src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse4_1_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     krf              = _mm_set1_ps(fr->ic->k_rf);
 125     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 126     crf              = _mm_set1_ps(fr->ic->c_rf);
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix0             = _mm_setzero_ps();
 172         fiy0             = _mm_setzero_ps();
 173         fiz0             = _mm_setzero_ps();
 174         fix1             = _mm_setzero_ps();
 175         fiy1             = _mm_setzero_ps();
 176         fiz1             = _mm_setzero_ps();
 177         fix2             = _mm_setzero_ps();
 178         fiy2             = _mm_setzero_ps();
 179         fiz2             = _mm_setzero_ps();
 180         fix3             = _mm_setzero_ps();
 181         fiy3             = _mm_setzero_ps();
 182         fiz3             = _mm_setzero_ps();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_ps();
 186         vvdwsum          = _mm_setzero_ps();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211             dx10             = _mm_sub_ps(ix1,jx0);
 212             dy10             = _mm_sub_ps(iy1,jy0);
 213             dz10             = _mm_sub_ps(iz1,jz0);
 214             dx20             = _mm_sub_ps(ix2,jx0);
 215             dy20             = _mm_sub_ps(iy2,jy0);
 216             dz20             = _mm_sub_ps(iz2,jz0);
 217             dx30             = _mm_sub_ps(ix3,jx0);
 218             dy30             = _mm_sub_ps(iy3,jy0);
 219             dz30             = _mm_sub_ps(iz3,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 225             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 226
 227             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 228             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 229             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 230
 231             rinvsq00         = gmx_mm_inv_ps(rsq00);
 232             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 234             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 238                                                               charge+jnrC+0,charge+jnrD+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243
 244             fjx0             = _mm_setzero_ps();
 245             fjy0             = _mm_setzero_ps();
 246             fjz0             = _mm_setzero_ps();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 254                                          vdwparam+vdwioffset0+vdwjidx0B,
 255                                          vdwparam+vdwioffset0+vdwjidx0C,
 256                                          vdwparam+vdwioffset0+vdwjidx0D,
 257                                          &c6_00,&c12_00);
 258
 259             /* LENNARD-JONES DISPERSION/REPULSION */
 260
 261             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 262             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 263             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 264             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 265             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 269
 270             fscal            = fvdw;
 271
 272             /* Calculate temporary vectorial force */
 273             tx               = _mm_mul_ps(fscal,dx00);
 274             ty               = _mm_mul_ps(fscal,dy00);
 275             tz               = _mm_mul_ps(fscal,dz00);
 276
 277             /* Update vectorial force */
 278             fix0             = _mm_add_ps(fix0,tx);
 279             fiy0             = _mm_add_ps(fiy0,ty);
 280             fiz0             = _mm_add_ps(fiz0,tz);
 281
 282             fjx0             = _mm_add_ps(fjx0,tx);
 283             fjy0             = _mm_add_ps(fjy0,ty);
 284             fjz0             = _mm_add_ps(fjz0,tz);
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq10             = _mm_mul_ps(iq1,jq0);
 292
 293             /* REACTION-FIELD ELECTROSTATICS */
 294             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 295             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_ps(fscal,dx10);
 304             ty               = _mm_mul_ps(fscal,dy10);
 305             tz               = _mm_mul_ps(fscal,dz10);
 306
 307             /* Update vectorial force */
 308             fix1             = _mm_add_ps(fix1,tx);
 309             fiy1             = _mm_add_ps(fiy1,ty);
 310             fiz1             = _mm_add_ps(fiz1,tz);
 311
 312             fjx0             = _mm_add_ps(fjx0,tx);
 313             fjy0             = _mm_add_ps(fjy0,ty);
 314             fjz0             = _mm_add_ps(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq20             = _mm_mul_ps(iq2,jq0);
 322
 323             /* REACTION-FIELD ELECTROSTATICS */
