src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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->ic->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     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 131     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 132     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* 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->ic->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->ic->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_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 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         fix3             = _mm_setzero_ps();
 186         fiy3             = _mm_setzero_ps();
 187         fiz3             = _mm_setzero_ps();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_ps();
 191         vvdwsum          = _mm_setzero_ps();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             jnrC             = jjnr[jidx+2];
 201             jnrD             = jjnr[jidx+3];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206
 207             /* load j atom coordinates */
 208             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 209                                               x+j_coord_offsetC,x+j_coord_offsetD,
 210                                               &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx00             = _mm_sub_ps(ix0,jx0);
 214             dy00             = _mm_sub_ps(iy0,jy0);
 215             dz00             = _mm_sub_ps(iz0,jz0);
 216             dx10             = _mm_sub_ps(ix1,jx0);
 217             dy10             = _mm_sub_ps(iy1,jy0);
 218             dz10             = _mm_sub_ps(iz1,jz0);
 219             dx20             = _mm_sub_ps(ix2,jx0);
 220             dy20             = _mm_sub_ps(iy2,jy0);
 221             dz20             = _mm_sub_ps(iz2,jz0);
 222             dx30             = _mm_sub_ps(ix3,jx0);
 223             dy30             = _mm_sub_ps(iy3,jy0);
 224             dz30             = _mm_sub_ps(iz3,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 228             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 229             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 230             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 231
 232             rinv10           = sse41_invsqrt_f(rsq10);
 233             rinv20           = sse41_invsqrt_f(rsq20);
 234             rinv30           = sse41_invsqrt_f(rsq30);
 235
 236             rinvsq00         = sse41_inv_f(rsq00);
 237             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 238             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 239             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                               charge+jnrC+0,charge+jnrD+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248
 249             fjx0             = _mm_setzero_ps();
 250             fjy0             = _mm_setzero_ps();
 251             fjz0             = _mm_setzero_ps();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             if (gmx_mm_any_lt(rsq00,rcutoff2))
 258             {
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 262                                          vdwparam+vdwioffset0+vdwjidx0B,
 263                                          vdwparam+vdwioffset0+vdwjidx0C,
 264                                          vdwparam+vdwioffset0+vdwjidx0D,
 265                                          &c6_00,&c12_00);
 266
 267             /* LENNARD-JONES DISPERSION/REPULSION */
 268
 269             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 270             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 271             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 272             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) ,
 273                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 274             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 275
 276             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 280             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 281
 282             fscal            = fvdw;
 283
 284             fscal            = _mm_and_ps(fscal,cutoff_mask);
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_ps(fscal,dx00);
 288             ty               = _mm_mul_ps(fscal,dy00);
 289             tz               = _mm_mul_ps(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_add_ps(fix0,tx);
 293             fiy0             = _mm_add_ps(fiy0,ty);
 294             fiz0             = _mm_add_ps(fiz0,tz);
 295
 296             fjx0             = _mm_add_ps(fjx0,tx);
 297             fjy0             = _mm_add_ps(fjy0,ty);
 298             fjz0             = _mm_add_ps(fjz0,tz);
 299
 300             }
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             if (gmx_mm_any_lt(rsq10,rcutoff2))
 307             {
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq10             = _mm_mul_ps(iq1,jq0);
 311
 312             /* REACTION-FIELD ELECTROSTATICS */
 313             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 314             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 315
 316             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm_and_ps(velec,cutoff_mask);
 320             velecsum         = _mm_add_ps(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             fscal            = _mm_and_ps(fscal,cutoff_mask);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm_mul_ps(fscal,dx10);
 328             ty               = _mm_mul_ps(fscal,dy10);
 329             tz               = _mm_mul_ps(fscal,dz10);
 330
 331             /* Update vectorial force */
 332             fix1             = _mm_add_ps(fix1,tx);
 333             fiy1             = _mm_add_ps(fiy1,ty);
 334             fiz1             = _mm_add_ps(fiz1,tz);
 335
 336             fjx0             = _mm_add_ps(fjx0,tx);
