src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_sse2_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 sse2_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_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse2_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse2_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              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 100     real             rswitch_scalar,d_scalar;
 101     __m128d          dummy_mask,cutoff_mask;
 102     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 103     __m128d          one     = _mm_set1_pd(1.0);
 104     __m128d          two     = _mm_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_pd(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     krf              = _mm_set1_pd(fr->ic->k_rf);
 119     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 120     crf              = _mm_set1_pd(fr->ic->c_rf);
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 128     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 129     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 130     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 131
 132     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 133     rcutoff_scalar   = fr->rcoulomb;
 134     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 135     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 136
 137     rswitch_scalar   = fr->rvdw_switch;
 138     rswitch          = _mm_set1_pd(rswitch_scalar);
 139     /* Setup switch parameters */
 140     d_scalar         = rcutoff_scalar-rswitch_scalar;
 141     d                = _mm_set1_pd(d_scalar);
 142     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 143     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 144     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 145     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 146     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 147     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 174
 175         fix0             = _mm_setzero_pd();
 176         fiy0             = _mm_setzero_pd();
 177         fiz0             = _mm_setzero_pd();
 178         fix1             = _mm_setzero_pd();
 179         fiy1             = _mm_setzero_pd();
 180         fiz1             = _mm_setzero_pd();
 181         fix2             = _mm_setzero_pd();
 182         fiy2             = _mm_setzero_pd();
 183         fiz2             = _mm_setzero_pd();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_pd();
 187         vvdwsum          = _mm_setzero_pd();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_pd(ix0,jx0);
 205             dy00             = _mm_sub_pd(iy0,jy0);
 206             dz00             = _mm_sub_pd(iz0,jz0);
 207             dx10             = _mm_sub_pd(ix1,jx0);
 208             dy10             = _mm_sub_pd(iy1,jy0);
 209             dz10             = _mm_sub_pd(iz1,jz0);
 210             dx20             = _mm_sub_pd(ix2,jx0);
 211             dy20             = _mm_sub_pd(iy2,jy0);
 212             dz20             = _mm_sub_pd(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 218
 219             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 220             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 221             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 222
 223             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 224             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 225             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231
 232             fjx0             = _mm_setzero_pd();
 233             fjy0             = _mm_setzero_pd();
 234             fjz0             = _mm_setzero_pd();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             r00              = _mm_mul_pd(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm_mul_pd(iq0,jq0);
 247             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 249
 250             /* REACTION-FIELD ELECTROSTATICS */
 251             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 252             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 253
 254             /* LENNARD-JONES DISPERSION/REPULSION */
 255
 256             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 257             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 258             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 259             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 260             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 261
 262             d                = _mm_sub_pd(r00,rswitch);
 263             d                = _mm_max_pd(d,_mm_setzero_pd());
 264             d2               = _mm_mul_pd(d,d);
 265             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)))))));
 266
 267             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 268
 269             /* Evaluate switch function */
 270             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 271             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 272             vvdw             = _mm_mul_pd(vvdw,sw);
 273             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm_and_pd(velec,cutoff_mask);
 277             velecsum         = _mm_add_pd(velecsum,velec);
 278             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 279             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_pd(felec,fvdw);
 282
 283             fscal            = _mm_and_pd(fscal,cutoff_mask);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_pd(fscal,dx00);
 287             ty               = _mm_mul_pd(fscal,dy00);
 288             tz               = _mm_mul_pd(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_pd(fix0,tx);
 292             fiy0             = _mm_add_pd(fiy0,ty);