 324             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 325             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm_add_ps(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_ps(fscal,dx20);
 334             ty               = _mm_mul_ps(fscal,dy20);
 335             tz               = _mm_mul_ps(fscal,dz20);
 336
 337             /* Update vectorial force */
 338             fix2             = _mm_add_ps(fix2,tx);
 339             fiy2             = _mm_add_ps(fiy2,ty);
 340             fiz2             = _mm_add_ps(fiz2,tz);
 341
 342             fjx0             = _mm_add_ps(fjx0,tx);
 343             fjy0             = _mm_add_ps(fjy0,ty);
 344             fjz0             = _mm_add_ps(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq30             = _mm_mul_ps(iq3,jq0);
 352
 353             /* REACTION-FIELD ELECTROSTATICS */
 354             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 355             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velecsum         = _mm_add_ps(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm_mul_ps(fscal,dx30);
 364             ty               = _mm_mul_ps(fscal,dy30);
 365             tz               = _mm_mul_ps(fscal,dz30);
 366
 367             /* Update vectorial force */
 368             fix3             = _mm_add_ps(fix3,tx);
 369             fiy3             = _mm_add_ps(fiy3,ty);
 370             fiz3             = _mm_add_ps(fiz3,tz);
 371
 372             fjx0             = _mm_add_ps(fjx0,tx);
 373             fjy0             = _mm_add_ps(fjy0,ty);
 374             fjz0             = _mm_add_ps(fjz0,tz);
 375
 376             fjptrA             = f+j_coord_offsetA;
 377             fjptrB             = f+j_coord_offsetB;
 378             fjptrC             = f+j_coord_offsetC;
 379             fjptrD             = f+j_coord_offsetD;
 380
 381             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 382
 383             /* Inner loop uses 128 flops */
 384         }
 385
 386         if(jidx<j_index_end)
 387         {
 388
 389             /* Get j neighbor index, and coordinate index */
 390             jnrlistA         = jjnr[jidx];
 391             jnrlistB         = jjnr[jidx+1];
 392             jnrlistC         = jjnr[jidx+2];
 393             jnrlistD         = jjnr[jidx+3];
 394             /* Sign of each element will be negative for non-real atoms.
 395              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 396              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 397              */
 398             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 399             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 400             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 401             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 402             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 403             j_coord_offsetA  = DIM*jnrA;
 404             j_coord_offsetB  = DIM*jnrB;
 405             j_coord_offsetC  = DIM*jnrC;
 406             j_coord_offsetD  = DIM*jnrD;
 407
 408             /* load j atom coordinates */
 409             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 410                                               x+j_coord_offsetC,x+j_coord_offsetD,
 411                                               &jx0,&jy0,&jz0);
 412
 413             /* Calculate displacement vector */
 414             dx00             = _mm_sub_ps(ix0,jx0);
 415             dy00             = _mm_sub_ps(iy0,jy0);
 416             dz00             = _mm_sub_ps(iz0,jz0);
 417             dx10             = _mm_sub_ps(ix1,jx0);
 418             dy10             = _mm_sub_ps(iy1,jy0);
 419             dz10             = _mm_sub_ps(iz1,jz0);
 420             dx20             = _mm_sub_ps(ix2,jx0);
 421             dy20             = _mm_sub_ps(iy2,jy0);
 422             dz20             = _mm_sub_ps(iz2,jz0);
 423             dx30             = _mm_sub_ps(ix3,jx0);
 424             dy30             = _mm_sub_ps(iy3,jy0);
 425             dz30             = _mm_sub_ps(iz3,jz0);
 426
 427             /* Calculate squared distance and things based on it */
 428             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 429             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 430             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 431             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 432
 433             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 434             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 435             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 436
 437             rinvsq00         = gmx_mm_inv_ps(rsq00);
 438             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 439             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 440             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 441
 442             /* Load parameters for j particles */
