src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_128_fma_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 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 102     real             rswitch_scalar,d_scalar;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     krf              = _mm_set1_pd(fr->ic->k_rf);
 121     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 122     crf              = _mm_set1_pd(fr->ic->c_rf);
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->rcoulomb;
 136     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 137     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 138
 139     rswitch_scalar   = fr->rvdw_switch;
 140     rswitch          = _mm_set1_pd(rswitch_scalar);
 141     /* Setup switch parameters */
 142     d_scalar         = rcutoff_scalar-rswitch_scalar;
 143     d                = _mm_set1_pd(d_scalar);
 144     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 145     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 146     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 147     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 148     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155
 156     outeriter        = 0;
 157     inneriter        = 0;
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm_setzero_pd();
 178         fiy0             = _mm_setzero_pd();
 179         fiz0             = _mm_setzero_pd();
 180         fix1             = _mm_setzero_pd();
 181         fiy1             = _mm_setzero_pd();
 182         fiz1             = _mm_setzero_pd();
 183         fix2             = _mm_setzero_pd();
 184         fiy2             = _mm_setzero_pd();
 185         fiz2             = _mm_setzero_pd();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm_setzero_pd();
 189         vvdwsum          = _mm_setzero_pd();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200
 201             /* load j atom coordinates */
 202             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_pd(ix0,jx0);
 207             dy00             = _mm_sub_pd(iy0,jy0);
 208             dz00             = _mm_sub_pd(iz0,jz0);
 209             dx10             = _mm_sub_pd(ix1,jx0);
 210             dy10             = _mm_sub_pd(iy1,jy0);
 211             dz10             = _mm_sub_pd(iz1,jz0);
 212             dx20             = _mm_sub_pd(ix2,jx0);
 213             dy20             = _mm_sub_pd(iy2,jy0);
 214             dz20             = _mm_sub_pd(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 222             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 223             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 224
 225             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 226             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 227             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 228
 229             /* Load parameters for j particles */
 230             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233
 234             fjx0             = _mm_setzero_pd();
 235             fjy0             = _mm_setzero_pd();
 236             fjz0             = _mm_setzero_pd();
 237
 238             /**************************
 239              * CALCULATE INTERACTIONS *
 240              **************************/
 241
 242             if (gmx_mm_any_lt(rsq00,rcutoff2))
 243             {
 244
 245             r00              = _mm_mul_pd(rsq00,rinv00);
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             qq00             = _mm_mul_pd(iq0,jq0);
 249             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 251
 252             /* REACTION-FIELD ELECTROSTATICS */
 253             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 254             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 260             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 261             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 262             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 263
 264             d                = _mm_sub_pd(r00,rswitch);
 265             d                = _mm_max_pd(d,_mm_setzero_pd());
 266             d2               = _mm_mul_pd(d,d);
 267             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 268
 269             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 270
 271             /* Evaluate switch function */
 272             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 273             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 274             vvdw             = _mm_mul_pd(vvdw,sw);
 275             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velec            = _mm_and_pd(velec,cutoff_mask);
 279             velecsum         = _mm_add_pd(velecsum,velec);
 280             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 281             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 282
 283             fscal            = _mm_add_pd(felec,fvdw);
 284
 285             fscal            = _mm_and_pd(fscal,cutoff_mask);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 289             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 290             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 291
 292             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 293             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 294             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 295
 296             }
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             if (gmx_mm_any_lt(rsq10,rcutoff2))
 303             {
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_pd(iq1,jq0);
 307
 308             /* REACTION-FIELD ELECTROSTATICS */
 309             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 310             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 311
 312             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm_and_pd(velec,cutoff_mask);
 316             velecsum         = _mm_add_pd(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             fscal            = _mm_and_pd(fscal,cutoff_mask);
 321
 322             /* Update vectorial force */
