src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  89     real             rswitch_scalar,d_scalar;
  90     __m128           dummy_mask,cutoff_mask;
  91     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  92     __m128           one     = _mm_set1_ps(1.0);
  93     __m128           two     = _mm_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     krf              = _mm_set1_ps(fr->ic->k_rf);
 108     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 109     crf              = _mm_set1_ps(fr->ic->c_rf);
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 117     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 118     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 119     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->rcoulomb;
 123     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 125
 126     rswitch_scalar   = fr->rvdw_switch;
 127     rswitch          = _mm_set1_ps(rswitch_scalar);
 128     /* Setup switch parameters */
 129     d_scalar         = rcutoff_scalar-rswitch_scalar;
 130     d                = _mm_set1_ps(d_scalar);
 131     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 132     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 133     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 134     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 135     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _mm_setzero_ps();
 172         fiy0             = _mm_setzero_ps();
 173         fiz0             = _mm_setzero_ps();
 174         fix1             = _mm_setzero_ps();
 175         fiy1             = _mm_setzero_ps();
 176         fiz1             = _mm_setzero_ps();
 177         fix2             = _mm_setzero_ps();
 178         fiy2             = _mm_setzero_ps();
 179         fiz2             = _mm_setzero_ps();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_ps();
 183         vvdwsum          = _mm_setzero_ps();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208             dx10             = _mm_sub_ps(ix1,jx0);
 209             dy10             = _mm_sub_ps(iy1,jy0);
 210             dz10             = _mm_sub_ps(iz1,jz0);
 211             dx20             = _mm_sub_ps(ix2,jx0);
 212             dy20             = _mm_sub_ps(iy2,jy0);
 213             dz20             = _mm_sub_ps(iz2,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 217             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 218             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 219
 220             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 221             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 222             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 223
 224             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 225             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 226             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 230                                                               charge+jnrC+0,charge+jnrD+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235
 236             fjx0             = _mm_setzero_ps();
 237             fjy0             = _mm_setzero_ps();
 238             fjz0             = _mm_setzero_ps();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             if (gmx_mm_any_lt(rsq00,rcutoff2))
 245             {
 246
 247             r00              = _mm_mul_ps(rsq00,rinv00);
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             qq00             = _mm_mul_ps(iq0,jq0);
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* REACTION-FIELD ELECTROSTATICS */
 258             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 259             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 265             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 266             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 267             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 268
 269             d                = _mm_sub_ps(r00,rswitch);
 270             d                = _mm_max_ps(d,_mm_setzero_ps());
 271             d2               = _mm_mul_ps(d,d);
 272             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 273
 274             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 275
 276             /* Evaluate switch function */
 277             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 278             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 279             vvdw             = _mm_mul_ps(vvdw,sw);
 280             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velec            = _mm_and_ps(velec,cutoff_mask);
 284             velecsum         = _mm_add_ps(velecsum,velec);
 285             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 286             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 287
 288             fscal            = _mm_add_ps(felec,fvdw);
 289
 290             fscal            = _mm_and_ps(fscal,cutoff_mask);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm_mul_ps(fscal,dx00);
 294             ty               = _mm_mul_ps(fscal,dy00);
 295             tz               = _mm_mul_ps(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm_add_ps(fix0,tx);
 299             fiy0             = _mm_add_ps(fiy0,ty);
 300             fiz0             = _mm_add_ps(fiz0,tz);
 301
 302             fjx0             = _mm_add_ps(fjx0,tx);
 303             fjy0             = _mm_add_ps(fjy0,ty);
 304             fjz0             = _mm_add_ps(fjz0,tz);
 305
 306             }
