src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 103     real             rswitch_scalar,d_scalar;
 104     __m128d          dummy_mask,cutoff_mask;
 105     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 106     __m128d          one     = _mm_set1_pd(1.0);
 107     __m128d          two     = _mm_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm_set1_pd(fr->ic->k_rf);
 122     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 123     crf              = _mm_set1_pd(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->rcoulomb;
 137     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 138     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 139
 140     rswitch_scalar   = fr->rvdw_switch;
 141     rswitch          = _mm_set1_pd(rswitch_scalar);
 142     /* Setup switch parameters */
 143     d_scalar         = rcutoff_scalar-rswitch_scalar;
 144     d                = _mm_set1_pd(d_scalar);
 145     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 146     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 147     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 148     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 149     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm_setzero_pd();
 179         fiy0             = _mm_setzero_pd();
 180         fiz0             = _mm_setzero_pd();
 181         fix1             = _mm_setzero_pd();
 182         fiy1             = _mm_setzero_pd();
 183         fiz1             = _mm_setzero_pd();
 184         fix2             = _mm_setzero_pd();
 185         fiy2             = _mm_setzero_pd();
 186         fiz2             = _mm_setzero_pd();
 187         fix3             = _mm_setzero_pd();
 188         fiy3             = _mm_setzero_pd();
 189         fiz3             = _mm_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_pd();
 193         vvdwsum          = _mm_setzero_pd();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_pd(ix0,jx0);
 211             dy00             = _mm_sub_pd(iy0,jy0);
 212             dz00             = _mm_sub_pd(iz0,jz0);
 213             dx10             = _mm_sub_pd(ix1,jx0);
 214             dy10             = _mm_sub_pd(iy1,jy0);
 215             dz10             = _mm_sub_pd(iz1,jz0);
 216             dx20             = _mm_sub_pd(ix2,jx0);
 217             dy20             = _mm_sub_pd(iy2,jy0);
 218             dz20             = _mm_sub_pd(iz2,jz0);
 219             dx30             = _mm_sub_pd(ix3,jx0);
 220             dy30             = _mm_sub_pd(iy3,jy0);
 221             dz30             = _mm_sub_pd(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 228
 229             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 230             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 231             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 232             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 233
 234             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 235             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 237             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243
 244             fjx0             = _mm_setzero_pd();
 245             fjy0             = _mm_setzero_pd();
 246             fjz0             = _mm_setzero_pd();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             if (gmx_mm_any_lt(rsq00,rcutoff2))
 253             {
 254
 255             r00              = _mm_mul_pd(rsq00,rinv00);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 259                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 265             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 266             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 267             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 268
 269             d                = _mm_sub_pd(r00,rswitch);
 270             d                = _mm_max_pd(d,_mm_setzero_pd());
 271             d2               = _mm_mul_pd(d,d);
 272             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 273
 274             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(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_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 279             vvdw             = _mm_mul_pd(vvdw,sw);
 280             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 284             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 285
 286             fscal            = fvdw;
 287
 288             fscal            = _mm_and_pd(fscal,cutoff_mask);
 289
 290             /* Calculate temporary vectorial force */
 291             tx               = _mm_mul_pd(fscal,dx00);
 292             ty               = _mm_mul_pd(fscal,dy00);
 293             tz               = _mm_mul_pd(fscal,dz00);
 294
 295             /* Update vectorial force */
 296             fix0             = _mm_add_pd(fix0,tx);
 297             fiy0             = _mm_add_pd(fiy0,ty);
 298             fiz0             = _mm_add_pd(fiz0,tz);
 299
 300             fjx0             = _mm_add_pd(fjx0,tx);
 301             fjy0             = _mm_add_pd(fjy0,ty);
 302             fjz0             = _mm_add_pd(fjz0,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_pd(iq1,jq0);
 315
 316             /* REACTION-FIELD ELECTROSTATICS */
