src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_sse4_1_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169         fix1             = _mm_setzero_ps();
 170         fiy1             = _mm_setzero_ps();
 171         fiz1             = _mm_setzero_ps();
 172         fix2             = _mm_setzero_ps();
 173         fiy2             = _mm_setzero_ps();
 174         fiz2             = _mm_setzero_ps();
 175         fix3             = _mm_setzero_ps();
 176         fiy3             = _mm_setzero_ps();
 177         fiz3             = _mm_setzero_ps();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_ps();
 181         vvdwsum          = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206             dx10             = _mm_sub_ps(ix1,jx0);
 207             dy10             = _mm_sub_ps(iy1,jy0);
 208             dz10             = _mm_sub_ps(iz1,jz0);
 209             dx20             = _mm_sub_ps(ix2,jx0);
 210             dy20             = _mm_sub_ps(iy2,jy0);
 211             dz20             = _mm_sub_ps(iz2,jz0);
 212             dx30             = _mm_sub_ps(ix3,jx0);
 213             dy30             = _mm_sub_ps(iy3,jy0);
 214             dz30             = _mm_sub_ps(iz3,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 221
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 225
 226             rinvsq00         = gmx_mm_inv_ps(rsq00);
 227             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 229             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                               charge+jnrC+0,charge+jnrD+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236             vdwjidx0C        = 2*vdwtype[jnrC+0];
 237             vdwjidx0D        = 2*vdwtype[jnrD+0];
 238
 239             fjx0             = _mm_setzero_ps();
 240             fjy0             = _mm_setzero_ps();
 241             fjz0             = _mm_setzero_ps();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 249                                          vdwparam+vdwioffset0+vdwjidx0B,
 250                                          vdwparam+vdwioffset0+vdwjidx0C,
 251                                          vdwparam+vdwioffset0+vdwjidx0D,
 252                                          &c6_00,&c12_00);
 253
 254             /* LENNARD-JONES DISPERSION/REPULSION */
 255
 256             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 257             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 258             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 259             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 260             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 264
 265             fscal            = fvdw;
 266
 267             /* Calculate temporary vectorial force */
 268             tx               = _mm_mul_ps(fscal,dx00);
 269             ty               = _mm_mul_ps(fscal,dy00);
 270             tz               = _mm_mul_ps(fscal,dz00);
 271
 272             /* Update vectorial force */
 273             fix0             = _mm_add_ps(fix0,tx);
 274             fiy0             = _mm_add_ps(fiy0,ty);
 275             fiz0             = _mm_add_ps(fiz0,tz);
 276
 277             fjx0             = _mm_add_ps(fjx0,tx);
 278             fjy0             = _mm_add_ps(fjy0,ty);
 279             fjz0             = _mm_add_ps(fjz0,tz);
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm_mul_ps(iq1,jq0);
 287
 288             /* COULOMB ELECTROSTATICS */
 289             velec            = _mm_mul_ps(qq10,rinv10);
 290             felec            = _mm_mul_ps(velec,rinvsq10);
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velecsum         = _mm_add_ps(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_ps(fscal,dx10);
 299             ty               = _mm_mul_ps(fscal,dy10);
 300             tz               = _mm_mul_ps(fscal,dz10);
 301
 302             /* Update vectorial force */
 303             fix1             = _mm_add_ps(fix1,tx);
 304             fiy1             = _mm_add_ps(fiy1,ty);
 305             fiz1             = _mm_add_ps(fiz1,tz);
 306
 307             fjx0             = _mm_add_ps(fjx0,tx);
 308             fjy0             = _mm_add_ps(fjy0,ty);
 309             fjz0             = _mm_add_ps(fjz0,tz);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq20             = _mm_mul_ps(iq2,jq0);
 317
 318             /* COULOMB ELECTROSTATICS */
 319             velec            = _mm_mul_ps(qq20,rinv20);
 320             felec            = _mm_mul_ps(velec,rinvsq20);
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velecsum         = _mm_add_ps(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm_mul_ps(fscal,dx20);
 329             ty               = _mm_mul_ps(fscal,dy20);
 330             tz               = _mm_mul_ps(fscal,dz20);
