src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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              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           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm_set1_ps(fr->ic->k_rf);
 122     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 123     crf              = _mm_set1_ps(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+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     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->rcoulomb;
 137     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 138     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 139
 140     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 141     rvdw             = _mm_set1_ps(fr->rvdw);
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179         fix1             = _mm_setzero_ps();
 180         fiy1             = _mm_setzero_ps();
 181         fiz1             = _mm_setzero_ps();
 182         fix2             = _mm_setzero_ps();
 183         fiy2             = _mm_setzero_ps();
 184         fiz2             = _mm_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 222             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 223             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 224
 225             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 226             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 227             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 228
 229             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 230             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 231             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             fjx0             = _mm_setzero_ps();
 242             fjy0             = _mm_setzero_ps();
 243             fjz0             = _mm_setzero_ps();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             if (gmx_mm_any_lt(rsq00,rcutoff2))
 250             {
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm_mul_ps(iq0,jq0);
 254             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,
 256                                          vdwparam+vdwioffset0+vdwjidx0C,
 257                                          vdwparam+vdwioffset0+vdwjidx0D,
 258                                          &c6_00,&c12_00);
 259
 260             /* REACTION-FIELD ELECTROSTATICS */
 261             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 262             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 268             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 269             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 270                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 271             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 272
 273             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm_and_ps(velec,cutoff_mask);
 277             velecsum         = _mm_add_ps(velecsum,velec);
 278             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 279             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_ps(felec,fvdw);
 282
 283             fscal            = _mm_and_ps(fscal,cutoff_mask);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_ps(fscal,dx00);
 287             ty               = _mm_mul_ps(fscal,dy00);
 288             tz               = _mm_mul_ps(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_ps(fix0,tx);
 292             fiy0             = _mm_add_ps(fiy0,ty);
 293             fiz0             = _mm_add_ps(fiz0,tz);
 294
 295             fjx0             = _mm_add_ps(fjx0,tx);
 296             fjy0             = _mm_add_ps(fjy0,ty);
 297             fjz0             = _mm_add_ps(fjz0,tz);
 298
 299             }
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             if (gmx_mm_any_lt(rsq10,rcutoff2))
 306             {
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm_mul_ps(iq1,jq0);
 310
 311             /* REACTION-FIELD ELECTROSTATICS */
 312             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 313             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 314
 315             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm_and_ps(velec,cutoff_mask);
 319             velecsum         = _mm_add_ps(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             fscal            = _mm_and_ps(fscal,cutoff_mask);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm_mul_ps(fscal,dx10);
 327             ty               = _mm_mul_ps(fscal,dy10);
 328             tz               = _mm_mul_ps(fscal,dz10);
 329
 330             /* Update vectorial force */
 331             fix1             = _mm_add_ps(fix1,tx);
 332             fiy1             = _mm_add_ps(fiy1,ty);
 333             fiz1             = _mm_add_ps(fiz1,tz);
 334
 335             fjx0             = _mm_add_ps(fjx0,tx);
 336             fjy0             = _mm_add_ps(fjy0,ty);
 337             fjz0             = _mm_add_ps(fjz0,tz);
 338
 339             }
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm_any_lt(rsq20,rcutoff2))
 346             {
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq20             = _mm_mul_ps(iq2,jq0);
 350
 351             /* REACTION-FIELD ELECTROSTATICS */
 352             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 353             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 354
 355             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velec            = _mm_and_ps(velec,cutoff_mask);
 359             velecsum         = _mm_add_ps(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             fscal            = _mm_and_ps(fscal,cutoff_mask);
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm_mul_ps(fscal,dx20);
 367             ty               = _mm_mul_ps(fscal,dy20);
 368             tz               = _mm_mul_ps(fscal,dz20);
 369
 370             /* Update vectorial force */
 371             fix2             = _mm_add_ps(fix2,tx);
 372             fiy2             = _mm_add_ps(fiy2,ty);
 373             fiz2             = _mm_add_ps(fiz2,tz);
 374
 375             fjx0             = _mm_add_ps(fjx0,tx);
 376             fjy0             = _mm_add_ps(fjy0,ty);
 377             fjz0             = _mm_add_ps(fjz0,tz);
 378
 379             }
 380
 381             fjptrA             = f+j_coord_offsetA;
 382             fjptrB             = f+j_coord_offsetB;
 383             fjptrC             = f+j_coord_offsetC;
 384             fjptrD             = f+j_coord_offsetD;
 385
 386             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 387
 388             /* Inner loop uses 126 flops */
 389         }
 390
 391         if(jidx<j_index_end)
 392         {
 393
 394             /* Get j neighbor index, and coordinate index */
 395             jnrlistA         = jjnr[jidx];
 396             jnrlistB         = jjnr[jidx+1];
 397             jnrlistC         = jjnr[jidx+2];
 398             jnrlistD         = jjnr[jidx+3];
 399             /* Sign of each element will be negative for non-real atoms.