 323             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 324             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 325             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 326
 327             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 328             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 329             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 330
 331             }
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             if (gmx_mm_any_lt(rsq20,rcutoff2))
 338             {
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq20             = _mm_mul_pd(iq2,jq0);
 342
 343             /* REACTION-FIELD ELECTROSTATICS */
 344             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 345             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 346
 347             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm_and_pd(velec,cutoff_mask);
 351             velecsum         = _mm_add_pd(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm_and_pd(fscal,cutoff_mask);
 356
 357             /* Update vectorial force */
 358             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 359             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 360             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 361
 362             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 363             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 364             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 365
 366             }
 367
 368             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 369
 370             /* Inner loop uses 154 flops */
 371         }
 372
 373         if(jidx<j_index_end)
 374         {
 375
 376             jnrA             = jjnr[jidx];
 377             j_coord_offsetA  = DIM*jnrA;
 378
 379             /* load j atom coordinates */
 380             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 381                                               &jx0,&jy0,&jz0);
 382
 383             /* Calculate displacement vector */
 384             dx00             = _mm_sub_pd(ix0,jx0);
 385             dy00             = _mm_sub_pd(iy0,jy0);
 386             dz00             = _mm_sub_pd(iz0,jz0);
 387             dx10             = _mm_sub_pd(ix1,jx0);
 388             dy10             = _mm_sub_pd(iy1,jy0);
 389             dz10             = _mm_sub_pd(iz1,jz0);
 390             dx20             = _mm_sub_pd(ix2,jx0);
 391             dy20             = _mm_sub_pd(iy2,jy0);
 392             dz20             = _mm_sub_pd(iz2,jz0);
 393
 394             /* Calculate squared distance and things based on it */
 395             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 396             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 397             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 398
 399             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 400             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 401             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 402
 403             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 404             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 405             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 406
 407             /* Load parameters for j particles */
 408             jq0              = _mm_load_sd(charge+jnrA+0);
 409             vdwjidx0A        = 2*vdwtype[jnrA+0];
 410
 411             fjx0             = _mm_setzero_pd();
 412             fjy0             = _mm_setzero_pd();
 413             fjz0             = _mm_setzero_pd();
 414
 415             /**************************
 416              * CALCULATE INTERACTIONS *
 417              **************************/
 418
 419             if (gmx_mm_any_lt(rsq00,rcutoff2))
 420             {
 421
 422             r00              = _mm_mul_pd(rsq00,rinv00);
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq00             = _mm_mul_pd(iq0,jq0);
 426             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 427
 428             /* REACTION-FIELD ELECTROSTATICS */
 429             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 430             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 431
 432             /* LENNARD-JONES DISPERSION/REPULSION */
 433
 434             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 435             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 436             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 437             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 438             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 439
 440             d                = _mm_sub_pd(r00,rswitch);
 441             d                = _mm_max_pd(d,_mm_setzero_pd());
 442             d2               = _mm_mul_pd(d,d);
 443             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 444
 445             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 446
 447             /* Evaluate switch function */
 448             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 449             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 450             vvdw             = _mm_mul_pd(vvdw,sw);
 451             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velec            = _mm_and_pd(velec,cutoff_mask);
 455             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 456             velecsum         = _mm_add_pd(velecsum,velec);
 457             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 458             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 459             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 460
 461             fscal            = _mm_add_pd(felec,fvdw);
 462
 463             fscal            = _mm_and_pd(fscal,cutoff_mask);
 464
 465             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 466
 467             /* Update vectorial force */
 468             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 469             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 470             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 471