 331
 332             /* Update vectorial force */
 333             fix2             = _mm_add_ps(fix2,tx);
 334             fiy2             = _mm_add_ps(fiy2,ty);
 335             fiz2             = _mm_add_ps(fiz2,tz);
 336
 337             fjx0             = _mm_add_ps(fjx0,tx);
 338             fjy0             = _mm_add_ps(fjy0,ty);
 339             fjz0             = _mm_add_ps(fjz0,tz);
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq30             = _mm_mul_ps(iq3,jq0);
 347
 348             /* COULOMB ELECTROSTATICS */
 349             velec            = _mm_mul_ps(qq30,rinv30);
 350             felec            = _mm_mul_ps(velec,rinvsq30);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velecsum         = _mm_add_ps(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm_mul_ps(fscal,dx30);
 359             ty               = _mm_mul_ps(fscal,dy30);
 360             tz               = _mm_mul_ps(fscal,dz30);
 361
 362             /* Update vectorial force */
 363             fix3             = _mm_add_ps(fix3,tx);
 364             fiy3             = _mm_add_ps(fiy3,ty);
 365             fiz3             = _mm_add_ps(fiz3,tz);
 366
 367             fjx0             = _mm_add_ps(fjx0,tx);
 368             fjy0             = _mm_add_ps(fjy0,ty);
 369             fjz0             = _mm_add_ps(fjz0,tz);
 370
 371             fjptrA             = f+j_coord_offsetA;
 372             fjptrB             = f+j_coord_offsetB;
 373             fjptrC             = f+j_coord_offsetC;
 374             fjptrD             = f+j_coord_offsetD;
 375
 376             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 377
 378             /* Inner loop uses 116 flops */
 379         }
 380
 381         if(jidx<j_index_end)
 382         {
 383
 384             /* Get j neighbor index, and coordinate index */
 385             jnrlistA         = jjnr[jidx];
 386             jnrlistB         = jjnr[jidx+1];
 387             jnrlistC         = jjnr[jidx+2];
 388             jnrlistD         = jjnr[jidx+3];
 389             /* Sign of each element will be negative for non-real atoms.
 390              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 391              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 392              */
 393             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 394             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 395             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 396             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 397             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 398             j_coord_offsetA  = DIM*jnrA;
 399             j_coord_offsetB  = DIM*jnrB;
 400             j_coord_offsetC  = DIM*jnrC;
 401             j_coord_offsetD  = DIM*jnrD;
 402
 403             /* load j atom coordinates */
 404             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 405                                               x+j_coord_offsetC,x+j_coord_offsetD,
 406                                               &jx0,&jy0,&jz0);
 407
 408             /* Calculate displacement vector */
 409             dx00             = _mm_sub_ps(ix0,jx0);
 410             dy00             = _mm_sub_ps(iy0,jy0);
 411             dz00             = _mm_sub_ps(iz0,jz0);
 412             dx10             = _mm_sub_ps(ix1,jx0);
 413             dy10             = _mm_sub_ps(iy1,jy0);
 414             dz10             = _mm_sub_ps(iz1,jz0);
 415             dx20             = _mm_sub_ps(ix2,jx0);
 416             dy20             = _mm_sub_ps(iy2,jy0);
 417             dz20             = _mm_sub_ps(iz2,jz0);
 418             dx30             = _mm_sub_ps(ix3,jx0);
 419             dy30             = _mm_sub_ps(iy3,jy0);
 420             dz30             = _mm_sub_ps(iz3,jz0);
 421
 422             /* Calculate squared distance and things based on it */
 423             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 424             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 425             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 426             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 427
 428             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 429             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 430             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 431
 432             rinvsq00         = gmx_mm_inv_ps(rsq00);
 433             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 434             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 435             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 436
 437             /* Load parameters for j particles */
