);
 367             fiz3             = _mm_add_ps(fiz3,tz);
 368
 369             fjx0             = _mm_add_ps(fjx0,tx);
 370             fjy0             = _mm_add_ps(fjy0,ty);
 371             fjz0             = _mm_add_ps(fjz0,tz);
 372
 373             fjptrA             = f+j_coord_offsetA;
 374             fjptrB             = f+j_coord_offsetB;
 375             fjptrC             = f+j_coord_offsetC;
 376             fjptrD             = f+j_coord_offsetD;
 377
 378             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 379
 380             /* Inner loop uses 116 flops */
 381         }
 382
 383         if(jidx<j_index_end)
 384         {
 385
 386             /* Get j neighbor index, and coordinate index */
 387             jnrlistA         = jjnr[jidx];
 388             jnrlistB         = jjnr[jidx+1];
 389             jnrlistC         = jjnr[jidx+2];
 390             jnrlistD         = jjnr[jidx+3];
 391             /* Sign of each element will be negative for non-real atoms.
 392              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 393              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 394              */
 395             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 396             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 397             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 398             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 399             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 400             j_coord_offsetA  = DIM*jnrA;
 401             j_coord_offsetB  = DIM*jnrB;
 402             j_coord_offsetC  = DIM*jnrC;
 403             j_coord_offsetD  = DIM*jnrD;
 404
 405             /* load j atom coordinates */
 406             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 407                                               x+j_coord_offsetC,x+j_coord_offsetD,
 408                                               &jx0,&jy0,&jz0);
 409
 410             /* Calculate displacement vector */
 411             dx00             = _mm_sub_ps(ix0,jx0);
 412             dy00             = _mm_sub_ps(iy0,jy0);
 413             dz00             = _mm_sub_ps(iz0,jz0);
 414             dx10             = _mm_sub_ps(ix1,jx0);
 415             dy10             = _mm_sub_ps(iy1,jy0);
 416             dz10             = _mm_sub_ps(iz1,jz0);
 417             dx20             = _mm_sub_ps(ix2,jx0);
 418             dy20             = _mm_sub_ps(iy2,jy0);
 419             dz20             = _mm_sub_ps(iz2,jz0);
 420             dx30             = _mm_sub_ps(ix3,jx0);
 421             dy30             = _mm_sub_ps(iy3,jy0);
 422             dz30             = _mm_sub_ps(iz3,jz0);
 423
 424             /* Calculate squared distance and things based on it */
 425             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 426             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 427             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 428             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 429
 430             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 431             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 432             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 433
 434             rinvsq00         = gmx_mm_inv_ps(rsq00);
 435             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 436             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 437             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 438
 439             /* Load parameters for j particles */
 440             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 441                                                               charge+jnrC+0,charge+jnrD+0);
 442             vdwjidx0A        = 2*vdwtype[jnrA+0];
 443             vdwjidx0B        = 2*vdwtype[jnrB+0];
 444             vdwjidx0C        = 2*vdwtype[jnrC+0];
 445             vdwjidx0D        = 2*vdwtype[jnrD+0];
 446
 447             fjx0             = _mm_setzero_ps();
 448             fjy0             = _mm_setzero_ps();
 449             fjz0             = _mm_setzero_ps();
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 457                                          vdwparam+vdwioffset0+vdwjidx0B,
 458                                          vdwparam+vdwioffset0+vdwjidx0C,
 459                                          vdwparam+vdwioffset0+vdwjidx0D,
 460                                          &c6_00,&c12_00);
 461
 462             /* LENNARD-JONES DISPERSION/REPULSION */
 463
 464             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 465             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 466             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 467             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 468             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 472             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 473
 474             fscal            = fvdw;
 475
 476             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 477
 478             /* Calculate temporary vectorial force */
 479             tx               = _mm_mul_ps(fscal,dx00);
 480             ty               = _mm_mul_ps(fscal,dy00);
 481             tz               = _mm_mul_ps(fscal,dz00);
 482
 483             /* Update vectorial force */
 484             fix0             = _mm_add_ps(fix0,tx);
 485             fiy0             = _mm_add_ps(fiy0,ty);
 486             fiz0             = _mm_add_ps(fiz0,tz);