 337             fjy0             = _mm_add_ps(fjy0,ty);
 338             fjz0             = _mm_add_ps(fjz0,tz);
 339
 340             }
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             if (gmx_mm_any_lt(rsq20,rcutoff2))
 347             {
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq20             = _mm_mul_ps(iq2,jq0);
 351
 352             /* REACTION-FIELD ELECTROSTATICS */
 353             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 354             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 355
 356             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velec            = _mm_and_ps(velec,cutoff_mask);
 360             velecsum         = _mm_add_ps(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             fscal            = _mm_and_ps(fscal,cutoff_mask);
 365
 366             /* Calculate temporary vectorial force */
 367             tx               = _mm_mul_ps(fscal,dx20);
 368             ty               = _mm_mul_ps(fscal,dy20);
 369             tz               = _mm_mul_ps(fscal,dz20);
 370
 371             /* Update vectorial force */
 372             fix2             = _mm_add_ps(fix2,tx);
 373             fiy2             = _mm_add_ps(fiy2,ty);
 374             fiz2             = _mm_add_ps(fiz2,tz);
 375
 376             fjx0             = _mm_add_ps(fjx0,tx);
 377             fjy0             = _mm_add_ps(fjy0,ty);
 378             fjz0             = _mm_add_ps(fjz0,tz);
 379
 380             }
 381
 382             /**************************
 383              * CALCULATE INTERACTIONS *
 384              **************************/
 385
 386             if (gmx_mm_any_lt(rsq30,rcutoff2))
 387             {
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq30             = _mm_mul_ps(iq3,jq0);
 391
 392             /* REACTION-FIELD ELECTROSTATICS */
 393             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 394             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 395
 396             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velec            = _mm_and_ps(velec,cutoff_mask);
 400             velecsum         = _mm_add_ps(velecsum,velec);
 401
 402             fscal            = felec;
 403
 404             fscal            = _mm_and_ps(fscal,cutoff_mask);
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm_mul_ps(fscal,dx30);
 408             ty               = _mm_mul_ps(fscal,dy30);
 409             tz               = _mm_mul_ps(fscal,dz30);
 410
 411             /* Update vectorial force */
 412             fix3             = _mm_add_ps(fix3,tx);
 413             fiy3             = _mm_add_ps(fiy3,ty);
 414             fiz3             = _mm_add_ps(fiz3,tz);
 415
 416             fjx0             = _mm_add_ps(fjx0,tx);
 417             fjy0             = _mm_add_ps(fjy0,ty);
 418             fjz0             = _mm_add_ps(fjz0,tz);
 419
 420             }
 421
 422             fjptrA             = f+j_coord_offsetA;
 423             fjptrB             = f+j_coord_offsetB;
 424             fjptrC             = f+j_coord_offsetC;
 425             fjptrD             = f+j_coord_offsetD;
 426
 427             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 428
 429             /* Inner loop uses 149 flops */
 430         }
 431
 432         if(jidx<j_index_end)
 433         {
 434
 435             /* Get j neighbor index, and coordinate index */
 436             jnrlistA         = jjnr[jidx];
 437             jnrlistB         = jjnr[jidx+1];
 438             jnrlistC         = jjnr[jidx+2];
 439             jnrlistD         = jjnr[jidx+3];
 440             /* Sign of each element will be negative for non-real atoms.
 441              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 442              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 443              */
 444             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 445             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 446             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 447             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 448             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 449             j_coord_offsetA  = DIM*jnrA;
 450             j_coord_offsetB  = DIM*jnrB;
 451             j_coord_offsetC  = DIM*jnrC;
 452             j_coord_offsetD  = DIM*jnrD;
 453
 454             /* load j atom coordinates */
 455             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 456                                               x+j_coord_offsetC,x+j_coord_offsetD,
 457                                               &jx0,&jy0,&jz0);
 458
 459             /* Calculate displacement vector */
 460             dx00             = _mm_sub_ps(ix0,jx0);
 461             dy00             = _mm_sub_ps(iy0,jy0);
 462             dz00             = _mm_sub_ps(iz0,jz0);
 463             dx10             = _mm_sub_ps(ix1,jx0);
 464             dy10             = _mm_sub_ps(iy1,jy0);
 465             dz10             = _mm_sub_ps(iz1,jz0);
 466             dx20             = _mm_sub_ps(ix2,jx0);
 467             dy20             = _mm_sub_ps(iy2,jy0);
 468             dz20             = _mm_sub_ps(iz2,jz0);
 469             dx30             = _mm_sub_ps(ix3,jx0);
 470             dy30             = _mm_sub_ps(iy3,jy0);
 471             dz30             = _mm_sub_ps(iz3,jz0);
 472
 473             /* Calculate squared distance and things based on it */
 474             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 475             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 476             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 477             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 478
 479             rinv10           = sse41_invsqrt_f(rsq10);
 480             rinv20           = sse41_invsqrt_f(rsq20);
 481             rinv30           = sse41_invsqrt_f(rsq30);
 482