 438             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 439                                                               charge+jnrC+0,charge+jnrD+0);
 440             vdwjidx0A        = 2*vdwtype[jnrA+0];
 441             vdwjidx0B        = 2*vdwtype[jnrB+0];
 442             vdwjidx0C        = 2*vdwtype[jnrC+0];
 443             vdwjidx0D        = 2*vdwtype[jnrD+0];
 444
 445             fjx0             = _mm_setzero_ps();
 446             fjy0             = _mm_setzero_ps();
 447             fjz0             = _mm_setzero_ps();
 448
 449             /**************************
 450              * CALCULATE INTERACTIONS *
 451              **************************/
 452
 453             /* Compute parameters for interactions between i and j atoms */
 454             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 455                                          vdwparam+vdwioffset0+vdwjidx0B,
 456                                          vdwparam+vdwioffset0+vdwjidx0C,
 457                                          vdwparam+vdwioffset0+vdwjidx0D,
 458                                          &c6_00,&c12_00);
 459
 460             /* LENNARD-JONES DISPERSION/REPULSION */
 461
 462             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 463             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 464             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 465             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 466             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 467
 468             /* Update potential sum for this i atom from the interaction with this j atom. */
 469             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 470             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 471
 472             fscal            = fvdw;
 473
 474             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 475
 476             /* Calculate temporary vectorial force */
 477             tx               = _mm_mul_ps(fscal,dx00);
 478             ty               = _mm_mul_ps(fscal,dy00);
 479             tz               = _mm_mul_ps(fscal,dz00);
 480
 481             /* Update vectorial force */
 482             fix0             = _mm_add_ps(fix0,tx);
 483             fiy0             = _mm_add_ps(fiy0,ty);
 484             fiz0             = _mm_add_ps(fiz0,tz);
 485
 486             fjx0             = _mm_add_ps(fjx0,tx);
 487             fjy0             = _mm_add_ps(fjy0,ty);
 488             fjz0             = _mm_add_ps(fjz0,tz);
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq10             = _mm_mul_ps(iq1,jq0);
 496
 497             /* COULOMB ELECTROSTATICS */
 498             velec            = _mm_mul_ps(qq10,rinv10);
 499             felec            = _mm_mul_ps(velec,rinvsq10);
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velec            = _mm_andnot_ps(dummy_mask,velec);
 503             velecsum         = _mm_add_ps(velecsum,velec);
 504
 505             fscal            = felec;
 506
 507             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_ps(fscal,dx10);
 511             ty               = _mm_mul_ps(fscal,dy10);
 512             tz               = _mm_mul_ps(fscal,dz10);
 513
 514             /* Update vectorial force */
 515             fix1             = _mm_add_ps(fix1,tx);
 516             fiy1             = _mm_add_ps(fiy1,ty);
 517             fiz1             = _mm_add_ps(fiz1,tz);
 518
 519             fjx0             = _mm_add_ps(fjx0,tx);
 520             fjy0             = _mm_add_ps(fjy0,ty);
 521             fjz0             = _mm_add_ps(fjz0,tz);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             /* Compute parameters for interactions between i and j atoms */
 528             qq20             = _mm_mul_ps(iq2,jq0);
 529
 530             /* COULOMB ELECTROSTATICS */
 531             velec            = _mm_mul_ps(qq20,rinv20);
 532             felec            = _mm_mul_ps(velec,rinvsq20);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_andnot_ps(dummy_mask,velec);
 536             velecsum         = _mm_add_ps(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx20);
 544             ty               = _mm_mul_ps(fscal,dy20);
 545             tz               = _mm_mul_ps(fscal,dz20);
 546
 547             /* Update vectorial force */
 548             fix2             = _mm_add_ps(fix2,tx);
 549             fiy2             = _mm_add_ps(fiy2,ty);
 550             fiz2             = _mm_add_ps(fiz2,tz);
 551
 552             fjx0             = _mm_add_ps(fjx0,tx);
 553             fjy0             = _mm_add_ps(fjy0,ty);
 554             fjz0             = _mm_add_ps(fjz0,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             qq30             = _mm_mul_ps(iq3,jq0);
 562
 563             /* COULOMB ELECTROSTATICS */