 317             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 318             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 319
 320             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velec            = _mm_and_pd(velec,cutoff_mask);
 324             velecsum         = _mm_add_pd(velecsum,velec);
 325
 326             fscal            = felec;
 327
 328             fscal            = _mm_and_pd(fscal,cutoff_mask);
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_pd(fscal,dx10);
 332             ty               = _mm_mul_pd(fscal,dy10);
 333             tz               = _mm_mul_pd(fscal,dz10);
 334
 335             /* Update vectorial force */
 336             fix1             = _mm_add_pd(fix1,tx);
 337             fiy1             = _mm_add_pd(fiy1,ty);
 338             fiz1             = _mm_add_pd(fiz1,tz);
 339
 340             fjx0             = _mm_add_pd(fjx0,tx);
 341             fjy0             = _mm_add_pd(fjy0,ty);
 342             fjz0             = _mm_add_pd(fjz0,tz);
 343
 344             }
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             if (gmx_mm_any_lt(rsq20,rcutoff2))
 351             {
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_pd(iq2,jq0);
 355
 356             /* REACTION-FIELD ELECTROSTATICS */
 357             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 358             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 359
 360             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velec            = _mm_and_pd(velec,cutoff_mask);
 364             velecsum         = _mm_add_pd(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             fscal            = _mm_and_pd(fscal,cutoff_mask);
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm_mul_pd(fscal,dx20);
 372             ty               = _mm_mul_pd(fscal,dy20);
 373             tz               = _mm_mul_pd(fscal,dz20);
 374
 375             /* Update vectorial force */
 376             fix2             = _mm_add_pd(fix2,tx);
 377             fiy2             = _mm_add_pd(fiy2,ty);
 378             fiz2             = _mm_add_pd(fiz2,tz);
 379
 380             fjx0             = _mm_add_pd(fjx0,tx);
 381             fjy0             = _mm_add_pd(fjy0,ty);
 382             fjz0             = _mm_add_pd(fjz0,tz);
 383
 384             }
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             if (gmx_mm_any_lt(rsq30,rcutoff2))
 391             {
 392
 393             /* Compute parameters for interactions between i and j atoms */
 394             qq30             = _mm_mul_pd(iq3,jq0);
 395
 396             /* REACTION-FIELD ELECTROSTATICS */
 397             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 398             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 399
 400             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velec            = _mm_and_pd(velec,cutoff_mask);
 404             velecsum         = _mm_add_pd(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             fscal            = _mm_and_pd(fscal,cutoff_mask);
 409
 410             /* Calculate temporary vectorial force */
 411             tx               = _mm_mul_pd(fscal,dx30);
 412             ty               = _mm_mul_pd(fscal,dy30);
 413             tz               = _mm_mul_pd(fscal,dz30);
 414
 415             /* Update vectorial force */
 416             fix3             = _mm_add_pd(fix3,tx);
 417             fiy3             = _mm_add_pd(fiy3,ty);
 418             fiz3             = _mm_add_pd(fiz3,tz);
 419
 420             fjx0             = _mm_add_pd(fjx0,tx);
 421             fjy0             = _mm_add_pd(fjy0,ty);
 422             fjz0             = _mm_add_pd(fjz0,tz);
 423
 424             }
 425
 426             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 427
 428             /* Inner loop uses 170 flops */
 429         }
 430
 431         if(jidx<j_index_end)
 432         {
 433
 434             jnrA             = jjnr[jidx];
 435             j_coord_offsetA  = DIM*jnrA;
 436
 437             /* load j atom coordinates */
 438             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 439                                               &jx0,&jy0,&jz0);
 440
 441             /* Calculate displacement vector */
 442             dx00             = _mm_sub_pd(ix0,jx0);
 443             dy00             = _mm_sub_pd(iy0,jy0);
 444             dz00             = _mm_sub_pd(iz0,jz0);
 445             dx10             = _mm_sub_pd(ix1,jx0);
 446             dy10             = _mm_sub_pd(iy1,jy0);
 447             dz10             = _mm_sub_pd(iz1,jz0);
 448             dx20             = _mm_sub_pd(ix2,jx0);
 449             dy20             = _mm_sub_pd(iy2,jy0);
 450             dz20             = _mm_sub_pd(iz2,jz0);
 451             dx30             = _mm_sub_pd(ix3,jx0);
 452             dy30             = _mm_sub_pd(iy3,jy0);
 453             dz30             = _mm_sub_pd(iz3,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 457             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 458             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 459             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 460
 461             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 462             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 463             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 464             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 465