 483             rinvsq00         = sse41_inv_f(rsq00);
 484             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 485             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 486             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 490                                                               charge+jnrC+0,charge+jnrD+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493             vdwjidx0C        = 2*vdwtype[jnrC+0];
 494             vdwjidx0D        = 2*vdwtype[jnrD+0];
 495
 496             fjx0             = _mm_setzero_ps();
 497             fjy0             = _mm_setzero_ps();
 498             fjz0             = _mm_setzero_ps();
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             if (gmx_mm_any_lt(rsq00,rcutoff2))
 505             {
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 509                                          vdwparam+vdwioffset0+vdwjidx0B,
 510                                          vdwparam+vdwioffset0+vdwjidx0C,
 511                                          vdwparam+vdwioffset0+vdwjidx0D,
 512                                          &c6_00,&c12_00);
 513
 514             /* LENNARD-JONES DISPERSION/REPULSION */
 515
 516             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 517             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 518             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 519             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) ,
 520                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 521             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 522
 523             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 527             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 528             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 529
 530             fscal            = fvdw;
 531
 532             fscal            = _mm_and_ps(fscal,cutoff_mask);
 533
 534             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 535
 536             /* Calculate temporary vectorial force */
 537             tx               = _mm_mul_ps(fscal,dx00);
 538             ty               = _mm_mul_ps(fscal,dy00);
 539             tz               = _mm_mul_ps(fscal,dz00);
 540
 541             /* Update vectorial force */
 542             fix0             = _mm_add_ps(fix0,tx);
 543             fiy0             = _mm_add_ps(fiy0,ty);
 544             fiz0             = _mm_add_ps(fiz0,tz);
 545
 546             fjx0             = _mm_add_ps(fjx0,tx);
 547             fjy0             = _mm_add_ps(fjy0,ty);
 548             fjz0             = _mm_add_ps(fjz0,tz);
 549
 550             }
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             if (gmx_mm_any_lt(rsq10,rcutoff2))
 557             {
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq10             = _mm_mul_ps(iq1,jq0);
 561
 562             /* REACTION-FIELD ELECTROSTATICS */
 563             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 564             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 565
 566             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm_and_ps(velec,cutoff_mask);
 570             velec            = _mm_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_and_ps(fscal,cutoff_mask);
 576
 577             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_ps(fscal,dx10);
 581             ty               = _mm_mul_ps(fscal,dy10);
 582             tz               = _mm_mul_ps(fscal,dz10);
 583
 584             /* Update vectorial force */
 585             fix1             = _mm_add_ps(fix1,tx);
 586             fiy1             = _mm_add_ps(fiy1,ty);
 587             fiz1             = _mm_add_ps(fiz1,tz);
 588
 589             fjx0             = _mm_add_ps(fjx0,tx);
 590             fjy0             = _mm_add_ps(fjy0,ty);
 591             fjz0             = _mm_add_ps(fjz0,tz);
 592
 593             }
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             if (gmx_mm_any_lt(rsq20,rcutoff2))
 600             {
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq20             = _mm_mul_ps(iq2,jq0);
 604
 605             /* REACTION-FIELD ELECTROSTATICS */
 606             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 607             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 608
 609             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 610
 611             /* Update potential sum for this i atom from the interaction with this j atom. */
 612             velec            = _mm_and_ps(velec,cutoff_mask);
 613             velec            = _mm_andnot_ps(dummy_mask,velec);
 614             velecsum         = _mm_add_ps(velecsum,velec);
 615
 616             fscal            = felec;
 617
 618             fscal            = _mm_and_ps(fscal,cutoff_mask);
 619
 620             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm_mul_ps(fscal,dx20);
 624             ty               = _mm_mul_ps(fscal,dy20);
 625             tz               = _mm_mul_ps(fscal,dz20);
 626
 627             /* Update vectorial force */
 628             fix2             = _mm_add_ps(fix2,tx);
 629             fiy2             = _mm_add_ps(fiy2,ty);
 630             fiz2             = _mm_add_ps(fiz2,tz);
 631
 632             fjx0             = _mm_add_ps(fjx0,tx);
 633             fjy0             = _mm_add_ps(fjy0,ty);
 634             fjz0             = _mm_add_ps(fjz0,tz);
 635
 636             }
 637
 638             /**************************
 639              * CALCULATE INTERACTIONS *
 640              **************************/
 641
 642             if (gmx_mm_any_lt(rsq30,rcutoff2))
 643             {
 644
 645             /* Compute parameters for interactions between i and j atoms */
 646             qq30             = _mm_mul_ps(iq3,jq0);
 647
 648             /* REACTION-FIELD ELECTROSTATICS */
