src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse2_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     krf              = _mm_set1_ps(fr->ic->k_rf);
 125     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 126     crf              = _mm_set1_ps(fr->ic->c_rf);
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 139     rcutoff_scalar   = fr->rcoulomb;
 140     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 141     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 142
 143     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 144     rvdw             = _mm_set1_ps(fr->rvdw);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     for(iidx=0;iidx<4*DIM;iidx++)
 157     {
 158         scratch[iidx] = 0.0;
 159     }
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 177                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 178
 179         fix0             = _mm_setzero_ps();
 180         fiy0             = _mm_setzero_ps();
 181         fiz0             = _mm_setzero_ps();
 182         fix1             = _mm_setzero_ps();
 183         fiy1             = _mm_setzero_ps();
 184         fiz1             = _mm_setzero_ps();
 185         fix2             = _mm_setzero_ps();
 186         fiy2             = _mm_setzero_ps();
 187         fiz2             = _mm_setzero_ps();
 188         fix3             = _mm_setzero_ps();
 189         fiy3             = _mm_setzero_ps();
 190         fiz3             = _mm_setzero_ps();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_ps();
 194         vvdwsum          = _mm_setzero_ps();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209
 210             /* load j atom coordinates */
 211             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 212                                               x+j_coord_offsetC,x+j_coord_offsetD,
 213                                               &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm_sub_ps(ix0,jx0);
 217             dy00             = _mm_sub_ps(iy0,jy0);
 218             dz00             = _mm_sub_ps(iz0,jz0);
 219             dx10             = _mm_sub_ps(ix1,jx0);
 220             dy10             = _mm_sub_ps(iy1,jy0);
 221             dz10             = _mm_sub_ps(iz1,jz0);
 222             dx20             = _mm_sub_ps(ix2,jx0);
 223             dy20             = _mm_sub_ps(iy2,jy0);
 224             dz20             = _mm_sub_ps(iz2,jz0);
 225             dx30             = _mm_sub_ps(ix3,jx0);
 226             dy30             = _mm_sub_ps(iy3,jy0);
 227             dz30             = _mm_sub_ps(iz3,jz0);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 231             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 232             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 233             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 234
 235             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 236             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 237             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 238
 239             rinvsq00         = gmx_mm_inv_ps(rsq00);
 240             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 241             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 242             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 246                                                               charge+jnrC+0,charge+jnrD+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251
 252             fjx0             = _mm_setzero_ps();
 253             fjy0             = _mm_setzero_ps();
 254             fjz0             = _mm_setzero_ps();
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm_any_lt(rsq00,rcutoff2))
 261             {
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 265                                          vdwparam+vdwioffset0+vdwjidx0B,
 266                                          vdwparam+vdwioffset0+vdwjidx0C,
 267                                          vdwparam+vdwioffset0+vdwjidx0D,
 268                                          &c6_00,&c12_00);
 269
 270             /* LENNARD-JONES DISPERSION/REPULSION */
 271
 272             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 273             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 274             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 275             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) ,
 276                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 277             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 278
 279             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 283             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 284
 285             fscal            = fvdw;
 286
 287             fscal            = _mm_and_ps(fscal,cutoff_mask);
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm_mul_ps(fscal,dx00);
 291             ty               = _mm_mul_ps(fscal,dy00);
