src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwLJ_GeomW4P1_sse2_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 sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse2_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse2_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 123     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_ps(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 132     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 133     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 167                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 168
 169         fix0             = _mm_setzero_ps();
 170         fiy0             = _mm_setzero_ps();
 171         fiz0             = _mm_setzero_ps();
 172         fix1             = _mm_setzero_ps();
 173         fiy1             = _mm_setzero_ps();
 174         fiz1             = _mm_setzero_ps();
 175         fix2             = _mm_setzero_ps();
 176         fiy2             = _mm_setzero_ps();
 177         fiz2             = _mm_setzero_ps();
 178         fix3             = _mm_setzero_ps();
 179         fiy3             = _mm_setzero_ps();
 180         fiz3             = _mm_setzero_ps();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184         vvdwsum          = _mm_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               x+j_coord_offsetC,x+j_coord_offsetD,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_ps(ix0,jx0);
 207             dy00             = _mm_sub_ps(iy0,jy0);
 208             dz00             = _mm_sub_ps(iz0,jz0);
 209             dx10             = _mm_sub_ps(ix1,jx0);
 210             dy10             = _mm_sub_ps(iy1,jy0);
 211             dz10             = _mm_sub_ps(iz1,jz0);
 212             dx20             = _mm_sub_ps(ix2,jx0);
 213             dy20             = _mm_sub_ps(iy2,jy0);
 214             dz20             = _mm_sub_ps(iz2,jz0);
 215             dx30             = _mm_sub_ps(ix3,jx0);
 216             dy30             = _mm_sub_ps(iy3,jy0);
 217             dz30             = _mm_sub_ps(iz3,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 221             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 222             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 223             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 224
 225             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 226             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 227             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 228
 229             rinvsq00         = gmx_mm_inv_ps(rsq00);
 230             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 231             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 232             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                               charge+jnrC+0,charge+jnrD+0);
 237             vdwjidx0A        = 2*vdwtype[jnrA+0];
 238             vdwjidx0B        = 2*vdwtype[jnrB+0];
 239             vdwjidx0C        = 2*vdwtype[jnrC+0];
 240             vdwjidx0D        = 2*vdwtype[jnrD+0];
 241
 242             fjx0             = _mm_setzero_ps();
 243             fjy0             = _mm_setzero_ps();
 244             fjz0             = _mm_setzero_ps();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* LENNARD-JONES DISPERSION/REPULSION */
 258
 259             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 260             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 261             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 262             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 263             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 267
 268             fscal            = fvdw;
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm_mul_ps(fscal,dx00);
 272             ty               = _mm_mul_ps(fscal,dy00);
 273             tz               = _mm_mul_ps(fscal,dz00);
 274
 275             /* Update vectorial force */
 276             fix0             = _mm_add_ps(fix0,tx);
 277             fiy0             = _mm_add_ps(fiy0,ty);
 278             fiz0             = _mm_add_ps(fiz0,tz);
 279
 280             fjx0             = _mm_add_ps(fjx0,tx);
 281             fjy0             = _mm_add_ps(fjy0,ty);
 282             fjz0             = _mm_add_ps(fjz0,tz);
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq10             = _mm_mul_ps(iq1,jq0);
 290
 291             /* REACTION-FIELD ELECTROSTATICS */
 292             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 293             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_ps(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_ps(fscal,dx10);
 302             ty               = _mm_mul_ps(fscal,dy10);
 303             tz               = _mm_mul_ps(fscal,dz10);
 304
 305             /* Update vectorial force */
 306             fix1             = _mm_add_ps(fix1,tx);
 307             fiy1             = _mm_add_ps(fiy1,ty);
 308             fiz1             = _mm_add_ps(fiz1,tz);
 309
 310             fjx0             = _mm_add_ps(fjx0,tx);