 466             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 467             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 468             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 469             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 470
 471             /* Load parameters for j particles */
 472             jq0              = _mm_load_sd(charge+jnrA+0);
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474
 475             fjx0             = _mm_setzero_pd();
 476             fjy0             = _mm_setzero_pd();
 477             fjz0             = _mm_setzero_pd();
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             if (gmx_mm_any_lt(rsq00,rcutoff2))
 484             {
 485
 486             r00              = _mm_mul_pd(rsq00,rinv00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 490
 491             /* LENNARD-JONES DISPERSION/REPULSION */
 492
 493             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 494             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 495             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 496             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 497             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 498
 499             d                = _mm_sub_pd(r00,rswitch);
 500             d                = _mm_max_pd(d,_mm_setzero_pd());
 501             d2               = _mm_mul_pd(d,d);
 502             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 503
 504             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 505
 506             /* Evaluate switch function */
 507             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 508             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 509             vvdw             = _mm_mul_pd(vvdw,sw);
 510             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 514             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 515             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 516
 517             fscal            = fvdw;
 518
 519             fscal            = _mm_and_pd(fscal,cutoff_mask);
 520
 521             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm_mul_pd(fscal,dx00);
 525             ty               = _mm_mul_pd(fscal,dy00);
 526             tz               = _mm_mul_pd(fscal,dz00);
 527
 528             /* Update vectorial force */
 529             fix0             = _mm_add_pd(fix0,tx);
 530             fiy0             = _mm_add_pd(fiy0,ty);
 531             fiz0             = _mm_add_pd(fiz0,tz);
 532
 533             fjx0             = _mm_add_pd(fjx0,tx);
 534             fjy0             = _mm_add_pd(fjy0,ty);
 535             fjz0             = _mm_add_pd(fjz0,tz);
 536
 537             }
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             if (gmx_mm_any_lt(rsq10,rcutoff2))
 544             {
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq10             = _mm_mul_pd(iq1,jq0);
 548
 549             /* REACTION-FIELD ELECTROSTATICS */
 550             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 551             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 552
 553             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             velec            = _mm_and_pd(velec,cutoff_mask);
 557             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 558             velecsum         = _mm_add_pd(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_and_pd(fscal,cutoff_mask);
 563
 564             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_pd(fscal,dx10);
 568             ty               = _mm_mul_pd(fscal,dy10);
 569             tz               = _mm_mul_pd(fscal,dz10);
 570
 571             /* Update vectorial force */
 572             fix1             = _mm_add_pd(fix1,tx);
 573             fiy1             = _mm_add_pd(fiy1,ty);
 574             fiz1             = _mm_add_pd(fiz1,tz);
 575
 576             fjx0             = _mm_add_pd(fjx0,tx);
 577             fjy0             = _mm_add_pd(fjy0,ty);
 578             fjz0             = _mm_add_pd(fjz0,tz);
 579
 580             }
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             if (gmx_mm_any_lt(rsq20,rcutoff2))
 587             {
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq20             = _mm_mul_pd(iq2,jq0);
 591
 592             /* REACTION-FIELD ELECTROSTATICS */
 593             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 594             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 595
 596             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velec            = _mm_and_pd(velec,cutoff_mask);
 600             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 601             velecsum         = _mm_add_pd(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm_and_pd(fscal,cutoff_mask);
 606
 607             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm_mul_pd(fscal,dx20);
 611             ty               = _mm_mul_pd(fscal,dy20);
 612             tz               = _mm_mul_pd(fscal,dz20);
 613
 614             /* Update vectorial force */
 615             fix2             = _mm_add_pd(fix2,tx);
 616             fiy2             = _mm_add_pd(fiy2,ty);