 293             fiz0             = _mm_add_pd(fiz0,tz);
 294
 295             fjx0             = _mm_add_pd(fjx0,tx);
 296             fjy0             = _mm_add_pd(fjy0,ty);
 297             fjz0             = _mm_add_pd(fjz0,tz);
 298
 299             }
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             if (gmx_mm_any_lt(rsq10,rcutoff2))
 306             {
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm_mul_pd(iq1,jq0);
 310
 311             /* REACTION-FIELD ELECTROSTATICS */
 312             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 313             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 314
 315             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm_and_pd(velec,cutoff_mask);
 319             velecsum         = _mm_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             fscal            = _mm_and_pd(fscal,cutoff_mask);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm_mul_pd(fscal,dx10);
 327             ty               = _mm_mul_pd(fscal,dy10);
 328             tz               = _mm_mul_pd(fscal,dz10);
 329
 330             /* Update vectorial force */
 331             fix1             = _mm_add_pd(fix1,tx);
 332             fiy1             = _mm_add_pd(fiy1,ty);
 333             fiz1             = _mm_add_pd(fiz1,tz);
 334
 335             fjx0             = _mm_add_pd(fjx0,tx);
 336             fjy0             = _mm_add_pd(fjy0,ty);
 337             fjz0             = _mm_add_pd(fjz0,tz);
 338
 339             }
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm_any_lt(rsq20,rcutoff2))
 346             {
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq20             = _mm_mul_pd(iq2,jq0);
 350
 351             /* REACTION-FIELD ELECTROSTATICS */
 352             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 353             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 354
 355             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velec            = _mm_and_pd(velec,cutoff_mask);
 359             velecsum         = _mm_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             fscal            = _mm_and_pd(fscal,cutoff_mask);
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm_mul_pd(fscal,dx20);
 367             ty               = _mm_mul_pd(fscal,dy20);
 368             tz               = _mm_mul_pd(fscal,dz20);
 369
 370             /* Update vectorial force */
 371             fix2             = _mm_add_pd(fix2,tx);
 372             fiy2             = _mm_add_pd(fiy2,ty);
 373             fiz2             = _mm_add_pd(fiz2,tz);
 374
 375             fjx0             = _mm_add_pd(fjx0,tx);
 376             fjy0             = _mm_add_pd(fjy0,ty);
 377             fjz0             = _mm_add_pd(fjz0,tz);
 378
 379             }
 380
 381             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 382
 383             /* Inner loop uses 145 flops */
 384         }
 385
 386         if(jidx<j_index_end)
 387         {
 388
 389             jnrA             = jjnr[jidx];
 390             j_coord_offsetA  = DIM*jnrA;
 391
 392             /* load j atom coordinates */
 393             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 394                                               &jx0,&jy0,&jz0);
 395
 396             /* Calculate displacement vector */
 397             dx00             = _mm_sub_pd(ix0,jx0);
 398             dy00             = _mm_sub_pd(iy0,jy0);
 399             dz00             = _mm_sub_pd(iz0,jz0);
 400             dx10             = _mm_sub_pd(ix1,jx0);
 401             dy10             = _mm_sub_pd(iy1,jy0);
 402             dz10             = _mm_sub_pd(iz1,jz0);
 403             dx20             = _mm_sub_pd(ix2,jx0);
 404             dy20             = _mm_sub_pd(iy2,jy0);
 405             dz20             = _mm_sub_pd(iz2,jz0);
 406
 407             /* Calculate squared distance and things based on it */
 408             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 409             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 410             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 411
 412             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 413             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 414             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 415
 416             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 417             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 418             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 419
 420             /* Load parameters for j particles */
 421             jq0              = _mm_load_sd(charge+jnrA+0);
 422             vdwjidx0A        = 2*vdwtype[jnrA+0];
 423
 424             fjx0             = _mm_setzero_pd();
 425             fjy0             = _mm_setzero_pd();
 426             fjz0             = _mm_setzero_pd();
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             if (gmx_mm_any_lt(rsq00,rcutoff2))
 433             {
 434
 435             r00              = _mm_mul_pd(rsq00,rinv00);
 436
 437             /* Compute parameters for interactions between i and j atoms */
 438             qq00             = _mm_mul_pd(iq0,jq0);
 439             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 440
 441             /* REACTION-FIELD ELECTROSTATICS */
 442             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 443             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 444
 445             /* LENNARD-JONES DISPERSION/REPULSION */
 446
 447             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 448             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 449             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 450             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 451             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 452