 311             fjy0             = _mm_add_ps(fjy0,ty);
 312             fjz0             = _mm_add_ps(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq20             = _mm_mul_ps(iq2,jq0);
 320
 321             /* REACTION-FIELD ELECTROSTATICS */
 322             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 323             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velecsum         = _mm_add_ps(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_ps(fscal,dx20);
 332             ty               = _mm_mul_ps(fscal,dy20);
 333             tz               = _mm_mul_ps(fscal,dz20);
 334
 335             /* Update vectorial force */
 336             fix2             = _mm_add_ps(fix2,tx);
 337             fiy2             = _mm_add_ps(fiy2,ty);
 338             fiz2             = _mm_add_ps(fiz2,tz);
 339
 340             fjx0             = _mm_add_ps(fjx0,tx);
 341             fjy0             = _mm_add_ps(fjy0,ty);
 342             fjz0             = _mm_add_ps(fjz0,tz);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq30             = _mm_mul_ps(iq3,jq0);
 350
 351             /* REACTION-FIELD ELECTROSTATICS */
 352             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 353             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm_mul_ps(fscal,dx30);
 362             ty               = _mm_mul_ps(fscal,dy30);
 363             tz               = _mm_mul_ps(fscal,dz30);
 364
 365             /* Update vectorial force */
 366             fix3             = _mm_add_ps(fix3,tx);
 367             fiy3             = _mm_add_ps(fiy3,ty);
 368             fiz3             = _mm_add_ps(fiz3,tz);
 369
 370             fjx0             = _mm_add_ps(fjx0,tx);
 371             fjy0             = _mm_add_ps(fjy0,ty);
 372             fjz0             = _mm_add_ps(fjz0,tz);
 373
 374             fjptrA             = f+j_coord_offsetA;
 375             fjptrB             = f+j_coord_offsetB;
 376             fjptrC             = f+j_coord_offsetC;
 377             fjptrD             = f+j_coord_offsetD;
 378
 379             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 380
 381             /* Inner loop uses 128 flops */
 382         }
 383
 384         if(jidx<j_index_end)
 385         {
 386
 387             /* Get j neighbor index, and coordinate index */
 388             jnrlistA         = jjnr[jidx];
 389             jnrlistB         = jjnr[jidx+1];
 390             jnrlistC         = jjnr[jidx+2];
 391             jnrlistD         = jjnr[jidx+3];
 392             /* Sign of each element will be negative for non-real atoms.
 393              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 394              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 395              */
 396             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 397             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 398             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 399             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 400             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 401             j_coord_offsetA  = DIM*jnrA;
 402             j_coord_offsetB  = DIM*jnrB;
 403             j_coord_offsetC  = DIM*jnrC;
 404             j_coord_offsetD  = DIM*jnrD;
 405
 406             /* load j atom coordinates */
 407             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 408                                               x+j_coord_offsetC,x+j_coord_offsetD,
 409                                               &jx0,&jy0,&jz0);
 410
 411             /* Calculate displacement vector */
 412             dx00             = _mm_sub_ps(ix0,jx0);
 413             dy00             = _mm_sub_ps(iy0,jy0);
 414             dz00             = _mm_sub_ps(iz0,jz0);
 415             dx10             = _mm_sub_ps(ix1,jx0);
 416             dy10             = _mm_sub_ps(iy1,jy0);
 417             dz10             = _mm_sub_ps(iz1,jz0);
 418             dx20             = _mm_sub_ps(ix2,jx0);
 419             dy20             = _mm_sub_ps(iy2,jy0);
 420             dz20             = _mm_sub_ps(iz2,jz0);
 421             dx30             = _mm_sub_ps(ix3,jx0);
 422             dy30             = _mm_sub_ps(iy3,jy0);
 423             dz30             = _mm_sub_ps(iz3,jz0);
 424
 425             /* Calculate squared distance and things based on it */
 426             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 427             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 428             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 429             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 430
 431             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 432             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 433             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 434
 435             rinvsq00         = gmx_mm_inv_ps(rsq00);
 436             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 437             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 438             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 439