 487
 488             fjx0             = _mm_add_ps(fjx0,tx);
 489             fjy0             = _mm_add_ps(fjy0,ty);
 490             fjz0             = _mm_add_ps(fjz0,tz);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq10             = _mm_mul_ps(iq1,jq0);
 498
 499             /* COULOMB ELECTROSTATICS */
 500             velec            = _mm_mul_ps(qq10,rinv10);
 501             felec            = _mm_mul_ps(velec,rinvsq10);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm_andnot_ps(dummy_mask,velec);
 505             velecsum         = _mm_add_ps(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm_mul_ps(fscal,dx10);
 513             ty               = _mm_mul_ps(fscal,dy10);
 514             tz               = _mm_mul_ps(fscal,dz10);
 515
 516             /* Update vectorial force */
 517             fix1             = _mm_add_ps(fix1,tx);
 518             fiy1             = _mm_add_ps(fiy1,ty);
 519             fiz1             = _mm_add_ps(fiz1,tz);
 520
 521             fjx0             = _mm_add_ps(fjx0,tx);
 522             fjy0             = _mm_add_ps(fjy0,ty);
 523             fjz0             = _mm_add_ps(fjz0,tz);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq20             = _mm_mul_ps(iq2,jq0);
 531
 532             /* COULOMB ELECTROSTATICS */
 533             velec            = _mm_mul_ps(qq20,rinv20);
 534             felec            = _mm_mul_ps(velec,rinvsq20);
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             velec            = _mm_andnot_ps(dummy_mask,velec);
 538             velecsum         = _mm_add_ps(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm_mul_ps(fscal,dx20);
 546             ty               = _mm_mul_ps(fscal,dy20);
 547             tz               = _mm_mul_ps(fscal,dz20);
 548
 549             /* Update vectorial force */
 550             fix2             = _mm_add_ps(fix2,tx);
 551             fiy2             = _mm_add_ps(fiy2,ty);
 552             fiz2             = _mm_add_ps(fiz2,tz);
 553
 554             fjx0             = _mm_add_ps(fjx0,tx);
 555             fjy0             = _mm_add_ps(fjy0,ty);
 556             fjz0             = _mm_add_ps(fjz0,tz);
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq30             = _mm_mul_ps(iq3,jq0);
 564
 565             /* COULOMB ELECTROSTATICS */
 566             velec            = _mm_mul_ps(qq30,rinv30);
 567             felec            = _mm_mul_ps(velec,rinvsq30);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 576
 577             /* Calculate temporary vectorial force */
 578             tx               = _mm_mul_ps(fscal,dx30);
 579             ty               = _mm_mul_ps(fscal,dy30);
 580             tz               = _mm_mul_ps(fscal,dz30);
 581
 582             /* Update vectorial force */
 583             fix3             = _mm_add_ps(fix3,tx);
 584             fiy3             = _mm_add_ps(fiy3,ty);
 585             fiz3             = _mm_add_ps(fiz3,tz);
 586
 587             fjx0             = _mm_add_ps(fjx0,tx);
 588             fjy0             = _mm_add_ps(fjy0,ty);
 589             fjz0             = _mm_add_ps(fjz0,tz);
 590
 591             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 592             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 593             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 594             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 595
 596             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 597
 598             /* Inner loop uses 116 flops */
 599         }
 600
 601         /* End of innermost loop */
 602
 603         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 604                                               f+i_coord_offset,fshift+i_shift_offset);
 605
 606         ggid                        = gid[iidx];
 607         /* Update potential energies */
 608         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 609         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 610
 611         /* Increment number of inner iterations */
 612         inneriter                  += j_index_end - j_index_start;
 613
 614         /* Outer loop uses 26 flops */
 615     }
 616
 617     /* Increment number of outer iterations */
 618     outeriter        += nri;
 619
 620     /* Update outer/inner flops */
 621
 622     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 623 }
 624 /*
 625  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse2_single
 626  * Electrostatics interaction: Coulomb
 627  * VdW interaction:            LennardJones
 628  * Geometry:                   Water4-Particle
 629  * Calculate force/pot:        Force
 630  */
 631 void
 632 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse2_single
 633                     (t_nblist                    * gmx_restrict       nlist,
 634                      rvec                        * gmx_restrict          xx,
 635                      rvec                        * gmx_restrict          ff,
 636                      t_forcerec                  * gmx_restrict          fr,
 637                      t_mdatoms                   * gmx_restrict     mdatoms,
 638                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 639                      t_nrnb                      * gmx_restrict        nrnb)
 640 {
 641     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 642      * just 0 for non-waters.
 643      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 644      * jnr indices corresponding to data put in the four positions in the SIMD register.