 649             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 650             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 651
 652             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 653
 654             /* Update potential sum for this i atom from the interaction with this j atom. */
 655             velec            = _mm_and_ps(velec,cutoff_mask);
 656             velec            = _mm_andnot_ps(dummy_mask,velec);
 657             velecsum         = _mm_add_ps(velecsum,velec);
 658
 659             fscal            = felec;
 660
 661             fscal            = _mm_and_ps(fscal,cutoff_mask);
 662
 663             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 664
 665             /* Calculate temporary vectorial force */
 666             tx               = _mm_mul_ps(fscal,dx30);
 667             ty               = _mm_mul_ps(fscal,dy30);
 668             tz               = _mm_mul_ps(fscal,dz30);
 669
 670             /* Update vectorial force */
 671             fix3             = _mm_add_ps(fix3,tx);
 672             fiy3             = _mm_add_ps(fiy3,ty);
 673             fiz3             = _mm_add_ps(fiz3,tz);
 674
 675             fjx0             = _mm_add_ps(fjx0,tx);
 676             fjy0             = _mm_add_ps(fjy0,ty);
 677             fjz0             = _mm_add_ps(fjz0,tz);
 678
 679             }
 680
 681             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 682             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 683             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 684             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 685
 686             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 687
 688             /* Inner loop uses 149 flops */
 689         }
 690
 691         /* End of innermost loop */
 692
 693         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 694                                               f+i_coord_offset,fshift+i_shift_offset);
 695
 696         ggid                        = gid[iidx];
 697         /* Update potential energies */
 698         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 699         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 700
 701         /* Increment number of inner iterations */
 702         inneriter                  += j_index_end - j_index_start;
 703
 704         /* Outer loop uses 26 flops */
 705     }
 706
 707     /* Increment number of outer iterations */
 708     outeriter        += nri;
 709
 710     /* Update outer/inner flops */
 711
 712     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*149);
 713 }
 714 /*
 715  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_single
 716  * Electrostatics interaction: ReactionField
 717  * VdW interaction:            LennardJones
 718  * Geometry:                   Water4-Particle
 719  * Calculate force/pot:        Force
 720  */
 721 void
 722 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_single
 723                     (t_nblist                    * gmx_restrict       nlist,
 724                      rvec                        * gmx_restrict          xx,
 725                      rvec                        * gmx_restrict          ff,
 726                      struct t_forcerec           * gmx_restrict          fr,
 727                      t_mdatoms                   * gmx_restrict     mdatoms,
 728                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 729                      t_nrnb                      * gmx_restrict        nrnb)
 730 {
 731     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 732      * just 0 for non-waters.
 733      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 734      * jnr indices corresponding to data put in the four positions in the SIMD register.
 735      */
 736     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 737     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 738     int              jnrA,jnrB,jnrC,jnrD;
 739     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 740     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 741     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 742     real             rcutoff_scalar;
 743     real             *shiftvec,*fshift,*x,*f;
 744     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 745     real             scratch[4*DIM];
 746     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 747     int              vdwioffset0;
 748     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 749     int              vdwioffset1;
 750     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 751     int              vdwioffset2;
 752     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 753     int              vdwioffset3;
 754     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 755     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 756     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 757     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 758     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 759     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 760     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 761     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 762     real             *charge;
 763     int              nvdwtype;
 764     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 765     int              *vdwtype;
 766     real             *vdwparam;
 767     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 768     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 769     __m128           dummy_mask,cutoff_mask;
 770     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 771     __m128           one     = _mm_set1_ps(1.0);
 772     __m128           two     = _mm_set1_ps(2.0);
 773     x                = xx[0];
 774     f                = ff[0];
 775
 776     nri              = nlist->nri;
 777     iinr             = nlist->iinr;
 778     jindex           = nlist->jindex;