 443             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 444                                                               charge+jnrC+0,charge+jnrD+0);
 445             vdwjidx0A        = 2*vdwtype[jnrA+0];
 446             vdwjidx0B        = 2*vdwtype[jnrB+0];
 447             vdwjidx0C        = 2*vdwtype[jnrC+0];
 448             vdwjidx0D        = 2*vdwtype[jnrD+0];
 449
 450             fjx0             = _mm_setzero_ps();
 451             fjy0             = _mm_setzero_ps();
 452             fjz0             = _mm_setzero_ps();
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             /* Compute parameters for interactions between i and j atoms */
 459             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 460                                          vdwparam+vdwioffset0+vdwjidx0B,
 461                                          vdwparam+vdwioffset0+vdwjidx0C,
 462                                          vdwparam+vdwioffset0+vdwjidx0D,
 463                                          &c6_00,&c12_00);
 464
 465             /* LENNARD-JONES DISPERSION/REPULSION */
 466
 467             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 468             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 469             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 470             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 471             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 475             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 476
 477             fscal            = fvdw;
 478
 479             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 480
 481             /* Calculate temporary vectorial force */
 482             tx               = _mm_mul_ps(fscal,dx00);
 483             ty               = _mm_mul_ps(fscal,dy00);
 484             tz               = _mm_mul_ps(fscal,dz00);
 485
 486             /* Update vectorial force */
 487             fix0             = _mm_add_ps(fix0,tx);
 488             fiy0             = _mm_add_ps(fiy0,ty);
 489             fiz0             = _mm_add_ps(fiz0,tz);
 490
 491             fjx0             = _mm_add_ps(fjx0,tx);
 492             fjy0             = _mm_add_ps(fjy0,ty);
 493             fjz0             = _mm_add_ps(fjz0,tz);
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq10             = _mm_mul_ps(iq1,jq0);
 501
 502             /* REACTION-FIELD ELECTROSTATICS */
 503             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 504             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             velec            = _mm_andnot_ps(dummy_mask,velec);
 508             velecsum         = _mm_add_ps(velecsum,velec);
 509
 510             fscal            = felec;
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm_mul_ps(fscal,dx10);
 516             ty               = _mm_mul_ps(fscal,dy10);
 517             tz               = _mm_mul_ps(fscal,dz10);
 518
 519             /* Update vectorial force */
 520             fix1             = _mm_add_ps(fix1,tx);
 521             fiy1             = _mm_add_ps(fiy1,ty);
 522             fiz1             = _mm_add_ps(fiz1,tz);
 523
 524             fjx0             = _mm_add_ps(fjx0,tx);
 525             fjy0             = _mm_add_ps(fjy0,ty);
 526             fjz0             = _mm_add_ps(fjz0,tz);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq20             = _mm_mul_ps(iq2,jq0);
 534
 535             /* REACTION-FIELD ELECTROSTATICS */
 536             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 537             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 538
 539             /* Update potential sum for this i atom from the interaction with this j atom. */
 540             velec            = _mm_andnot_ps(dummy_mask,velec);
 541             velecsum         = _mm_add_ps(velecsum,velec);
 542
 543             fscal            = felec;
 544
 545             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 546
 547             /* Calculate temporary vectorial force */
 548             tx               = _mm_mul_ps(fscal,dx20);
 549             ty               = _mm_mul_ps(fscal,dy20);
 550             tz               = _mm_mul_ps(fscal,dz20);
 551
 552             /* Update vectorial force */
 553             fix2             = _mm_add_ps(fix2,tx);
 554             fiy2             = _mm_add_ps(fiy2,ty);
 555             fiz2             = _mm_add_ps(fiz2,tz);
 556
 557             fjx0             = _mm_add_ps(fjx0,tx);
 558             fjy0             = _mm_add_ps(fjy0,ty);
 559             fjz0             = _mm_add_ps(fjz0,tz);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             qq30             = _mm_mul_ps(iq3,jq0);
 567
 568             /* REACTION-FIELD ELECTROSTATICS */
 569             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 570             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 571