 440             /* Load parameters for j particles */
 441             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 442                                                               charge+jnrC+0,charge+jnrD+0);
 443             vdwjidx0A        = 2*vdwtype[jnrA+0];
 444             vdwjidx0B        = 2*vdwtype[jnrB+0];
 445             vdwjidx0C        = 2*vdwtype[jnrC+0];
 446             vdwjidx0D        = 2*vdwtype[jnrD+0];
 447
 448             fjx0             = _mm_setzero_ps();
 449             fjy0             = _mm_setzero_ps();
 450             fjz0             = _mm_setzero_ps();
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             /* Compute parameters for interactions between i and j atoms */
 457             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 458                                          vdwparam+vdwioffset0+vdwjidx0B,
 459                                          vdwparam+vdwioffset0+vdwjidx0C,
 460                                          vdwparam+vdwioffset0+vdwjidx0D,
 461                                          &c6_00,&c12_00);
 462
 463             /* LENNARD-JONES DISPERSION/REPULSION */
 464
 465             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 466             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 467             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 468             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 469             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 473             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 474
 475             fscal            = fvdw;
 476
 477             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 478
 479             /* Calculate temporary vectorial force */
 480             tx               = _mm_mul_ps(fscal,dx00);
 481             ty               = _mm_mul_ps(fscal,dy00);
 482             tz               = _mm_mul_ps(fscal,dz00);
 483
 484             /* Update vectorial force */
 485             fix0             = _mm_add_ps(fix0,tx);
 486             fiy0             = _mm_add_ps(fiy0,ty);
 487             fiz0             = _mm_add_ps(fiz0,tz);
 488
 489             fjx0             = _mm_add_ps(fjx0,tx);
 490             fjy0             = _mm_add_ps(fjy0,ty);
 491             fjz0             = _mm_add_ps(fjz0,tz);
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq10             = _mm_mul_ps(iq1,jq0);
 499
 500             /* REACTION-FIELD ELECTROSTATICS */
 501             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 502             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm_andnot_ps(dummy_mask,velec);
 506             velecsum         = _mm_add_ps(velecsum,velec);
 507
 508             fscal            = felec;
 509
 510             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 511
 512             /* Calculate temporary vectorial force */
 513             tx               = _mm_mul_ps(fscal,dx10);
 514             ty               = _mm_mul_ps(fscal,dy10);
 515             tz               = _mm_mul_ps(fscal,dz10);
 516
 517             /* Update vectorial force */
 518             fix1             = _mm_add_ps(fix1,tx);
 519             fiy1             = _mm_add_ps(fiy1,ty);
 520             fiz1             = _mm_add_ps(fiz1,tz);
 521
 522             fjx0             = _mm_add_ps(fjx0,tx);
 523             fjy0             = _mm_add_ps(fjy0,ty);
 524             fjz0             = _mm_add_ps(fjz0,tz);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq20             = _mm_mul_ps(iq2,jq0);
 532
 533             /* REACTION-FIELD ELECTROSTATICS */
 534             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 535             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm_mul_ps(fscal,dx20);
 547             ty               = _mm_mul_ps(fscal,dy20);
 548             tz               = _mm_mul_ps(fscal,dz20);
 549
 550             /* Update vectorial force */
 551             fix2             = _mm_add_ps(fix2,tx);
 552             fiy2             = _mm_add_ps(fiy2,ty);
 553             fiz2             = _mm_add_ps(fiz2,tz);
 554
 555             fjx0             = _mm_add_ps(fjx0,tx);
 556             fjy0             = _mm_add_ps(fjy0,ty);
 557             fjz0             = _mm_add_ps(fjz0,tz);
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq30             = _mm_mul_ps(iq3,jq0);
 565
 566             /* REACTION-FIELD ELECTROSTATICS */
 567             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 568             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_andnot_ps(dummy_mask,velec);
 572             velecsum         = _mm_add_ps(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 577
 578             /* Calculate temporary vectorial force */
 579             tx               = _mm_mul_ps(fscal,dx30);
 580             ty               = _mm_mul_ps(fscal,dy30);
 581             tz               = _mm_mul_ps(fscal,dz30);
 582
 583             /* Update vectorial force */
 584             fix3             = _mm_add_ps(fix3,tx);