 400              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 401              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 402              */
 403             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 404             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 405             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 406             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 407             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 408             j_coord_offsetA  = DIM*jnrA;
 409             j_coord_offsetB  = DIM*jnrB;
 410             j_coord_offsetC  = DIM*jnrC;
 411             j_coord_offsetD  = DIM*jnrD;
 412
 413             /* load j atom coordinates */
 414             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 415                                               x+j_coord_offsetC,x+j_coord_offsetD,
 416                                               &jx0,&jy0,&jz0);
 417
 418             /* Calculate displacement vector */
 419             dx00             = _mm_sub_ps(ix0,jx0);
 420             dy00             = _mm_sub_ps(iy0,jy0);
 421             dz00             = _mm_sub_ps(iz0,jz0);
 422             dx10             = _mm_sub_ps(ix1,jx0);
 423             dy10             = _mm_sub_ps(iy1,jy0);
 424             dz10             = _mm_sub_ps(iz1,jz0);
 425             dx20             = _mm_sub_ps(ix2,jx0);
 426             dy20             = _mm_sub_ps(iy2,jy0);
 427             dz20             = _mm_sub_ps(iz2,jz0);
 428
 429             /* Calculate squared distance and things based on it */
 430             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 431             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 432             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 433
 434             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 435             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 436             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 437
 438             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 439             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 440             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 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             if (gmx_mm_any_lt(rsq00,rcutoff2))
 459             {
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             qq00             = _mm_mul_ps(iq0,jq0);
 463             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 464                                          vdwparam+vdwioffset0+vdwjidx0B,
 465                                          vdwparam+vdwioffset0+vdwjidx0C,
 466                                          vdwparam+vdwioffset0+vdwjidx0D,
 467                                          &c6_00,&c12_00);
 468
 469             /* REACTION-FIELD ELECTROSTATICS */
 470             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 471             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 472
 473             /* LENNARD-JONES DISPERSION/REPULSION */
 474
 475             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 476             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 477             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 478             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 479                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 480             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 481
 482             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 483
 484             /* Update potential sum for this i atom from the interaction with this j atom. */
 485             velec            = _mm_and_ps(velec,cutoff_mask);
 486             velec            = _mm_andnot_ps(dummy_mask,velec);
 487             velecsum         = _mm_add_ps(velecsum,velec);
 488             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 489             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 490             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 491
 492             fscal            = _mm_add_ps(felec,fvdw);
 493
 494             fscal            = _mm_and_ps(fscal,cutoff_mask);
 495
 496             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 497
 498             /* Calculate temporary vectorial force */
 499             tx               = _mm_mul_ps(fscal,dx00);
 500             ty               = _mm_mul_ps(fscal,dy00);
 501             tz               = _mm_mul_ps(fscal,dz00);
 502
 503             /* Update vectorial force */
 504             fix0             = _mm_add_ps(fix0,tx);
 505             fiy0             = _mm_add_ps(fiy0,ty);
 506             fiz0             = _mm_add_ps(fiz0,tz);
 507
 508             fjx0             = _mm_add_ps(fjx0,tx);
 509             fjy0             = _mm_add_ps(fjy0,ty);
 510             fjz0             = _mm_add_ps(fjz0,tz);
 511
 512             }
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             if (gmx_mm_any_lt(rsq10,rcutoff2))
 519             {