 572             /* Update potential sum for this i atom from the interaction with this j atom. */
 573             velec            = _mm_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm_mul_ps(fscal,dx30);
 582             ty               = _mm_mul_ps(fscal,dy30);
 583             tz               = _mm_mul_ps(fscal,dz30);
 584
 585             /* Update vectorial force */
 586             fix3             = _mm_add_ps(fix3,tx);
 587             fiy3             = _mm_add_ps(fiy3,ty);
 588             fiz3             = _mm_add_ps(fiz3,tz);
 589
 590             fjx0             = _mm_add_ps(fjx0,tx);
 591             fjy0             = _mm_add_ps(fjy0,ty);
 592             fjz0             = _mm_add_ps(fjz0,tz);
 593
 594             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 595             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 596             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 597             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 598
 599             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 600
 601             /* Inner loop uses 128 flops */
 602         }
 603
 604         /* End of innermost loop */
 605
 606         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 607                                               f+i_coord_offset,fshift+i_shift_offset);
 608
 609         ggid                        = gid[iidx];
 610         /* Update potential energies */
 611         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 612         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 613
 614         /* Increment number of inner iterations */
 615         inneriter                  += j_index_end - j_index_start;
 616
 617         /* Outer loop uses 26 flops */
 618     }
 619
 620     /* Increment number of outer iterations */
 621     outeriter        += nri;
 622
 623     /* Update outer/inner flops */
 624
 625     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*128);
 626 }
 627 /*
 628  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse4_1_single
 629  * Electrostatics interaction: ReactionField
 630  * VdW interaction:            LennardJones
 631  * Geometry:                   Water4-Particle
 632  * Calculate force/pot:        Force
 633  */
 634 void
 635 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse4_1_single
 636                     (t_nblist                    * gmx_restrict       nlist,
 637                      rvec                        * gmx_restrict          xx,
 638                      rvec                        * gmx_restrict          ff,
 639                      t_forcerec                  * gmx_restrict          fr,
 640                      t_mdatoms                   * gmx_restrict     mdatoms,
 641                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 642                      t_nrnb                      * gmx_restrict        nrnb)
 643 {
 644     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 645      * just 0 for non-waters.
 646      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 647      * jnr indices corresponding to data put in the four positions in the SIMD register.
 648      */
 649     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 650     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 651     int              jnrA,jnrB,jnrC,jnrD;
 652     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 653     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 654     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 655     real             rcutoff_scalar;
 656     real             *shiftvec,*fshift,*x,*f;
 657     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 658     real             scratch[4*DIM];
 659     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 660     int              vdwioffset0;
 661     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 662     int              vdwioffset1;
 663     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 664     int              vdwioffset2;
 665     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 666     int              vdwioffset3;
 667     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 668     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 669     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 670     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 671     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 672     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 673     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 674     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 675     real             *charge;
 676     int              nvdwtype;
 677     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 678     int              *vdwtype;
 679     real             *vdwparam;