 453             d                = _mm_sub_pd(r00,rswitch);
 454             d                = _mm_max_pd(d,_mm_setzero_pd());
 455             d2               = _mm_mul_pd(d,d);
 456             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)))))));
 457
 458             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 459
 460             /* Evaluate switch function */
 461             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 462             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 463             vvdw             = _mm_mul_pd(vvdw,sw);
 464             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velec            = _mm_and_pd(velec,cutoff_mask);
 468             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 469             velecsum         = _mm_add_pd(velecsum,velec);
 470             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 471             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 472             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 473
 474             fscal            = _mm_add_pd(felec,fvdw);
 475
 476             fscal            = _mm_and_pd(fscal,cutoff_mask);
 477
 478             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm_mul_pd(fscal,dx00);
 482             ty               = _mm_mul_pd(fscal,dy00);
 483             tz               = _mm_mul_pd(fscal,dz00);
 484
 485             /* Update vectorial force */
 486             fix0             = _mm_add_pd(fix0,tx);
 487             fiy0             = _mm_add_pd(fiy0,ty);
 488             fiz0             = _mm_add_pd(fiz0,tz);
 489
 490             fjx0             = _mm_add_pd(fjx0,tx);
 491             fjy0             = _mm_add_pd(fjy0,ty);
 492             fjz0             = _mm_add_pd(fjz0,tz);
 493
 494             }
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             if (gmx_mm_any_lt(rsq10,rcutoff2))
 501             {
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             qq10             = _mm_mul_pd(iq1,jq0);
 505
 506             /* REACTION-FIELD ELECTROSTATICS */
 507             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 508             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 509
 510             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velec            = _mm_and_pd(velec,cutoff_mask);
 514             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 515             velecsum         = _mm_add_pd(velecsum,velec);
 516
 517             fscal            = felec;
 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,dx10);
 525             ty               = _mm_mul_pd(fscal,dy10);
 526             tz               = _mm_mul_pd(fscal,dz10);
 527
 528             /* Update vectorial force */
 529             fix1             = _mm_add_pd(fix1,tx);
 530             fiy1             = _mm_add_pd(fiy1,ty);
 531             fiz1             = _mm_add_pd(fiz1,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(rsq20,rcutoff2))
 544             {
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq20             = _mm_mul_pd(iq2,jq0);
 548
 549             /* REACTION-FIELD ELECTROSTATICS */
 550             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 551             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 552
 553             cutoff_mask      = _mm_cmplt_pd(rsq20,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,dx20);
 568             ty               = _mm_mul_pd(fscal,dy20);
 569             tz               = _mm_mul_pd(fscal,dz20);
 570
 571             /* Update vectorial force */
 572             fix2             = _mm_add_pd(fix2,tx);
 573             fiy2             = _mm_add_pd(fiy2,ty);
 574             fiz2             = _mm_add_pd(fiz2,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             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 583
 584             /* Inner loop uses 145 flops */
 585         }
 586
 587         /* End of innermost loop */
 588
 589         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 590                                               f+i_coord_offset,fshift+i_shift_offset);
 591
 592         ggid                        = gid[iidx];
 593         /* Update potential energies */
 594         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 595         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 596
 597         /* Increment number of inner iterations */
 598         inneriter                  += j_index_end - j_index_start;
 599
 600         /* Outer loop uses 20 flops */
 601     }
 602
 603     /* Increment number of outer iterations */
 604     outeriter        += nri;
 605
 606     /* Update outer/inner flops */
 607
 608     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 609 }
 610 /*
 611  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_double
 612  * Electrostatics interaction: ReactionField
 613  * VdW interaction:            LennardJones
 614  * Geometry:                   Water3-Particle
 615  * Calculate force/pot:        Force
 616  */
 617 void
 618 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_double
 619                     (t_nblist                    * gmx_restrict       nlist,
 620                      rvec                        * gmx_restrict          xx,
 621                      rvec                        * gmx_restrict          ff,
 622                      t_forcerec                  * gmx_restrict          fr,
 623                      t_mdatoms                   * gmx_restrict     mdatoms,
 624                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 625                      t_nrnb                      * gmx_restrict        nrnb)
 626 {
 627     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 628      * just 0 for non-waters.