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq10             = _mm_mul_ps(iq1,jq0);
 523
 524             /* REACTION-FIELD ELECTROSTATICS */
 525             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 526             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 527
 528             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_and_ps(velec,cutoff_mask);
 532             velec            = _mm_andnot_ps(dummy_mask,velec);
 533             velecsum         = _mm_add_ps(velecsum,velec);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm_and_ps(fscal,cutoff_mask);
 538
 539             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_ps(fscal,dx10);
 543             ty               = _mm_mul_ps(fscal,dy10);
 544             tz               = _mm_mul_ps(fscal,dz10);
 545
 546             /* Update vectorial force */
 547             fix1             = _mm_add_ps(fix1,tx);
 548             fiy1             = _mm_add_ps(fiy1,ty);
 549             fiz1             = _mm_add_ps(fiz1,tz);
 550
 551             fjx0             = _mm_add_ps(fjx0,tx);
 552             fjy0             = _mm_add_ps(fjy0,ty);
 553             fjz0             = _mm_add_ps(fjz0,tz);
 554
 555             }
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             if (gmx_mm_any_lt(rsq20,rcutoff2))
 562             {
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq20             = _mm_mul_ps(iq2,jq0);
 566
 567             /* REACTION-FIELD ELECTROSTATICS */
 568             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 569             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 570
 571             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 572
 573             /* Update potential sum for this i atom from the interaction with this j atom. */
 574             velec            = _mm_and_ps(velec,cutoff_mask);
 575             velec            = _mm_andnot_ps(dummy_mask,velec);
 576             velecsum         = _mm_add_ps(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm_and_ps(fscal,cutoff_mask);
 581
 582             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm_mul_ps(fscal,dx20);
 586             ty               = _mm_mul_ps(fscal,dy20);
 587             tz               = _mm_mul_ps(fscal,dz20);
 588
 589             /* Update vectorial force */
 590             fix2             = _mm_add_ps(fix2,tx);
 591             fiy2             = _mm_add_ps(fiy2,ty);
 592             fiz2             = _mm_add_ps(fiz2,tz);
 593
 594             fjx0             = _mm_add_ps(fjx0,tx);
 595             fjy0             = _mm_add_ps(fjy0,ty);
 596             fjz0             = _mm_add_ps(fjz0,tz);
 597
 598             }
 599
 600             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 601             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 602             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 603             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 604
 605             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 606
 607             /* Inner loop uses 126 flops */
 608         }
 609
 610         /* End of innermost loop */
 611
 612         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 613                                               f+i_coord_offset,fshift+i_shift_offset);
 614
 615         ggid                        = gid[iidx];
 616         /* Update potential energies */
 617         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 618         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 619
 620         /* Increment number of inner iterations */
 621         inneriter                  += j_index_end - j_index_start;
 622
 623         /* Outer loop uses 20 flops */
 624     }
 625
 626     /* Increment number of outer iterations */
 627     outeriter        += nri;
 628
 629     /* Update outer/inner flops */
 630
 631     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*126);
 632 }
 633 /*
 634  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse4_1_single
 635  * Electrostatics interaction: ReactionField
 636  * VdW interaction:            LennardJones
 637  * Geometry:                   Water3-Particle
 638  * Calculate force/pot:        Force
 639  */
 640 void
 641 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse4_1_single
 642                     (t_nblist                    * gmx_restrict       nlist,
 643                      rvec                        * gmx_restrict          xx,
 644                      rvec                        * gmx_restrict          ff,
 645                      t_forcerec                  * gmx_restrict          fr,
 646                      t_mdatoms                   * gmx_restrict     mdatoms,
 647                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 648                      t_nrnb                      * gmx_restrict        nrnb)
 649 {
 650     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 651      * just 0 for non-waters.
 652      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 653      * jnr indices corresponding to data put in the four positions in the SIMD register.