 564             velec            = _mm_mul_ps(qq30,rinv30);
 565             felec            = _mm_mul_ps(velec,rinvsq30);
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm_andnot_ps(dummy_mask,velec);
 569             velecsum         = _mm_add_ps(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_ps(fscal,dx30);
 577             ty               = _mm_mul_ps(fscal,dy30);
 578             tz               = _mm_mul_ps(fscal,dz30);
 579
 580             /* Update vectorial force */
 581             fix3             = _mm_add_ps(fix3,tx);
 582             fiy3             = _mm_add_ps(fiy3,ty);
 583             fiz3             = _mm_add_ps(fiz3,tz);
 584
 585             fjx0             = _mm_add_ps(fjx0,tx);
 586             fjy0             = _mm_add_ps(fjy0,ty);
 587             fjz0             = _mm_add_ps(fjz0,tz);
 588
 589             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 590             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 591             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 592             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 593
 594             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 595
 596             /* Inner loop uses 116 flops */
 597         }
 598
 599         /* End of innermost loop */
 600
 601         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 602                                               f+i_coord_offset,fshift+i_shift_offset);
 603
 604         ggid                        = gid[iidx];
 605         /* Update potential energies */
 606         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 607         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 608
 609         /* Increment number of inner iterations */
 610         inneriter                  += j_index_end - j_index_start;
 611
 612         /* Outer loop uses 26 flops */
 613     }
 614
 615     /* Increment number of outer iterations */
 616     outeriter        += nri;
 617
 618     /* Update outer/inner flops */
 619
 620     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 621 }
 622 /*
 623  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse4_1_single
 624  * Electrostatics interaction: Coulomb
 625  * VdW interaction:            LennardJones
 626  * Geometry:                   Water4-Particle
 627  * Calculate force/pot:        Force
 628  */
 629 void
 630 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse4_1_single
 631                     (t_nblist                    * gmx_restrict       nlist,
 632                      rvec                        * gmx_restrict          xx,
 633                      rvec                        * gmx_restrict          ff,
 634                      t_forcerec                  * gmx_restrict          fr,
 635                      t_mdatoms                   * gmx_restrict     mdatoms,
 636                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 637                      t_nrnb                      * gmx_restrict        nrnb)
 638 {
 639     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 640      * just 0 for non-waters.
 641      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 642      * jnr indices corresponding to data put in the four positions in the SIMD register.
 643      */
 644     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 645     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 646     int              jnrA,jnrB,jnrC,jnrD;
 647     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 648     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 649     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 650     real             rcutoff_scalar;
 651     real             *shiftvec,*fshift,*x,*f;
 652     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 653     real             scratch[4*DIM];
 654     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 655     int              vdwioffset0;
 656     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 657     int              vdwioffset1;
 658     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 659     int              vdwioffset2;
 660     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 661     int              vdwioffset3;
 662     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 663     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 664     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 665     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 666     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 667     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 668     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 669     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 670     real             *charge;
 671     int              nvdwtype;