 292             tz               = _mm_mul_ps(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm_add_ps(fix0,tx);
 296             fiy0             = _mm_add_ps(fiy0,ty);
 297             fiz0             = _mm_add_ps(fiz0,tz);
 298
 299             fjx0             = _mm_add_ps(fjx0,tx);
 300             fjy0             = _mm_add_ps(fjy0,ty);
 301             fjz0             = _mm_add_ps(fjz0,tz);
 302
 303             }
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm_any_lt(rsq10,rcutoff2))
 310             {
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm_mul_ps(iq1,jq0);
 314
 315             /* REACTION-FIELD ELECTROSTATICS */
 316             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 317             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 318
 319             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velec            = _mm_and_ps(velec,cutoff_mask);
 323             velecsum         = _mm_add_ps(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             fscal            = _mm_and_ps(fscal,cutoff_mask);
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm_mul_ps(fscal,dx10);
 331             ty               = _mm_mul_ps(fscal,dy10);
 332             tz               = _mm_mul_ps(fscal,dz10);
 333
 334             /* Update vectorial force */
 335             fix1             = _mm_add_ps(fix1,tx);
 336             fiy1             = _mm_add_ps(fiy1,ty);
 337             fiz1             = _mm_add_ps(fiz1,tz);
 338
 339             fjx0             = _mm_add_ps(fjx0,tx);
 340             fjy0             = _mm_add_ps(fjy0,ty);
 341             fjz0             = _mm_add_ps(fjz0,tz);
 342
 343             }
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             if (gmx_mm_any_lt(rsq20,rcutoff2))
 350             {
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm_mul_ps(iq2,jq0);
 354
 355             /* REACTION-FIELD ELECTROSTATICS */
 356             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 357             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 358
 359             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_and_ps(velec,cutoff_mask);
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             fscal            = _mm_and_ps(fscal,cutoff_mask);
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm_mul_ps(fscal,dx20);
 371             ty               = _mm_mul_ps(fscal,dy20);
 372             tz               = _mm_mul_ps(fscal,dz20);
 373
 374             /* Update vectorial force */
 375             fix2             = _mm_add_ps(fix2,tx);
 376             fiy2             = _mm_add_ps(fiy2,ty);
 377             fiz2             = _mm_add_ps(fiz2,tz);
 378
 379             fjx0             = _mm_add_ps(fjx0,tx);
 380             fjy0             = _mm_add_ps(fjy0,ty);
 381             fjz0             = _mm_add_ps(fjz0,tz);
 382
 383             }
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             if (gmx_mm_any_lt(rsq30,rcutoff2))
 390             {
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq30             = _mm_mul_ps(iq3,jq0);
 394
 395             /* REACTION-FIELD ELECTROSTATICS */
 396             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 397             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 398
 399             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velec            = _mm_and_ps(velec,cutoff_mask);
 403             velecsum         = _mm_add_ps(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             fscal            = _mm_and_ps(fscal,cutoff_mask);
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm_mul_ps(fscal,dx30);
 411             ty               = _mm_mul_ps(fscal,dy30);
 412             tz               = _mm_mul_ps(fscal,dz30);
 413
 414             /* Update vectorial force */
 415             fix3             = _mm_add_ps(fix3,tx);
 416             fiy3             = _mm_add_ps(fiy3,ty);
 417             fiz3             = _mm_add_ps(fiz3,tz);
 418
 419             fjx0             = _mm_add_ps(fjx0,tx);
 420             fjy0             = _mm_add_ps(fjy0,ty);
 421             fjz0             = _mm_add_ps(fjz0,tz);
 422
 423             }
 424
 425             fjptrA             = f+j_coord_offsetA;
 426             fjptrB             = f+j_coord_offsetB;
 427             fjptrC             = f+j_coord_offsetC;
 428             fjptrD             = f+j_coord_offsetD;
 429
 430             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 431
 432             /* Inner loop uses 149 flops */
 433         }
 434
 435         if(jidx<j_index_end)
 436         {
 437
 438             /* Get j neighbor index, and coordinate index */
 439             jnrlistA         = jjnr[jidx];
 440             jnrlistB         = jjnr[jidx+1];
 441             jnrlistC         = jjnr[jidx+2];
 442             jnrlistD         = jjnr[jidx+3];
 443             /* Sign of each element will be negative for non-real atoms.