 617             fiz2             = _mm_add_pd(fiz2,tz);
 618
 619             fjx0             = _mm_add_pd(fjx0,tx);
 620             fjy0             = _mm_add_pd(fjy0,ty);
 621             fjz0             = _mm_add_pd(fjz0,tz);
 622
 623             }
 624
 625             /**************************
 626              * CALCULATE INTERACTIONS *
 627              **************************/
 628
 629             if (gmx_mm_any_lt(rsq30,rcutoff2))
 630             {
 631
 632             /* Compute parameters for interactions between i and j atoms */
 633             qq30             = _mm_mul_pd(iq3,jq0);
 634
 635             /* REACTION-FIELD ELECTROSTATICS */
 636             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 637             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 638
 639             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm_and_pd(velec,cutoff_mask);
 643             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 644             velecsum         = _mm_add_pd(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm_and_pd(fscal,cutoff_mask);
 649
 650             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm_mul_pd(fscal,dx30);
 654             ty               = _mm_mul_pd(fscal,dy30);
 655             tz               = _mm_mul_pd(fscal,dz30);
 656
 657             /* Update vectorial force */
 658             fix3             = _mm_add_pd(fix3,tx);
 659             fiy3             = _mm_add_pd(fiy3,ty);
 660             fiz3             = _mm_add_pd(fiz3,tz);
 661
 662             fjx0             = _mm_add_pd(fjx0,tx);
 663             fjy0             = _mm_add_pd(fjy0,ty);
 664             fjz0             = _mm_add_pd(fjz0,tz);
 665
 666             }
 667
 668             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 669
 670             /* Inner loop uses 170 flops */
 671         }
 672
 673         /* End of innermost loop */
 674
 675         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 676                                               f+i_coord_offset,fshift+i_shift_offset);
 677
 678         ggid                        = gid[iidx];
 679         /* Update potential energies */
 680         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 681         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 682
 683         /* Increment number of inner iterations */
 684         inneriter                  += j_index_end - j_index_start;
 685
 686         /* Outer loop uses 26 flops */
 687     }
 688
 689     /* Increment number of outer iterations */
 690     outeriter        += nri;
 691
 692     /* Update outer/inner flops */
 693
 694     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*170);
 695 }
 696 /*
 697  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_double
 698  * Electrostatics interaction: ReactionField
 699  * VdW interaction:            LennardJones
 700  * Geometry:                   Water4-Particle
 701  * Calculate force/pot:        Force
 702  */
 703 void
 704 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_double
 705                     (t_nblist                    * gmx_restrict       nlist,
 706                      rvec                        * gmx_restrict          xx,
 707                      rvec                        * gmx_restrict          ff,
 708                      t_forcerec                  * gmx_restrict          fr,
 709                      t_mdatoms                   * gmx_restrict     mdatoms,
 710                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 711                      t_nrnb                      * gmx_restrict        nrnb)
 712 {
 713     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 714      * just 0 for non-waters.
 715      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 716      * jnr indices corresponding to data put in the four positions in the SIMD register.
 717      */
 718     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 719     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 720     int              jnrA,jnrB;
 721     int              j_coord_offsetA,j_coord_offsetB;
 722     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 723     real             rcutoff_scalar;
 724     real             *shiftvec,*fshift,*x,*f;
 725     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 726     int              vdwioffset0;
 727     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 728     int              vdwioffset1;
 729     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 730     int              vdwioffset2;
 731     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 732     int              vdwioffset3;
 733     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 734     int              vdwjidx0A,vdwjidx0B;
 735     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 736     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 737     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 738     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 739     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 740     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 741     real             *charge;
 742     int              nvdwtype;
 743     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 744     int              *vdwtype;
