src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  99     real             rswitch_scalar,d_scalar;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->ic->epsfac);
 116     charge           = mdatoms->chargeA;
 117     krf              = _mm_set1_pd(fr->ic->k_rf);
 118     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 119     crf              = _mm_set1_pd(fr->ic->c_rf);
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 127     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 128     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 129     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 130
 131     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 132     rcutoff_scalar   = fr->ic->rcoulomb;
 133     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 134     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 135
 136     rswitch_scalar   = fr->ic->rvdw_switch;
 137     rswitch          = _mm_set1_pd(rswitch_scalar);
 138     /* Setup switch parameters */
 139     d_scalar         = rcutoff_scalar-rswitch_scalar;
 140     d                = _mm_set1_pd(d_scalar);
 141     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 142     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 143     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 144     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 145     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 146     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm_setzero_pd();
 175         fiy0             = _mm_setzero_pd();
 176         fiz0             = _mm_setzero_pd();
 177         fix1             = _mm_setzero_pd();
 178         fiy1             = _mm_setzero_pd();
 179         fiz1             = _mm_setzero_pd();
 180         fix2             = _mm_setzero_pd();
 181         fiy2             = _mm_setzero_pd();
 182         fiz2             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186         vvdwsum          = _mm_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_pd(ix0,jx0);
 204             dy00             = _mm_sub_pd(iy0,jy0);
 205             dz00             = _mm_sub_pd(iz0,jz0);
 206             dx10             = _mm_sub_pd(ix1,jx0);
 207             dy10             = _mm_sub_pd(iy1,jy0);
 208             dz10             = _mm_sub_pd(iz1,jz0);
 209             dx20             = _mm_sub_pd(ix2,jx0);
 210             dy20             = _mm_sub_pd(iy2,jy0);
 211             dz20             = _mm_sub_pd(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 217
 218             rinv00           = avx128fma_invsqrt_d(rsq00);
 219             rinv10           = avx128fma_invsqrt_d(rsq10);
 220             rinv20           = avx128fma_invsqrt_d(rsq20);
 221
 222             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 223             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             if (gmx_mm_any_lt(rsq00,rcutoff2))
 240             {
 241
 242             r00              = _mm_mul_pd(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             qq00             = _mm_mul_pd(iq0,jq0);
 246             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 247                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 248
 249             /* REACTION-FIELD ELECTROSTATICS */
 250             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 251             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 252
 253             /* LENNARD-JONES DISPERSION/REPULSION */
 254
 255             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 256             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 257             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 258             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 259             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 260
 261             d                = _mm_sub_pd(r00,rswitch);
 262             d                = _mm_max_pd(d,_mm_setzero_pd());
 263             d2               = _mm_mul_pd(d,d);
 264             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 265
 266             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 267
 268             /* Evaluate switch function */
 269             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 270             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 271             vvdw             = _mm_mul_pd(vvdw,sw);
 272             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velec            = _mm_and_pd(velec,cutoff_mask);
 276             velecsum         = _mm_add_pd(velecsum,velec);
 277             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 278             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 279
 280             fscal            = _mm_add_pd(felec,fvdw);
 281
 282             fscal            = _mm_and_pd(fscal,cutoff_mask);
 283
 284             /* Update vectorial force */
 285             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 286             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 287             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 288
 289             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 290             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 291             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 292
 293             }
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             if (gmx_mm_any_lt(rsq10,rcutoff2))
 300             {
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq10             = _mm_mul_pd(iq1,jq0);
 304
 305             /* REACTION-FIELD ELECTROSTATICS */
 306             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 307             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 308
 309             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velec            = _mm_and_pd(velec,cutoff_mask);
 313             velecsum         = _mm_add_pd(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             fscal            = _mm_and_pd(fscal,cutoff_mask);
 318
 319             /* Update vectorial force */
 320             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 321             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 322             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 323
 324             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 325             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 326             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 327
 328             }
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             if (gmx_mm_any_lt(rsq20,rcutoff2))
 335             {
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm_mul_pd(iq2,jq0);
 339
 340             /* REACTION-FIELD ELECTROSTATICS */
 341             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 342             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 343
 344             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _mm_and_pd(velec,cutoff_mask);
 348             velecsum         = _mm_add_pd(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             fscal            = _mm_and_pd(fscal,cutoff_mask);
 353
 354             /* Update vectorial force */
 355             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 356             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 357             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 358
 359             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 360             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 361             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 362
 363             }
 364
 365             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 366
 367             /* Inner loop uses 154 flops */
 368         }
 369
 370         if(jidx<j_index_end)
 371         {
 372
 373             jnrA             = jjnr[jidx];
 374             j_coord_offsetA  = DIM*jnrA;
 375
 376             /* load j atom coordinates */
 377             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 378                                               &jx0,&jy0,&jz0);
 379
 380             /* Calculate displacement vector */
 381             dx00             = _mm_sub_pd(ix0,jx0);
 382             dy00             = _mm_sub_pd(iy0,jy0);
 383             dz00             = _mm_sub_pd(iz0,jz0);
 384             dx10             = _mm_sub_pd(ix1,jx0);
 385             dy10             = _mm_sub_pd(iy1,jy0);
 386             dz10             = _mm_sub_pd(iz1,jz0);
 387             dx20             = _mm_sub_pd(ix2,jx0);