 585             fiy3             = _mm_add_ps(fiy3,ty);
 586             fiz3             = _mm_add_ps(fiz3,tz);
 587
 588             fjx0             = _mm_add_ps(fjx0,tx);
 589             fjy0             = _mm_add_ps(fjy0,ty);
 590             fjz0             = _mm_add_ps(fjz0,tz);
 591
 592             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 593             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 594             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 595             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 596
 597             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 598
 599             /* Inner loop uses 128 flops */
 600         }
 601
 602         /* End of innermost loop */
 603
 604         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 605                                               f+i_coord_offset,fshift+i_shift_offset);
 606
 607         ggid                        = gid[iidx];
 608         /* Update potential energies */
 609         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 610         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 611
 612         /* Increment number of inner iterations */
 613         inneriter                  += j_index_end - j_index_start;
 614
 615         /* Outer loop uses 26 flops */
 616     }
 617
 618     /* Increment number of outer iterations */
 619     outeriter        += nri;
 620
 621     /* Update outer/inner flops */
 622
 623     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*128);
 624 }
 625 /*
 626  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse2_single
 627  * Electrostatics interaction: ReactionField
 628  * VdW interaction:            LennardJones
 629  * Geometry:                   Water4-Particle
 630  * Calculate force/pot:        Force
 631  */
 632 void
 633 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse2_single
 634                     (t_nblist                    * gmx_restrict       nlist,
 635                      rvec                        * gmx_restrict          xx,
 636                      rvec                        * gmx_restrict          ff,
 637                      t_forcerec                  * gmx_restrict          fr,
 638                      t_mdatoms                   * gmx_restrict     mdatoms,
 639                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 640                      t_nrnb                      * gmx_restrict        nrnb)
 641 {
 642     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 643      * just 0 for non-waters.
 644      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 645      * jnr indices corresponding to data put in the four positions in the SIMD register.
 646      */
 647     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 648     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 649     int              jnrA,jnrB,jnrC,jnrD;
 650     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 651     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 652     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 653     real             rcutoff_scalar;
 654     real             *shiftvec,*fshift,*x,*f;
 655     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 656     real             scratch[4*DIM];
 657     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 658     int              vdwioffset0;
 659     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 660     int              vdwioffset1;
 661     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 662     int              vdwioffset2;
 663     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 664     int              vdwioffset3;
 665     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 666     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 667     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 668     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 669     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 670     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 671     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 672     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 673     real             *charge;
 674     int              nvdwtype;
 675     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 676     int              *vdwtype;
 677     real             *vdwparam;
 678     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 679     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 680     __m128           dummy_mask,cutoff_mask;
 681     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 682     __m128           one     = _mm_set1_ps(1.0);
 683     __m128           two     = _mm_set1_ps(2.0);
 684     x                = xx[0];
 685     f                = ff[0];
 686
 687     nri              = nlist->nri;
 688     iinr             = nlist->iinr;
 689     jindex           = nlist->jindex;
 690     jjnr             = nlist->jjnr;
 691     shiftidx         = nlist->shift;
 692     gid              = nlist->gid;
 693     shiftvec         = fr->shift_vec[0];
 694     fshift           = fr->fshift[0];