 645      */
 646     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 647     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 648     int              jnrA,jnrB,jnrC,jnrD;
 649     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 650     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 651     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 652     real             rcutoff_scalar;
 653     real             *shiftvec,*fshift,*x,*f;
 654     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 655     real             scratch[4*DIM];
 656     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 657     int              vdwioffset0;
 658     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 659     int              vdwioffset1;
 660     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 661     int              vdwioffset2;
 662     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 663     int              vdwioffset3;
 664     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 665     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 666     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 667     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 668     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 669     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 670     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 671     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 672     real             *charge;
 673     int              nvdwtype;
 674     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 675     int              *vdwtype;
 676     real             *vdwparam;
 677     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 678     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 679     __m128           dummy_mask,cutoff_mask;
 680     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 681     __m128           one     = _mm_set1_ps(1.0);
 682     __m128           two     = _mm_set1_ps(2.0);
 683     x                = xx[0];
 684     f                = ff[0];
 685
 686     nri              = nlist->nri;
 687     iinr             = nlist->iinr;
 688     jindex           = nlist->jindex;
 689     jjnr             = nlist->jjnr;
 690     shiftidx         = nlist->shift;
 691     gid              = nlist->gid;
 692     shiftvec         = fr->shift_vec[0];
 693     fshift           = fr->fshift[0];
 694     facel            = _mm_set1_ps(fr->epsfac);
 695     charge           = mdatoms->chargeA;
 696     nvdwtype         = fr->ntype;
 697     vdwparam         = fr->nbfp;
 698     vdwtype          = mdatoms->typeA;
 699
 700     /* Setup water-specific parameters */
 701     inr              = nlist->iinr[0];
 702     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 703     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 704     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 705     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 706
 707     /* Avoid stupid compiler warnings */
 708     jnrA = jnrB = jnrC = jnrD = 0;
 709     j_coord_offsetA = 0;
 710     j_coord_offsetB = 0;
 711     j_coord_offsetC = 0;
 712     j_coord_offsetD = 0;
 713
 714     outeriter        = 0;
 715     inneriter        = 0;
 716
 717     for(iidx=0;iidx<4*DIM;iidx++)
 718     {
 719         scratch[iidx] = 0.0;
 720     }
 721
 722     /* Start outer loop over neighborlists */
 723     for(iidx=0; iidx<nri; iidx++)
 724     {
 725         /* Load shift vector for this list */
 726         i_shift_offset   = DIM*shiftidx[iidx];
 727
 728         /* Load limits for loop over neighbors */
 729         j_index_start    = jindex[iidx];
 730         j_index_end      = jindex[iidx+1];
 731
 732         /* Get outer coordinate index */
 733         inr              = iinr[iidx];
 734         i_coord_offset   = DIM*inr;
 735
 736         /* Load i particle coords and add shift vector */
 737         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 738                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 739
 740         fix0             = _mm_setzero_ps();
 741         fiy0             = _mm_setzero_ps();
 742         fiz0             = _mm_setzero_ps();
 743         fix1             = _mm_setzero_ps();
 744         fiy1             = _mm_setzero_ps();
 745         fiz1             = _mm_setzero_ps();
 746         fix2             = _mm_setzero_ps();
 747         fiy2             = _mm_setzero_ps();
 748         fiz2             = _mm_setzero_ps();
 749         fix3             = _mm_setzero_ps();
 750         fiy3             = _mm_setzero_ps();
 751         fiz3             = _mm_setzero_ps();
 752
 753         /* Start inner kernel loop */
 754         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 755         {
 756
 757             /* Get j neighbor index, and coordinate index */
 758             jnrA             = jjnr[jidx];
 759             jnrB             = jjnr[jidx+1];
 760             jnrC             = jjnr[jidx+2];
 761             jnrD             = jjnr[jidx+3];
 762             j_coord_offsetA  = DIM*jnrA;
 763             j_coord_offsetB  = DIM*jnrB;
 764             j_coord_offsetC  = DIM*jnrC;
 765             j_coord_offsetD  = DIM*jnrD;
 766
 767             /* load j atom coordinates */