 672     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 673     int              *vdwtype;
 674     real             *vdwparam;
 675     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 676     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 677     __m128           dummy_mask,cutoff_mask;
 678     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 679     __m128           one     = _mm_set1_ps(1.0);
 680     __m128           two     = _mm_set1_ps(2.0);
 681     x                = xx[0];
 682     f                = ff[0];
 683
 684     nri              = nlist->nri;
 685     iinr             = nlist->iinr;
 686     jindex           = nlist->jindex;
 687     jjnr             = nlist->jjnr;
 688     shiftidx         = nlist->shift;
 689     gid              = nlist->gid;
 690     shiftvec         = fr->shift_vec[0];
 691     fshift           = fr->fshift[0];
 692     facel            = _mm_set1_ps(fr->epsfac);
 693     charge           = mdatoms->chargeA;
 694     nvdwtype         = fr->ntype;
 695     vdwparam         = fr->nbfp;
 696     vdwtype          = mdatoms->typeA;
 697
 698     /* Setup water-specific parameters */
 699     inr              = nlist->iinr[0];
 700     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 701     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 702     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 703     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 704
 705     /* Avoid stupid compiler warnings */
 706     jnrA = jnrB = jnrC = jnrD = 0;
 707     j_coord_offsetA = 0;
 708     j_coord_offsetB = 0;
 709     j_coord_offsetC = 0;
 710     j_coord_offsetD = 0;
 711
 712     outeriter        = 0;
 713     inneriter        = 0;
 714
 715     for(iidx=0;iidx<4*DIM;iidx++)
 716     {
 717         scratch[iidx] = 0.0;
 718     }
 719
 720     /* Start outer loop over neighborlists */
 721     for(iidx=0; iidx<nri; iidx++)
 722     {
 723         /* Load shift vector for this list */
 724         i_shift_offset   = DIM*shiftidx[iidx];
 725
 726         /* Load limits for loop over neighbors */
 727         j_index_start    = jindex[iidx];
 728         j_index_end      = jindex[iidx+1];
 729
 730         /* Get outer coordinate index */
 731         inr              = iinr[iidx];
 732         i_coord_offset   = DIM*inr;
 733
 734         /* Load i particle coords and add shift vector */
 735         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 736                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 737
 738         fix0             = _mm_setzero_ps();
 739         fiy0             = _mm_setzero_ps();
 740         fiz0             = _mm_setzero_ps();
 741         fix1             = _mm_setzero_ps();
 742         fiy1             = _mm_setzero_ps();
 743         fiz1             = _mm_setzero_ps();
 744         fix2             = _mm_setzero_ps();
 745         fiy2             = _mm_setzero_ps();
 746         fiz2             = _mm_setzero_ps();
 747         fix3             = _mm_setzero_ps();
 748         fiy3             = _mm_setzero_ps();
 749         fiz3             = _mm_setzero_ps();
 750
 751         /* Start inner kernel loop */
 752         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 753         {
 754
 755             /* Get j neighbor index, and coordinate index */
 756             jnrA             = jjnr[jidx];
 757             jnrB             = jjnr[jidx+1];
 758             jnrC             = jjnr[jidx+2];
 759             jnrD             = jjnr[jidx+3];
 760             j_coord_offsetA  = DIM*jnrA;
 761             j_coord_offsetB  = DIM*jnrB;
 762             j_coord_offsetC  = DIM*jnrC;
 763             j_coord_offsetD  = DIM*jnrD;
 764
 765             /* load j atom coordinates */
 766             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 767                                               x+j_coord_offsetC,x+j_coord_offsetD,
 768                                               &jx0,&jy0,&jz0);
 769
 770             /* Calculate displacement vector */
 771             dx00             = _mm_sub_ps(ix0,jx0);
 772             dy00             = _mm_sub_ps(iy0,jy0);
 773             dz00             = _mm_sub_ps(iz0,jz0);
 774             dx10             = _mm_sub_ps(ix1,jx0);
 775             dy10             = _mm_sub_ps(iy1,jy0);
 776             dz10             = _mm_sub_ps(iz1,jz0);
 777             dx20             = _mm_sub_ps(ix2,jx0);
 778             dy20             = _mm_sub_ps(iy2,jy0);
 779             dz20             = _mm_sub_ps(iz2,jz0);
 780             dx30             = _mm_sub_ps(ix3,jx0);
 781             dy30             = _mm_sub_ps(iy3,jy0);
 782             dz30             = _mm_sub_ps(iz3,jz0);
 783
 784             /* Calculate squared distance and things based on it */
