src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse4_1_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_sse4_1_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             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             r00              = _mm_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm_mul_ps(iq0,jq0);
 247             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,
 249                                          vdwparam+vdwioffset0+vdwjidx0C,
 250                                          vdwparam+vdwioffset0+vdwjidx0D,
 251                                          &c6_00,&c12_00);
 252
 253             /* REACTION-FIELD ELECTROSTATICS */
 254             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 255             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 256
 257             /* LENNARD-JONES DISPERSION/REPULSION */
 258
 259             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 260             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 261             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 262             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 263             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 264
 265             d                = _mm_sub_ps(r00,rswitch);
 266             d                = _mm_max_ps(d,_mm_setzero_ps());
 267             d2               = _mm_mul_ps(d,d);
 268             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)))))));
 269
 270             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 271
 272             /* Evaluate switch function */
 273             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 274             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 275             vvdw             = _mm_mul_ps(vvdw,sw);
 276             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velec            = _mm_and_ps(velec,cutoff_mask);
 280             velecsum         = _mm_add_ps(velecsum,velec);
 281             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 282             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_ps(felec,fvdw);
 285
 286             fscal            = _mm_and_ps(fscal,cutoff_mask);
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm_mul_ps(fscal,dx00);
 290             ty               = _mm_mul_ps(fscal,dy00);
 291             tz               = _mm_mul_ps(fscal,dz00);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm_add_ps(fix0,tx);
 295             fiy0             = _mm_add_ps(fiy0,ty);
 296             fiz0             = _mm_add_ps(fiz0,tz);
 297
 298             fjptrA             = f+j_coord_offsetA;
 299             fjptrB             = f+j_coord_offsetB;
 300             fjptrC             = f+j_coord_offsetC;
 301             fjptrD             = f+j_coord_offsetD;
 302             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 303
 304             }
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             if (gmx_mm_any_lt(rsq10,rcutoff2))
 311             {
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq10             = _mm_mul_ps(iq1,jq0);
 315
 316             /* REACTION-FIELD ELECTROSTATICS */
 317             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 318             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 319
 320             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velec            = _mm_and_ps(velec,cutoff_mask);
 324             velecsum         = _mm_add_ps(velecsum,velec);
 325
 326             fscal            = felec;
 327
 328             fscal            = _mm_and_ps(fscal,cutoff_mask);
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_ps(fscal,dx10);
 332             ty               = _mm_mul_ps(fscal,dy10);
 333             tz               = _mm_mul_ps(fscal,dz10);
 334
 335             /* Update vectorial force */
 336             fix1             = _mm_add_ps(fix1,tx);
 337             fiy1             = _mm_add_ps(fiy1,ty);
 338             fiz1             = _mm_add_ps(fiz1,tz);
 339
 340             fjptrA             = f+j_coord_offsetA;
 341             fjptrB             = f+j_coord_offsetB;
 342             fjptrC             = f+j_coord_offsetC;
 343             fjptrD             = f+j_coord_offsetD;
 344             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,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             fjptrA             = f+j_coord_offsetA;
 383             fjptrB             = f+j_coord_offsetB;
 384             fjptrC             = f+j_coord_offsetC;
 385             fjptrD             = f+j_coord_offsetD;
 386             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 387
 388             }
 389
 390             /* Inner loop uses 142 flops */
 391         }
 392
 393         if(jidx<j_index_end)
 394         {
 395
 396             /* Get j neighbor index, and coordinate index */
 397             jnrlistA         = jjnr[jidx];
 398             jnrlistB         = jjnr[jidx+1];
 399             jnrlistC         = jjnr[jidx+2];
 400             jnrlistD         = jjnr[jidx+3];
 401             /* Sign of each element will be negative for non-real atoms.