 768             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 769                                               x+j_coord_offsetC,x+j_coord_offsetD,
 770                                               &jx0,&jy0,&jz0);
 771
 772             /* Calculate displacement vector */
 773             dx00             = _mm_sub_ps(ix0,jx0);
 774             dy00             = _mm_sub_ps(iy0,jy0);
 775             dz00             = _mm_sub_ps(iz0,jz0);
 776             dx10             = _mm_sub_ps(ix1,jx0);
 777             dy10             = _mm_sub_ps(iy1,jy0);
 778             dz10             = _mm_sub_ps(iz1,jz0);
 779             dx20             = _mm_sub_ps(ix2,jx0);
 780             dy20             = _mm_sub_ps(iy2,jy0);
 781             dz20             = _mm_sub_ps(iz2,jz0);
 782             dx30             = _mm_sub_ps(ix3,jx0);
 783             dy30             = _mm_sub_ps(iy3,jy0);
 784             dz30             = _mm_sub_ps(iz3,jz0);
 785
 786             /* Calculate squared distance and things based on it */
 787             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 788             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 789             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 790             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 791
 792             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 793             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 794             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 795
 796             rinvsq00         = gmx_mm_inv_ps(rsq00);
 797             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 798             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 799             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 800
 801             /* Load parameters for j particles */
 802             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 803                                                               charge+jnrC+0,charge+jnrD+0);
 804             vdwjidx0A        = 2*vdwtype[jnrA+0];
 805             vdwjidx0B        = 2*vdwtype[jnrB+0];
 806             vdwjidx0C        = 2*vdwtype[jnrC+0];
 807             vdwjidx0D        = 2*vdwtype[jnrD+0];
 808
 809             fjx0             = _mm_setzero_ps();
 810             fjy0             = _mm_setzero_ps();
 811             fjz0             = _mm_setzero_ps();
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             /* Compute parameters for interactions between i and j atoms */
 818             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 819                                          vdwparam+vdwioffset0+vdwjidx0B,
 820                                          vdwparam+vdwioffset0+vdwjidx0C,
 821                                          vdwparam+vdwioffset0+vdwjidx0D,
 822                                          &c6_00,&c12_00);
 823
 824             /* LENNARD-JONES DISPERSION/REPULSION */
 825
 826             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 827             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 828
 829             fscal            = fvdw;
 830
 831             /* Calculate temporary vectorial force */
 832             tx               = _mm_mul_ps(fscal,dx00);
 833             ty               = _mm_mul_ps(fscal,dy00);
 834             tz               = _mm_mul_ps(fscal,dz00);
 835
 836             /* Update vectorial force */
 837             fix0             = _mm_add_ps(fix0,tx);
 838             fiy0             = _mm_add_ps(fiy0,ty);
 839             fiz0             = _mm_add_ps(fiz0,tz);
 840
 841             fjx0             = _mm_add_ps(fjx0,tx);
 842             fjy0             = _mm_add_ps(fjy0,ty);
 843             fjz0             = _mm_add_ps(fjz0,tz);
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq10             = _mm_mul_ps(iq1,jq0);
 851
 852             /* COULOMB ELECTROSTATICS */
 853             velec            = _mm_mul_ps(qq10,rinv10);
 854             felec            = _mm_mul_ps(velec,rinvsq10);
 855
 856             fscal            = felec;
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm_mul_ps(fscal,dx10);
 860             ty               = _mm_mul_ps(fscal,dy10);
 861             tz               = _mm_mul_ps(fscal,dz10);
 862
 863             /* Update vectorial force */
 864             fix1             = _mm_add_ps(fix1,tx);
 865             fiy1             = _mm_add_ps(fiy1,ty);
 866             fiz1             = _mm_add_ps(fiz1,tz);
 867
 868             fjx0             = _mm_add_ps(fjx0,tx);
 869             fjy0             = _mm_add_ps(fjy0,ty);
 870             fjz0             = _mm_add_ps(fjz0,tz);
 871
 872             /**************************
 873              * CALCULATE INTERACTIONS *
 874              **************************/
 875
 876             /* Compute parameters for interactions between i and j atoms */
 877             qq20             = _mm_mul_ps(iq2,jq0);
 878
 879             /* COULOMB ELECTROSTATICS */
 880             velec            = _mm_mul_ps(qq20,rinv20);
 881             felec            = _mm_mul_ps(velec,rinvsq20);
 882
 883             fscal            = felec;
 884
 885             /* Calculate temporary vectorial force */