 695     facel            = _mm_set1_ps(fr->epsfac);
 696     charge           = mdatoms->chargeA;
 697     krf              = _mm_set1_ps(fr->ic->k_rf);
 698     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 699     crf              = _mm_set1_ps(fr->ic->c_rf);
 700     nvdwtype         = fr->ntype;
 701     vdwparam         = fr->nbfp;
 702     vdwtype          = mdatoms->typeA;
 703
 704     /* Setup water-specific parameters */
 705     inr              = nlist->iinr[0];
 706     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 707     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 708     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 709     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 710
 711     /* Avoid stupid compiler warnings */
 712     jnrA = jnrB = jnrC = jnrD = 0;
 713     j_coord_offsetA = 0;
 714     j_coord_offsetB = 0;
 715     j_coord_offsetC = 0;
 716     j_coord_offsetD = 0;
 717
 718     outeriter        = 0;
 719     inneriter        = 0;
 720
 721     for(iidx=0;iidx<4*DIM;iidx++)
 722     {
 723         scratch[iidx] = 0.0;
 724     }
 725
 726     /* Start outer loop over neighborlists */
 727     for(iidx=0; iidx<nri; iidx++)
 728     {
 729         /* Load shift vector for this list */
 730         i_shift_offset   = DIM*shiftidx[iidx];
 731
 732         /* Load limits for loop over neighbors */
 733         j_index_start    = jindex[iidx];
 734         j_index_end      = jindex[iidx+1];
 735
 736         /* Get outer coordinate index */
 737         inr              = iinr[iidx];
 738         i_coord_offset   = DIM*inr;
 739
 740         /* Load i particle coords and add shift vector */
 741         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 742                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 743
 744         fix0             = _mm_setzero_ps();
 745         fiy0             = _mm_setzero_ps();
 746         fiz0             = _mm_setzero_ps();
 747         fix1             = _mm_setzero_ps();
 748         fiy1             = _mm_setzero_ps();
 749         fiz1             = _mm_setzero_ps();
 750         fix2             = _mm_setzero_ps();
 751         fiy2             = _mm_setzero_ps();
 752         fiz2             = _mm_setzero_ps();
 753         fix3             = _mm_setzero_ps();
 754         fiy3             = _mm_setzero_ps();
 755         fiz3             = _mm_setzero_ps();
 756
 757         /* Start inner kernel loop */
 758         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 759         {
 760
 761             /* Get j neighbor index, and coordinate index */
 762             jnrA             = jjnr[jidx];
 763             jnrB             = jjnr[jidx+1];
 764             jnrC             = jjnr[jidx+2];
 765             jnrD             = jjnr[jidx+3];
 766             j_coord_offsetA  = DIM*jnrA;
 767             j_coord_offsetB  = DIM*jnrB;
 768             j_coord_offsetC  = DIM*jnrC;
 769             j_coord_offsetD  = DIM*jnrD;
 770
 771             /* load j atom coordinates */
 772             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 773                                               x+j_coord_offsetC,x+j_coord_offsetD,
 774                                               &jx0,&jy0,&jz0);
 775
 776             /* Calculate displacement vector */
 777             dx00             = _mm_sub_ps(ix0,jx0);
 778             dy00             = _mm_sub_ps(iy0,jy0);
 779             dz00             = _mm_sub_ps(iz0,jz0);
 780             dx10             = _mm_sub_ps(ix1,jx0);
 781             dy10             = _mm_sub_ps(iy1,jy0);
 782             dz10             = _mm_sub_ps(iz1,jz0);
 783             dx20             = _mm_sub_ps(ix2,jx0);
 784             dy20             = _mm_sub_ps(iy2,jy0);
 785             dz20             = _mm_sub_ps(iz2,jz0);
 786             dx30             = _mm_sub_ps(ix3,jx0);
 787             dy30             = _mm_sub_ps(iy3,jy0);
 788             dz30             = _mm_sub_ps(iz3,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 792             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 793             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 794             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 795
 796             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 797             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 798             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 799
 800             rinvsq00         = gmx_mm_inv_ps(rsq00);
 801             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 802             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 803             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 804
 805             /* Load parameters for j particles */
 806             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 807                                                               charge+jnrC+0,charge+jnrD+0);
 808             vdwjidx0A        = 2*vdwtype[jnrA+0];
 809             vdwjidx0B        = 2*vdwtype[jnrB+0];
 810             vdwjidx0C        = 2*vdwtype[jnrC+0];