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_mm_any_lt(rsq10,rcutoff2))
 313             {
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq10             = _mm_mul_ps(iq1,jq0);
 317
 318             /* REACTION-FIELD ELECTROSTATICS */
 319             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 320             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 321
 322             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velec            = _mm_and_ps(velec,cutoff_mask);
 326             velecsum         = _mm_add_ps(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             fscal            = _mm_and_ps(fscal,cutoff_mask);
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_ps(fscal,dx10);
 334             ty               = _mm_mul_ps(fscal,dy10);
 335             tz               = _mm_mul_ps(fscal,dz10);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm_add_ps(fix1,tx);
 339             fiy1             = _mm_add_ps(fiy1,ty);
 340             fiz1             = _mm_add_ps(fiz1,tz);
 341
 342             fjx0             = _mm_add_ps(fjx0,tx);
 343             fjy0             = _mm_add_ps(fjy0,ty);
 344             fjz0             = _mm_add_ps(fjz0,tz);
 345
 346             }
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             if (gmx_mm_any_lt(rsq20,rcutoff2))
 353             {
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq20             = _mm_mul_ps(iq2,jq0);
 357
 358             /* REACTION-FIELD ELECTROSTATICS */
 359             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 360             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 361
 362             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm_and_ps(velec,cutoff_mask);
 366             velecsum         = _mm_add_ps(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             fscal            = _mm_and_ps(fscal,cutoff_mask);
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm_mul_ps(fscal,dx20);
 374             ty               = _mm_mul_ps(fscal,dy20);
 375             tz               = _mm_mul_ps(fscal,dz20);
 376
 377             /* Update vectorial force */
 378             fix2             = _mm_add_ps(fix2,tx);
 379             fiy2             = _mm_add_ps(fiy2,ty);
 380             fiz2             = _mm_add_ps(fiz2,tz);
 381
 382             fjx0             = _mm_add_ps(fjx0,tx);
 383             fjy0             = _mm_add_ps(fjy0,ty);
 384             fjz0             = _mm_add_ps(fjz0,tz);
 385
 386             }
 387
 388             fjptrA             = f+j_coord_offsetA;
 389             fjptrB             = f+j_coord_offsetB;
 390             fjptrC             = f+j_coord_offsetC;
 391             fjptrD             = f+j_coord_offsetD;
 392
 393             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 394
 395             /* Inner loop uses 142 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             /* Get j neighbor index, and coordinate index */
 402             jnrlistA         = jjnr[jidx];
 403             jnrlistB         = jjnr[jidx+1];
 404             jnrlistC         = jjnr[jidx+2];
 405             jnrlistD         = jjnr[jidx+3];
 406             /* Sign of each element will be negative for non-real atoms.
 407              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 408              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 409              */
 410             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 411             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 412             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 413             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 414             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 415             j_coord_offsetA  = DIM*jnrA;
 416             j_coord_offsetB  = DIM*jnrB;
 417             j_coord_offsetC  = DIM*jnrC;
 418             j_coord_offsetD  = DIM*jnrD;
 419
 420             /* load j atom coordinates */
 421             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 422                                               x+j_coord_offsetC,x+j_coord_offsetD,
 423                                               &jx0,&jy0,&jz0);
 424
 425             /* Calculate displacement vector */
 426             dx00             = _mm_sub_ps(ix0,jx0);
 427             dy00             = _mm_sub_ps(iy0,jy0);
 428             dz00             = _mm_sub_ps(iz0,jz0);
 429             dx10             = _mm_sub_ps(ix1,jx0);
 430             dy10             = _mm_sub_ps(iy1,jy0);
 431             dz10             = _mm_sub_ps(iz1,jz0);
 432             dx20             = _mm_sub_ps(ix2,jx0);
 433             dy20             = _mm_sub_ps(iy2,jy0);
 434             dz20             = _mm_sub_ps(iz2,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 438             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 439             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 440
 441             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 442             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 443             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 444
 445             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 446             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 447             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 448
 449             /* Load parameters for j particles */
 450             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 451                                                               charge+jnrC+0,charge+jnrD+0);