 886             tx               = _mm_mul_ps(fscal,dx20);
 887             ty               = _mm_mul_ps(fscal,dy20);
 888             tz               = _mm_mul_ps(fscal,dz20);
 889
 890             /* Update vectorial force */
 891             fix2             = _mm_add_ps(fix2,tx);
 892             fiy2             = _mm_add_ps(fiy2,ty);
 893             fiz2             = _mm_add_ps(fiz2,tz);
 894
 895             fjx0             = _mm_add_ps(fjx0,tx);
 896             fjy0             = _mm_add_ps(fjy0,ty);
 897             fjz0             = _mm_add_ps(fjz0,tz);
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             /* Compute parameters for interactions between i and j atoms */
 904             qq30             = _mm_mul_ps(iq3,jq0);
 905
 906             /* COULOMB ELECTROSTATICS */
 907             velec            = _mm_mul_ps(qq30,rinv30);
 908             felec            = _mm_mul_ps(velec,rinvsq30);
 909
 910             fscal            = felec;
 911
 912             /* Calculate temporary vectorial force */
 913             tx               = _mm_mul_ps(fscal,dx30);
 914             ty               = _mm_mul_ps(fscal,dy30);
 915             tz               = _mm_mul_ps(fscal,dz30);
 916
 917             /* Update vectorial force */
 918             fix3             = _mm_add_ps(fix3,tx);
 919             fiy3             = _mm_add_ps(fiy3,ty);
 920             fiz3             = _mm_add_ps(fiz3,tz);
 921
 922             fjx0             = _mm_add_ps(fjx0,tx);
 923             fjy0             = _mm_add_ps(fjy0,ty);
 924             fjz0             = _mm_add_ps(fjz0,tz);
 925
 926             fjptrA             = f+j_coord_offsetA;
 927             fjptrB             = f+j_coord_offsetB;
 928             fjptrC             = f+j_coord_offsetC;
 929             fjptrD             = f+j_coord_offsetD;
 930
 931             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 932
 933             /* Inner loop uses 108 flops */
 934         }
 935
 936         if(jidx<j_index_end)
 937         {
 938
 939             /* Get j neighbor index, and coordinate index */
 940             jnrlistA         = jjnr[jidx];
 941             jnrlistB         = jjnr[jidx+1];
 942             jnrlistC         = jjnr[jidx+2];
 943             jnrlistD         = jjnr[jidx+3];
 944             /* Sign of each element will be negative for non-real atoms.
 945              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 946              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 947              */
 948             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 949             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 950             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 951             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 952             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 953             j_coord_offsetA  = DIM*jnrA;
 954             j_coord_offsetB  = DIM*jnrB;
 955             j_coord_offsetC  = DIM*jnrC;
 956             j_coord_offsetD  = DIM*jnrD;
 957
 958             /* load j atom coordinates */
 959             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 960                                               x+j_coord_offsetC,x+j_coord_offsetD,
 961                                               &jx0,&jy0,&jz0);
 962
 963             /* Calculate displacement vector */
 964             dx00             = _mm_sub_ps(ix0,jx0);
 965             dy00             = _mm_sub_ps(iy0,jy0);
 966             dz00             = _mm_sub_ps(iz0,jz0);
 967             dx10             = _mm_sub_ps(ix1,jx0);
 968             dy10             = _mm_sub_ps(iy1,jy0);
 969             dz10             = _mm_sub_ps(iz1,jz0);
 970             dx20             = _mm_sub_ps(ix2,jx0);
 971             dy20             = _mm_sub_ps(iy2,jy0);
 972             dz20             = _mm_sub_ps(iz2,jz0);
 973             dx30             = _mm_sub_ps(ix3,jx0);
 974             dy30             = _mm_sub_ps(iy3,jy0);
 975             dz30             = _mm_sub_ps(iz3,jz0);
 976
 977             /* Calculate squared distance and things based on it */
 978             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 979             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 980             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 981             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 982
 983             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 984             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 985             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 986
 987             rinvsq00         = gmx_mm_inv_ps(rsq00);
 988             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 989             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 990             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 991
 992             /* Load parameters for j particles */
 993             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 994                                                               charge+jnrC+0,charge+jnrD+0);
 995             vdwjidx0A        = 2*vdwtype[jnrA+0];
 996             vdwjidx0B        = 2*vdwtype[jnrB+0];