 444              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 445              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 446              */
 447             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 448             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 449             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 450             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 451             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 452             j_coord_offsetA  = DIM*jnrA;
 453             j_coord_offsetB  = DIM*jnrB;
 454             j_coord_offsetC  = DIM*jnrC;
 455             j_coord_offsetD  = DIM*jnrD;
 456
 457             /* load j atom coordinates */
 458             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 459                                               x+j_coord_offsetC,x+j_coord_offsetD,
 460                                               &jx0,&jy0,&jz0);
 461
 462             /* Calculate displacement vector */
 463             dx00             = _mm_sub_ps(ix0,jx0);
 464             dy00             = _mm_sub_ps(iy0,jy0);
 465             dz00             = _mm_sub_ps(iz0,jz0);
 466             dx10             = _mm_sub_ps(ix1,jx0);
 467             dy10             = _mm_sub_ps(iy1,jy0);
 468             dz10             = _mm_sub_ps(iz1,jz0);
 469             dx20             = _mm_sub_ps(ix2,jx0);
 470             dy20             = _mm_sub_ps(iy2,jy0);
 471             dz20             = _mm_sub_ps(iz2,jz0);
 472             dx30             = _mm_sub_ps(ix3,jx0);
 473             dy30             = _mm_sub_ps(iy3,jy0);
 474             dz30             = _mm_sub_ps(iz3,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 478             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 479             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 480             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 481
 482             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 483             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 484             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 485
 486             rinvsq00         = gmx_mm_inv_ps(rsq00);
 487             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 488             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 489             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 490
 491             /* Load parameters for j particles */
 492             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 493                                                               charge+jnrC+0,charge+jnrD+0);
 494             vdwjidx0A        = 2*vdwtype[jnrA+0];
 495             vdwjidx0B        = 2*vdwtype[jnrB+0];
 496             vdwjidx0C        = 2*vdwtype[jnrC+0];
 497             vdwjidx0D        = 2*vdwtype[jnrD+0];
 498
 499             fjx0             = _mm_setzero_ps();
 500             fjy0             = _mm_setzero_ps();
 501             fjz0             = _mm_setzero_ps();
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             if (gmx_mm_any_lt(rsq00,rcutoff2))
 508             {
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 512                                          vdwparam+vdwioffset0+vdwjidx0B,
 513                                          vdwparam+vdwioffset0+vdwjidx0C,
 514                                          vdwparam+vdwioffset0+vdwjidx0D,
 515                                          &c6_00,&c12_00);
 516
 517             /* LENNARD-JONES DISPERSION/REPULSION */
 518
 519             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 520             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 521             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 522             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) ,
 523                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 524             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 525
 526             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 527
 528             /* Update potential sum for this i atom from the interaction with this j atom. */
 529             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 530             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 531             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 532
 533             fscal            = fvdw;
 534
 535             fscal            = _mm_and_ps(fscal,cutoff_mask);
 536
 537             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm_mul_ps(fscal,dx00);
 541             ty               = _mm_mul_ps(fscal,dy00);
 542             tz               = _mm_mul_ps(fscal,dz00);
 543
 544             /* Update vectorial force */
 545             fix0             = _mm_add_ps(fix0,tx);
 546             fiy0             = _mm_add_ps(fiy0,ty);
 547             fiz0             = _mm_add_ps(fiz0,tz);
 548
 549             fjx0             = _mm_add_ps(fjx0,tx);
 550             fjy0             = _mm_add_ps(fjy0,ty);
 551             fjz0             = _mm_add_ps(fjz0,tz);
 552
 553             }
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm_any_lt(rsq10,rcutoff2))
 560             {
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq10             = _mm_mul_ps(iq1,jq0);
 564
 565             /* REACTION-FIELD ELECTROSTATICS */
 566             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 567             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 568
 569             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm_and_ps(velec,cutoff_mask);
 573             velec            = _mm_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_and_ps(fscal,cutoff_mask);
 579
 580             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm_mul_ps(fscal,dx10);
 584             ty               = _mm_mul_ps(fscal,dy10);
 585             tz               = _mm_mul_ps(fscal,dz10);
 586
 587             /* Update vectorial force */
 588             fix1             = _mm_add_ps(fix1,tx);
 589             fiy1             = _mm_add_ps(fiy1,ty);
 590             fiz1             = _mm_add_ps(fiz1,tz);
 591
 592             fjx0             = _mm_add_ps(fjx0,tx);
 593             fjy0             = _mm_add_ps(fjy0,ty);
 594             fjz0             = _mm_add_ps(fjz0,tz);