 654      */
 655     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 656     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 657     int              jnrA,jnrB,jnrC,jnrD;
 658     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 659     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 660     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 661     real             rcutoff_scalar;
 662     real             *shiftvec,*fshift,*x,*f;
 663     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 664     real             scratch[4*DIM];
 665     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 666     int              vdwioffset0;
 667     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 668     int              vdwioffset1;
 669     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 670     int              vdwioffset2;
 671     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 672     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 673     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 674     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 675     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 676     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 677     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 678     real             *charge;
 679     int              nvdwtype;
 680     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 681     int              *vdwtype;
 682     real             *vdwparam;
 683     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 684     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 685     __m128           dummy_mask,cutoff_mask;
 686     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 687     __m128           one     = _mm_set1_ps(1.0);
 688     __m128           two     = _mm_set1_ps(2.0);
 689     x                = xx[0];
 690     f                = ff[0];
 691
 692     nri              = nlist->nri;
 693     iinr             = nlist->iinr;
 694     jindex           = nlist->jindex;
 695     jjnr             = nlist->jjnr;
 696     shiftidx         = nlist->shift;
 697     gid              = nlist->gid;
 698     shiftvec         = fr->shift_vec[0];
 699     fshift           = fr->fshift[0];
 700     facel            = _mm_set1_ps(fr->epsfac);
 701     charge           = mdatoms->chargeA;
 702     krf              = _mm_set1_ps(fr->ic->k_rf);
 703     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 704     crf              = _mm_set1_ps(fr->ic->c_rf);
 705     nvdwtype         = fr->ntype;
 706     vdwparam         = fr->nbfp;
 707     vdwtype          = mdatoms->typeA;
 708
 709     /* Setup water-specific parameters */
 710     inr              = nlist->iinr[0];
 711     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 712     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 713     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 714     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 715
 716     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 717     rcutoff_scalar   = fr->rcoulomb;
 718     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 719     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 720
 721     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 722     rvdw             = _mm_set1_ps(fr->rvdw);
 723
 724     /* Avoid stupid compiler warnings */
 725     jnrA = jnrB = jnrC = jnrD = 0;
 726     j_coord_offsetA = 0;
 727     j_coord_offsetB = 0;
 728     j_coord_offsetC = 0;
 729     j_coord_offsetD = 0;
 730
 731     outeriter        = 0;
 732     inneriter        = 0;
 733
 734     for(iidx=0;iidx<4*DIM;iidx++)
 735     {
 736         scratch[iidx] = 0.0;
 737     }
 738
 739     /* Start outer loop over neighborlists */
 740     for(iidx=0; iidx<nri; iidx++)
 741     {
 742         /* Load shift vector for this list */
 743         i_shift_offset   = DIM*shiftidx[iidx];
 744
 745         /* Load limits for loop over neighbors */
 746         j_index_start    = jindex[iidx];
 747         j_index_end      = jindex[iidx+1];
 748
 749         /* Get outer coordinate index */
 750         inr              = iinr[iidx];
 751         i_coord_offset   = DIM*inr;
 752
 753         /* Load i particle coords and add shift vector */
 754         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 755                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 756
 757         fix0             = _mm_setzero_ps();
 758         fiy0             = _mm_setzero_ps();
 759         fiz0             = _mm_setzero_ps();
 760         fix1             = _mm_setzero_ps();
 761         fiy1             = _mm_setzero_ps();
 762         fiz1             = _mm_setzero_ps();
 763         fix2             = _mm_setzero_ps();
 764         fiy2             = _mm_setzero_ps();
 765         fiz2             = _mm_setzero_ps();
 766
 767         /* Start inner kernel loop */
 768         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 769         {
 770
 771             /* Get j neighbor index, and coordinate index */
 772             jnrA             = jjnr[jidx];
 773             jnrB             = jjnr[jidx+1];
 774             jnrC             = jjnr[jidx+2];
 775             jnrD             = jjnr[jidx+3];