 402              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 403              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 404              */
 405             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 406             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 407             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 408             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 409             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 410             j_coord_offsetA  = DIM*jnrA;
 411             j_coord_offsetB  = DIM*jnrB;
 412             j_coord_offsetC  = DIM*jnrC;
 413             j_coord_offsetD  = DIM*jnrD;
 414
 415             /* load j atom coordinates */
 416             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 417                                               x+j_coord_offsetC,x+j_coord_offsetD,
 418                                               &jx0,&jy0,&jz0);
 419
 420             /* Calculate displacement vector */
 421             dx00             = _mm_sub_ps(ix0,jx0);
 422             dy00             = _mm_sub_ps(iy0,jy0);
 423             dz00             = _mm_sub_ps(iz0,jz0);
 424             dx10             = _mm_sub_ps(ix1,jx0);
 425             dy10             = _mm_sub_ps(iy1,jy0);
 426             dz10             = _mm_sub_ps(iz1,jz0);
 427             dx20             = _mm_sub_ps(ix2,jx0);
 428             dy20             = _mm_sub_ps(iy2,jy0);
 429             dz20             = _mm_sub_ps(iz2,jz0);
 430
 431             /* Calculate squared distance and things based on it */
 432             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 433             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 434             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 435
 436             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 437             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 438             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 439
 440             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 441             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 442             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 443
 444             /* Load parameters for j particles */
 445             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 446                                                               charge+jnrC+0,charge+jnrD+0);
 447             vdwjidx0A        = 2*vdwtype[jnrA+0];
 448             vdwjidx0B        = 2*vdwtype[jnrB+0];
 449             vdwjidx0C        = 2*vdwtype[jnrC+0];
 450             vdwjidx0D        = 2*vdwtype[jnrD+0];
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             if (gmx_mm_any_lt(rsq00,rcutoff2))
 457             {
 458
 459             r00              = _mm_mul_ps(rsq00,rinv00);
 460             r00              = _mm_andnot_ps(dummy_mask,r00);
 461
 462             /* Compute parameters for interactions between i and j atoms */
 463             qq00             = _mm_mul_ps(iq0,jq0);
 464             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 465                                          vdwparam+vdwioffset0+vdwjidx0B,
 466                                          vdwparam+vdwioffset0+vdwjidx0C,
 467                                          vdwparam+vdwioffset0+vdwjidx0D,
 468                                          &c6_00,&c12_00);
 469
 470             /* REACTION-FIELD ELECTROSTATICS */
 471             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 472             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 473
 474             /* LENNARD-JONES DISPERSION/REPULSION */
 475
 476             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 477             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 478             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 479             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 480             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 481
 482             d                = _mm_sub_ps(r00,rswitch);
 483             d                = _mm_max_ps(d,_mm_setzero_ps());
 484             d2               = _mm_mul_ps(d,d);
 485             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)))))));
 486
 487             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 488
 489             /* Evaluate switch function */
 490             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 491             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 492             vvdw             = _mm_mul_ps(vvdw,sw);
 493             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 494
 495             /* Update potential sum for this i atom from the interaction with this j atom. */
 496             velec            = _mm_and_ps(velec,cutoff_mask);
 497             velec            = _mm_andnot_ps(dummy_mask,velec);
 498             velecsum         = _mm_add_ps(velecsum,velec);
 499             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 500             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 501             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 502
 503             fscal            = _mm_add_ps(felec,fvdw);
 504
 505             fscal            = _mm_and_ps(fscal,cutoff_mask);
 506
 507             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_ps(fscal,dx00);
 511             ty               = _mm_mul_ps(fscal,dy00);
 512             tz               = _mm_mul_ps(fscal,dz00);
 513
 514             /* Update vectorial force */