 997             vdwjidx0C        = 2*vdwtype[jnrC+0];
 998             vdwjidx0D        = 2*vdwtype[jnrD+0];
 999
1000             fjx0             = _mm_setzero_ps();
1001             fjy0             = _mm_setzero_ps();
1002             fjz0             = _mm_setzero_ps();
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1010                                          vdwparam+vdwioffset0+vdwjidx0B,
1011                                          vdwparam+vdwioffset0+vdwjidx0C,
1012                                          vdwparam+vdwioffset0+vdwjidx0D,
1013                                          &c6_00,&c12_00);
1014
1015             /* LENNARD-JONES DISPERSION/REPULSION */
1016
1017             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1018             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1019
1020             fscal            = fvdw;
1021
1022             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm_mul_ps(fscal,dx00);
1026             ty               = _mm_mul_ps(fscal,dy00);
1027             tz               = _mm_mul_ps(fscal,dz00);
1028
1029             /* Update vectorial force */
1030             fix0             = _mm_add_ps(fix0,tx);
1031             fiy0             = _mm_add_ps(fiy0,ty);
1032             fiz0             = _mm_add_ps(fiz0,tz);
1033
1034             fjx0             = _mm_add_ps(fjx0,tx);
1035             fjy0             = _mm_add_ps(fjy0,ty);
1036             fjz0             = _mm_add_ps(fjz0,tz);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             qq10             = _mm_mul_ps(iq1,jq0);
1044
1045             /* COULOMB ELECTROSTATICS */
1046             velec            = _mm_mul_ps(qq10,rinv10);
1047             felec            = _mm_mul_ps(velec,rinvsq10);
1048
1049             fscal            = felec;
1050
1051             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm_mul_ps(fscal,dx10);
1055             ty               = _mm_mul_ps(fscal,dy10);
1056             tz               = _mm_mul_ps(fscal,dz10);
1057
1058             /* Update vectorial force */
1059             fix1             = _mm_add_ps(fix1,tx);
1060             fiy1             = _mm_add_ps(fiy1,ty);
1061             fiz1             = _mm_add_ps(fiz1,tz);
1062
1063             fjx0             = _mm_add_ps(fjx0,tx);
1064             fjy0             = _mm_add_ps(fjy0,ty);
1065             fjz0             = _mm_add_ps(fjz0,tz);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq20             = _mm_mul_ps(iq2,jq0);
1073
1074             /* COULOMB ELECTROSTATICS */
1075             velec            = _mm_mul_ps(qq20,rinv20);
1076             felec            = _mm_mul_ps(velec,rinvsq20);
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm_mul_ps(fscal,dx20);
1084             ty               = _mm_mul_ps(fscal,dy20);
1085             tz               = _mm_mul_ps(fscal,dz20);
1086
1087             /* Update vectorial force */
1088             fix2             = _mm_add_ps(fix2,tx);
1089             fiy2             = _mm_add_ps(fiy2,ty);
1090             fiz2             = _mm_add_ps(fiz2,tz);
1091
1092             fjx0             = _mm_add_ps(fjx0,tx);
1093             fjy0             = _mm_add_ps(fjy0,ty);
1094             fjz0             = _mm_add_ps(fjz0,tz);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq30             = _mm_mul_ps(iq3,jq0);
1102
1103             /* COULOMB ELECTROSTATICS */
1104             velec            = _mm_mul_ps(qq30,rinv30);
1105             felec            = _mm_mul_ps(velec,rinvsq30);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1110
1111             /* Calculate temporary vectorial force */
1112             tx               = _mm_mul_ps(fscal,dx30);
1113             ty               = _mm_mul_ps(fscal,dy30);
1114             tz               = _mm_mul_ps(fscal,dz30);
1115
1116             /* Update vectorial force */
1117             fix3             = _mm_add_ps(fix3,tx);
1118             fiy3             = _mm_add_ps(fiy3,ty);
1119             fiz3             = _mm_add_ps(fiz3,tz);
1120
1121             fjx0             = _mm_add_ps(fjx0,tx);
1122             fjy0             = _mm_add_ps(fjy0,ty);
1123             fjz0             = _mm_add_ps(fjz0,tz);
1124
1125             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1126             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1127             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1128             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1129
1130             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1131
1132             /* Inner loop uses 108 flops */
1133         }
1134
1135         /* End of innermost loop */
1136
1137         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1138                                               f+i_coord_offset,fshift+i_shift_offset);
1139
1140         /* Increment number of inner iterations */
1141         inneriter                  += j_index_end - j_index_start;
1142
1143         /* Outer loop uses 24 flops */
1144     }
1145
1146     /* Increment number of outer iterations */
1147     outeriter        += nri;
1148
1149     /* Update outer/inner flops */
1150
1151     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1152 }