 776             j_coord_offsetA  = DIM*jnrA;
 777             j_coord_offsetB  = DIM*jnrB;
 778             j_coord_offsetC  = DIM*jnrC;
 779             j_coord_offsetD  = DIM*jnrD;
 780
 781             /* load j atom coordinates */
 782             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 783                                               x+j_coord_offsetC,x+j_coord_offsetD,
 784                                               &jx0,&jy0,&jz0);
 785
 786             /* Calculate displacement vector */
 787             dx00             = _mm_sub_ps(ix0,jx0);
 788             dy00             = _mm_sub_ps(iy0,jy0);
 789             dz00             = _mm_sub_ps(iz0,jz0);
 790             dx10             = _mm_sub_ps(ix1,jx0);
 791             dy10             = _mm_sub_ps(iy1,jy0);
 792             dz10             = _mm_sub_ps(iz1,jz0);
 793             dx20             = _mm_sub_ps(ix2,jx0);
 794             dy20             = _mm_sub_ps(iy2,jy0);
 795             dz20             = _mm_sub_ps(iz2,jz0);
 796
 797             /* Calculate squared distance and things based on it */
 798             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 799             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 800             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 801
 802             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 803             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 804             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 805
 806             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 807             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 808             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 809
 810             /* Load parameters for j particles */
 811             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 812                                                               charge+jnrC+0,charge+jnrD+0);
 813             vdwjidx0A        = 2*vdwtype[jnrA+0];
 814             vdwjidx0B        = 2*vdwtype[jnrB+0];
 815             vdwjidx0C        = 2*vdwtype[jnrC+0];
 816             vdwjidx0D        = 2*vdwtype[jnrD+0];
 817
 818             fjx0             = _mm_setzero_ps();
 819             fjy0             = _mm_setzero_ps();
 820             fjz0             = _mm_setzero_ps();
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             if (gmx_mm_any_lt(rsq00,rcutoff2))
 827             {
 828
 829             /* Compute parameters for interactions between i and j atoms */
 830             qq00             = _mm_mul_ps(iq0,jq0);
 831             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 832                                          vdwparam+vdwioffset0+vdwjidx0B,
 833                                          vdwparam+vdwioffset0+vdwjidx0C,
 834                                          vdwparam+vdwioffset0+vdwjidx0D,
 835                                          &c6_00,&c12_00);
 836
 837             /* REACTION-FIELD ELECTROSTATICS */
 838             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 839
 840             /* LENNARD-JONES DISPERSION/REPULSION */
 841
 842             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 843             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 844
 845             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 846
 847             fscal            = _mm_add_ps(felec,fvdw);
 848
 849             fscal            = _mm_and_ps(fscal,cutoff_mask);
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm_mul_ps(fscal,dx00);
 853             ty               = _mm_mul_ps(fscal,dy00);
 854             tz               = _mm_mul_ps(fscal,dz00);
 855
 856             /* Update vectorial force */
 857             fix0             = _mm_add_ps(fix0,tx);
 858             fiy0             = _mm_add_ps(fiy0,ty);
 859             fiz0             = _mm_add_ps(fiz0,tz);
 860
 861             fjx0             = _mm_add_ps(fjx0,tx);
 862             fjy0             = _mm_add_ps(fjy0,ty);
 863             fjz0             = _mm_add_ps(fjz0,tz);
 864
 865             }
 866
 867             /**************************
 868              * CALCULATE INTERACTIONS *
 869              **************************/
 870
 871             if (gmx_mm_any_lt(rsq10,rcutoff2))
 872             {
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq10             = _mm_mul_ps(iq1,jq0);
 876
 877             /* REACTION-FIELD ELECTROSTATICS */
 878             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 879
 880             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 881
 882             fscal            = felec;
 883
 884             fscal            = _mm_and_ps(fscal,cutoff_mask);
 885
 886             /* Calculate temporary vectorial force */
 887             tx               = _mm_mul_ps(fscal,dx10);
 888             ty               = _mm_mul_ps(fscal,dy10);
 889             tz               = _mm_mul_ps(fscal,dz10);
 890
 891             /* Update vectorial force */
 892             fix1             = _mm_add_ps(fix1,tx);
 893             fiy1             = _mm_add_ps(fiy1,ty);
 894             fiz1             = _mm_add_ps(fiz1,tz);
 895
 896             fjx0             = _mm_add_ps(fjx0,tx);