 680     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 681     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 682     __m128           dummy_mask,cutoff_mask;
 683     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 684     __m128           one     = _mm_set1_ps(1.0);
 685     __m128           two     = _mm_set1_ps(2.0);
 686     x                = xx[0];
 687     f                = ff[0];
 688
 689     nri              = nlist->nri;
 690     iinr             = nlist->iinr;
 691     jindex           = nlist->jindex;
 692     jjnr             = nlist->jjnr;
 693     shiftidx         = nlist->shift;
 694     gid              = nlist->gid;
 695     shiftvec         = fr->shift_vec[0];
 696     fshift           = fr->fshift[0];
 697     facel            = _mm_set1_ps(fr->epsfac);
 698     charge           = mdatoms->chargeA;
 699     krf              = _mm_set1_ps(fr->ic->k_rf);
 700     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 701     crf              = _mm_set1_ps(fr->ic->c_rf);
 702     nvdwtype         = fr->ntype;
 703     vdwparam         = fr->nbfp;
 704     vdwtype          = mdatoms->typeA;
 705
 706     /* Setup water-specific parameters */
 707     inr              = nlist->iinr[0];
 708     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 709     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 710     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 711     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 712
 713     /* Avoid stupid compiler warnings */
 714     jnrA = jnrB = jnrC = jnrD = 0;
 715     j_coord_offsetA = 0;
 716     j_coord_offsetB = 0;
 717     j_coord_offsetC = 0;
 718     j_coord_offsetD = 0;
 719
 720     outeriter        = 0;
 721     inneriter        = 0;
 722
 723     for(iidx=0;iidx<4*DIM;iidx++)
 724     {
 725         scratch[iidx] = 0.0;
 726     }
 727
 728     /* Start outer loop over neighborlists */
 729     for(iidx=0; iidx<nri; iidx++)
 730     {
 731         /* Load shift vector for this list */
 732         i_shift_offset   = DIM*shiftidx[iidx];
 733
 734         /* Load limits for loop over neighbors */
 735         j_index_start    = jindex[iidx];
 736         j_index_end      = jindex[iidx+1];
 737
 738         /* Get outer coordinate index */
 739         inr              = iinr[iidx];
 740         i_coord_offset   = DIM*inr;
 741
 742         /* Load i particle coords and add shift vector */
 743         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 744                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 745
 746         fix0             = _mm_setzero_ps();
 747         fiy0             = _mm_setzero_ps();
 748         fiz0             = _mm_setzero_ps();
 749         fix1             = _mm_setzero_ps();
 750         fiy1             = _mm_setzero_ps();
 751         fiz1             = _mm_setzero_ps();
 752         fix2             = _mm_setzero_ps();
 753         fiy2             = _mm_setzero_ps();
 754         fiz2             = _mm_setzero_ps();
 755         fix3             = _mm_setzero_ps();
 756         fiy3             = _mm_setzero_ps();
 757         fiz3             = _mm_setzero_ps();
 758
 759         /* Start inner kernel loop */
 760         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 761         {
 762
 763             /* Get j neighbor index, and coordinate index */
 764             jnrA             = jjnr[jidx];
 765             jnrB             = jjnr[jidx+1];
 766             jnrC             = jjnr[jidx+2];
 767             jnrD             = jjnr[jidx+3];
 768             j_coord_offsetA  = DIM*jnrA;
 769             j_coord_offsetB  = DIM*jnrB;
 770             j_coord_offsetC  = DIM*jnrC;
 771             j_coord_offsetD  = DIM*jnrD;
 772
 773             /* load j atom coordinates */
 774             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 775                                               x+j_coord_offsetC,x+j_coord_offsetD,
 776                                               &jx0,&jy0,&jz0);
 777
 778             /* Calculate displacement vector */
 779             dx00             = _mm_sub_ps(ix0,jx0);
 780             dy00             = _mm_sub_ps(iy0,jy0);
 781             dz00             = _mm_sub_ps(iz0,jz0);
 782             dx10             = _mm_sub_ps(ix1,jx0);
 783             dy10             = _mm_sub_ps(iy1,jy0);
 784             dz10             = _mm_sub_ps(iz1,jz0);
 785             dx20             = _mm_sub_ps(ix2,jx0);
 786             dy20             = _mm_sub_ps(iy2,jy0);
 787             dz20             = _mm_sub_ps(iz2,jz0);
 788             dx30             = _mm_sub_ps(ix3,jx0);
 789             dy30             = _mm_sub_ps(iy3,jy0);
 790             dz30             = _mm_sub_ps(iz3,jz0);
 791
 792             /* Calculate squared distance and things based on it */
 793             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 794             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 795             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 796             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 797