 745     real             *vdwparam;
 746     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 747     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 748     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 749     real             rswitch_scalar,d_scalar;
 750     __m128d          dummy_mask,cutoff_mask;
 751     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 752     __m128d          one     = _mm_set1_pd(1.0);
 753     __m128d          two     = _mm_set1_pd(2.0);
 754     x                = xx[0];
 755     f                = ff[0];
 756
 757     nri              = nlist->nri;
 758     iinr             = nlist->iinr;
 759     jindex           = nlist->jindex;
 760     jjnr             = nlist->jjnr;
 761     shiftidx         = nlist->shift;
 762     gid              = nlist->gid;
 763     shiftvec         = fr->shift_vec[0];
 764     fshift           = fr->fshift[0];
 765     facel            = _mm_set1_pd(fr->epsfac);
 766     charge           = mdatoms->chargeA;
 767     krf              = _mm_set1_pd(fr->ic->k_rf);
 768     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 769     crf              = _mm_set1_pd(fr->ic->c_rf);
 770     nvdwtype         = fr->ntype;
 771     vdwparam         = fr->nbfp;
 772     vdwtype          = mdatoms->typeA;
 773
 774     /* Setup water-specific parameters */
 775     inr              = nlist->iinr[0];
 776     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 777     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 778     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 779     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 780
 781     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 782     rcutoff_scalar   = fr->rcoulomb;
 783     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 784     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 785
 786     rswitch_scalar   = fr->rvdw_switch;
 787     rswitch          = _mm_set1_pd(rswitch_scalar);
 788     /* Setup switch parameters */
 789     d_scalar         = rcutoff_scalar-rswitch_scalar;
 790     d                = _mm_set1_pd(d_scalar);
 791     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 792     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 793     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 794     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 795     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 796     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 797
 798     /* Avoid stupid compiler warnings */
 799     jnrA = jnrB = 0;
 800     j_coord_offsetA = 0;
 801     j_coord_offsetB = 0;
 802
 803     outeriter        = 0;
 804     inneriter        = 0;
 805
 806     /* Start outer loop over neighborlists */
 807     for(iidx=0; iidx<nri; iidx++)
 808     {
 809         /* Load shift vector for this list */
 810         i_shift_offset   = DIM*shiftidx[iidx];
 811
 812         /* Load limits for loop over neighbors */
 813         j_index_start    = jindex[iidx];
 814         j_index_end      = jindex[iidx+1];
 815
 816         /* Get outer coordinate index */
 817         inr              = iinr[iidx];
 818         i_coord_offset   = DIM*inr;
 819
 820         /* Load i particle coords and add shift vector */
 821         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 822                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 823
 824         fix0             = _mm_setzero_pd();
 825         fiy0             = _mm_setzero_pd();
 826         fiz0             = _mm_setzero_pd();
 827         fix1             = _mm_setzero_pd();
 828         fiy1             = _mm_setzero_pd();
 829         fiz1             = _mm_setzero_pd();
 830         fix2             = _mm_setzero_pd();
 831         fiy2             = _mm_setzero_pd();
 832         fiz2             = _mm_setzero_pd();
 833         fix3             = _mm_setzero_pd();
 834         fiy3             = _mm_setzero_pd();
 835         fiz3             = _mm_setzero_pd();
 836
 837         /* Start inner kernel loop */
 838         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 839         {
 840
 841             /* Get j neighbor index, and coordinate index */
 842             jnrA             = jjnr[jidx];
 843             jnrB             = jjnr[jidx+1];
 844             j_coord_offsetA  = DIM*jnrA;
 845             j_coord_offsetB  = DIM*jnrB;
 846
 847             /* load j atom coordinates */
 848             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 849                                               &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_pd(ix0,jx0);
 853             dy00             = _mm_sub_pd(iy0,jy0);
 854             dz00             = _mm_sub_pd(iz0,jz0);
 855             dx10             = _mm_sub_pd(ix1,jx0);
 856             dy10             = _mm_sub_pd(iy1,jy0);
 857             dz10             = _mm_sub_pd(iz1,jz0);
 858             dx20             = _mm_sub_pd(ix2,jx0);
 859             dy20             = _mm_sub_pd(iy2,jy0);
 860             dz20             = _mm_sub_pd(iz2,jz0);
 861             dx30             = _mm_sub_pd(ix3,jx0);
 862             dy30             = _mm_sub_pd(iy3,jy0);
 863             dz30             = _mm_sub_pd(iz3,jz0);
 864
 865             /* Calculate squared distance and things based on it */