 472             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 473             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 474             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 475
 476             }
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm_any_lt(rsq10,rcutoff2))
 483             {
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             qq10             = _mm_mul_pd(iq1,jq0);
 487
 488             /* REACTION-FIELD ELECTROSTATICS */
 489             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 490             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 491
 492             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm_and_pd(velec,cutoff_mask);
 496             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 497             velecsum         = _mm_add_pd(velecsum,velec);
 498
 499             fscal            = felec;
 500
 501             fscal            = _mm_and_pd(fscal,cutoff_mask);
 502
 503             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 504
 505             /* Update vectorial force */
 506             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 507             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 508             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 509
 510             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 511             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 512             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 513
 514             }
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             if (gmx_mm_any_lt(rsq20,rcutoff2))
 521             {
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq20             = _mm_mul_pd(iq2,jq0);
 525
 526             /* REACTION-FIELD ELECTROSTATICS */
 527             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 528             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 529
 530             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm_and_pd(velec,cutoff_mask);
 534             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 535             velecsum         = _mm_add_pd(velecsum,velec);
 536
 537             fscal            = felec;
 538
 539             fscal            = _mm_and_pd(fscal,cutoff_mask);
 540
 541             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 542
 543             /* Update vectorial force */
 544             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 545             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 546             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 547
 548             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 549             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 550             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 551
 552             }
 553
 554             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 555
 556             /* Inner loop uses 154 flops */
 557         }
 558
 559         /* End of innermost loop */
 560
 561         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 562                                               f+i_coord_offset,fshift+i_shift_offset);
 563
 564         ggid                        = gid[iidx];
 565         /* Update potential energies */
 566         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 567         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 568
 569         /* Increment number of inner iterations */
 570         inneriter                  += j_index_end - j_index_start;
 571
 572         /* Outer loop uses 20 flops */
 573     }
 574
 575     /* Increment number of outer iterations */
 576     outeriter        += nri;
 577
 578     /* Update outer/inner flops */
 579
 580     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 581 }
 582 /*
 583  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 584  * Electrostatics interaction: ReactionField
 585  * VdW interaction:            LennardJones
 586  * Geometry:                   Water3-Particle
 587  * Calculate force/pot:        Force
 588  */
 589 void
 590 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 591                     (t_nblist                    * gmx_restrict       nlist,
 592                      rvec                        * gmx_restrict          xx,
 593                      rvec                        * gmx_restrict          ff,
 594                      t_forcerec                  * gmx_restrict          fr,
 595                      t_mdatoms                   * gmx_restrict     mdatoms,
 596                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 597                      t_nrnb                      * gmx_restrict        nrnb)
 598 {
 599     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 600      * just 0 for non-waters.
 601      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 602      * jnr indices corresponding to data put in the four positions in the SIMD register.
 603      */
 604     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 605     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 606     int              jnrA,jnrB;
 607     int              j_coord_offsetA,j_coord_offsetB;
 608     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 609     real             rcutoff_scalar;
 610     real             *shiftvec,*fshift,*x,*f;
 611     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 612     int              vdwioffset0;
 613     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 614     int              vdwioffset1;
 615     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 616     int              vdwioffset2;
 617     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 618     int              vdwjidx0A,vdwjidx0B;
 619     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 620     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 621     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 622     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 623     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 624     real             *charge;
 625     int              nvdwtype;