 452             vdwjidx0A        = 2*vdwtype[jnrA+0];
 453             vdwjidx0B        = 2*vdwtype[jnrB+0];
 454             vdwjidx0C        = 2*vdwtype[jnrC+0];
 455             vdwjidx0D        = 2*vdwtype[jnrD+0];
 456
 457             fjx0             = _mm_setzero_ps();
 458             fjy0             = _mm_setzero_ps();
 459             fjz0             = _mm_setzero_ps();
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm_any_lt(rsq00,rcutoff2))
 466             {
 467
 468             r00              = _mm_mul_ps(rsq00,rinv00);
 469             r00              = _mm_andnot_ps(dummy_mask,r00);
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq00             = _mm_mul_ps(iq0,jq0);
 473             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 474                                          vdwparam+vdwioffset0+vdwjidx0B,
 475                                          vdwparam+vdwioffset0+vdwjidx0C,
 476                                          vdwparam+vdwioffset0+vdwjidx0D,
 477                                          &c6_00,&c12_00);
 478
 479             /* REACTION-FIELD ELECTROSTATICS */
 480             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 481             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 482
 483             /* LENNARD-JONES DISPERSION/REPULSION */
 484
 485             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 486             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 487             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 488             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 489             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 490
 491             d                = _mm_sub_ps(r00,rswitch);
 492             d                = _mm_max_ps(d,_mm_setzero_ps());
 493             d2               = _mm_mul_ps(d,d);
 494             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 495
 496             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 497
 498             /* Evaluate switch function */
 499             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 500             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 501             vvdw             = _mm_mul_ps(vvdw,sw);
 502             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm_and_ps(velec,cutoff_mask);
 506             velec            = _mm_andnot_ps(dummy_mask,velec);
 507             velecsum         = _mm_add_ps(velecsum,velec);
 508             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 509             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 510             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 511
 512             fscal            = _mm_add_ps(felec,fvdw);
 513
 514             fscal            = _mm_and_ps(fscal,cutoff_mask);
 515
 516             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_ps(fscal,dx00);
 520             ty               = _mm_mul_ps(fscal,dy00);
 521             tz               = _mm_mul_ps(fscal,dz00);
 522
 523             /* Update vectorial force */
 524             fix0             = _mm_add_ps(fix0,tx);
 525             fiy0             = _mm_add_ps(fiy0,ty);
 526             fiz0             = _mm_add_ps(fiz0,tz);
 527
 528             fjx0             = _mm_add_ps(fjx0,tx);
 529             fjy0             = _mm_add_ps(fjy0,ty);
 530             fjz0             = _mm_add_ps(fjz0,tz);
 531
 532             }
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             if (gmx_mm_any_lt(rsq10,rcutoff2))
 539             {
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq10             = _mm_mul_ps(iq1,jq0);
 543
 544             /* REACTION-FIELD ELECTROSTATICS */
 545             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 546             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 547
 548             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_and_ps(velec,cutoff_mask);
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_and_ps(fscal,cutoff_mask);
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm_mul_ps(fscal,dx10);
 563             ty               = _mm_mul_ps(fscal,dy10);
 564             tz               = _mm_mul_ps(fscal,dz10);
 565
 566             /* Update vectorial force */
 567             fix1             = _mm_add_ps(fix1,tx);
 568             fiy1             = _mm_add_ps(fiy1,ty);
 569             fiz1             = _mm_add_ps(fiz1,tz);
 570
 571             fjx0             = _mm_add_ps(fjx0,tx);
 572             fjy0             = _mm_add_ps(fjy0,ty);
 573             fjz0             = _mm_add_ps(fjz0,tz);
 574
 575             }
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             if (gmx_mm_any_lt(rsq20,rcutoff2))
 582             {
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq20             = _mm_mul_ps(iq2,jq0);
 586
 587             /* REACTION-FIELD ELECTROSTATICS */
 588             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 589             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 590
 591             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm_and_ps(velec,cutoff_mask);
 595             velec            = _mm_andnot_ps(dummy_mask,velec);
 596             velecsum         = _mm_add_ps(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm_and_ps(fscal,cutoff_mask);
 601
 602             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm_mul_ps(fscal,dx20);
 606             ty               = _mm_mul_ps(fscal,dy20);
 607             tz               = _mm_mul_ps(fscal,dz20);