 388             dy20             = _mm_sub_pd(iy2,jy0);
 389             dz20             = _mm_sub_pd(iz2,jz0);
 390
 391             /* Calculate squared distance and things based on it */
 392             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 393             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 394             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 395
 396             rinv00           = avx128fma_invsqrt_d(rsq00);
 397             rinv10           = avx128fma_invsqrt_d(rsq10);
 398             rinv20           = avx128fma_invsqrt_d(rsq20);
 399
 400             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 401             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 402             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 403
 404             /* Load parameters for j particles */
 405             jq0              = _mm_load_sd(charge+jnrA+0);
 406             vdwjidx0A        = 2*vdwtype[jnrA+0];
 407
 408             fjx0             = _mm_setzero_pd();
 409             fjy0             = _mm_setzero_pd();
 410             fjz0             = _mm_setzero_pd();
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             if (gmx_mm_any_lt(rsq00,rcutoff2))
 417             {
 418
 419             r00              = _mm_mul_pd(rsq00,rinv00);
 420
 421             /* Compute parameters for interactions between i and j atoms */
 422             qq00             = _mm_mul_pd(iq0,jq0);
 423             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 424
 425             /* REACTION-FIELD ELECTROSTATICS */
 426             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 427             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 428
 429             /* LENNARD-JONES DISPERSION/REPULSION */
 430
 431             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 432             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 433             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 434             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 435             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 436
 437             d                = _mm_sub_pd(r00,rswitch);
 438             d                = _mm_max_pd(d,_mm_setzero_pd());
 439             d2               = _mm_mul_pd(d,d);
 440             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 441
 442             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 443
 444             /* Evaluate switch function */
 445             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 446             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 447             vvdw             = _mm_mul_pd(vvdw,sw);
 448             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velec            = _mm_and_pd(velec,cutoff_mask);
 452             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 453             velecsum         = _mm_add_pd(velecsum,velec);
 454             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 455             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 456             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 457
 458             fscal            = _mm_add_pd(felec,fvdw);
 459
 460             fscal            = _mm_and_pd(fscal,cutoff_mask);
 461
 462             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 463
 464             /* Update vectorial force */
 465             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 466             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 467             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 468
 469             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 470             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 471             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 472
 473             }
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             if (gmx_mm_any_lt(rsq10,rcutoff2))
 480             {
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq10             = _mm_mul_pd(iq1,jq0);
 484
 485             /* REACTION-FIELD ELECTROSTATICS */
 486             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 487             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 488
 489             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 490
 491             /* Update potential sum for this i atom from the interaction with this j atom. */
 492             velec            = _mm_and_pd(velec,cutoff_mask);
 493             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 494             velecsum         = _mm_add_pd(velecsum,velec);
 495
 496             fscal            = felec;
 497
 498             fscal            = _mm_and_pd(fscal,cutoff_mask);
 499
 500             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 501
 502             /* Update vectorial force */
 503             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 504             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 505             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 506
 507             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 508             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 509             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 510
 511             }
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             if (gmx_mm_any_lt(rsq20,rcutoff2))
 518             {
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq20             = _mm_mul_pd(iq2,jq0);
 522
 523             /* REACTION-FIELD ELECTROSTATICS */
 524             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 525             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 526
 527             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm_and_pd(velec,cutoff_mask);
 531             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 532             velecsum         = _mm_add_pd(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm_and_pd(fscal,cutoff_mask);
 537
 538             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 539
 540             /* Update vectorial force */
 541             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 542             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 543             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 544
 545             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 546             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 547             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 548
 549             }
 550
 551             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 552
 553             /* Inner loop uses 154 flops */
 554         }
 555
 556         /* End of innermost loop */
 557
 558         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 559                                               f+i_coord_offset,fshift+i_shift_offset);
 560
 561         ggid                        = gid[iidx];
 562         /* Update potential energies */
 563         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 564         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 565
 566         /* Increment number of inner iterations */
 567         inneriter                  += j_index_end - j_index_start;
 568
 569         /* Outer loop uses 20 flops */
 570     }
 571
 572     /* Increment number of outer iterations */
 573     outeriter        += nri;
 574
 575     /* Update outer/inner flops */
 576
 577     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 578 }
 579 /*
 580  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 581  * Electrostatics interaction: ReactionField
 582  * VdW interaction:            LennardJones
 583  * Geometry:                   Water3-Particle
 584  * Calculate force/pot:        Force
 585  */
 586 void
 587 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 588                     (t_nblist                    * gmx_restrict       nlist,
 589                      rvec                        * gmx_restrict          xx,
 590                      rvec                        * gmx_restrict          ff,
 591                      struct t_forcerec           * gmx_restrict          fr,
 592                      t_mdatoms                   * gmx_restrict     mdatoms,
 593                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 594                      t_nrnb                      * gmx_restrict        nrnb)
 595 {
 596     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 597      * just 0 for non-waters.