 811             vdwjidx0D        = 2*vdwtype[jnrD+0];
 812
 813             fjx0             = _mm_setzero_ps();
 814             fjy0             = _mm_setzero_ps();
 815             fjz0             = _mm_setzero_ps();
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             /* Compute parameters for interactions between i and j atoms */
 822             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 823                                          vdwparam+vdwioffset0+vdwjidx0B,
 824                                          vdwparam+vdwioffset0+vdwjidx0C,
 825                                          vdwparam+vdwioffset0+vdwjidx0D,
 826                                          &c6_00,&c12_00);
 827
 828             /* LENNARD-JONES DISPERSION/REPULSION */
 829
 830             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 831             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 832
 833             fscal            = fvdw;
 834
 835             /* Calculate temporary vectorial force */
 836             tx               = _mm_mul_ps(fscal,dx00);
 837             ty               = _mm_mul_ps(fscal,dy00);
 838             tz               = _mm_mul_ps(fscal,dz00);
 839
 840             /* Update vectorial force */
 841             fix0             = _mm_add_ps(fix0,tx);
 842             fiy0             = _mm_add_ps(fiy0,ty);
 843             fiz0             = _mm_add_ps(fiz0,tz);
 844
 845             fjx0             = _mm_add_ps(fjx0,tx);
 846             fjy0             = _mm_add_ps(fjy0,ty);
 847             fjz0             = _mm_add_ps(fjz0,tz);
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq10             = _mm_mul_ps(iq1,jq0);
 855
 856             /* REACTION-FIELD ELECTROSTATICS */
 857             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 858
 859             fscal            = felec;
 860
 861             /* Calculate temporary vectorial force */
 862             tx               = _mm_mul_ps(fscal,dx10);
 863             ty               = _mm_mul_ps(fscal,dy10);
 864             tz               = _mm_mul_ps(fscal,dz10);
 865
 866             /* Update vectorial force */
 867             fix1             = _mm_add_ps(fix1,tx);
 868             fiy1             = _mm_add_ps(fiy1,ty);
 869             fiz1             = _mm_add_ps(fiz1,tz);
 870
 871             fjx0             = _mm_add_ps(fjx0,tx);
 872             fjy0             = _mm_add_ps(fjy0,ty);
 873             fjz0             = _mm_add_ps(fjz0,tz);
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             /* Compute parameters for interactions between i and j atoms */
 880             qq20             = _mm_mul_ps(iq2,jq0);
 881
 882             /* REACTION-FIELD ELECTROSTATICS */
 883             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 884
 885             fscal            = felec;
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm_mul_ps(fscal,dx20);
 889             ty               = _mm_mul_ps(fscal,dy20);
 890             tz               = _mm_mul_ps(fscal,dz20);
 891
 892             /* Update vectorial force */
 893             fix2             = _mm_add_ps(fix2,tx);
 894             fiy2             = _mm_add_ps(fiy2,ty);
 895             fiz2             = _mm_add_ps(fiz2,tz);
 896
 897             fjx0             = _mm_add_ps(fjx0,tx);
 898             fjy0             = _mm_add_ps(fjy0,ty);
 899             fjz0             = _mm_add_ps(fjz0,tz);
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             qq30             = _mm_mul_ps(iq3,jq0);
 907
 908             /* REACTION-FIELD ELECTROSTATICS */
 909             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 910
 911             fscal            = felec;
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_ps(fscal,dx30);
 915             ty               = _mm_mul_ps(fscal,dy30);
 916             tz               = _mm_mul_ps(fscal,dz30);
 917
 918             /* Update vectorial force */
 919             fix3             = _mm_add_ps(fix3,tx);
 920             fiy3             = _mm_add_ps(fiy3,ty);
 921             fiz3             = _mm_add_ps(fiz3,tz);
 922
 923             fjx0             = _mm_add_ps(fjx0,tx);
 924             fjy0             = _mm_add_ps(fjy0,ty);
 925             fjz0             = _mm_add_ps(fjz0,tz);
 926
 927             fjptrA             = f+j_coord_offsetA;
 928             fjptrB             = f+j_coord_offsetB;
 929             fjptrC             = f+j_coord_offsetC;
 930             fjptrD             = f+j_coord_offsetD;
 931
 932             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 933
 934             /* Inner loop uses 108 flops */
 935         }
 936
 937         if(jidx<j_index_end)
 938         {
 939
 940             /* Get j neighbor index, and coordinate index */
 941             jnrlistA         = jjnr[jidx];
 942             jnrlistB         = jjnr[jidx+1];
 943             jnrlistC         = jjnr[jidx+2];
 944             jnrlistD         = jjnr[jidx+3];
 945             /* Sign of each element will be negative for non-real atoms.