 798             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 799             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 800             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 801
 802             rinvsq00         = gmx_mm_inv_ps(rsq00);
 803             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 804             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 805             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 806
 807             /* Load parameters for j particles */
 808             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 809                                                               charge+jnrC+0,charge+jnrD+0);
 810             vdwjidx0A        = 2*vdwtype[jnrA+0];
 811             vdwjidx0B        = 2*vdwtype[jnrB+0];
 812             vdwjidx0C        = 2*vdwtype[jnrC+0];
 813             vdwjidx0D        = 2*vdwtype[jnrD+0];
 814
 815             fjx0             = _mm_setzero_ps();
 816             fjy0             = _mm_setzero_ps();
 817             fjz0             = _mm_setzero_ps();
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             /* Compute parameters for interactions between i and j atoms */
 824             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 825                                          vdwparam+vdwioffset0+vdwjidx0B,
 826                                          vdwparam+vdwioffset0+vdwjidx0C,
 827                                          vdwparam+vdwioffset0+vdwjidx0D,
 828                                          &c6_00,&c12_00);
 829
 830             /* LENNARD-JONES DISPERSION/REPULSION */
 831
 832             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 833             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 834
 835             fscal            = fvdw;
 836
 837             /* Calculate temporary vectorial force */
 838             tx               = _mm_mul_ps(fscal,dx00);
 839             ty               = _mm_mul_ps(fscal,dy00);
 840             tz               = _mm_mul_ps(fscal,dz00);
 841
 842             /* Update vectorial force */
 843             fix0             = _mm_add_ps(fix0,tx);
 844             fiy0             = _mm_add_ps(fiy0,ty);
 845             fiz0             = _mm_add_ps(fiz0,tz);
 846
 847             fjx0             = _mm_add_ps(fjx0,tx);
 848             fjy0             = _mm_add_ps(fjy0,ty);
 849             fjz0             = _mm_add_ps(fjz0,tz);
 850
 851             /**************************
 852              * CALCULATE INTERACTIONS *
 853              **************************/
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             qq10             = _mm_mul_ps(iq1,jq0);
 857
 858             /* REACTION-FIELD ELECTROSTATICS */
 859             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 860
 861             fscal            = felec;
 862
 863             /* Calculate temporary vectorial force */
 864             tx               = _mm_mul_ps(fscal,dx10);
 865             ty               = _mm_mul_ps(fscal,dy10);
 866             tz               = _mm_mul_ps(fscal,dz10);
 867
 868             /* Update vectorial force */
 869             fix1             = _mm_add_ps(fix1,tx);
 870             fiy1             = _mm_add_ps(fiy1,ty);
 871             fiz1             = _mm_add_ps(fiz1,tz);
 872
 873             fjx0             = _mm_add_ps(fjx0,tx);
 874             fjy0             = _mm_add_ps(fjy0,ty);
 875             fjz0             = _mm_add_ps(fjz0,tz);
 876
 877             /**************************
 878              * CALCULATE INTERACTIONS *
 879              **************************/
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             qq20             = _mm_mul_ps(iq2,jq0);
 883
 884             /* REACTION-FIELD ELECTROSTATICS */
 885             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 886
 887             fscal            = felec;
 888
 889             /* Calculate temporary vectorial force */
 890             tx               = _mm_mul_ps(fscal,dx20);
 891             ty               = _mm_mul_ps(fscal,dy20);
 892             tz               = _mm_mul_ps(fscal,dz20);
 893
 894             /* Update vectorial force */
 895             fix2             = _mm_add_ps(fix2,tx);
 896             fiy2             = _mm_add_ps(fiy2,ty);
 897             fiz2             = _mm_add_ps(fiz2,tz);
 898
 899             fjx0             = _mm_add_ps(fjx0,tx);
 900             fjy0             = _mm_add_ps(fjy0,ty);
 901             fjz0             = _mm_add_ps(fjz0,tz);
 902
 903             /**************************
 904              * CALCULATE INTERACTIONS *
 905              **************************/
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq30             = _mm_mul_ps(iq3,jq0);