 595
 596             }
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             if (gmx_mm_any_lt(rsq20,rcutoff2))
 603             {
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq20             = _mm_mul_ps(iq2,jq0);
 607
 608             /* REACTION-FIELD ELECTROSTATICS */
 609             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 610             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 611
 612             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm_and_ps(velec,cutoff_mask);
 616             velec            = _mm_andnot_ps(dummy_mask,velec);
 617             velecsum         = _mm_add_ps(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm_and_ps(fscal,cutoff_mask);
 622
 623             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm_mul_ps(fscal,dx20);
 627             ty               = _mm_mul_ps(fscal,dy20);
 628             tz               = _mm_mul_ps(fscal,dz20);
 629
 630             /* Update vectorial force */
 631             fix2             = _mm_add_ps(fix2,tx);
 632             fiy2             = _mm_add_ps(fiy2,ty);
 633             fiz2             = _mm_add_ps(fiz2,tz);
 634
 635             fjx0             = _mm_add_ps(fjx0,tx);
 636             fjy0             = _mm_add_ps(fjy0,ty);
 637             fjz0             = _mm_add_ps(fjz0,tz);
 638
 639             }
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             if (gmx_mm_any_lt(rsq30,rcutoff2))
 646             {
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             qq30             = _mm_mul_ps(iq3,jq0);
 650
 651             /* REACTION-FIELD ELECTROSTATICS */
 652             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 653             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 654
 655             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 656
 657             /* Update potential sum for this i atom from the interaction with this j atom. */
 658             velec            = _mm_and_ps(velec,cutoff_mask);
 659             velec            = _mm_andnot_ps(dummy_mask,velec);
 660             velecsum         = _mm_add_ps(velecsum,velec);
 661
 662             fscal            = felec;
 663
 664             fscal            = _mm_and_ps(fscal,cutoff_mask);
 665
 666             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm_mul_ps(fscal,dx30);
 670             ty               = _mm_mul_ps(fscal,dy30);
 671             tz               = _mm_mul_ps(fscal,dz30);
 672
 673             /* Update vectorial force */
 674             fix3             = _mm_add_ps(fix3,tx);
 675             fiy3             = _mm_add_ps(fiy3,ty);
 676             fiz3             = _mm_add_ps(fiz3,tz);
 677
 678             fjx0             = _mm_add_ps(fjx0,tx);
 679             fjy0             = _mm_add_ps(fjy0,ty);
 680             fjz0             = _mm_add_ps(fjz0,tz);
 681
 682             }
 683
 684             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 685             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 686             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 687             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 688
 689             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 690
 691             /* Inner loop uses 149 flops */
 692         }
 693
 694         /* End of innermost loop */
 695
 696         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 697                                               f+i_coord_offset,fshift+i_shift_offset);
 698
 699         ggid                        = gid[iidx];
 700         /* Update potential energies */
 701         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 702         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 703
 704         /* Increment number of inner iterations */
 705         inneriter                  += j_index_end - j_index_start;
 706
 707         /* Outer loop uses 26 flops */
 708     }
 709
 710     /* Increment number of outer iterations */
 711     outeriter        += nri;
 712
 713     /* Update outer/inner flops */
 714
 715     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*149);
 716 }
 717 /*
 718  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse2_single
 719  * Electrostatics interaction: ReactionField
 720  * VdW interaction:            LennardJones
 721  * Geometry:                   Water4-Particle
 722  * Calculate force/pot:        Force
 723  */
 724 void
 725 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse2_single
 726                     (t_nblist                    * gmx_restrict       nlist,
 727                      rvec                        * gmx_restrict          xx,
 728                      rvec                        * gmx_restrict          ff,
 729                      t_forcerec                  * gmx_restrict          fr,
 730                      t_mdatoms                   * gmx_restrict     mdatoms,
 731                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 732                      t_nrnb                      * gmx_restrict        nrnb)
 733 {
 734     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 735      * just 0 for non-waters.
 736      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 737      * jnr indices corresponding to data put in the four positions in the SIMD register.
 738      */
 739     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 740     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 741     int              jnrA,jnrB,jnrC,jnrD;
 742     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 743     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 744     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 745     real             rcutoff_scalar;
 746     real             *shiftvec,*fshift,*x,*f;
 747     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 748     real             scratch[4*DIM];
 749     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 750     int              vdwioffset0;
 751     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 752     int              vdwioffset1;
 753     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 754     int              vdwioffset2;
 755     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 756     int              vdwioffset3;