 897             fjy0             = _mm_add_ps(fjy0,ty);
 898             fjz0             = _mm_add_ps(fjz0,tz);
 899
 900             }
 901
 902             /**************************
 903              * CALCULATE INTERACTIONS *
 904              **************************/
 905
 906             if (gmx_mm_any_lt(rsq20,rcutoff2))
 907             {
 908
 909             /* Compute parameters for interactions between i and j atoms */
 910             qq20             = _mm_mul_ps(iq2,jq0);
 911
 912             /* REACTION-FIELD ELECTROSTATICS */
 913             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 914
 915             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 916
 917             fscal            = felec;
 918
 919             fscal            = _mm_and_ps(fscal,cutoff_mask);
 920
 921             /* Calculate temporary vectorial force */
 922             tx               = _mm_mul_ps(fscal,dx20);
 923             ty               = _mm_mul_ps(fscal,dy20);
 924             tz               = _mm_mul_ps(fscal,dz20);
 925
 926             /* Update vectorial force */
 927             fix2             = _mm_add_ps(fix2,tx);
 928             fiy2             = _mm_add_ps(fiy2,ty);
 929             fiz2             = _mm_add_ps(fiz2,tz);
 930
 931             fjx0             = _mm_add_ps(fjx0,tx);
 932             fjy0             = _mm_add_ps(fjy0,ty);
 933             fjz0             = _mm_add_ps(fjz0,tz);
 934
 935             }
 936
 937             fjptrA             = f+j_coord_offsetA;
 938             fjptrB             = f+j_coord_offsetB;
 939             fjptrC             = f+j_coord_offsetC;
 940             fjptrD             = f+j_coord_offsetD;
 941
 942             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 943
 944             /* Inner loop uses 97 flops */
 945         }
 946
 947         if(jidx<j_index_end)
 948         {
 949
 950             /* Get j neighbor index, and coordinate index */
 951             jnrlistA         = jjnr[jidx];
 952             jnrlistB         = jjnr[jidx+1];
 953             jnrlistC         = jjnr[jidx+2];
 954             jnrlistD         = jjnr[jidx+3];
 955             /* Sign of each element will be negative for non-real atoms.
 956              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 957              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 958              */
 959             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 960             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 961             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 962             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 963             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 964             j_coord_offsetA  = DIM*jnrA;
 965             j_coord_offsetB  = DIM*jnrB;
 966             j_coord_offsetC  = DIM*jnrC;
 967             j_coord_offsetD  = DIM*jnrD;
 968
 969             /* load j atom coordinates */
 970             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 971                                               x+j_coord_offsetC,x+j_coord_offsetD,
 972                                               &jx0,&jy0,&jz0);
 973
 974             /* Calculate displacement vector */
 975             dx00             = _mm_sub_ps(ix0,jx0);
 976             dy00             = _mm_sub_ps(iy0,jy0);
 977             dz00             = _mm_sub_ps(iz0,jz0);
 978             dx10             = _mm_sub_ps(ix1,jx0);
 979             dy10             = _mm_sub_ps(iy1,jy0);
 980             dz10             = _mm_sub_ps(iz1,jz0);
 981             dx20             = _mm_sub_ps(ix2,jx0);
 982             dy20             = _mm_sub_ps(iy2,jy0);
 983             dz20             = _mm_sub_ps(iz2,jz0);
 984
 985             /* Calculate squared distance and things based on it */
 986             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 987             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 988             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 989
 990             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 991             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 992             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 993
 994             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 995             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 996             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 997
 998             /* Load parameters for j particles */
 999             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1000                                                               charge+jnrC+0,charge+jnrD+0);
1001             vdwjidx0A        = 2*vdwtype[jnrA+0];
1002             vdwjidx0B        = 2*vdwtype[jnrB+0];
1003             vdwjidx0C        = 2*vdwtype[jnrC+0];
1004             vdwjidx0D        = 2*vdwtype[jnrD+0];
1005
1006             fjx0             = _mm_setzero_ps();
1007             fjy0             = _mm_setzero_ps();
1008             fjz0             = _mm_setzero_ps();
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             if (gmx_mm_any_lt(rsq00,rcutoff2))
1015             {
1016