 779     jjnr             = nlist->jjnr;
 780     shiftidx         = nlist->shift;
 781     gid              = nlist->gid;
 782     shiftvec         = fr->shift_vec[0];
 783     fshift           = fr->fshift[0];
 784     facel            = _mm_set1_ps(fr->ic->epsfac);
 785     charge           = mdatoms->chargeA;
 786     krf              = _mm_set1_ps(fr->ic->k_rf);
 787     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 788     crf              = _mm_set1_ps(fr->ic->c_rf);
 789     nvdwtype         = fr->ntype;
 790     vdwparam         = fr->nbfp;
 791     vdwtype          = mdatoms->typeA;
 792
 793     /* Setup water-specific parameters */
 794     inr              = nlist->iinr[0];
 795     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 796     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 797     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 798     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 799
 800     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 801     rcutoff_scalar   = fr->ic->rcoulomb;
 802     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 803     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 804
 805     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 806     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 807
 808     /* Avoid stupid compiler warnings */
 809     jnrA = jnrB = jnrC = jnrD = 0;
 810     j_coord_offsetA = 0;
 811     j_coord_offsetB = 0;
 812     j_coord_offsetC = 0;
 813     j_coord_offsetD = 0;
 814
 815     outeriter        = 0;
 816     inneriter        = 0;
 817
 818     for(iidx=0;iidx<4*DIM;iidx++)
 819     {
 820         scratch[iidx] = 0.0;
 821     }
 822
 823     /* Start outer loop over neighborlists */
 824     for(iidx=0; iidx<nri; iidx++)
 825     {
 826         /* Load shift vector for this list */
 827         i_shift_offset   = DIM*shiftidx[iidx];
 828
 829         /* Load limits for loop over neighbors */
 830         j_index_start    = jindex[iidx];
 831         j_index_end      = jindex[iidx+1];
 832
 833         /* Get outer coordinate index */
 834         inr              = iinr[iidx];
 835         i_coord_offset   = DIM*inr;
 836
 837         /* Load i particle coords and add shift vector */
 838         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 839                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 840
 841         fix0             = _mm_setzero_ps();
 842         fiy0             = _mm_setzero_ps();
 843         fiz0             = _mm_setzero_ps();
 844         fix1             = _mm_setzero_ps();
 845         fiy1             = _mm_setzero_ps();
 846         fiz1             = _mm_setzero_ps();
 847         fix2             = _mm_setzero_ps();
 848         fiy2             = _mm_setzero_ps();
 849         fiz2             = _mm_setzero_ps();
 850         fix3             = _mm_setzero_ps();
 851         fiy3             = _mm_setzero_ps();
 852         fiz3             = _mm_setzero_ps();
 853
 854         /* Start inner kernel loop */
 855         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 856         {
 857
 858             /* Get j neighbor index, and coordinate index */
 859             jnrA             = jjnr[jidx];
 860             jnrB             = jjnr[jidx+1];
 861             jnrC             = jjnr[jidx+2];
 862             jnrD             = jjnr[jidx+3];
 863             j_coord_offsetA  = DIM*jnrA;
 864             j_coord_offsetB  = DIM*jnrB;
 865             j_coord_offsetC  = DIM*jnrC;
 866             j_coord_offsetD  = DIM*jnrD;
 867
 868             /* load j atom coordinates */
 869             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 870                                               x+j_coord_offsetC,x+j_coord_offsetD,
 871                                               &jx0,&jy0,&jz0);
 872
 873             /* Calculate displacement vector */
 874             dx00             = _mm_sub_ps(ix0,jx0);
 875             dy00             = _mm_sub_ps(iy0,jy0);
 876             dz00             = _mm_sub_ps(iz0,jz0);
 877             dx10             = _mm_sub_ps(ix1,jx0);
 878             dy10             = _mm_sub_ps(iy1,jy0);
 879             dz10             = _mm_sub_ps(iz1,jz0);
 880             dx20             = _mm_sub_ps(ix2,jx0);
 881             dy20             = _mm_sub_ps(iy2,jy0);
 882             dz20             = _mm_sub_ps(iz2,jz0);
 883             dx30             = _mm_sub_ps(ix3,jx0);
 884             dy30             = _mm_sub_ps(iy3,jy0);
 885             dz30             = _mm_sub_ps(iz3,jz0);
 886
 887             /* Calculate squared distance and things based on it */
 888             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 889             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 890             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 891             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 892
 893             rinv10           = sse41_invsqrt_f(rsq10);
 894             rinv20           = sse41_invsqrt_f(rsq20);
 895             rinv30           = sse41_invsqrt_f(rsq30);
 896
 897             rinvsq00         = sse41_inv_f(rsq00);
 898             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 899             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 900             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 901
 902             /* Load parameters for j particles */
 903             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 904                                                               charge+jnrC+0,charge+jnrD+0);
 905             vdwjidx0A        = 2*vdwtype[jnrA+0];
 906             vdwjidx0B        = 2*vdwtype[jnrB+0];
 907             vdwjidx0C        = 2*vdwtype[jnrC+0];
 908             vdwjidx0D        = 2*vdwtype[jnrD+0];
 909
 910             fjx0             = _mm_setzero_ps();
 911             fjy0             = _mm_setzero_ps();
 912             fjz0             = _mm_setzero_ps();