 757     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 758     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 759     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 760     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 761     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 762     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 763     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 764     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 765     real             *charge;
 766     int              nvdwtype;
 767     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 768     int              *vdwtype;
 769     real             *vdwparam;
 770     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 771     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 772     __m128           dummy_mask,cutoff_mask;
 773     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 774     __m128           one     = _mm_set1_ps(1.0);
 775     __m128           two     = _mm_set1_ps(2.0);
 776     x                = xx[0];
 777     f                = ff[0];
 778
 779     nri              = nlist->nri;
 780     iinr             = nlist->iinr;
 781     jindex           = nlist->jindex;
 782     jjnr             = nlist->jjnr;
 783     shiftidx         = nlist->shift;
 784     gid              = nlist->gid;
 785     shiftvec         = fr->shift_vec[0];
 786     fshift           = fr->fshift[0];
 787     facel            = _mm_set1_ps(fr->epsfac);
 788     charge           = mdatoms->chargeA;
 789     krf              = _mm_set1_ps(fr->ic->k_rf);
 790     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 791     crf              = _mm_set1_ps(fr->ic->c_rf);
 792     nvdwtype         = fr->ntype;
 793     vdwparam         = fr->nbfp;
 794     vdwtype          = mdatoms->typeA;
 795
 796     /* Setup water-specific parameters */
 797     inr              = nlist->iinr[0];
 798     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 799     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 800     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 801     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 802
 803     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 804     rcutoff_scalar   = fr->rcoulomb;
 805     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 806     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 807
 808     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 809     rvdw             = _mm_set1_ps(fr->rvdw);
 810
 811     /* Avoid stupid compiler warnings */
 812     jnrA = jnrB = jnrC = jnrD = 0;
 813     j_coord_offsetA = 0;
 814     j_coord_offsetB = 0;
 815     j_coord_offsetC = 0;
 816     j_coord_offsetD = 0;
 817
 818     outeriter        = 0;
 819     inneriter        = 0;
 820
 821     for(iidx=0;iidx<4*DIM;iidx++)
 822     {
 823         scratch[iidx] = 0.0;
 824     }
 825
 826     /* Start outer loop over neighborlists */
 827     for(iidx=0; iidx<nri; iidx++)
 828     {
 829         /* Load shift vector for this list */
 830         i_shift_offset   = DIM*shiftidx[iidx];
 831
 832         /* Load limits for loop over neighbors */
 833         j_index_start    = jindex[iidx];
 834         j_index_end      = jindex[iidx+1];
 835
 836         /* Get outer coordinate index */
 837         inr              = iinr[iidx];
 838         i_coord_offset   = DIM*inr;
 839
 840         /* Load i particle coords and add shift vector */
 841         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 842                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 843
 844         fix0             = _mm_setzero_ps();
 845         fiy0             = _mm_setzero_ps();
 846         fiz0             = _mm_setzero_ps();
 847         fix1             = _mm_setzero_ps();
 848         fiy1             = _mm_setzero_ps();
 849         fiz1             = _mm_setzero_ps();
 850         fix2             = _mm_setzero_ps();
 851         fiy2             = _mm_setzero_ps();
 852         fiz2             = _mm_setzero_ps();
 853         fix3             = _mm_setzero_ps();
 854         fiy3             = _mm_setzero_ps();
 855         fiz3             = _mm_setzero_ps();
 856
 857         /* Start inner kernel loop */
 858         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 859         {
 860
 861             /* Get j neighbor index, and coordinate index */
 862             jnrA             = jjnr[jidx];
 863             jnrB             = jjnr[jidx+1];
 864             jnrC             = jjnr[jidx+2];
 865             jnrD             = jjnr[jidx+3];
 866             j_coord_offsetA  = DIM*jnrA;
 867             j_coord_offsetB  = DIM*jnrB;
 868             j_coord_offsetC  = DIM*jnrC;
 869             j_coord_offsetD  = DIM*jnrD;
 870
 871             /* load j atom coordinates */
 872             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 873                                               x+j_coord_offsetC,x+j_coord_offsetD,
 874                                               &jx0,&jy0,&jz0);
 875
 876             /* Calculate displacement vector */
 877             dx00             = _mm_sub_ps(ix0,jx0);
 878             dy00             = _mm_sub_ps(iy0,jy0);
 879             dz00             = _mm_sub_ps(iz0,jz0);
 880             dx10             = _mm_sub_ps(ix1,jx0);
 881             dy10             = _mm_sub_ps(iy1,jy0);
 882             dz10             = _mm_sub_ps(iz1,jz0);
 883             dx20             = _mm_sub_ps(ix2,jx0);
 884             dy20             = _mm_sub_ps(iy2,jy0);
 885             dz20             = _mm_sub_ps(iz2,jz0);
 886             dx30             = _mm_sub_ps(ix3,jx0);
 887             dy30             = _mm_sub_ps(iy3,jy0);
 888             dz30             = _mm_sub_ps(iz3,jz0);
 889
 890             /* Calculate squared distance and things based on it */
 891             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 892             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 893             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 894             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 895
 896             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 897             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 898             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 899