 608
 609             /* Update vectorial force */
 610             fix2             = _mm_add_ps(fix2,tx);
 611             fiy2             = _mm_add_ps(fiy2,ty);
 612             fiz2             = _mm_add_ps(fiz2,tz);
 613
 614             fjx0             = _mm_add_ps(fjx0,tx);
 615             fjy0             = _mm_add_ps(fjy0,ty);
 616             fjz0             = _mm_add_ps(fjz0,tz);
 617
 618             }
 619
 620             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 621             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 622             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 623             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 624
 625             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 626
 627             /* Inner loop uses 143 flops */
 628         }
 629
 630         /* End of innermost loop */
 631
 632         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 633                                               f+i_coord_offset,fshift+i_shift_offset);
 634
 635         ggid                        = gid[iidx];
 636         /* Update potential energies */
 637         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 638         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 639
 640         /* Increment number of inner iterations */
 641         inneriter                  += j_index_end - j_index_start;
 642
 643         /* Outer loop uses 20 flops */
 644     }
 645
 646     /* Increment number of outer iterations */
 647     outeriter        += nri;
 648
 649     /* Update outer/inner flops */
 650
 651     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*143);
 652 }
 653 /*
 654  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_single
 655  * Electrostatics interaction: ReactionField
 656  * VdW interaction:            LennardJones
 657  * Geometry:                   Water3-Particle
 658  * Calculate force/pot:        Force
 659  */
 660 void
 661 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_single
 662                     (t_nblist * gmx_restrict                nlist,
 663                      rvec * gmx_restrict                    xx,
 664                      rvec * gmx_restrict                    ff,
 665                      t_forcerec * gmx_restrict              fr,
 666                      t_mdatoms * gmx_restrict               mdatoms,
 667                      nb_kernel_data_t * gmx_restrict        kernel_data,
 668                      t_nrnb * gmx_restrict                  nrnb)
 669 {
 670     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 671      * just 0 for non-waters.
 672      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 673      * jnr indices corresponding to data put in the four positions in the SIMD register.
 674      */
 675     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 676     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 677     int              jnrA,jnrB,jnrC,jnrD;
 678     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 679     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 680     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 681     real             rcutoff_scalar;
 682     real             *shiftvec,*fshift,*x,*f;
 683     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 684     real             scratch[4*DIM];
 685     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 686     int              vdwioffset0;
 687     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 688     int              vdwioffset1;
 689     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 690     int              vdwioffset2;
 691     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 692     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 693     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 694     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 695     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 696     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 697     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 698     real             *charge;
 699     int              nvdwtype;
 700     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 701     int              *vdwtype;
 702     real             *vdwparam;
 703     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 704     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 705     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 706     real             rswitch_scalar,d_scalar;
 707     __m128           dummy_mask,cutoff_mask;
 708     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 709     __m128           one     = _mm_set1_ps(1.0);
 710     __m128           two     = _mm_set1_ps(2.0);
 711     x                = xx[0];
 712     f                = ff[0];
 713
 714     nri              = nlist->nri;
 715     iinr             = nlist->iinr;
 716     jindex           = nlist->jindex;
 717     jjnr             = nlist->jjnr;
 718     shiftidx         = nlist->shift;
 719     gid              = nlist->gid;
 720     shiftvec         = fr->shift_vec[0];
 721     fshift           = fr->fshift[0];
 722     facel            = _mm_set1_ps(fr->epsfac);
 723     charge           = mdatoms->chargeA;
 724     krf              = _mm_set1_ps(fr->ic->k_rf);
 725     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 726     crf              = _mm_set1_ps(fr->ic->c_rf);
 727     nvdwtype         = fr->ntype;
 728     vdwparam         = fr->nbfp;
 729     vdwtype          = mdatoms->typeA;
 730
 731     /* Setup water-specific parameters */