 866             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 867             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 868             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 869             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 870
 871             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 872             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 873             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 874             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 875
 876             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 877             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 878             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 879             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 880
 881             /* Load parameters for j particles */
 882             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 883             vdwjidx0A        = 2*vdwtype[jnrA+0];
 884             vdwjidx0B        = 2*vdwtype[jnrB+0];
 885
 886             fjx0             = _mm_setzero_pd();
 887             fjy0             = _mm_setzero_pd();
 888             fjz0             = _mm_setzero_pd();
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             if (gmx_mm_any_lt(rsq00,rcutoff2))
 895             {
 896
 897             r00              = _mm_mul_pd(rsq00,rinv00);
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 901                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 902
 903             /* LENNARD-JONES DISPERSION/REPULSION */
 904
 905             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 906             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 907             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 908             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 909             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 910
 911             d                = _mm_sub_pd(r00,rswitch);
 912             d                = _mm_max_pd(d,_mm_setzero_pd());
 913             d2               = _mm_mul_pd(d,d);
 914             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 915
 916             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 917
 918             /* Evaluate switch function */
 919             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 920             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 921             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 922
 923             fscal            = fvdw;
 924
 925             fscal            = _mm_and_pd(fscal,cutoff_mask);
 926
 927             /* Calculate temporary vectorial force */
 928             tx               = _mm_mul_pd(fscal,dx00);
 929             ty               = _mm_mul_pd(fscal,dy00);
 930             tz               = _mm_mul_pd(fscal,dz00);
 931
 932             /* Update vectorial force */
 933             fix0             = _mm_add_pd(fix0,tx);
 934             fiy0             = _mm_add_pd(fiy0,ty);
 935             fiz0             = _mm_add_pd(fiz0,tz);
 936
 937             fjx0             = _mm_add_pd(fjx0,tx);
 938             fjy0             = _mm_add_pd(fjy0,ty);
 939             fjz0             = _mm_add_pd(fjz0,tz);
 940
 941             }
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             if (gmx_mm_any_lt(rsq10,rcutoff2))
 948             {
 949
 950             /* Compute parameters for interactions between i and j atoms */
 951             qq10             = _mm_mul_pd(iq1,jq0);
 952
 953             /* REACTION-FIELD ELECTROSTATICS */
 954             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 955
 956             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 957
 958             fscal            = felec;
 959
 960             fscal            = _mm_and_pd(fscal,cutoff_mask);
 961
 962             /* Calculate temporary vectorial force */
 963             tx               = _mm_mul_pd(fscal,dx10);
 964             ty               = _mm_mul_pd(fscal,dy10);
 965             tz               = _mm_mul_pd(fscal,dz10);
 966
 967             /* Update vectorial force */
 968             fix1             = _mm_add_pd(fix1,tx);
 969             fiy1             = _mm_add_pd(fiy1,ty);
 970             fiz1             = _mm_add_pd(fiz1,tz);
 971
 972             fjx0             = _mm_add_pd(fjx0,tx);
 973             fjy0             = _mm_add_pd(fjy0,ty);
 974             fjz0             = _mm_add_pd(fjz0,tz);
 975
 976             }
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             if (gmx_mm_any_lt(rsq20,rcutoff2))
 983             {
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             qq20             = _mm_mul_pd(iq2,jq0);
 987
 988             /* REACTION-FIELD ELECTROSTATICS */
 989             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 990
 991             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 992
 993             fscal            = felec;
 994
 995             fscal            = _mm_and_pd(fscal,cutoff_mask);
 996
 997             /* Calculate temporary vectorial force */
 998             tx               = _mm_mul_pd(fscal,dx20);
 999             ty               = _mm_mul_pd(fscal,dy20);
1000             tz               = _mm_mul_pd(fscal,dz20);
1001
1002             /* Update vectorial force */