 629      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 630      * jnr indices corresponding to data put in the four positions in the SIMD register.
 631      */
 632     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 633     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 634     int              jnrA,jnrB;
 635     int              j_coord_offsetA,j_coord_offsetB;
 636     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 637     real             rcutoff_scalar;
 638     real             *shiftvec,*fshift,*x,*f;
 639     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 640     int              vdwioffset0;
 641     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 642     int              vdwioffset1;
 643     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 644     int              vdwioffset2;
 645     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 646     int              vdwjidx0A,vdwjidx0B;
 647     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 648     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 649     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 650     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 651     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 652     real             *charge;
 653     int              nvdwtype;
 654     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 655     int              *vdwtype;
 656     real             *vdwparam;
 657     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 658     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 659     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 660     real             rswitch_scalar,d_scalar;
 661     __m128d          dummy_mask,cutoff_mask;
 662     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 663     __m128d          one     = _mm_set1_pd(1.0);
 664     __m128d          two     = _mm_set1_pd(2.0);
 665     x                = xx[0];
 666     f                = ff[0];
 667
 668     nri              = nlist->nri;
 669     iinr             = nlist->iinr;
 670     jindex           = nlist->jindex;
 671     jjnr             = nlist->jjnr;
 672     shiftidx         = nlist->shift;
 673     gid              = nlist->gid;
 674     shiftvec         = fr->shift_vec[0];
 675     fshift           = fr->fshift[0];
 676     facel            = _mm_set1_pd(fr->epsfac);
 677     charge           = mdatoms->chargeA;
 678     krf              = _mm_set1_pd(fr->ic->k_rf);
 679     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 680     crf              = _mm_set1_pd(fr->ic->c_rf);
 681     nvdwtype         = fr->ntype;
 682     vdwparam         = fr->nbfp;
 683     vdwtype          = mdatoms->typeA;
 684
 685     /* Setup water-specific parameters */
 686     inr              = nlist->iinr[0];
 687     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 688     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 689     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 690     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 691
 692     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 693     rcutoff_scalar   = fr->rcoulomb;
 694     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 695     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 696
 697     rswitch_scalar   = fr->rvdw_switch;
 698     rswitch          = _mm_set1_pd(rswitch_scalar);
 699     /* Setup switch parameters */
 700     d_scalar         = rcutoff_scalar-rswitch_scalar;
 701     d                = _mm_set1_pd(d_scalar);
 702     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 703     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 704     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 705     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 706     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 707     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 708
 709     /* Avoid stupid compiler warnings */
 710     jnrA = jnrB = 0;
 711     j_coord_offsetA = 0;
 712     j_coord_offsetB = 0;
 713
 714     outeriter        = 0;
 715     inneriter        = 0;
 716
 717     /* Start outer loop over neighborlists */
 718     for(iidx=0; iidx<nri; iidx++)
 719     {
 720         /* Load shift vector for this list */
 721         i_shift_offset   = DIM*shiftidx[iidx];
 722
 723         /* Load limits for loop over neighbors */
 724         j_index_start    = jindex[iidx];
 725         j_index_end      = jindex[iidx+1];
 726
 727         /* Get outer coordinate index */
 728         inr              = iinr[iidx];
 729         i_coord_offset   = DIM*inr;
 730
 731         /* Load i particle coords and add shift vector */
 732         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 733                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 734
 735         fix0             = _mm_setzero_pd();
 736         fiy0             = _mm_setzero_pd();