 732     inr              = nlist->iinr[0];
 733     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 734     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 735     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 736     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 737
 738     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 739     rcutoff_scalar   = fr->rcoulomb;
 740     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 741     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 742
 743     rswitch_scalar   = fr->rvdw_switch;
 744     rswitch          = _mm_set1_ps(rswitch_scalar);
 745     /* Setup switch parameters */
 746     d_scalar         = rcutoff_scalar-rswitch_scalar;
 747     d                = _mm_set1_ps(d_scalar);
 748     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 749     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 750     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 751     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 752     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 753     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 754
 755     /* Avoid stupid compiler warnings */
 756     jnrA = jnrB = jnrC = jnrD = 0;
 757     j_coord_offsetA = 0;
 758     j_coord_offsetB = 0;
 759     j_coord_offsetC = 0;
 760     j_coord_offsetD = 0;
 761
 762     outeriter        = 0;
 763     inneriter        = 0;
 764
 765     for(iidx=0;iidx<4*DIM;iidx++)
 766     {
 767         scratch[iidx] = 0.0;
 768     }
 769
 770     /* Start outer loop over neighborlists */
 771     for(iidx=0; iidx<nri; iidx++)
 772     {
 773         /* Load shift vector for this list */
 774         i_shift_offset   = DIM*shiftidx[iidx];
 775
 776         /* Load limits for loop over neighbors */
 777         j_index_start    = jindex[iidx];
 778         j_index_end      = jindex[iidx+1];
 779
 780         /* Get outer coordinate index */
 781         inr              = iinr[iidx];
 782         i_coord_offset   = DIM*inr;
 783
 784         /* Load i particle coords and add shift vector */
 785         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 786                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 787
 788         fix0             = _mm_setzero_ps();
 789         fiy0             = _mm_setzero_ps();
 790         fiz0             = _mm_setzero_ps();
 791         fix1             = _mm_setzero_ps();
 792         fiy1             = _mm_setzero_ps();
 793         fiz1             = _mm_setzero_ps();
 794         fix2             = _mm_setzero_ps();
 795         fiy2             = _mm_setzero_ps();
 796         fiz2             = _mm_setzero_ps();
 797
 798         /* Start inner kernel loop */
 799         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 800         {
 801
 802             /* Get j neighbor index, and coordinate index */
 803             jnrA             = jjnr[jidx];
 804             jnrB             = jjnr[jidx+1];
 805             jnrC             = jjnr[jidx+2];
 806             jnrD             = jjnr[jidx+3];
 807             j_coord_offsetA  = DIM*jnrA;
 808             j_coord_offsetB  = DIM*jnrB;
 809             j_coord_offsetC  = DIM*jnrC;
 810             j_coord_offsetD  = DIM*jnrD;
 811
 812             /* load j atom coordinates */
 813             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 814                                               x+j_coord_offsetC,x+j_coord_offsetD,
 815                                               &jx0,&jy0,&jz0);
 816
 817             /* Calculate displacement vector */
 818             dx00             = _mm_sub_ps(ix0,jx0);
 819             dy00             = _mm_sub_ps(iy0,jy0);
 820             dz00             = _mm_sub_ps(iz0,jz0);
 821             dx10             = _mm_sub_ps(ix1,jx0);
 822             dy10             = _mm_sub_ps(iy1,jy0);
 823             dz10             = _mm_sub_ps(iz1,jz0);
 824             dx20             = _mm_sub_ps(ix2,jx0);
 825             dy20             = _mm_sub_ps(iy2,jy0);
 826             dz20             = _mm_sub_ps(iz2,jz0);
 827
 828             /* Calculate squared distance and things based on it */
 829             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 830             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 831             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 832
 833             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 834             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 835             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 836
 837             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 838             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 839             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 840
 841             /* Load parameters for j particles */
 842             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 843                                                               charge+jnrC+0,charge+jnrD+0);
 844             vdwjidx0A        = 2*vdwtype[jnrA+0];
 845             vdwjidx0B        = 2*vdwtype[jnrB+0];
 846             vdwjidx0C        = 2*vdwtype[jnrC+0];
 847             vdwjidx0D        = 2*vdwtype[jnrD+0];
 848
 849             fjx0             = _mm_setzero_ps();
 850             fjy0             = _mm_setzero_ps();
 851             fjz0             = _mm_setzero_ps();
 852
 853             /**************************
 854              * CALCULATE INTERACTIONS *
 855              **************************/
 856