1017             /* Compute parameters for interactions between i and j atoms */
1018             qq00             = _mm_mul_ps(iq0,jq0);
1019             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1020                                          vdwparam+vdwioffset0+vdwjidx0B,
1021                                          vdwparam+vdwioffset0+vdwjidx0C,
1022                                          vdwparam+vdwioffset0+vdwjidx0D,
1023                                          &c6_00,&c12_00);
1024
1025             /* REACTION-FIELD ELECTROSTATICS */
1026             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1027
1028             /* LENNARD-JONES DISPERSION/REPULSION */
1029
1030             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1031             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1032
1033             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1034
1035             fscal            = _mm_add_ps(felec,fvdw);
1036
1037             fscal            = _mm_and_ps(fscal,cutoff_mask);
1038
1039             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm_mul_ps(fscal,dx00);
1043             ty               = _mm_mul_ps(fscal,dy00);
1044             tz               = _mm_mul_ps(fscal,dz00);
1045
1046             /* Update vectorial force */
1047             fix0             = _mm_add_ps(fix0,tx);
1048             fiy0             = _mm_add_ps(fiy0,ty);
1049             fiz0             = _mm_add_ps(fiz0,tz);
1050
1051             fjx0             = _mm_add_ps(fjx0,tx);
1052             fjy0             = _mm_add_ps(fjy0,ty);
1053             fjz0             = _mm_add_ps(fjz0,tz);
1054
1055             }
1056
1057             /**************************
1058              * CALCULATE INTERACTIONS *
1059              **************************/
1060
1061             if (gmx_mm_any_lt(rsq10,rcutoff2))
1062             {
1063
1064             /* Compute parameters for interactions between i and j atoms */
1065             qq10             = _mm_mul_ps(iq1,jq0);
1066
1067             /* REACTION-FIELD ELECTROSTATICS */
1068             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1069
1070             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1071
1072             fscal            = felec;
1073
1074             fscal            = _mm_and_ps(fscal,cutoff_mask);
1075
1076             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm_mul_ps(fscal,dx10);
1080             ty               = _mm_mul_ps(fscal,dy10);
1081             tz               = _mm_mul_ps(fscal,dz10);
1082
1083             /* Update vectorial force */
1084             fix1             = _mm_add_ps(fix1,tx);
1085             fiy1             = _mm_add_ps(fiy1,ty);
1086             fiz1             = _mm_add_ps(fiz1,tz);
1087
1088             fjx0             = _mm_add_ps(fjx0,tx);
1089             fjy0             = _mm_add_ps(fjy0,ty);
1090             fjz0             = _mm_add_ps(fjz0,tz);
1091
1092             }
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             if (gmx_mm_any_lt(rsq20,rcutoff2))
1099             {
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq20             = _mm_mul_ps(iq2,jq0);
1103
1104             /* REACTION-FIELD ELECTROSTATICS */
1105             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1106
1107             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1108
1109             fscal            = felec;
1110
1111             fscal            = _mm_and_ps(fscal,cutoff_mask);
1112
1113             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1114
1115             /* Calculate temporary vectorial force */
1116             tx               = _mm_mul_ps(fscal,dx20);
1117             ty               = _mm_mul_ps(fscal,dy20);
1118             tz               = _mm_mul_ps(fscal,dz20);
1119
1120             /* Update vectorial force */
1121             fix2             = _mm_add_ps(fix2,tx);
1122             fiy2             = _mm_add_ps(fiy2,ty);
1123             fiz2             = _mm_add_ps(fiz2,tz);
1124
1125             fjx0             = _mm_add_ps(fjx0,tx);
1126             fjy0             = _mm_add_ps(fjy0,ty);
1127             fjz0             = _mm_add_ps(fjz0,tz);
1128
1129             }
1130
1131             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1132             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1133             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1134             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1135
1136             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1137
1138             /* Inner loop uses 97 flops */
1139         }
1140
1141         /* End of innermost loop */
1142
1143         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1144                                               f+i_coord_offset,fshift+i_shift_offset);
1145
1146         /* Increment number of inner iterations */
1147         inneriter                  += j_index_end - j_index_start;
1148
1149         /* Outer loop uses 18 flops */
1150     }
1151
1152     /* Increment number of outer iterations */
1153     outeriter        += nri;
1154
1155     /* Update outer/inner flops */
1156
1157     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);
1158 }