 737         fiz0             = _mm_setzero_pd();
 738         fix1             = _mm_setzero_pd();
 739         fiy1             = _mm_setzero_pd();
 740         fiz1             = _mm_setzero_pd();
 741         fix2             = _mm_setzero_pd();
 742         fiy2             = _mm_setzero_pd();
 743         fiz2             = _mm_setzero_pd();
 744
 745         /* Start inner kernel loop */
 746         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 747         {
 748
 749             /* Get j neighbor index, and coordinate index */
 750             jnrA             = jjnr[jidx];
 751             jnrB             = jjnr[jidx+1];
 752             j_coord_offsetA  = DIM*jnrA;
 753             j_coord_offsetB  = DIM*jnrB;
 754
 755             /* load j atom coordinates */
 756             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 757                                               &jx0,&jy0,&jz0);
 758
 759             /* Calculate displacement vector */
 760             dx00             = _mm_sub_pd(ix0,jx0);
 761             dy00             = _mm_sub_pd(iy0,jy0);
 762             dz00             = _mm_sub_pd(iz0,jz0);
 763             dx10             = _mm_sub_pd(ix1,jx0);
 764             dy10             = _mm_sub_pd(iy1,jy0);
 765             dz10             = _mm_sub_pd(iz1,jz0);
 766             dx20             = _mm_sub_pd(ix2,jx0);
 767             dy20             = _mm_sub_pd(iy2,jy0);
 768             dz20             = _mm_sub_pd(iz2,jz0);
 769
 770             /* Calculate squared distance and things based on it */
 771             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 772             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 773             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 774
 775             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 776             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 777             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 778
 779             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 780             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 781             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 782
 783             /* Load parameters for j particles */
 784             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 785             vdwjidx0A        = 2*vdwtype[jnrA+0];
 786             vdwjidx0B        = 2*vdwtype[jnrB+0];
 787
 788             fjx0             = _mm_setzero_pd();
 789             fjy0             = _mm_setzero_pd();
 790             fjz0             = _mm_setzero_pd();
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             if (gmx_mm_any_lt(rsq00,rcutoff2))
 797             {
 798
 799             r00              = _mm_mul_pd(rsq00,rinv00);
 800
 801             /* Compute parameters for interactions between i and j atoms */
 802             qq00             = _mm_mul_pd(iq0,jq0);
 803             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 804                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 805
 806             /* REACTION-FIELD ELECTROSTATICS */
 807             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 808
 809             /* LENNARD-JONES DISPERSION/REPULSION */
 810
 811             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 812             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 813             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 814             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 815             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 816
 817             d                = _mm_sub_pd(r00,rswitch);
 818             d                = _mm_max_pd(d,_mm_setzero_pd());
 819             d2               = _mm_mul_pd(d,d);
 820             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)))))));
 821
 822             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 823
 824             /* Evaluate switch function */
 825             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 826             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 827             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 828
 829             fscal            = _mm_add_pd(felec,fvdw);
 830
 831             fscal            = _mm_and_pd(fscal,cutoff_mask);
 832
 833             /* Calculate temporary vectorial force */
 834             tx               = _mm_mul_pd(fscal,dx00);
 835             ty               = _mm_mul_pd(fscal,dy00);
 836             tz               = _mm_mul_pd(fscal,dz00);
 837
 838             /* Update vectorial force */
 839             fix0             = _mm_add_pd(fix0,tx);
 840             fiy0             = _mm_add_pd(fiy0,ty);
 841             fiz0             = _mm_add_pd(fiz0,tz);
 842
 843             fjx0             = _mm_add_pd(fjx0,tx);
 844             fjy0             = _mm_add_pd(fjy0,ty);