 785             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 786             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 787             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 788             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 789
 790             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 791             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 792             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 793
 794             rinvsq00         = gmx_mm_inv_ps(rsq00);
 795             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 796             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 797             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 798
 799             /* Load parameters for j particles */
 800             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 801                                                               charge+jnrC+0,charge+jnrD+0);
 802             vdwjidx0A        = 2*vdwtype[jnrA+0];
 803             vdwjidx0B        = 2*vdwtype[jnrB+0];
 804             vdwjidx0C        = 2*vdwtype[jnrC+0];
 805             vdwjidx0D        = 2*vdwtype[jnrD+0];
 806
 807             fjx0             = _mm_setzero_ps();
 808             fjy0             = _mm_setzero_ps();
 809             fjz0             = _mm_setzero_ps();
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 817                                          vdwparam+vdwioffset0+vdwjidx0B,
 818                                          vdwparam+vdwioffset0+vdwjidx0C,
 819                                          vdwparam+vdwioffset0+vdwjidx0D,
 820                                          &c6_00,&c12_00);
 821
 822             /* LENNARD-JONES DISPERSION/REPULSION */
 823
 824             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 825             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 826
 827             fscal            = fvdw;
 828
 829             /* Calculate temporary vectorial force */
 830             tx               = _mm_mul_ps(fscal,dx00);
 831             ty               = _mm_mul_ps(fscal,dy00);
 832             tz               = _mm_mul_ps(fscal,dz00);
 833
 834             /* Update vectorial force */
 835             fix0             = _mm_add_ps(fix0,tx);
 836             fiy0             = _mm_add_ps(fiy0,ty);
 837             fiz0             = _mm_add_ps(fiz0,tz);
 838
 839             fjx0             = _mm_add_ps(fjx0,tx);
 840             fjy0             = _mm_add_ps(fjy0,ty);
 841             fjz0             = _mm_add_ps(fjz0,tz);
 842
 843             /**************************
 844              * CALCULATE INTERACTIONS *
 845              **************************/
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             qq10             = _mm_mul_ps(iq1,jq0);
 849
 850             /* COULOMB ELECTROSTATICS */
 851             velec            = _mm_mul_ps(qq10,rinv10);
 852             felec            = _mm_mul_ps(velec,rinvsq10);
 853
 854             fscal            = felec;
 855
 856             /* Calculate temporary vectorial force */
 857             tx               = _mm_mul_ps(fscal,dx10);
 858             ty               = _mm_mul_ps(fscal,dy10);
 859             tz               = _mm_mul_ps(fscal,dz10);
 860
 861             /* Update vectorial force */
 862             fix1             = _mm_add_ps(fix1,tx);
 863             fiy1             = _mm_add_ps(fiy1,ty);
 864             fiz1             = _mm_add_ps(fiz1,tz);
 865
 866             fjx0             = _mm_add_ps(fjx0,tx);
 867             fjy0             = _mm_add_ps(fjy0,ty);
 868             fjz0             = _mm_add_ps(fjz0,tz);
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq20             = _mm_mul_ps(iq2,jq0);
 876
 877             /* COULOMB ELECTROSTATICS */
 878             velec            = _mm_mul_ps(qq20,rinv20);
 879             felec            = _mm_mul_ps(velec,rinvsq20);
 880
 881             fscal            = felec;
 882
 883             /* Calculate temporary vectorial force */
 884             tx               = _mm_mul_ps(fscal,dx20);
 885             ty               = _mm_mul_ps(fscal,dy20);
 886             tz               = _mm_mul_ps(fscal,dz20);
 887
 888             /* Update vectorial force */
 889             fix2             = _mm_add_ps(fix2,tx);
 890             fiy2             = _mm_add_ps(fiy2,ty);
 891             fiz2             = _mm_add_ps(fiz2,tz);
 892
 893             fjx0             = _mm_add_ps(fjx0,tx);
 894             fjy0             = _mm_add_ps(fjy0,ty);
 895             fjz0             = _mm_add_ps(fjz0,tz);
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq30             = _mm_mul_ps(iq3,jq0);
 903
 904             /* COULOMB ELECTROSTATICS */