1003             fix2             = _mm_add_pd(fix2,tx);
1004             fiy2             = _mm_add_pd(fiy2,ty);
1005             fiz2             = _mm_add_pd(fiz2,tz);
1006
1007             fjx0             = _mm_add_pd(fjx0,tx);
1008             fjy0             = _mm_add_pd(fjy0,ty);
1009             fjz0             = _mm_add_pd(fjz0,tz);
1010
1011             }
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             if (gmx_mm_any_lt(rsq30,rcutoff2))
1018             {
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq30             = _mm_mul_pd(iq3,jq0);
1022
1023             /* REACTION-FIELD ELECTROSTATICS */
1024             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1025
1026             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1027
1028             fscal            = felec;
1029
1030             fscal            = _mm_and_pd(fscal,cutoff_mask);
1031
1032             /* Calculate temporary vectorial force */
1033             tx               = _mm_mul_pd(fscal,dx30);
1034             ty               = _mm_mul_pd(fscal,dy30);
1035             tz               = _mm_mul_pd(fscal,dz30);
1036
1037             /* Update vectorial force */
1038             fix3             = _mm_add_pd(fix3,tx);
1039             fiy3             = _mm_add_pd(fiy3,ty);
1040             fiz3             = _mm_add_pd(fiz3,tz);
1041
1042             fjx0             = _mm_add_pd(fjx0,tx);
1043             fjy0             = _mm_add_pd(fjy0,ty);
1044             fjz0             = _mm_add_pd(fjz0,tz);
1045
1046             }
1047
1048             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1049
1050             /* Inner loop uses 149 flops */
1051         }
1052
1053         if(jidx<j_index_end)
1054         {
1055
1056             jnrA             = jjnr[jidx];
1057             j_coord_offsetA  = DIM*jnrA;
1058
1059             /* load j atom coordinates */
1060             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1061                                               &jx0,&jy0,&jz0);
1062
1063             /* Calculate displacement vector */
1064             dx00             = _mm_sub_pd(ix0,jx0);
1065             dy00             = _mm_sub_pd(iy0,jy0);
1066             dz00             = _mm_sub_pd(iz0,jz0);
1067             dx10             = _mm_sub_pd(ix1,jx0);
1068             dy10             = _mm_sub_pd(iy1,jy0);
1069             dz10             = _mm_sub_pd(iz1,jz0);
1070             dx20             = _mm_sub_pd(ix2,jx0);
1071             dy20             = _mm_sub_pd(iy2,jy0);
1072             dz20             = _mm_sub_pd(iz2,jz0);
1073             dx30             = _mm_sub_pd(ix3,jx0);
1074             dy30             = _mm_sub_pd(iy3,jy0);
1075             dz30             = _mm_sub_pd(iz3,jz0);
1076
1077             /* Calculate squared distance and things based on it */
1078             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1079             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1080             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1081             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1082
1083             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1084             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1085             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1086             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1087
1088             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1089             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1090             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1091             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1092
1093             /* Load parameters for j particles */
1094             jq0              = _mm_load_sd(charge+jnrA+0);
1095             vdwjidx0A        = 2*vdwtype[jnrA+0];
1096
1097             fjx0             = _mm_setzero_pd();
1098             fjy0             = _mm_setzero_pd();
1099             fjz0             = _mm_setzero_pd();
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             if (gmx_mm_any_lt(rsq00,rcutoff2))
1106             {
1107
1108             r00              = _mm_mul_pd(rsq00,rinv00);
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1112
1113             /* LENNARD-JONES DISPERSION/REPULSION */
1114
1115             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1116             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1117             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1118             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1119             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1120
1121             d                = _mm_sub_pd(r00,rswitch);
1122             d                = _mm_max_pd(d,_mm_setzero_pd());
1123             d2               = _mm_mul_pd(d,d);
1124             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1125
1126             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1127
1128             /* Evaluate switch function */
1129             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1130             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1131             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1132
1133             fscal            = fvdw;
1134