 845             fjz0             = _mm_add_pd(fjz0,tz);
 846
 847             }
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             if (gmx_mm_any_lt(rsq10,rcutoff2))
 854             {
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             qq10             = _mm_mul_pd(iq1,jq0);
 858
 859             /* REACTION-FIELD ELECTROSTATICS */
 860             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 861
 862             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 863
 864             fscal            = felec;
 865
 866             fscal            = _mm_and_pd(fscal,cutoff_mask);
 867
 868             /* Calculate temporary vectorial force */
 869             tx               = _mm_mul_pd(fscal,dx10);
 870             ty               = _mm_mul_pd(fscal,dy10);
 871             tz               = _mm_mul_pd(fscal,dz10);
 872
 873             /* Update vectorial force */
 874             fix1             = _mm_add_pd(fix1,tx);
 875             fiy1             = _mm_add_pd(fiy1,ty);
 876             fiz1             = _mm_add_pd(fiz1,tz);
 877
 878             fjx0             = _mm_add_pd(fjx0,tx);
 879             fjy0             = _mm_add_pd(fjy0,ty);
 880             fjz0             = _mm_add_pd(fjz0,tz);
 881
 882             }
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             if (gmx_mm_any_lt(rsq20,rcutoff2))
 889             {
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq20             = _mm_mul_pd(iq2,jq0);
 893
 894             /* REACTION-FIELD ELECTROSTATICS */
 895             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 896
 897             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 898
 899             fscal            = felec;
 900
 901             fscal            = _mm_and_pd(fscal,cutoff_mask);
 902
 903             /* Calculate temporary vectorial force */
 904             tx               = _mm_mul_pd(fscal,dx20);
 905             ty               = _mm_mul_pd(fscal,dy20);
 906             tz               = _mm_mul_pd(fscal,dz20);
 907
 908             /* Update vectorial force */
 909             fix2             = _mm_add_pd(fix2,tx);
 910             fiy2             = _mm_add_pd(fiy2,ty);
 911             fiz2             = _mm_add_pd(fiz2,tz);
 912
 913             fjx0             = _mm_add_pd(fjx0,tx);
 914             fjy0             = _mm_add_pd(fjy0,ty);
 915             fjz0             = _mm_add_pd(fjz0,tz);
 916
 917             }
 918
 919             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 920
 921             /* Inner loop uses 124 flops */
 922         }
 923
 924         if(jidx<j_index_end)
 925         {
 926
 927             jnrA             = jjnr[jidx];
 928             j_coord_offsetA  = DIM*jnrA;
 929
 930             /* load j atom coordinates */
 931             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 932                                               &jx0,&jy0,&jz0);
 933
 934             /* Calculate displacement vector */
 935             dx00             = _mm_sub_pd(ix0,jx0);
 936             dy00             = _mm_sub_pd(iy0,jy0);
 937             dz00             = _mm_sub_pd(iz0,jz0);
 938             dx10             = _mm_sub_pd(ix1,jx0);
 939             dy10             = _mm_sub_pd(iy1,jy0);
 940             dz10             = _mm_sub_pd(iz1,jz0);
 941             dx20             = _mm_sub_pd(ix2,jx0);
 942             dy20             = _mm_sub_pd(iy2,jy0);
 943             dz20             = _mm_sub_pd(iz2,jz0);
 944
 945             /* Calculate squared distance and things based on it */
 946             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 947             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 948             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 949
 950             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 951             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 952             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 953
 954             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 955             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 956             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 957
 958             /* Load parameters for j particles */
 959             jq0              = _mm_load_sd(charge+jnrA+0);
 960             vdwjidx0A        = 2*vdwtype[jnrA+0];
 961
 962             fjx0             = _mm_setzero_pd();
 963             fjy0             = _mm_setzero_pd();
 964             fjz0             = _mm_setzero_pd();
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             if (gmx_mm_any_lt(rsq00,rcutoff2))
 971             {
 972
 973             r00              = _mm_mul_pd(rsq00,rinv00);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq00             = _mm_mul_pd(iq0,jq0);
 977             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 978