 515             fix0             = _mm_add_ps(fix0,tx);
 516             fiy0             = _mm_add_ps(fiy0,ty);
 517             fiz0             = _mm_add_ps(fiz0,tz);
 518
 519             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 520             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 521             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 522             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 523             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 524
 525             }
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm_any_lt(rsq10,rcutoff2))
 532             {
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq10             = _mm_mul_ps(iq1,jq0);
 536
 537             /* REACTION-FIELD ELECTROSTATICS */
 538             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 539             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 540
 541             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_and_ps(velec,cutoff_mask);
 545             velec            = _mm_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm_and_ps(fscal,cutoff_mask);
 551
 552             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm_mul_ps(fscal,dx10);
 556             ty               = _mm_mul_ps(fscal,dy10);
 557             tz               = _mm_mul_ps(fscal,dz10);
 558
 559             /* Update vectorial force */
 560             fix1             = _mm_add_ps(fix1,tx);
 561             fiy1             = _mm_add_ps(fiy1,ty);
 562             fiz1             = _mm_add_ps(fiz1,tz);
 563
 564             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 565             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 566             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 567             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 568             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 569
 570             }
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             if (gmx_mm_any_lt(rsq20,rcutoff2))
 577             {
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             qq20             = _mm_mul_ps(iq2,jq0);
 581
 582             /* REACTION-FIELD ELECTROSTATICS */
 583             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 584             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 585
 586             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             velec            = _mm_and_ps(velec,cutoff_mask);
 590             velec            = _mm_andnot_ps(dummy_mask,velec);
 591             velecsum         = _mm_add_ps(velecsum,velec);
 592
 593             fscal            = felec;
 594
 595             fscal            = _mm_and_ps(fscal,cutoff_mask);
 596
 597             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm_mul_ps(fscal,dx20);
 601             ty               = _mm_mul_ps(fscal,dy20);
 602             tz               = _mm_mul_ps(fscal,dz20);
 603
 604             /* Update vectorial force */
 605             fix2             = _mm_add_ps(fix2,tx);
 606             fiy2             = _mm_add_ps(fiy2,ty);
 607             fiz2             = _mm_add_ps(fiz2,tz);
 608
 609             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 610             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 611             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 612             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 613             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 614
 615             }
 616
 617             /* Inner loop uses 143 flops */
 618         }
 619
 620         /* End of innermost loop */
 621
 622         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 623                                               f+i_coord_offset,fshift+i_shift_offset);
 624
 625         ggid                        = gid[iidx];
 626         /* Update potential energies */
 627         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 628         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 629
 630         /* Increment number of inner iterations */
 631         inneriter                  += j_index_end - j_index_start;
 632
 633         /* Outer loop uses 20 flops */
 634     }
 635
 636     /* Increment number of outer iterations */
 637     outeriter        += nri;
 638
 639     /* Update outer/inner flops */
 640
 641     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*143);
 642 }
 643 /*
 644  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse4_1_single
 645  * Electrostatics interaction: ReactionField
 646  * VdW interaction:            LennardJones
 647  * Geometry:                   Water3-Particle
 648  * Calculate force/pot:        Force
 649  */
 650 void
 651 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse4_1_single
 652                     (t_nblist * gmx_restrict                nlist,
 653                      rvec * gmx_restrict                    xx,
 654                      rvec * gmx_restrict                    ff,
 655                      t_forcerec * gmx_restrict              fr,
 656                      t_mdatoms * gmx_restrict               mdatoms,
 657                      nb_kernel_data_t * gmx_restrict        kernel_data,
 658                      t_nrnb * gmx_restrict                  nrnb)
 659 {
 660     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 661      * just 0 for non-waters.
 662      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 663      * jnr indices corresponding to data put in the four positions in the SIMD register.