 946              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 947              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 948              */
 949             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 950             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 951             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 952             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 953             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 954             j_coord_offsetA  = DIM*jnrA;
 955             j_coord_offsetB  = DIM*jnrB;
 956             j_coord_offsetC  = DIM*jnrC;
 957             j_coord_offsetD  = DIM*jnrD;
 958
 959             /* load j atom coordinates */
 960             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 961                                               x+j_coord_offsetC,x+j_coord_offsetD,
 962                                               &jx0,&jy0,&jz0);
 963
 964             /* Calculate displacement vector */
 965             dx00             = _mm_sub_ps(ix0,jx0);
 966             dy00             = _mm_sub_ps(iy0,jy0);
 967             dz00             = _mm_sub_ps(iz0,jz0);
 968             dx10             = _mm_sub_ps(ix1,jx0);
 969             dy10             = _mm_sub_ps(iy1,jy0);
 970             dz10             = _mm_sub_ps(iz1,jz0);
 971             dx20             = _mm_sub_ps(ix2,jx0);
 972             dy20             = _mm_sub_ps(iy2,jy0);
 973             dz20             = _mm_sub_ps(iz2,jz0);
 974             dx30             = _mm_sub_ps(ix3,jx0);
 975             dy30             = _mm_sub_ps(iy3,jy0);
 976             dz30             = _mm_sub_ps(iz3,jz0);
 977
 978             /* Calculate squared distance and things based on it */
 979             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 980             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 981             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 982             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 983
 984             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 985             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 986             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 987
 988             rinvsq00         = gmx_mm_inv_ps(rsq00);
 989             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 990             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 991             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 992
 993             /* Load parameters for j particles */
 994             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 995                                                               charge+jnrC+0,charge+jnrD+0);
 996             vdwjidx0A        = 2*vdwtype[jnrA+0];
 997             vdwjidx0B        = 2*vdwtype[jnrB+0];
 998             vdwjidx0C        = 2*vdwtype[jnrC+0];
 999             vdwjidx0D        = 2*vdwtype[jnrD+0];
1000
1001             fjx0             = _mm_setzero_ps();
1002             fjy0             = _mm_setzero_ps();
1003             fjz0             = _mm_setzero_ps();
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1011                                          vdwparam+vdwioffset0+vdwjidx0B,
1012                                          vdwparam+vdwioffset0+vdwjidx0C,
1013                                          vdwparam+vdwioffset0+vdwjidx0D,
1014                                          &c6_00,&c12_00);
1015
1016             /* LENNARD-JONES DISPERSION/REPULSION */
1017
1018             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1019             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1020
1021             fscal            = fvdw;
1022
1023             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm_mul_ps(fscal,dx00);
1027             ty               = _mm_mul_ps(fscal,dy00);
1028             tz               = _mm_mul_ps(fscal,dz00);
1029
1030             /* Update vectorial force */
1031             fix0             = _mm_add_ps(fix0,tx);
1032             fiy0             = _mm_add_ps(fiy0,ty);
1033             fiz0             = _mm_add_ps(fiz0,tz);
1034
1035             fjx0             = _mm_add_ps(fjx0,tx);
1036             fjy0             = _mm_add_ps(fjy0,ty);
1037             fjz0             = _mm_add_ps(fjz0,tz);
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq10             = _mm_mul_ps(iq1,jq0);
1045
1046             /* REACTION-FIELD ELECTROSTATICS */
1047             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
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             /* REACTION-FIELD ELECTROSTATICS */
1075             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1076
1077             fscal            = felec;
1078
1079             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm_mul_ps(fscal,dx20);
1083             ty               = _mm_mul_ps(fscal,dy20);
1084             tz               = _mm_mul_ps(fscal,dz20);
1085
1086             /* Update vectorial force */
1087             fix2             = _mm_add_ps(fix2,tx);
1088             fiy2             = _mm_add_ps(fiy2,ty);
1089             fiz2             = _mm_add_ps(fiz2,tz);
1090
1091             fjx0             = _mm_add_ps(fjx0,tx);
1092             fjy0             = _mm_add_ps(fjy0,ty);
1093             fjz0             = _mm_add_ps(fjz0,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq30             = _mm_mul_ps(iq3,jq0);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
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 }