 900             rinvsq00         = gmx_mm_inv_ps(rsq00);
 901             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 902             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 903             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 904
 905             /* Load parameters for j particles */
 906             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 907                                                               charge+jnrC+0,charge+jnrD+0);
 908             vdwjidx0A        = 2*vdwtype[jnrA+0];
 909             vdwjidx0B        = 2*vdwtype[jnrB+0];
 910             vdwjidx0C        = 2*vdwtype[jnrC+0];
 911             vdwjidx0D        = 2*vdwtype[jnrD+0];
 912
 913             fjx0             = _mm_setzero_ps();
 914             fjy0             = _mm_setzero_ps();
 915             fjz0             = _mm_setzero_ps();
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             if (gmx_mm_any_lt(rsq00,rcutoff2))
 922             {
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 926                                          vdwparam+vdwioffset0+vdwjidx0B,
 927                                          vdwparam+vdwioffset0+vdwjidx0C,
 928                                          vdwparam+vdwioffset0+vdwjidx0D,
 929                                          &c6_00,&c12_00);
 930
 931             /* LENNARD-JONES DISPERSION/REPULSION */
 932
 933             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 934             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 935
 936             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 937
 938             fscal            = fvdw;
 939
 940             fscal            = _mm_and_ps(fscal,cutoff_mask);
 941
 942             /* Calculate temporary vectorial force */
 943             tx               = _mm_mul_ps(fscal,dx00);
 944             ty               = _mm_mul_ps(fscal,dy00);
 945             tz               = _mm_mul_ps(fscal,dz00);
 946
 947             /* Update vectorial force */
 948             fix0             = _mm_add_ps(fix0,tx);
 949             fiy0             = _mm_add_ps(fiy0,ty);
 950             fiz0             = _mm_add_ps(fiz0,tz);
 951
 952             fjx0             = _mm_add_ps(fjx0,tx);
 953             fjy0             = _mm_add_ps(fjy0,ty);
 954             fjz0             = _mm_add_ps(fjz0,tz);
 955
 956             }
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             if (gmx_mm_any_lt(rsq10,rcutoff2))
 963             {
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             qq10             = _mm_mul_ps(iq1,jq0);
 967
 968             /* REACTION-FIELD ELECTROSTATICS */
 969             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 970
 971             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm_and_ps(fscal,cutoff_mask);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm_mul_ps(fscal,dx10);
 979             ty               = _mm_mul_ps(fscal,dy10);
 980             tz               = _mm_mul_ps(fscal,dz10);
 981
 982             /* Update vectorial force */
 983             fix1             = _mm_add_ps(fix1,tx);
 984             fiy1             = _mm_add_ps(fiy1,ty);
 985             fiz1             = _mm_add_ps(fiz1,tz);
 986
 987             fjx0             = _mm_add_ps(fjx0,tx);
 988             fjy0             = _mm_add_ps(fjy0,ty);
 989             fjz0             = _mm_add_ps(fjz0,tz);
 990
 991             }
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             if (gmx_mm_any_lt(rsq20,rcutoff2))
 998             {
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq20             = _mm_mul_ps(iq2,jq0);
1002
1003             /* REACTION-FIELD ELECTROSTATICS */
1004             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1005
1006             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm_and_ps(fscal,cutoff_mask);
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm_mul_ps(fscal,dx20);
1014             ty               = _mm_mul_ps(fscal,dy20);
1015             tz               = _mm_mul_ps(fscal,dz20);
1016
1017             /* Update vectorial force */
1018             fix2             = _mm_add_ps(fix2,tx);
1019             fiy2             = _mm_add_ps(fiy2,ty);
1020             fiz2             = _mm_add_ps(fiz2,tz);
1021
1022             fjx0             = _mm_add_ps(fjx0,tx);
1023             fjy0             = _mm_add_ps(fjy0,ty);
1024             fjz0             = _mm_add_ps(fjz0,tz);
1025
1026             }
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             if (gmx_mm_any_lt(rsq30,rcutoff2))
1033             {
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq30             = _mm_mul_ps(iq3,jq0);
1037
1038             /* REACTION-FIELD ELECTROSTATICS */
1039             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1040
1041             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm_and_ps(fscal,cutoff_mask);
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm_mul_ps(fscal,dx30);
1049             ty               = _mm_mul_ps(fscal,dy30);
1050             tz               = _mm_mul_ps(fscal,dz30);
1051
1052             /* Update vectorial force */
1053             fix3             = _mm_add_ps(fix3,tx);
1054             fiy3             = _mm_add_ps(fiy3,ty);
1055             fiz3             = _mm_add_ps(fiz3,tz);
1056
1057             fjx0             = _mm_add_ps(fjx0,tx);
1058             fjy0             = _mm_add_ps(fjy0,ty);
1059             fjz0             = _mm_add_ps(fjz0,tz);
1060
1061             }
1062
1063             fjptrA             = f+j_coord_offsetA;
1064             fjptrB             = f+j_coord_offsetB;
1065             fjptrC             = f+j_coord_offsetC;
1066             fjptrD             = f+j_coord_offsetD;
1067
1068             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1069
1070             /* Inner loop uses 120 flops */
1071         }
1072
1073         if(jidx<j_index_end)
1074         {
1075
1076             /* Get j neighbor index, and coordinate index */
1077             jnrlistA         = jjnr[jidx];
1078             jnrlistB         = jjnr[jidx+1];
1079             jnrlistC         = jjnr[jidx+2];
1080             jnrlistD         = jjnr[jidx+3];
1081             /* Sign of each element will be negative for non-real atoms.