 857             if (gmx_mm_any_lt(rsq00,rcutoff2))
 858             {
 859
 860             r00              = _mm_mul_ps(rsq00,rinv00);
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             qq00             = _mm_mul_ps(iq0,jq0);
 864             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 865                                          vdwparam+vdwioffset0+vdwjidx0B,
 866                                          vdwparam+vdwioffset0+vdwjidx0C,
 867                                          vdwparam+vdwioffset0+vdwjidx0D,
 868                                          &c6_00,&c12_00);
 869
 870             /* REACTION-FIELD ELECTROSTATICS */
 871             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 872
 873             /* LENNARD-JONES DISPERSION/REPULSION */
 874
 875             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 876             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 877             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 878             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 879             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 880
 881             d                = _mm_sub_ps(r00,rswitch);
 882             d                = _mm_max_ps(d,_mm_setzero_ps());
 883             d2               = _mm_mul_ps(d,d);
 884             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 885
 886             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 887
 888             /* Evaluate switch function */
 889             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 890             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 891             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 892
 893             fscal            = _mm_add_ps(felec,fvdw);
 894
 895             fscal            = _mm_and_ps(fscal,cutoff_mask);
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm_mul_ps(fscal,dx00);
 899             ty               = _mm_mul_ps(fscal,dy00);
 900             tz               = _mm_mul_ps(fscal,dz00);
 901
 902             /* Update vectorial force */
 903             fix0             = _mm_add_ps(fix0,tx);
 904             fiy0             = _mm_add_ps(fiy0,ty);
 905             fiz0             = _mm_add_ps(fiz0,tz);
 906
 907             fjx0             = _mm_add_ps(fjx0,tx);
 908             fjy0             = _mm_add_ps(fjy0,ty);
 909             fjz0             = _mm_add_ps(fjz0,tz);
 910
 911             }
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             if (gmx_mm_any_lt(rsq10,rcutoff2))
 918             {
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             qq10             = _mm_mul_ps(iq1,jq0);
 922
 923             /* REACTION-FIELD ELECTROSTATICS */
 924             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 925
 926             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 927
 928             fscal            = felec;
 929
 930             fscal            = _mm_and_ps(fscal,cutoff_mask);
 931
 932             /* Calculate temporary vectorial force */
 933             tx               = _mm_mul_ps(fscal,dx10);
 934             ty               = _mm_mul_ps(fscal,dy10);
 935             tz               = _mm_mul_ps(fscal,dz10);
 936
 937             /* Update vectorial force */
 938             fix1             = _mm_add_ps(fix1,tx);
 939             fiy1             = _mm_add_ps(fiy1,ty);
 940             fiz1             = _mm_add_ps(fiz1,tz);
 941
 942             fjx0             = _mm_add_ps(fjx0,tx);
 943             fjy0             = _mm_add_ps(fjy0,ty);
 944             fjz0             = _mm_add_ps(fjz0,tz);
 945
 946             }
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             if (gmx_mm_any_lt(rsq20,rcutoff2))
 953             {
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq20             = _mm_mul_ps(iq2,jq0);
 957
 958             /* REACTION-FIELD ELECTROSTATICS */
 959             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 960
 961             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 962
 963             fscal            = felec;
 964
 965             fscal            = _mm_and_ps(fscal,cutoff_mask);
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm_mul_ps(fscal,dx20);
 969             ty               = _mm_mul_ps(fscal,dy20);
 970             tz               = _mm_mul_ps(fscal,dz20);
 971
 972             /* Update vectorial force */
 973             fix2             = _mm_add_ps(fix2,tx);
 974             fiy2             = _mm_add_ps(fiy2,ty);
 975             fiz2             = _mm_add_ps(fiz2,tz);
 976
 977             fjx0             = _mm_add_ps(fjx0,tx);
 978             fjy0             = _mm_add_ps(fjy0,ty);
 979             fjz0             = _mm_add_ps(fjz0,tz);
 980
 981             }
 982
 983             fjptrA             = f+j_coord_offsetA;
 984             fjptrB             = f+j_coord_offsetB;
 985             fjptrC             = f+j_coord_offsetC;
 986             fjptrD             = f+j_coord_offsetD;
 987
 988             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 989
 990             /* Inner loop uses 121 flops */
 991         }
 992
 993         if(jidx<j_index_end)
 994         {
 995
 996             /* Get j neighbor index, and coordinate index */
 997             jnrlistA         = jjnr[jidx];
 998             jnrlistB         = jjnr[jidx+1];
 999             jnrlistC         = jjnr[jidx+2];
1000             jnrlistD         = jjnr[jidx+3];
1001             /* Sign of each element will be negative for non-real atoms.