 905             velec            = _mm_mul_ps(qq30,rinv30);
 906             felec            = _mm_mul_ps(velec,rinvsq30);
 907
 908             fscal            = felec;
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm_mul_ps(fscal,dx30);
 912             ty               = _mm_mul_ps(fscal,dy30);
 913             tz               = _mm_mul_ps(fscal,dz30);
 914
 915             /* Update vectorial force */
 916             fix3             = _mm_add_ps(fix3,tx);
 917             fiy3             = _mm_add_ps(fiy3,ty);
 918             fiz3             = _mm_add_ps(fiz3,tz);
 919
 920             fjx0             = _mm_add_ps(fjx0,tx);
 921             fjy0             = _mm_add_ps(fjy0,ty);
 922             fjz0             = _mm_add_ps(fjz0,tz);
 923
 924             fjptrA             = f+j_coord_offsetA;
 925             fjptrB             = f+j_coord_offsetB;
 926             fjptrC             = f+j_coord_offsetC;
 927             fjptrD             = f+j_coord_offsetD;
 928
 929             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 930
 931             /* Inner loop uses 108 flops */
 932         }
 933
 934         if(jidx<j_index_end)
 935         {
 936
 937             /* Get j neighbor index, and coordinate index */
 938             jnrlistA         = jjnr[jidx];
 939             jnrlistB         = jjnr[jidx+1];
 940             jnrlistC         = jjnr[jidx+2];
 941             jnrlistD         = jjnr[jidx+3];
 942             /* Sign of each element will be negative for non-real atoms.
 943              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 944              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 945              */
 946             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 947             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 948             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 949             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 950             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 951             j_coord_offsetA  = DIM*jnrA;
 952             j_coord_offsetB  = DIM*jnrB;
 953             j_coord_offsetC  = DIM*jnrC;
 954             j_coord_offsetD  = DIM*jnrD;
 955
 956             /* load j atom coordinates */
 957             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 958                                               x+j_coord_offsetC,x+j_coord_offsetD,
 959                                               &jx0,&jy0,&jz0);
 960
 961             /* Calculate displacement vector */
 962             dx00             = _mm_sub_ps(ix0,jx0);
 963             dy00             = _mm_sub_ps(iy0,jy0);
 964             dz00             = _mm_sub_ps(iz0,jz0);
 965             dx10             = _mm_sub_ps(ix1,jx0);
 966             dy10             = _mm_sub_ps(iy1,jy0);
 967             dz10             = _mm_sub_ps(iz1,jz0);
 968             dx20             = _mm_sub_ps(ix2,jx0);
 969             dy20             = _mm_sub_ps(iy2,jy0);
 970             dz20             = _mm_sub_ps(iz2,jz0);
 971             dx30             = _mm_sub_ps(ix3,jx0);
 972             dy30             = _mm_sub_ps(iy3,jy0);
 973             dz30             = _mm_sub_ps(iz3,jz0);
 974
 975             /* Calculate squared distance and things based on it */
 976             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 977             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 978             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 979             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 980
 981             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 982             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 983             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 984
 985             rinvsq00         = gmx_mm_inv_ps(rsq00);
 986             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 987             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 988             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 989
 990             /* Load parameters for j particles */
 991             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 992                                                               charge+jnrC+0,charge+jnrD+0);
 993             vdwjidx0A        = 2*vdwtype[jnrA+0];
 994             vdwjidx0B        = 2*vdwtype[jnrB+0];
 995             vdwjidx0C        = 2*vdwtype[jnrC+0];
 996             vdwjidx0D        = 2*vdwtype[jnrD+0];
 997
 998             fjx0             = _mm_setzero_ps();
 999             fjy0             = _mm_setzero_ps();
1000             fjz0             = _mm_setzero_ps();
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1008                                          vdwparam+vdwioffset0+vdwjidx0B,