1082              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1083              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1084              */
1085             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1086             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1087             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1088             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1089             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1090             j_coord_offsetA  = DIM*jnrA;
1091             j_coord_offsetB  = DIM*jnrB;
1092             j_coord_offsetC  = DIM*jnrC;
1093             j_coord_offsetD  = DIM*jnrD;
1094
1095             /* load j atom coordinates */
1096             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1097                                               x+j_coord_offsetC,x+j_coord_offsetD,
1098                                               &jx0,&jy0,&jz0);
1099
1100             /* Calculate displacement vector */
1101             dx00             = _mm_sub_ps(ix0,jx0);
1102             dy00             = _mm_sub_ps(iy0,jy0);
1103             dz00             = _mm_sub_ps(iz0,jz0);
1104             dx10             = _mm_sub_ps(ix1,jx0);
1105             dy10             = _mm_sub_ps(iy1,jy0);
1106             dz10             = _mm_sub_ps(iz1,jz0);
1107             dx20             = _mm_sub_ps(ix2,jx0);
1108             dy20             = _mm_sub_ps(iy2,jy0);
1109             dz20             = _mm_sub_ps(iz2,jz0);
1110             dx30             = _mm_sub_ps(ix3,jx0);
1111             dy30             = _mm_sub_ps(iy3,jy0);
1112             dz30             = _mm_sub_ps(iz3,jz0);
1113
1114             /* Calculate squared distance and things based on it */
1115             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1116             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1117             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1118             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1119
1120             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1121             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1122             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1123
1124             rinvsq00         = gmx_mm_inv_ps(rsq00);
1125             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1126             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1127             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1128
1129             /* Load parameters for j particles */
1130             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1131                                                               charge+jnrC+0,charge+jnrD+0);
1132             vdwjidx0A        = 2*vdwtype[jnrA+0];
1133             vdwjidx0B        = 2*vdwtype[jnrB+0];
1134             vdwjidx0C        = 2*vdwtype[jnrC+0];
1135             vdwjidx0D        = 2*vdwtype[jnrD+0];
1136
1137             fjx0             = _mm_setzero_ps();
1138             fjy0             = _mm_setzero_ps();
1139             fjz0             = _mm_setzero_ps();
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             if (gmx_mm_any_lt(rsq00,rcutoff2))
1146             {
1147
1148             /* Compute parameters for interactions between i and j atoms */
1149             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1150                                          vdwparam+vdwioffset0+vdwjidx0B,
1151                                          vdwparam+vdwioffset0+vdwjidx0C,
1152                                          vdwparam+vdwioffset0+vdwjidx0D,
1153                                          &c6_00,&c12_00);
1154
1155             /* LENNARD-JONES DISPERSION/REPULSION */
1156
1157             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1158             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1159
1160             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1161
1162             fscal            = fvdw;
1163
1164             fscal            = _mm_and_ps(fscal,cutoff_mask);
1165
1166             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1167
1168             /* Calculate temporary vectorial force */
1169             tx               = _mm_mul_ps(fscal,dx00);
1170             ty               = _mm_mul_ps(fscal,dy00);
1171             tz               = _mm_mul_ps(fscal,dz00);
1172
1173             /* Update vectorial force */
1174             fix0             = _mm_add_ps(fix0,tx);
1175             fiy0             = _mm_add_ps(fiy0,ty);
1176             fiz0             = _mm_add_ps(fiz0,tz);
1177
1178             fjx0             = _mm_add_ps(fjx0,tx);
1179             fjy0             = _mm_add_ps(fjy0,ty);
1180             fjz0             = _mm_add_ps(fjz0,tz);
1181
1182             }
1183
1184             /**************************
1185              * CALCULATE INTERACTIONS *
1186              **************************/
1187