 664      */
 665     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 666     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 667     int              jnrA,jnrB,jnrC,jnrD;
 668     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 669     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 670     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 671     real             rcutoff_scalar;
 672     real             *shiftvec,*fshift,*x,*f;
 673     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 674     real             scratch[4*DIM];
 675     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 676     int              vdwioffset0;
 677     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 678     int              vdwioffset1;
 679     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 680     int              vdwioffset2;
 681     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 682     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 683     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 684     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 685     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 686     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 687     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 688     real             *charge;
 689     int              nvdwtype;
 690     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 691     int              *vdwtype;
 692     real             *vdwparam;
 693     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 694     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 695     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 696     real             rswitch_scalar,d_scalar;
 697     __m128           dummy_mask,cutoff_mask;
 698     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 699     __m128           one     = _mm_set1_ps(1.0);
 700     __m128           two     = _mm_set1_ps(2.0);
 701     x                = xx[0];
 702     f                = ff[0];
 703
 704     nri              = nlist->nri;
 705     iinr             = nlist->iinr;
 706     jindex           = nlist->jindex;
 707     jjnr             = nlist->jjnr;
 708     shiftidx         = nlist->shift;
 709     gid              = nlist->gid;
 710     shiftvec         = fr->shift_vec[0];
 711     fshift           = fr->fshift[0];
 712     facel            = _mm_set1_ps(fr->epsfac);
 713     charge           = mdatoms->chargeA;
 714     krf              = _mm_set1_ps(fr->ic->k_rf);
 715     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 716     crf              = _mm_set1_ps(fr->ic->c_rf);
 717     nvdwtype         = fr->ntype;
 718     vdwparam         = fr->nbfp;
 719     vdwtype          = mdatoms->typeA;
 720
 721     /* Setup water-specific parameters */
 722     inr              = nlist->iinr[0];
 723     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 724     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 725     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 726     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 727
 728     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 729     rcutoff_scalar   = fr->rcoulomb;
 730     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 731     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 732
 733     rswitch_scalar   = fr->rvdw_switch;
 734     rswitch          = _mm_set1_ps(rswitch_scalar);
 735     /* Setup switch parameters */
 736     d_scalar         = rcutoff_scalar-rswitch_scalar;
 737     d                = _mm_set1_ps(d_scalar);
 738     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 739     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 740     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 741     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 742     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 743     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 744
 745     /* Avoid stupid compiler warnings */
 746     jnrA = jnrB = jnrC = jnrD = 0;
 747     j_coord_offsetA = 0;
 748     j_coord_offsetB = 0;
 749     j_coord_offsetC = 0;
 750     j_coord_offsetD = 0;
 751
 752     outeriter        = 0;
 753     inneriter        = 0;
 754
 755     for(iidx=0;iidx<4*DIM;iidx++)
 756     {
 757         scratch[iidx] = 0.0;
 758     }
 759
 760     /* Start outer loop over neighborlists */
 761     for(iidx=0; iidx<nri; iidx++)
 762     {
 763         /* Load shift vector for this list */
 764         i_shift_offset   = DIM*shiftidx[iidx];
 765
 766         /* Load limits for loop over neighbors */
 767         j_index_start    = jindex[iidx];
 768         j_index_end      = jindex[iidx+1];
 769
 770         /* Get outer coordinate index */
 771         inr              = iinr[iidx];
 772         i_coord_offset   = DIM*inr;
 773
 774         /* Load i particle coords and add shift vector */
 775         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 776                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 777
 778         fix0             = _mm_setzero_ps();
 779         fiy0             = _mm_setzero_ps();
 780         fiz0             = _mm_setzero_ps();
 781         fix1             = _mm_setzero_ps();
 782         fiy1             = _mm_setzero_ps();
 783         fiz1             = _mm_setzero_ps();
 784         fix2             = _mm_setzero_ps();
 785         fiy2             = _mm_setzero_ps();
 786         fiz2             = _mm_setzero_ps();
 787
 788         /* Start inner kernel loop */
 789         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 790         {
 791
 792             /* Get j neighbor index, and coordinate index */
 793             jnrA             = jjnr[jidx];
 794             jnrB             = jjnr[jidx+1];