 598      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 599      * jnr indices corresponding to data put in the four positions in the SIMD register.
 600      */
 601     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 602     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 603     int              jnrA,jnrB;
 604     int              j_coord_offsetA,j_coord_offsetB;
 605     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 606     real             rcutoff_scalar;
 607     real             *shiftvec,*fshift,*x,*f;
 608     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 609     int              vdwioffset0;
 610     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 611     int              vdwioffset1;
 612     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 613     int              vdwioffset2;
 614     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 615     int              vdwjidx0A,vdwjidx0B;
 616     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 617     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 618     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 619     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 620     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 621     real             *charge;
 622     int              nvdwtype;
 623     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 624     int              *vdwtype;
 625     real             *vdwparam;
 626     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 627     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 628     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 629     real             rswitch_scalar,d_scalar;
 630     __m128d          dummy_mask,cutoff_mask;
 631     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 632     __m128d          one     = _mm_set1_pd(1.0);
 633     __m128d          two     = _mm_set1_pd(2.0);
 634     x                = xx[0];
 635     f                = ff[0];
 636
 637     nri              = nlist->nri;
 638     iinr             = nlist->iinr;
 639     jindex           = nlist->jindex;
 640     jjnr             = nlist->jjnr;
 641     shiftidx         = nlist->shift;
 642     gid              = nlist->gid;
 643     shiftvec         = fr->shift_vec[0];
 644     fshift           = fr->fshift[0];
 645     facel            = _mm_set1_pd(fr->ic->epsfac);
 646     charge           = mdatoms->chargeA;
 647     krf              = _mm_set1_pd(fr->ic->k_rf);
 648     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 649     crf              = _mm_set1_pd(fr->ic->c_rf);
 650     nvdwtype         = fr->ntype;
 651     vdwparam         = fr->nbfp;
 652     vdwtype          = mdatoms->typeA;
 653
 654     /* Setup water-specific parameters */
 655     inr              = nlist->iinr[0];
 656     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 657     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 658     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 659     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 660
 661     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 662     rcutoff_scalar   = fr->ic->rcoulomb;
 663     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 664     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 665
 666     rswitch_scalar   = fr->ic->rvdw_switch;
 667     rswitch          = _mm_set1_pd(rswitch_scalar);
 668     /* Setup switch parameters */
 669     d_scalar         = rcutoff_scalar-rswitch_scalar;
 670     d                = _mm_set1_pd(d_scalar);
 671     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 672     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 673     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 674     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 675     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 676     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 677
 678     /* Avoid stupid compiler warnings */
 679     jnrA = jnrB = 0;
 680     j_coord_offsetA = 0;
 681     j_coord_offsetB = 0;
 682
 683     outeriter        = 0;
 684     inneriter        = 0;
 685
 686     /* Start outer loop over neighborlists */
 687     for(iidx=0; iidx<nri; iidx++)
 688     {
 689         /* Load shift vector for this list */
 690         i_shift_offset   = DIM*shiftidx[iidx];
 691
 692         /* Load limits for loop over neighbors */
 693         j_index_start    = jindex[iidx];
 694         j_index_end      = jindex[iidx+1];
 695
 696         /* Get outer coordinate index */
 697         inr              = iinr[iidx];
 698         i_coord_offset   = DIM*inr;
 699
 700         /* Load i particle coords and add shift vector */
 701         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 702                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 703
 704         fix0             = _mm_setzero_pd();
 705         fiy0             = _mm_setzero_pd();
 706         fiz0             = _mm_setzero_pd();
 707         fix1             = _mm_setzero_pd();
 708         fiy1             = _mm_setzero_pd();
 709         fiz1             = _mm_setzero_pd();
 710         fix2             = _mm_setzero_pd();
 711         fiy2             = _mm_setzero_pd();
 712         fiz2             = _mm_setzero_pd();
 713
 714         /* Start inner kernel loop */
 715         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 716         {
 717
 718             /* Get j neighbor index, and coordinate index */
 719             jnrA             = jjnr[jidx];
 720             jnrB             = jjnr[jidx+1];
 721             j_coord_offsetA  = DIM*jnrA;
 722             j_coord_offsetB  = DIM*jnrB;
 723
 724             /* load j atom coordinates */
 725             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 726                                               &jx0,&jy0,&jz0);
 727
 728             /* Calculate displacement vector */
 729             dx00             = _mm_sub_pd(ix0,jx0);
 730             dy00             = _mm_sub_pd(iy0,jy0);
 731             dz00             = _mm_sub_pd(iz0,jz0);
 732             dx10             = _mm_sub_pd(ix1,jx0);
 733             dy10             = _mm_sub_pd(iy1,jy0);
 734             dz10             = _mm_sub_pd(iz1,jz0);
 735             dx20             = _mm_sub_pd(ix2,jx0);
 736             dy20             = _mm_sub_pd(iy2,jy0);
 737             dz20             = _mm_sub_pd(iz2,jz0);
 738
 739             /* Calculate squared distance and things based on it */