1188             if (gmx_mm_any_lt(rsq10,rcutoff2))
1189             {
1190
1191             /* Compute parameters for interactions between i and j atoms */
1192             qq10             = _mm_mul_ps(iq1,jq0);
1193
1194             /* REACTION-FIELD ELECTROSTATICS */
1195             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1196
1197             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm_and_ps(fscal,cutoff_mask);
1202
1203             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1204
1205             /* Calculate temporary vectorial force */
1206             tx               = _mm_mul_ps(fscal,dx10);
1207             ty               = _mm_mul_ps(fscal,dy10);
1208             tz               = _mm_mul_ps(fscal,dz10);
1209
1210             /* Update vectorial force */
1211             fix1             = _mm_add_ps(fix1,tx);
1212             fiy1             = _mm_add_ps(fiy1,ty);
1213             fiz1             = _mm_add_ps(fiz1,tz);
1214
1215             fjx0             = _mm_add_ps(fjx0,tx);
1216             fjy0             = _mm_add_ps(fjy0,ty);
1217             fjz0             = _mm_add_ps(fjz0,tz);
1218
1219             }
1220
1221             /**************************
1222              * CALCULATE INTERACTIONS *
1223              **************************/
1224
1225             if (gmx_mm_any_lt(rsq20,rcutoff2))
1226             {
1227
1228             /* Compute parameters for interactions between i and j atoms */
1229             qq20             = _mm_mul_ps(iq2,jq0);
1230
1231             /* REACTION-FIELD ELECTROSTATICS */
1232             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1233
1234             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1235
1236             fscal            = felec;
1237
1238             fscal            = _mm_and_ps(fscal,cutoff_mask);
1239
1240             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1241
1242             /* Calculate temporary vectorial force */
1243             tx               = _mm_mul_ps(fscal,dx20);
1244             ty               = _mm_mul_ps(fscal,dy20);
1245             tz               = _mm_mul_ps(fscal,dz20);
1246
1247             /* Update vectorial force */
1248             fix2             = _mm_add_ps(fix2,tx);
1249             fiy2             = _mm_add_ps(fiy2,ty);
1250             fiz2             = _mm_add_ps(fiz2,tz);
1251
1252             fjx0             = _mm_add_ps(fjx0,tx);
1253             fjy0             = _mm_add_ps(fjy0,ty);
1254             fjz0             = _mm_add_ps(fjz0,tz);
1255
1256             }
1257
1258             /**************************
1259              * CALCULATE INTERACTIONS *
1260              **************************/
1261
1262             if (gmx_mm_any_lt(rsq30,rcutoff2))
1263             {
1264
1265             /* Compute parameters for interactions between i and j atoms */
1266             qq30             = _mm_mul_ps(iq3,jq0);
1267
1268             /* REACTION-FIELD ELECTROSTATICS */
1269             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1270
1271             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1272
1273             fscal            = felec;
1274
1275             fscal            = _mm_and_ps(fscal,cutoff_mask);
1276
1277             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1278
1279             /* Calculate temporary vectorial force */
1280             tx               = _mm_mul_ps(fscal,dx30);
1281             ty               = _mm_mul_ps(fscal,dy30);
1282             tz               = _mm_mul_ps(fscal,dz30);
1283
1284             /* Update vectorial force */
1285             fix3             = _mm_add_ps(fix3,tx);
1286             fiy3             = _mm_add_ps(fiy3,ty);
1287             fiz3             = _mm_add_ps(fiz3,tz);
1288
1289             fjx0             = _mm_add_ps(fjx0,tx);
1290             fjy0             = _mm_add_ps(fjy0,ty);
1291             fjz0             = _mm_add_ps(fjz0,tz);
1292
1293             }
1294
1295             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1296             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1297             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1298             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1299
1300             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1301
1302             /* Inner loop uses 120 flops */
1303         }
1304
1305         /* End of innermost loop */
1306
1307         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1308                                               f+i_coord_offset,fshift+i_shift_offset);
1309
1310         /* Increment number of inner iterations */
1311         inneriter                  += j_index_end - j_index_start;
1312
1313         /* Outer loop uses 24 flops */
1314     }
1315
1316     /* Increment number of outer iterations */
1317     outeriter        += nri;
1318
1319     /* Update outer/inner flops */
1320
1321     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1322 }