1009                                          vdwparam+vdwioffset0+vdwjidx0C,
1010                                          vdwparam+vdwioffset0+vdwjidx0D,
1011                                          &c6_00,&c12_00);
1012
1013             /* LENNARD-JONES DISPERSION/REPULSION */
1014
1015             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1016             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1017
1018             fscal            = fvdw;
1019
1020             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm_mul_ps(fscal,dx00);
1024             ty               = _mm_mul_ps(fscal,dy00);
1025             tz               = _mm_mul_ps(fscal,dz00);
1026
1027             /* Update vectorial force */
1028             fix0             = _mm_add_ps(fix0,tx);
1029             fiy0             = _mm_add_ps(fiy0,ty);
1030             fiz0             = _mm_add_ps(fiz0,tz);
1031
1032             fjx0             = _mm_add_ps(fjx0,tx);
1033             fjy0             = _mm_add_ps(fjy0,ty);
1034             fjz0             = _mm_add_ps(fjz0,tz);
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             /* Compute parameters for interactions between i and j atoms */
1041             qq10             = _mm_mul_ps(iq1,jq0);
1042
1043             /* COULOMB ELECTROSTATICS */
1044             velec            = _mm_mul_ps(qq10,rinv10);
1045             felec            = _mm_mul_ps(velec,rinvsq10);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm_mul_ps(fscal,dx10);
1053             ty               = _mm_mul_ps(fscal,dy10);
1054             tz               = _mm_mul_ps(fscal,dz10);
1055
1056             /* Update vectorial force */
1057             fix1             = _mm_add_ps(fix1,tx);
1058             fiy1             = _mm_add_ps(fiy1,ty);
1059             fiz1             = _mm_add_ps(fiz1,tz);
1060
1061             fjx0             = _mm_add_ps(fjx0,tx);
1062             fjy0             = _mm_add_ps(fjy0,ty);
1063             fjz0             = _mm_add_ps(fjz0,tz);
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq20             = _mm_mul_ps(iq2,jq0);
1071
1072             /* COULOMB ELECTROSTATICS */
1073             velec            = _mm_mul_ps(qq20,rinv20);
1074             felec            = _mm_mul_ps(velec,rinvsq20);
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm_mul_ps(fscal,dx20);
1082             ty               = _mm_mul_ps(fscal,dy20);
1083             tz               = _mm_mul_ps(fscal,dz20);
1084
1085             /* Update vectorial force */
1086             fix2             = _mm_add_ps(fix2,tx);
1087             fiy2             = _mm_add_ps(fiy2,ty);
1088             fiz2             = _mm_add_ps(fiz2,tz);
1089
1090             fjx0             = _mm_add_ps(fjx0,tx);
1091             fjy0             = _mm_add_ps(fjy0,ty);
1092             fjz0             = _mm_add_ps(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq30             = _mm_mul_ps(iq3,jq0);
1100
1101             /* COULOMB ELECTROSTATICS */
1102             velec            = _mm_mul_ps(qq30,rinv30);
1103             felec            = _mm_mul_ps(velec,rinvsq30);
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm_mul_ps(fscal,dx30);
1111             ty               = _mm_mul_ps(fscal,dy30);
1112             tz               = _mm_mul_ps(fscal,dz30);
1113
1114             /* Update vectorial force */
1115             fix3             = _mm_add_ps(fix3,tx);
1116             fiy3             = _mm_add_ps(fiy3,ty);
1117             fiz3             = _mm_add_ps(fiz3,tz);
1118
1119             fjx0             = _mm_add_ps(fjx0,tx);
1120             fjy0             = _mm_add_ps(fjy0,ty);
1121             fjz0             = _mm_add_ps(fjz0,tz);
1122
1123             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1124             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1125             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1126             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1127
1128             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1129
1130             /* Inner loop uses 108 flops */
1131         }
1132
1133         /* End of innermost loop */
1134
1135         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1136                                               f+i_coord_offset,fshift+i_shift_offset);
1137
1138         /* Increment number of inner iterations */
1139         inneriter                  += j_index_end - j_index_start;
1140
1141         /* Outer loop uses 24 flops */
1142     }
1143
1144     /* Increment number of outer iterations */
1145     outeriter        += nri;
1146
1147     /* Update outer/inner flops */
1148
1149     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1150 }