 909
 910             /* REACTION-FIELD ELECTROSTATICS */
 911             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 912
 913             fscal            = felec;
 914
 915             /* Calculate temporary vectorial force */
 916             tx               = _mm_mul_ps(fscal,dx30);
 917             ty               = _mm_mul_ps(fscal,dy30);
 918             tz               = _mm_mul_ps(fscal,dz30);
 919
 920             /* Update vectorial force */
 921             fix3             = _mm_add_ps(fix3,tx);
 922             fiy3             = _mm_add_ps(fiy3,ty);
 923             fiz3             = _mm_add_ps(fiz3,tz);
 924
 925             fjx0             = _mm_add_ps(fjx0,tx);
 926             fjy0             = _mm_add_ps(fjy0,ty);
 927             fjz0             = _mm_add_ps(fjz0,tz);
 928
 929             fjptrA             = f+j_coord_offsetA;
 930             fjptrB             = f+j_coord_offsetB;
 931             fjptrC             = f+j_coord_offsetC;
 932             fjptrD             = f+j_coord_offsetD;
 933
 934             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 935
 936             /* Inner loop uses 108 flops */
 937         }
 938
 939         if(jidx<j_index_end)
 940         {
 941
 942             /* Get j neighbor index, and coordinate index */
 943             jnrlistA         = jjnr[jidx];
 944             jnrlistB         = jjnr[jidx+1];
 945             jnrlistC         = jjnr[jidx+2];
 946             jnrlistD         = jjnr[jidx+3];
 947             /* Sign of each element will be negative for non-real atoms.
 948              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 949              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 950              */
 951             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 952             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 953             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 954             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 955             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 956             j_coord_offsetA  = DIM*jnrA;
 957             j_coord_offsetB  = DIM*jnrB;
 958             j_coord_offsetC  = DIM*jnrC;
 959             j_coord_offsetD  = DIM*jnrD;
 960
 961             /* load j atom coordinates */
 962             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 963                                               x+j_coord_offsetC,x+j_coord_offsetD,
 964                                               &jx0,&jy0,&jz0);
 965
 966             /* Calculate displacement vector */
 967             dx00             = _mm_sub_ps(ix0,jx0);
 968             dy00             = _mm_sub_ps(iy0,jy0);
 969             dz00             = _mm_sub_ps(iz0,jz0);
 970             dx10             = _mm_sub_ps(ix1,jx0);
 971             dy10             = _mm_sub_ps(iy1,jy0);
 972             dz10             = _mm_sub_ps(iz1,jz0);
 973             dx20             = _mm_sub_ps(ix2,jx0);
 974             dy20             = _mm_sub_ps(iy2,jy0);
 975             dz20             = _mm_sub_ps(iz2,jz0);
 976             dx30             = _mm_sub_ps(ix3,jx0);
 977             dy30             = _mm_sub_ps(iy3,jy0);
 978             dz30             = _mm_sub_ps(iz3,jz0);
 979
 980             /* Calculate squared distance and things based on it */
 981             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 982             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 983             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 984             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 985
 986             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 987             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 988             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 989
 990             rinvsq00         = gmx_mm_inv_ps(rsq00);
 991             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 992             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 993             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 994
 995             /* Load parameters for j particles */
 996             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 997                                                               charge+jnrC+0,charge+jnrD+0);
 998             vdwjidx0A        = 2*vdwtype[jnrA+0];
 999             vdwjidx0B        = 2*vdwtype[jnrB+0];
1000             vdwjidx0C        = 2*vdwtype[jnrC+0];
1001             vdwjidx0D        = 2*vdwtype[jnrD+0];
1002
1003             fjx0             = _mm_setzero_ps();
1004             fjy0             = _mm_setzero_ps();
1005             fjz0             = _mm_setzero_ps();
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1013                                          vdwparam+vdwioffset0+vdwjidx0B,