 979             /* REACTION-FIELD ELECTROSTATICS */
 980             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 981
 982             /* LENNARD-JONES DISPERSION/REPULSION */
 983
 984             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 985             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 986             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 987             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 988             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 989
 990             d                = _mm_sub_pd(r00,rswitch);
 991             d                = _mm_max_pd(d,_mm_setzero_pd());
 992             d2               = _mm_mul_pd(d,d);
 993             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)))))));
 994
 995             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 996
 997             /* Evaluate switch function */
 998             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 999             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1000             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1001
1002             fscal            = _mm_add_pd(felec,fvdw);
1003
1004             fscal            = _mm_and_pd(fscal,cutoff_mask);
1005
1006             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm_mul_pd(fscal,dx00);
1010             ty               = _mm_mul_pd(fscal,dy00);
1011             tz               = _mm_mul_pd(fscal,dz00);
1012
1013             /* Update vectorial force */
1014             fix0             = _mm_add_pd(fix0,tx);
1015             fiy0             = _mm_add_pd(fiy0,ty);
1016             fiz0             = _mm_add_pd(fiz0,tz);
1017
1018             fjx0             = _mm_add_pd(fjx0,tx);
1019             fjy0             = _mm_add_pd(fjy0,ty);
1020             fjz0             = _mm_add_pd(fjz0,tz);
1021
1022             }
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             if (gmx_mm_any_lt(rsq10,rcutoff2))
1029             {
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq10             = _mm_mul_pd(iq1,jq0);
1033
1034             /* REACTION-FIELD ELECTROSTATICS */
1035             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1036
1037             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1038
1039             fscal            = felec;
1040
1041             fscal            = _mm_and_pd(fscal,cutoff_mask);
1042
1043             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm_mul_pd(fscal,dx10);
1047             ty               = _mm_mul_pd(fscal,dy10);
1048             tz               = _mm_mul_pd(fscal,dz10);
1049
1050             /* Update vectorial force */
1051             fix1             = _mm_add_pd(fix1,tx);
1052             fiy1             = _mm_add_pd(fiy1,ty);
1053             fiz1             = _mm_add_pd(fiz1,tz);
1054
1055             fjx0             = _mm_add_pd(fjx0,tx);
1056             fjy0             = _mm_add_pd(fjy0,ty);
1057             fjz0             = _mm_add_pd(fjz0,tz);
1058
1059             }
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             if (gmx_mm_any_lt(rsq20,rcutoff2))
1066             {
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq20             = _mm_mul_pd(iq2,jq0);
1070
1071             /* REACTION-FIELD ELECTROSTATICS */
1072             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1073
1074             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm_and_pd(fscal,cutoff_mask);
1079
1080             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm_mul_pd(fscal,dx20);
1084             ty               = _mm_mul_pd(fscal,dy20);
1085             tz               = _mm_mul_pd(fscal,dz20);
1086
1087             /* Update vectorial force */
1088             fix2             = _mm_add_pd(fix2,tx);
1089             fiy2             = _mm_add_pd(fiy2,ty);
1090             fiz2             = _mm_add_pd(fiz2,tz);
1091
1092             fjx0             = _mm_add_pd(fjx0,tx);
1093             fjy0             = _mm_add_pd(fjy0,ty);
1094             fjz0             = _mm_add_pd(fjz0,tz);
1095
1096             }
1097
1098             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1099
1100             /* Inner loop uses 124 flops */
1101         }
1102
1103         /* End of innermost loop */
1104
1105         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1106                                               f+i_coord_offset,fshift+i_shift_offset);
1107
1108         /* Increment number of inner iterations */
1109         inneriter                  += j_index_end - j_index_start;
1110
1111         /* Outer loop uses 18 flops */
1112     }
1113
1114     /* Increment number of outer iterations */
1115     outeriter        += nri;
1116
1117     /* Update outer/inner flops */
1118
1119     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*124);
1120 }