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             if (gmx_mm_any_lt(rsq00,rcutoff2))
 919             {
 920
 921             /* Compute parameters for interactions between i and j atoms */
 922             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 923                                          vdwparam+vdwioffset0+vdwjidx0B,
 924                                          vdwparam+vdwioffset0+vdwjidx0C,
 925                                          vdwparam+vdwioffset0+vdwjidx0D,
 926                                          &c6_00,&c12_00);
 927
 928             /* LENNARD-JONES DISPERSION/REPULSION */
 929
 930             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 931             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 932
 933             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 934
 935             fscal            = fvdw;
 936
 937             fscal            = _mm_and_ps(fscal,cutoff_mask);
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm_mul_ps(fscal,dx00);
 941             ty               = _mm_mul_ps(fscal,dy00);
 942             tz               = _mm_mul_ps(fscal,dz00);
 943
 944             /* Update vectorial force */
 945             fix0             = _mm_add_ps(fix0,tx);
 946             fiy0             = _mm_add_ps(fiy0,ty);
 947             fiz0             = _mm_add_ps(fiz0,tz);
 948
 949             fjx0             = _mm_add_ps(fjx0,tx);
 950             fjy0             = _mm_add_ps(fjy0,ty);
 951             fjz0             = _mm_add_ps(fjz0,tz);
 952
 953             }
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             if (gmx_mm_any_lt(rsq10,rcutoff2))
 960             {
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq10             = _mm_mul_ps(iq1,jq0);
 964
 965             /* REACTION-FIELD ELECTROSTATICS */
 966             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 967
 968             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 969
 970             fscal            = felec;
 971
 972             fscal            = _mm_and_ps(fscal,cutoff_mask);
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm_mul_ps(fscal,dx10);
 976             ty               = _mm_mul_ps(fscal,dy10);
 977             tz               = _mm_mul_ps(fscal,dz10);
 978
 979             /* Update vectorial force */
 980             fix1             = _mm_add_ps(fix1,tx);
 981             fiy1             = _mm_add_ps(fiy1,ty);
 982             fiz1             = _mm_add_ps(fiz1,tz);
 983
 984             fjx0             = _mm_add_ps(fjx0,tx);
 985             fjy0             = _mm_add_ps(fjy0,ty);
 986             fjz0             = _mm_add_ps(fjz0,tz);
 987
 988             }
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             if (gmx_mm_any_lt(rsq20,rcutoff2))
 995             {
 996
 997             /* Compute parameters for interactions between i and j atoms */
 998             qq20             = _mm_mul_ps(iq2,jq0);
 999
1000             /* REACTION-FIELD ELECTROSTATICS */
1001             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1002
1003             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1004
1005             fscal            = felec;
1006
1007             fscal            = _mm_and_ps(fscal,cutoff_mask);
1008
1009             /* Calculate temporary vectorial force */
1010             tx               = _mm_mul_ps(fscal,dx20);
1011             ty               = _mm_mul_ps(fscal,dy20);
1012             tz               = _mm_mul_ps(fscal,dz20);
1013
1014             /* Update vectorial force */
1015             fix2             = _mm_add_ps(fix2,tx);
1016             fiy2             = _mm_add_ps(fiy2,ty);
1017             fiz2             = _mm_add_ps(fiz2,tz);
1018
1019             fjx0             = _mm_add_ps(fjx0,tx);
1020             fjy0             = _mm_add_ps(fjy0,ty);
1021             fjz0             = _mm_add_ps(fjz0,tz);
1022
1023             }
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             if (gmx_mm_any_lt(rsq30,rcutoff2))
1030             {
1031
1032             /* Compute parameters for interactions between i and j atoms */
1033             qq30             = _mm_mul_ps(iq3,jq0);
1034
1035             /* REACTION-FIELD ELECTROSTATICS */
1036             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1037
1038             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1039
1040             fscal            = felec;
1041
1042             fscal            = _mm_and_ps(fscal,cutoff_mask);
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm_mul_ps(fscal,dx30);
1046             ty               = _mm_mul_ps(fscal,dy30);
1047             tz               = _mm_mul_ps(fscal,dz30);
1048
1049             /* Update vectorial force */
1050             fix3             = _mm_add_ps(fix3,tx);
1051             fiy3             = _mm_add_ps(fiy3,ty);
1052             fiz3             = _mm_add_ps(fiz3,tz);
1053
1054             fjx0             = _mm_add_ps(fjx0,tx);
1055             fjy0             = _mm_add_ps(fjy0,ty);
1056             fjz0             = _mm_add_ps(fjz0,tz);
1057
1058             }
1059
1060             fjptrA             = f+j_coord_offsetA;
1061             fjptrB             = f+j_coord_offsetB;
1062             fjptrC             = f+j_coord_offsetC;
1063             fjptrD             = f+j_coord_offsetD;
1064
1065             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1066
1067             /* Inner loop uses 120 flops */
1068         }
1069
1070         if(jidx<j_index_end)
1071         {
1072
1073             /* Get j neighbor index, and coordinate index */
1074             jnrlistA         = jjnr[jidx];
1075             jnrlistB         = jjnr[jidx+1];
1076             jnrlistC         = jjnr[jidx+2];
1077             jnrlistD         = jjnr[jidx+3];
1078             /* Sign of each element will be negative for non-real atoms.