 740             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 741             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 742             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 743
 744             rinv00           = avx128fma_invsqrt_d(rsq00);
 745             rinv10           = avx128fma_invsqrt_d(rsq10);
 746             rinv20           = avx128fma_invsqrt_d(rsq20);
 747
 748             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 749             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 750             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 751
 752             /* Load parameters for j particles */
 753             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 754             vdwjidx0A        = 2*vdwtype[jnrA+0];
 755             vdwjidx0B        = 2*vdwtype[jnrB+0];
 756
 757             fjx0             = _mm_setzero_pd();
 758             fjy0             = _mm_setzero_pd();
 759             fjz0             = _mm_setzero_pd();
 760
 761             /**************************
 762              * CALCULATE INTERACTIONS *
 763              **************************/
 764
 765             if (gmx_mm_any_lt(rsq00,rcutoff2))
 766             {
 767
 768             r00              = _mm_mul_pd(rsq00,rinv00);
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq00             = _mm_mul_pd(iq0,jq0);
 772             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 773                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 774
 775             /* REACTION-FIELD ELECTROSTATICS */
 776             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 777
 778             /* LENNARD-JONES DISPERSION/REPULSION */
 779
 780             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 781             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 782             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 783             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 784             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 785
 786             d                = _mm_sub_pd(r00,rswitch);
 787             d                = _mm_max_pd(d,_mm_setzero_pd());
 788             d2               = _mm_mul_pd(d,d);
 789             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 790
 791             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 792
 793             /* Evaluate switch function */
 794             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 795             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 796             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 797
 798             fscal            = _mm_add_pd(felec,fvdw);
 799
 800             fscal            = _mm_and_pd(fscal,cutoff_mask);
 801
 802             /* Update vectorial force */
 803             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 804             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 805             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 806
 807             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 808             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 809             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 810
 811             }
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             if (gmx_mm_any_lt(rsq10,rcutoff2))
 818             {
 819
 820             /* Compute parameters for interactions between i and j atoms */
 821             qq10             = _mm_mul_pd(iq1,jq0);
 822
 823             /* REACTION-FIELD ELECTROSTATICS */
 824             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 825
 826             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 827
 828             fscal            = felec;
 829
 830             fscal            = _mm_and_pd(fscal,cutoff_mask);
 831
 832             /* Update vectorial force */
 833             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 834             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 835             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 836
 837             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 838             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 839             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 840
 841             }
 842
 843             /**************************
 844              * CALCULATE INTERACTIONS *
 845              **************************/
 846
 847             if (gmx_mm_any_lt(rsq20,rcutoff2))
 848             {
 849
 850             /* Compute parameters for interactions between i and j atoms */
 851             qq20             = _mm_mul_pd(iq2,jq0);
 852
 853             /* REACTION-FIELD ELECTROSTATICS */
 854             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 855
 856             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 857
 858             fscal            = felec;
 859
 860             fscal            = _mm_and_pd(fscal,cutoff_mask);
 861
 862             /* Update vectorial force */
 863             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 864             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 865             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 866
 867             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 868             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 869             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 870
 871             }
 872
 873             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 874
 875             /* Inner loop uses 133 flops */
 876         }
 877
 878         if(jidx<j_index_end)
 879         {
 880
 881             jnrA             = jjnr[jidx];
 882             j_coord_offsetA  = DIM*jnrA;
 883
 884             /* load j atom coordinates */
 885             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 886                                               &jx0,&jy0,&jz0);
 887
 888             /* Calculate displacement vector */
 889             dx00             = _mm_sub_pd(ix0,jx0);
 890             dy00             = _mm_sub_pd(iy0,jy0);
 891             dz00             = _mm_sub_pd(iz0,jz0);
 892             dx10             = _mm_sub_pd(ix1,jx0);
 893             dy10             = _mm_sub_pd(iy1,jy0);
 894             dz10             = _mm_sub_pd(iz1,jz0);
 895             dx20             = _mm_sub_pd(ix2,jx0);