 795             jnrC             = jjnr[jidx+2];
 796             jnrD             = jjnr[jidx+3];
 797             j_coord_offsetA  = DIM*jnrA;
 798             j_coord_offsetB  = DIM*jnrB;
 799             j_coord_offsetC  = DIM*jnrC;
 800             j_coord_offsetD  = DIM*jnrD;
 801
 802             /* load j atom coordinates */
 803             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 804                                               x+j_coord_offsetC,x+j_coord_offsetD,
 805                                               &jx0,&jy0,&jz0);
 806
 807             /* Calculate displacement vector */
 808             dx00             = _mm_sub_ps(ix0,jx0);
 809             dy00             = _mm_sub_ps(iy0,jy0);
 810             dz00             = _mm_sub_ps(iz0,jz0);
 811             dx10             = _mm_sub_ps(ix1,jx0);
 812             dy10             = _mm_sub_ps(iy1,jy0);
 813             dz10             = _mm_sub_ps(iz1,jz0);
 814             dx20             = _mm_sub_ps(ix2,jx0);
 815             dy20             = _mm_sub_ps(iy2,jy0);
 816             dz20             = _mm_sub_ps(iz2,jz0);
 817
 818             /* Calculate squared distance and things based on it */
 819             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 820             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 821             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 822
 823             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 824             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 825             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 826
 827             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 828             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 829             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 830
 831             /* Load parameters for j particles */
 832             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 833                                                               charge+jnrC+0,charge+jnrD+0);
 834             vdwjidx0A        = 2*vdwtype[jnrA+0];
 835             vdwjidx0B        = 2*vdwtype[jnrB+0];
 836             vdwjidx0C        = 2*vdwtype[jnrC+0];
 837             vdwjidx0D        = 2*vdwtype[jnrD+0];
 838
 839             /**************************
 840              * CALCULATE INTERACTIONS *
 841              **************************/
 842
 843             if (gmx_mm_any_lt(rsq00,rcutoff2))
 844             {
 845
 846             r00              = _mm_mul_ps(rsq00,rinv00);
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq00             = _mm_mul_ps(iq0,jq0);
 850             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 851                                          vdwparam+vdwioffset0+vdwjidx0B,
 852                                          vdwparam+vdwioffset0+vdwjidx0C,
 853                                          vdwparam+vdwioffset0+vdwjidx0D,
 854                                          &c6_00,&c12_00);
 855
 856             /* REACTION-FIELD ELECTROSTATICS */
 857             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 858
 859             /* LENNARD-JONES DISPERSION/REPULSION */
 860
 861             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 862             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 863             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 864             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 865             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 866
 867             d                = _mm_sub_ps(r00,rswitch);
 868             d                = _mm_max_ps(d,_mm_setzero_ps());
 869             d2               = _mm_mul_ps(d,d);
 870             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)))))));
 871
 872             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 873
 874             /* Evaluate switch function */
 875             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 876             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 877             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 878
 879             fscal            = _mm_add_ps(felec,fvdw);
 880
 881             fscal            = _mm_and_ps(fscal,cutoff_mask);
 882
 883             /* Calculate temporary vectorial force */
 884             tx               = _mm_mul_ps(fscal,dx00);
 885             ty               = _mm_mul_ps(fscal,dy00);
 886             tz               = _mm_mul_ps(fscal,dz00);
 887
 888             /* Update vectorial force */
 889             fix0             = _mm_add_ps(fix0,tx);
 890             fiy0             = _mm_add_ps(fiy0,ty);
 891             fiz0             = _mm_add_ps(fiz0,tz);
 892
 893             fjptrA             = f+j_coord_offsetA;
 894             fjptrB             = f+j_coord_offsetB;
 895             fjptrC             = f+j_coord_offsetC;
 896             fjptrD             = f+j_coord_offsetD;
 897             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 898
 899             }
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             if (gmx_mm_any_lt(rsq10,rcutoff2))
 906             {
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             qq10             = _mm_mul_ps(iq1,jq0);
 910
 911             /* REACTION-FIELD ELECTROSTATICS */
 912             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 913
 914             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 915
 916             fscal            = felec;
 917
 918             fscal            = _mm_and_ps(fscal,cutoff_mask);
 919
 920             /* Calculate temporary vectorial force */
 921             tx               = _mm_mul_ps(fscal,dx10);