1079              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1080              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1081              */
1082             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1083             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1084             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1085             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1086             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1087             j_coord_offsetA  = DIM*jnrA;
1088             j_coord_offsetB  = DIM*jnrB;
1089             j_coord_offsetC  = DIM*jnrC;
1090             j_coord_offsetD  = DIM*jnrD;
1091
1092             /* load j atom coordinates */
1093             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1094                                               x+j_coord_offsetC,x+j_coord_offsetD,
1095                                               &jx0,&jy0,&jz0);
1096
1097             /* Calculate displacement vector */
1098             dx00             = _mm_sub_ps(ix0,jx0);
1099             dy00             = _mm_sub_ps(iy0,jy0);
1100             dz00             = _mm_sub_ps(iz0,jz0);
1101             dx10             = _mm_sub_ps(ix1,jx0);
1102             dy10             = _mm_sub_ps(iy1,jy0);
1103             dz10             = _mm_sub_ps(iz1,jz0);
1104             dx20             = _mm_sub_ps(ix2,jx0);
1105             dy20             = _mm_sub_ps(iy2,jy0);
1106             dz20             = _mm_sub_ps(iz2,jz0);
1107             dx30             = _mm_sub_ps(ix3,jx0);
1108             dy30             = _mm_sub_ps(iy3,jy0);
1109             dz30             = _mm_sub_ps(iz3,jz0);
1110
1111             /* Calculate squared distance and things based on it */
1112             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1113             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1114             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1115             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1116
1117             rinv10           = sse41_invsqrt_f(rsq10);
1118             rinv20           = sse41_invsqrt_f(rsq20);
1119             rinv30           = sse41_invsqrt_f(rsq30);
1120
1121             rinvsq00         = sse41_inv_f(rsq00);
1122             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1123             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1124             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1125
1126             /* Load parameters for j particles */
1127             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1128                                                               charge+jnrC+0,charge+jnrD+0);
1129             vdwjidx0A        = 2*vdwtype[jnrA+0];
1130             vdwjidx0B        = 2*vdwtype[jnrB+0];
1131             vdwjidx0C        = 2*vdwtype[jnrC+0];
1132             vdwjidx0D        = 2*vdwtype[jnrD+0];
1133
1134             fjx0             = _mm_setzero_ps();
1135             fjy0             = _mm_setzero_ps();
1136             fjz0             = _mm_setzero_ps();
1137
1138             /**************************
1139              * CALCULATE INTERACTIONS *
1140              **************************/
1141
1142             if (gmx_mm_any_lt(rsq00,rcutoff2))
1143             {
1144
1145             /* Compute parameters for interactions between i and j atoms */
1146             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1147                                          vdwparam+vdwioffset0+vdwjidx0B,
1148                                          vdwparam+vdwioffset0+vdwjidx0C,
1149                                          vdwparam+vdwioffset0+vdwjidx0D,
1150                                          &c6_00,&c12_00);
1151
1152             /* LENNARD-JONES DISPERSION/REPULSION */
1153
1154             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1155             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1156
1157             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1158
1159             fscal            = fvdw;
1160
1161             fscal            = _mm_and_ps(fscal,cutoff_mask);
1162
1163             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1164
1165             /* Calculate temporary vectorial force */
1166             tx               = _mm_mul_ps(fscal,dx00);
1167             ty               = _mm_mul_ps(fscal,dy00);
1168             tz               = _mm_mul_ps(fscal,dz00);
1169
1170             /* Update vectorial force */
1171             fix0             = _mm_add_ps(fix0,tx);
1172             fiy0             = _mm_add_ps(fiy0,ty);
1173             fiz0             = _mm_add_ps(fiz0,tz);
1174
1175             fjx0             = _mm_add_ps(fjx0,tx);
1176             fjy0             = _mm_add_ps(fjy0,ty);
1177             fjz0             = _mm_add_ps(fjz0,tz);
1178
1179             }