 896             dy20             = _mm_sub_pd(iy2,jy0);
 897             dz20             = _mm_sub_pd(iz2,jz0);
 898
 899             /* Calculate squared distance and things based on it */
 900             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 901             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 902             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 903
 904             rinv00           = avx128fma_invsqrt_d(rsq00);
 905             rinv10           = avx128fma_invsqrt_d(rsq10);
 906             rinv20           = avx128fma_invsqrt_d(rsq20);
 907
 908             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 909             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 910             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 911
 912             /* Load parameters for j particles */
 913             jq0              = _mm_load_sd(charge+jnrA+0);
 914             vdwjidx0A        = 2*vdwtype[jnrA+0];
 915
 916             fjx0             = _mm_setzero_pd();
 917             fjy0             = _mm_setzero_pd();
 918             fjz0             = _mm_setzero_pd();
 919
 920             /**************************
 921              * CALCULATE INTERACTIONS *
 922              **************************/
 923
 924             if (gmx_mm_any_lt(rsq00,rcutoff2))
 925             {
 926
 927             r00              = _mm_mul_pd(rsq00,rinv00);
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             qq00             = _mm_mul_pd(iq0,jq0);
 931             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 932
 933             /* REACTION-FIELD ELECTROSTATICS */
 934             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 935
 936             /* LENNARD-JONES DISPERSION/REPULSION */
 937
 938             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 939             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 940             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 941             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 942             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 943
 944             d                = _mm_sub_pd(r00,rswitch);
 945             d                = _mm_max_pd(d,_mm_setzero_pd());
 946             d2               = _mm_mul_pd(d,d);
 947             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 948
 949             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 950
 951             /* Evaluate switch function */
 952             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 953             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 954             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 955
 956             fscal            = _mm_add_pd(felec,fvdw);
 957
 958             fscal            = _mm_and_pd(fscal,cutoff_mask);
 959
 960             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 961
 962             /* Update vectorial force */
 963             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 964             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 965             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 966
 967             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 968             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 969             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 970
 971             }
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             if (gmx_mm_any_lt(rsq10,rcutoff2))
 978             {
 979
 980             /* Compute parameters for interactions between i and j atoms */
 981             qq10             = _mm_mul_pd(iq1,jq0);
 982
 983             /* REACTION-FIELD ELECTROSTATICS */
 984             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 985
 986             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 987
 988             fscal            = felec;
 989
 990             fscal            = _mm_and_pd(fscal,cutoff_mask);
 991
 992             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 993
 994             /* Update vectorial force */
 995             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 996             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 997             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 998
 999             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1000             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1001             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1002
1003             }
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             if (gmx_mm_any_lt(rsq20,rcutoff2))
1010             {
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             qq20             = _mm_mul_pd(iq2,jq0);
1014
1015             /* REACTION-FIELD ELECTROSTATICS */
1016             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1017
1018             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1019
1020             fscal            = felec;
1021
1022             fscal            = _mm_and_pd(fscal,cutoff_mask);
1023
1024             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1025
1026             /* Update vectorial force */
1027             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1028             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1029             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1030
1031             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1032             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1033             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1034
1035             }
1036
1037             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1038
1039             /* Inner loop uses 133 flops */
1040         }
1041
1042         /* End of innermost loop */
1043
1044         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1045                                               f+i_coord_offset,fshift+i_shift_offset);
1046
1047         /* Increment number of inner iterations */
1048         inneriter                  += j_index_end - j_index_start;
1049
1050         /* Outer loop uses 18 flops */
1051     }
1052
1053     /* Increment number of outer iterations */
1054     outeriter        += nri;
1055
1056     /* Update outer/inner flops */
1057
1058     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*133);
1059 }