 922             ty               = _mm_mul_ps(fscal,dy10);
 923             tz               = _mm_mul_ps(fscal,dz10);
 924
 925             /* Update vectorial force */
 926             fix1             = _mm_add_ps(fix1,tx);
 927             fiy1             = _mm_add_ps(fiy1,ty);
 928             fiz1             = _mm_add_ps(fiz1,tz);
 929
 930             fjptrA             = f+j_coord_offsetA;
 931             fjptrB             = f+j_coord_offsetB;
 932             fjptrC             = f+j_coord_offsetC;
 933             fjptrD             = f+j_coord_offsetD;
 934             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 935
 936             }
 937
 938             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             if (gmx_mm_any_lt(rsq20,rcutoff2))
 943             {
 944
 945             /* Compute parameters for interactions between i and j atoms */
 946             qq20             = _mm_mul_ps(iq2,jq0);
 947
 948             /* REACTION-FIELD ELECTROSTATICS */
 949             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 950
 951             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 952
 953             fscal            = felec;
 954
 955             fscal            = _mm_and_ps(fscal,cutoff_mask);
 956
 957             /* Calculate temporary vectorial force */
 958             tx               = _mm_mul_ps(fscal,dx20);
 959             ty               = _mm_mul_ps(fscal,dy20);
 960             tz               = _mm_mul_ps(fscal,dz20);
 961
 962             /* Update vectorial force */
 963             fix2             = _mm_add_ps(fix2,tx);
 964             fiy2             = _mm_add_ps(fiy2,ty);
 965             fiz2             = _mm_add_ps(fiz2,tz);
 966
 967             fjptrA             = f+j_coord_offsetA;
 968             fjptrB             = f+j_coord_offsetB;
 969             fjptrC             = f+j_coord_offsetC;
 970             fjptrD             = f+j_coord_offsetD;
 971             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 972
 973             }
 974
 975             /* Inner loop uses 121 flops */
 976         }
 977
 978         if(jidx<j_index_end)
 979         {
 980
 981             /* Get j neighbor index, and coordinate index */
 982             jnrlistA         = jjnr[jidx];
 983             jnrlistB         = jjnr[jidx+1];
 984             jnrlistC         = jjnr[jidx+2];
 985             jnrlistD         = jjnr[jidx+3];
 986             /* Sign of each element will be negative for non-real atoms.
 987              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 988              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 989              */
 990             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 991             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 992             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 993             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 994             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 995             j_coord_offsetA  = DIM*jnrA;
 996             j_coord_offsetB  = DIM*jnrB;
 997             j_coord_offsetC  = DIM*jnrC;
 998             j_coord_offsetD  = DIM*jnrD;
 999
1000             /* load j atom coordinates */
1001             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1002                                               x+j_coord_offsetC,x+j_coord_offsetD,
1003                                               &jx0,&jy0,&jz0);
1004
1005             /* Calculate displacement vector */
1006             dx00             = _mm_sub_ps(ix0,jx0);
1007             dy00             = _mm_sub_ps(iy0,jy0);
1008             dz00             = _mm_sub_ps(iz0,jz0);
1009             dx10             = _mm_sub_ps(ix1,jx0);
1010             dy10             = _mm_sub_ps(iy1,jy0);
1011             dz10             = _mm_sub_ps(iz1,jz0);
1012             dx20             = _mm_sub_ps(ix2,jx0);
1013             dy20             = _mm_sub_ps(iy2,jy0);
1014             dz20             = _mm_sub_ps(iz2,jz0);
1015
1016             /* Calculate squared distance and things based on it */
1017             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1018             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1019             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1020
1021             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1022             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1023             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1024
1025             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1026             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1027             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1028
1029             /* Load parameters for j particles */
1030             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1031                                                               charge+jnrC+0,charge+jnrD+0);
1032             vdwjidx0A        = 2*vdwtype[jnrA+0];
1033             vdwjidx0B        = 2*vdwtype[jnrB+0];
1034             vdwjidx0C        = 2*vdwtype[jnrC+0];
1035             vdwjidx0D        = 2*vdwtype[jnrD+0];
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             if (gmx_mm_any_lt(rsq00,rcutoff2))
1042             {
1043
1044             r00              = _mm_mul_ps(rsq00,rinv00);
1045             r00              = _mm_andnot_ps(dummy_mask,r00);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq00             = _mm_mul_ps(iq0,jq0);
1049             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1050                                          vdwparam+vdwioffset0+vdwjidx0B,
1051                                          vdwparam+vdwioffset0+vdwjidx0C,