1135             fscal            = _mm_and_pd(fscal,cutoff_mask);
1136
1137             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm_mul_pd(fscal,dx00);
1141             ty               = _mm_mul_pd(fscal,dy00);
1142             tz               = _mm_mul_pd(fscal,dz00);
1143
1144             /* Update vectorial force */
1145             fix0             = _mm_add_pd(fix0,tx);
1146             fiy0             = _mm_add_pd(fiy0,ty);
1147             fiz0             = _mm_add_pd(fiz0,tz);
1148
1149             fjx0             = _mm_add_pd(fjx0,tx);
1150             fjy0             = _mm_add_pd(fjy0,ty);
1151             fjz0             = _mm_add_pd(fjz0,tz);
1152
1153             }
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             if (gmx_mm_any_lt(rsq10,rcutoff2))
1160             {
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             qq10             = _mm_mul_pd(iq1,jq0);
1164
1165             /* REACTION-FIELD ELECTROSTATICS */
1166             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1167
1168             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm_and_pd(fscal,cutoff_mask);
1173
1174             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1175
1176             /* Calculate temporary vectorial force */
1177             tx               = _mm_mul_pd(fscal,dx10);
1178             ty               = _mm_mul_pd(fscal,dy10);
1179             tz               = _mm_mul_pd(fscal,dz10);
1180
1181             /* Update vectorial force */
1182             fix1             = _mm_add_pd(fix1,tx);
1183             fiy1             = _mm_add_pd(fiy1,ty);
1184             fiz1             = _mm_add_pd(fiz1,tz);
1185
1186             fjx0             = _mm_add_pd(fjx0,tx);
1187             fjy0             = _mm_add_pd(fjy0,ty);
1188             fjz0             = _mm_add_pd(fjz0,tz);
1189
1190             }
1191
1192             /**************************
1193              * CALCULATE INTERACTIONS *
1194              **************************/
1195
1196             if (gmx_mm_any_lt(rsq20,rcutoff2))
1197             {
1198
1199             /* Compute parameters for interactions between i and j atoms */
1200             qq20             = _mm_mul_pd(iq2,jq0);
1201
1202             /* REACTION-FIELD ELECTROSTATICS */
1203             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1204
1205             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1206
1207             fscal            = felec;
1208
1209             fscal            = _mm_and_pd(fscal,cutoff_mask);
1210
1211             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1212
1213             /* Calculate temporary vectorial force */
1214             tx               = _mm_mul_pd(fscal,dx20);
1215             ty               = _mm_mul_pd(fscal,dy20);
1216             tz               = _mm_mul_pd(fscal,dz20);
1217
1218             /* Update vectorial force */
1219             fix2             = _mm_add_pd(fix2,tx);
1220             fiy2             = _mm_add_pd(fiy2,ty);
1221             fiz2             = _mm_add_pd(fiz2,tz);
1222
1223             fjx0             = _mm_add_pd(fjx0,tx);
1224             fjy0             = _mm_add_pd(fjy0,ty);
1225             fjz0             = _mm_add_pd(fjz0,tz);
1226
1227             }
1228
1229             /**************************
1230              * CALCULATE INTERACTIONS *
1231              **************************/
1232
1233             if (gmx_mm_any_lt(rsq30,rcutoff2))
1234             {
1235
1236             /* Compute parameters for interactions between i and j atoms */
1237             qq30             = _mm_mul_pd(iq3,jq0);
1238
1239             /* REACTION-FIELD ELECTROSTATICS */
1240             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1241
1242             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1243
1244             fscal            = felec;
1245
1246             fscal            = _mm_and_pd(fscal,cutoff_mask);
1247
1248             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm_mul_pd(fscal,dx30);
1252             ty               = _mm_mul_pd(fscal,dy30);
1253             tz               = _mm_mul_pd(fscal,dz30);
1254
1255             /* Update vectorial force */
1256             fix3             = _mm_add_pd(fix3,tx);
1257             fiy3             = _mm_add_pd(fiy3,ty);
1258             fiz3             = _mm_add_pd(fiz3,tz);
1259
1260             fjx0             = _mm_add_pd(fjx0,tx);
1261             fjy0             = _mm_add_pd(fjy0,ty);
1262             fjz0             = _mm_add_pd(fjz0,tz);
1263
1264             }
1265
1266             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1267
1268             /* Inner loop uses 149 flops */
1269         }
1270
1271         /* End of innermost loop */
1272
1273         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1274                                               f+i_coord_offset,fshift+i_shift_offset);
1275
1276         /* Increment number of inner iterations */
1277         inneriter                  += j_index_end - j_index_start;
1278
1279         /* Outer loop uses 24 flops */
1280     }
1281
1282     /* Increment number of outer iterations */
1283     outeriter        += nri;
1284
1285     /* Update outer/inner flops */
1286
1287     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*149);
1288 }