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             if (gmx_mm_any_lt(rsq10,rcutoff2))
1186             {
1187
1188             /* Compute parameters for interactions between i and j atoms */
1189             qq10             = _mm_mul_ps(iq1,jq0);
1190
1191             /* REACTION-FIELD ELECTROSTATICS */
1192             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1193
1194             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1195
1196             fscal            = felec;
1197
1198             fscal            = _mm_and_ps(fscal,cutoff_mask);
1199
1200             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm_mul_ps(fscal,dx10);
1204             ty               = _mm_mul_ps(fscal,dy10);
1205             tz               = _mm_mul_ps(fscal,dz10);
1206
1207             /* Update vectorial force */
1208             fix1             = _mm_add_ps(fix1,tx);
1209             fiy1             = _mm_add_ps(fiy1,ty);
1210             fiz1             = _mm_add_ps(fiz1,tz);
1211
1212             fjx0             = _mm_add_ps(fjx0,tx);
1213             fjy0             = _mm_add_ps(fjy0,ty);
1214             fjz0             = _mm_add_ps(fjz0,tz);
1215
1216             }
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             if (gmx_mm_any_lt(rsq20,rcutoff2))
1223             {
1224
1225             /* Compute parameters for interactions between i and j atoms */
1226             qq20             = _mm_mul_ps(iq2,jq0);
1227
1228             /* REACTION-FIELD ELECTROSTATICS */
1229             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1230
1231             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1232
1233             fscal            = felec;
1234
1235             fscal            = _mm_and_ps(fscal,cutoff_mask);
1236
1237             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1238
1239             /* Calculate temporary vectorial force */
1240             tx               = _mm_mul_ps(fscal,dx20);
1241             ty               = _mm_mul_ps(fscal,dy20);
1242             tz               = _mm_mul_ps(fscal,dz20);
1243
1244             /* Update vectorial force */
1245             fix2             = _mm_add_ps(fix2,tx);
1246             fiy2             = _mm_add_ps(fiy2,ty);
1247             fiz2             = _mm_add_ps(fiz2,tz);
1248
1249             fjx0             = _mm_add_ps(fjx0,tx);
1250             fjy0             = _mm_add_ps(fjy0,ty);
1251             fjz0             = _mm_add_ps(fjz0,tz);
1252
1253             }
1254
1255             /**************************
1256              * CALCULATE INTERACTIONS *
1257              **************************/
1258
1259             if (gmx_mm_any_lt(rsq30,rcutoff2))
1260             {
1261
1262             /* Compute parameters for interactions between i and j atoms */
1263             qq30             = _mm_mul_ps(iq3,jq0);
1264
1265             /* REACTION-FIELD ELECTROSTATICS */
1266             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1267
1268             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1269
1270             fscal            = felec;
1271
1272             fscal            = _mm_and_ps(fscal,cutoff_mask);
1273
1274             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1275
1276             /* Calculate temporary vectorial force */
1277             tx               = _mm_mul_ps(fscal,dx30);
1278             ty               = _mm_mul_ps(fscal,dy30);
1279             tz               = _mm_mul_ps(fscal,dz30);
1280
1281             /* Update vectorial force */
1282             fix3             = _mm_add_ps(fix3,tx);
1283             fiy3             = _mm_add_ps(fiy3,ty);
1284             fiz3             = _mm_add_ps(fiz3,tz);
1285
1286             fjx0             = _mm_add_ps(fjx0,tx);
1287             fjy0             = _mm_add_ps(fjy0,ty);
1288             fjz0             = _mm_add_ps(fjz0,tz);
1289
1290             }
1291
1292             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1293             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1294             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1295             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1296
1297             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1298
1299             /* Inner loop uses 120 flops */
1300         }
1301
1302         /* End of innermost loop */
1303
1304         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1305                                               f+i_coord_offset,fshift+i_shift_offset);
1306
1307         /* Increment number of inner iterations */
1308         inneriter                  += j_index_end - j_index_start;
1309
1310         /* Outer loop uses 24 flops */
1311     }
1312
1313     /* Increment number of outer iterations */
1314     outeriter        += nri;
1315
1316     /* Update outer/inner flops */
1317
1318     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1319 }