1052                                          vdwparam+vdwioffset0+vdwjidx0D,
1053                                          &c6_00,&c12_00);
1054
1055             /* REACTION-FIELD ELECTROSTATICS */
1056             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1057
1058             /* LENNARD-JONES DISPERSION/REPULSION */
1059
1060             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1061             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1062             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1063             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1064             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1065
1066             d                = _mm_sub_ps(r00,rswitch);
1067             d                = _mm_max_ps(d,_mm_setzero_ps());
1068             d2               = _mm_mul_ps(d,d);
1069             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)))))));
1070
1071             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1072
1073             /* Evaluate switch function */
1074             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1075             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1076             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1077
1078             fscal            = _mm_add_ps(felec,fvdw);
1079
1080             fscal            = _mm_and_ps(fscal,cutoff_mask);
1081
1082             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm_mul_ps(fscal,dx00);
1086             ty               = _mm_mul_ps(fscal,dy00);
1087             tz               = _mm_mul_ps(fscal,dz00);
1088
1089             /* Update vectorial force */
1090             fix0             = _mm_add_ps(fix0,tx);
1091             fiy0             = _mm_add_ps(fiy0,ty);
1092             fiz0             = _mm_add_ps(fiz0,tz);
1093
1094             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1095             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1096             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1097             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1098             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1099
1100             }
1101
1102             /**************************
1103              * CALCULATE INTERACTIONS *
1104              **************************/
1105
1106             if (gmx_mm_any_lt(rsq10,rcutoff2))
1107             {
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             qq10             = _mm_mul_ps(iq1,jq0);
1111
1112             /* REACTION-FIELD ELECTROSTATICS */
1113             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1114
1115             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm_and_ps(fscal,cutoff_mask);
1120
1121             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm_mul_ps(fscal,dx10);
1125             ty               = _mm_mul_ps(fscal,dy10);
1126             tz               = _mm_mul_ps(fscal,dz10);
1127
1128             /* Update vectorial force */
1129             fix1             = _mm_add_ps(fix1,tx);
1130             fiy1             = _mm_add_ps(fiy1,ty);
1131             fiz1             = _mm_add_ps(fiz1,tz);
1132
1133             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1134             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1135             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1136             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1137             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1138
1139             }
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             if (gmx_mm_any_lt(rsq20,rcutoff2))
1146             {
1147
1148             /* Compute parameters for interactions between i and j atoms */
1149             qq20             = _mm_mul_ps(iq2,jq0);
1150
1151             /* REACTION-FIELD ELECTROSTATICS */
1152             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1153
1154             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1155
1156             fscal            = felec;
1157
1158             fscal            = _mm_and_ps(fscal,cutoff_mask);
1159
1160             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1161
1162             /* Calculate temporary vectorial force */
1163             tx               = _mm_mul_ps(fscal,dx20);
1164             ty               = _mm_mul_ps(fscal,dy20);
1165             tz               = _mm_mul_ps(fscal,dz20);
1166
1167             /* Update vectorial force */
1168             fix2             = _mm_add_ps(fix2,tx);
1169             fiy2             = _mm_add_ps(fiy2,ty);
1170             fiz2             = _mm_add_ps(fiz2,tz);
1171
1172             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1173             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1174             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1175             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1176             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1177
1178             }
1179
1180             /* Inner loop uses 122 flops */
1181         }
1182
1183         /* End of innermost loop */
1184
1185         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1186                                               f+i_coord_offset,fshift+i_shift_offset);
1187
1188         /* Increment number of inner iterations */
1189         inneriter                  += j_index_end - j_index_start;
1190
1191         /* Outer loop uses 18 flops */
1192     }
1193
1194     /* Increment number of outer iterations */
1195     outeriter        += nri;
1196
1197     /* Update outer/inner flops */
1198
1199     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*122);
1200 }