1014                                          vdwparam+vdwioffset0+vdwjidx0C,
1015                                          vdwparam+vdwioffset0+vdwjidx0D,
1016                                          &c6_00,&c12_00);
1017
1018             /* LENNARD-JONES DISPERSION/REPULSION */
1019
1020             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1021             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1022
1023             fscal            = fvdw;
1024
1025             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm_mul_ps(fscal,dx00);
1029             ty               = _mm_mul_ps(fscal,dy00);
1030             tz               = _mm_mul_ps(fscal,dz00);
1031
1032             /* Update vectorial force */
1033             fix0             = _mm_add_ps(fix0,tx);
1034             fiy0             = _mm_add_ps(fiy0,ty);
1035             fiz0             = _mm_add_ps(fiz0,tz);
1036
1037             fjx0             = _mm_add_ps(fjx0,tx);
1038             fjy0             = _mm_add_ps(fjy0,ty);
1039             fjz0             = _mm_add_ps(fjz0,tz);
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq10             = _mm_mul_ps(iq1,jq0);
1047
1048             /* REACTION-FIELD ELECTROSTATICS */
1049             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1050
1051             fscal            = felec;
1052
1053             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1054
1055             /* Calculate temporary vectorial force */
1056             tx               = _mm_mul_ps(fscal,dx10);
1057             ty               = _mm_mul_ps(fscal,dy10);
1058             tz               = _mm_mul_ps(fscal,dz10);
1059
1060             /* Update vectorial force */
1061             fix1             = _mm_add_ps(fix1,tx);
1062             fiy1             = _mm_add_ps(fiy1,ty);
1063             fiz1             = _mm_add_ps(fiz1,tz);
1064
1065             fjx0             = _mm_add_ps(fjx0,tx);
1066             fjy0             = _mm_add_ps(fjy0,ty);
1067             fjz0             = _mm_add_ps(fjz0,tz);
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq20             = _mm_mul_ps(iq2,jq0);
1075
1076             /* REACTION-FIELD ELECTROSTATICS */
1077             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1078
1079             fscal            = felec;
1080
1081             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1082
1083             /* Calculate temporary vectorial force */
1084             tx               = _mm_mul_ps(fscal,dx20);
1085             ty               = _mm_mul_ps(fscal,dy20);
1086             tz               = _mm_mul_ps(fscal,dz20);
1087
1088             /* Update vectorial force */
1089             fix2             = _mm_add_ps(fix2,tx);
1090             fiy2             = _mm_add_ps(fiy2,ty);
1091             fiz2             = _mm_add_ps(fiz2,tz);
1092
1093             fjx0             = _mm_add_ps(fjx0,tx);
1094             fjy0             = _mm_add_ps(fjy0,ty);
1095             fjz0             = _mm_add_ps(fjz0,tz);
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq30             = _mm_mul_ps(iq3,jq0);
1103
1104             /* REACTION-FIELD ELECTROSTATICS */
1105             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1110
1111             /* Calculate temporary vectorial force */
1112             tx               = _mm_mul_ps(fscal,dx30);
1113             ty               = _mm_mul_ps(fscal,dy30);
1114             tz               = _mm_mul_ps(fscal,dz30);
1115
1116             /* Update vectorial force */
1117             fix3             = _mm_add_ps(fix3,tx);
1118             fiy3             = _mm_add_ps(fiy3,ty);
1119             fiz3             = _mm_add_ps(fiz3,tz);
1120
1121             fjx0             = _mm_add_ps(fjx0,tx);
1122             fjy0             = _mm_add_ps(fjy0,ty);
1123             fjz0             = _mm_add_ps(fjz0,tz);
1124
1125             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1126             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1127             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1128             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1129
1130             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1131
1132             /* Inner loop uses 108 flops */
1133         }
1134
1135         /* End of innermost loop */
1136
1137         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1138                                               f+i_coord_offset,fshift+i_shift_offset);
1139
1140         /* Increment number of inner iterations */
1141         inneriter                  += j_index_end - j_index_start;
1142
1143         /* Outer loop uses 24 flops */
1144     }
1145
1146     /* Increment number of outer iterations */
1147     outeriter        += nri;
1148
1149     /* Update outer/inner flops */
1150
1151     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1152 }