1002              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1003              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1004              */
1005             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1006             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1007             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1008             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1009             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1010             j_coord_offsetA  = DIM*jnrA;
1011             j_coord_offsetB  = DIM*jnrB;
1012             j_coord_offsetC  = DIM*jnrC;
1013             j_coord_offsetD  = DIM*jnrD;
1014
1015             /* load j atom coordinates */
1016             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1017                                               x+j_coord_offsetC,x+j_coord_offsetD,
1018                                               &jx0,&jy0,&jz0);
1019
1020             /* Calculate displacement vector */
1021             dx00             = _mm_sub_ps(ix0,jx0);
1022             dy00             = _mm_sub_ps(iy0,jy0);
1023             dz00             = _mm_sub_ps(iz0,jz0);
1024             dx10             = _mm_sub_ps(ix1,jx0);
1025             dy10             = _mm_sub_ps(iy1,jy0);
1026             dz10             = _mm_sub_ps(iz1,jz0);
1027             dx20             = _mm_sub_ps(ix2,jx0);
1028             dy20             = _mm_sub_ps(iy2,jy0);
1029             dz20             = _mm_sub_ps(iz2,jz0);
1030
1031             /* Calculate squared distance and things based on it */
1032             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1033             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1034             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1035
1036             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1037             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1038             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1039
1040             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1041             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1042             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1043
1044             /* Load parameters for j particles */
1045             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1046                                                               charge+jnrC+0,charge+jnrD+0);
1047             vdwjidx0A        = 2*vdwtype[jnrA+0];
1048             vdwjidx0B        = 2*vdwtype[jnrB+0];
1049             vdwjidx0C        = 2*vdwtype[jnrC+0];
1050             vdwjidx0D        = 2*vdwtype[jnrD+0];
1051
1052             fjx0             = _mm_setzero_ps();
1053             fjy0             = _mm_setzero_ps();
1054             fjz0             = _mm_setzero_ps();
1055
1056             /**************************
1057              * CALCULATE INTERACTIONS *
1058              **************************/
1059
1060             if (gmx_mm_any_lt(rsq00,rcutoff2))
1061             {
1062
1063             r00              = _mm_mul_ps(rsq00,rinv00);
1064             r00              = _mm_andnot_ps(dummy_mask,r00);
1065
1066             /* Compute parameters for interactions between i and j atoms */
1067             qq00             = _mm_mul_ps(iq0,jq0);
1068             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1069                                          vdwparam+vdwioffset0+vdwjidx0B,
1070                                          vdwparam+vdwioffset0+vdwjidx0C,
1071                                          vdwparam+vdwioffset0+vdwjidx0D,
1072                                          &c6_00,&c12_00);
1073
1074             /* REACTION-FIELD ELECTROSTATICS */
1075             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1076
1077             /* LENNARD-JONES DISPERSION/REPULSION */
1078
1079             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1080             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1081             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1082             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1083             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1084
1085             d                = _mm_sub_ps(r00,rswitch);
1086             d                = _mm_max_ps(d,_mm_setzero_ps());
1087             d2               = _mm_mul_ps(d,d);
1088             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1089
1090             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1091
1092             /* Evaluate switch function */
1093             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1094             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1095             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1096
1097             fscal            = _mm_add_ps(felec,fvdw);
1098
1099             fscal            = _mm_and_ps(fscal,cutoff_mask);
1100
1101             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1102
1103             /* Calculate temporary vectorial force */
1104             tx               = _mm_mul_ps(fscal,dx00);
1105             ty               = _mm_mul_ps(fscal,dy00);
1106             tz               = _mm_mul_ps(fscal,dz00);
1107
1108             /* Update vectorial force */
1109             fix0             = _mm_add_ps(fix0,tx);
1110             fiy0             = _mm_add_ps(fiy0,ty);
1111             fiz0             = _mm_add_ps(fiz0,tz);
1112
1113             fjx0             = _mm_add_ps(fjx0,tx);
1114             fjy0             = _mm_add_ps(fjy0,ty);
1115             fjz0             = _mm_add_ps(fjz0,tz);
1116
1117             }
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             if (gmx_mm_any_lt(rsq10,rcutoff2))
1124             {
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq10             = _mm_mul_ps(iq1,jq0);
1128
1129             /* REACTION-FIELD ELECTROSTATICS */
1130             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1131
1132             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1133
1134             fscal            = felec;
1135
1136             fscal            = _mm_and_ps(fscal,cutoff_mask);
1137
1138             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm_mul_ps(fscal,dx10);
1142             ty               = _mm_mul_ps(fscal,dy10);
1143             tz               = _mm_mul_ps(fscal,dz10);
1144
1145             /* Update vectorial force */
1146             fix1             = _mm_add_ps(fix1,tx);
1147             fiy1             = _mm_add_ps(fiy1,ty);
1148             fiz1             = _mm_add_ps(fiz1,tz);
1149
1150             fjx0             = _mm_add_ps(fjx0,tx);
1151             fjy0             = _mm_add_ps(fjy0,ty);
1152             fjz0             = _mm_add_ps(fjz0,tz);
1153
1154             }
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             if (gmx_mm_any_lt(rsq20,rcutoff2))
1161             {
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq20             = _mm_mul_ps(iq2,jq0);
1165
1166             /* REACTION-FIELD ELECTROSTATICS */
1167             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1168
1169             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1170
1171             fscal            = felec;
1172
1173             fscal            = _mm_and_ps(fscal,cutoff_mask);
1174
1175             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1176
1177             /* Calculate temporary vectorial force */
1178             tx               = _mm_mul_ps(fscal,dx20);
1179             ty               = _mm_mul_ps(fscal,dy20);
1180             tz               = _mm_mul_ps(fscal,dz20);
1181
1182             /* Update vectorial force */
1183             fix2             = _mm_add_ps(fix2,tx);
1184             fiy2             = _mm_add_ps(fiy2,ty);
1185             fiz2             = _mm_add_ps(fiz2,tz);
1186
1187             fjx0             = _mm_add_ps(fjx0,tx);
1188             fjy0             = _mm_add_ps(fjy0,ty);
1189             fjz0             = _mm_add_ps(fjz0,tz);
1190
1191             }
1192
1193             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1194             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1195             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1196             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1197
1198             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1199
1200             /* Inner loop uses 122 flops */
1201         }
1202
1203         /* End of innermost loop */
1204
1205         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1206                                               f+i_coord_offset,fshift+i_shift_offset);
1207
1208         /* Increment number of inner iterations */
1209         inneriter                  += j_index_end - j_index_start;
1210
1211         /* Outer loop uses 18 flops */
1212     }
1213
1214     /* Increment number of outer iterations */
1215     outeriter        += nri;
1216
1217     /* Update outer/inner flops */
1218
1219     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*122);
1220 }