 626     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 627     int              *vdwtype;
 628     real             *vdwparam;
 629     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 630     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 631     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 632     real             rswitch_scalar,d_scalar;
 633     __m128d          dummy_mask,cutoff_mask;
 634     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 635     __m128d          one     = _mm_set1_pd(1.0);
 636     __m128d          two     = _mm_set1_pd(2.0);
 637     x                = xx[0];
 638     f                = ff[0];
 639
 640     nri              = nlist->nri;
 641     iinr             = nlist->iinr;
 642     jindex           = nlist->jindex;
 643     jjnr             = nlist->jjnr;
 644     shiftidx         = nlist->shift;
 645     gid              = nlist->gid;
 646     shiftvec         = fr->shift_vec[0];
 647     fshift           = fr->fshift[0];
 648     facel            = _mm_set1_pd(fr->epsfac);
 649     charge           = mdatoms->chargeA;
 650     krf              = _mm_set1_pd(fr->ic->k_rf);
 651     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 652     crf              = _mm_set1_pd(fr->ic->c_rf);
 653     nvdwtype         = fr->ntype;
 654     vdwparam         = fr->nbfp;
 655     vdwtype          = mdatoms->typeA;
 656
 657     /* Setup water-specific parameters */
 658     inr              = nlist->iinr[0];
 659     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 660     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 661     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 662     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 663
 664     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 665     rcutoff_scalar   = fr->rcoulomb;
 666     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 667     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 668
 669     rswitch_scalar   = fr->rvdw_switch;
 670     rswitch          = _mm_set1_pd(rswitch_scalar);
 671     /* Setup switch parameters */
 672     d_scalar         = rcutoff_scalar-rswitch_scalar;
 673     d                = _mm_set1_pd(d_scalar);
 674     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 675     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 676     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 677     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 678     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 679     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 680
 681     /* Avoid stupid compiler warnings */
 682     jnrA = jnrB = 0;
 683     j_coord_offsetA = 0;
 684     j_coord_offsetB = 0;
 685
 686     outeriter        = 0;
 687     inneriter        = 0;
 688
 689     /* Start outer loop over neighborlists */
 690     for(iidx=0; iidx<nri; iidx++)
 691     {
 692         /* Load shift vector for this list */
 693         i_shift_offset   = DIM*shiftidx[iidx];
 694
 695         /* Load limits for loop over neighbors */
 696         j_index_start    = jindex[iidx];
 697         j_index_end      = jindex[iidx+1];
 698
 699         /* Get outer coordinate index */
 700         inr              = iinr[iidx];
 701         i_coord_offset   = DIM*inr;
 702
 703         /* Load i particle coords and add shift vector */
 704         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 705                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 706
 707         fix0             = _mm_setzero_pd();
 708         fiy0             = _mm_setzero_pd();
 709         fiz0             = _mm_setzero_pd();
 710         fix1             = _mm_setzero_pd();
 711         fiy1             = _mm_setzero_pd();
 712         fiz1             = _mm_setzero_pd();
 713         fix2             = _mm_setzero_pd();
 714         fiy2             = _mm_setzero_pd();
 715         fiz2             = _mm_setzero_pd();
 716
 717         /* Start inner kernel loop */
 718         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 719         {
 720
 721             /* Get j neighbor index, and coordinate index */
 722             jnrA             = jjnr[jidx];
 723             jnrB             = jjnr[jidx+1];
 724             j_coord_offsetA  = DIM*jnrA;
 725             j_coord_offsetB  = DIM*jnrB;
 726
 727             /* load j atom coordinates */
 728             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 729                                               &jx0,&jy0,&jz0);
 730
 731             /* Calculate displacement vector */
 732             dx00             = _mm_sub_pd(ix0,jx0);
 733             dy00             = _mm_sub_pd(iy0,jy0);
 734             dz00             = _mm_sub_pd(iz0,jz0);
 735             dx10             = _mm_sub_pd(ix1,jx0);
 736             dy10             = _mm_sub_pd(iy1,jy0);
 737             dz10             = _mm_sub_pd(iz1,jz0);
 738             dx20             = _mm_sub_pd(ix2,jx0);
 739             dy20             = _mm_sub_pd(iy2,jy0);
 740             dz20             = _mm_sub_pd(iz2,jz0);
 741
 742             /* Calculate squared distance and things based on it */
 743             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 744             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 745             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 746
 747             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 748             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 749             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 750
 751             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 752             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 753             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 754
 755             /* Load parameters for j particles */
 756             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 757             vdwjidx0A        = 2*vdwtype[jnrA+0];
 758             vdwjidx0B        = 2*vdwtype[jnrB+0];
 759
 760             fjx0             = _mm_setzero_pd();
 761             fjy0             = _mm_setzero_pd();
 762             fjz0             = _mm_setzero_pd();
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             if (gmx_mm_any_lt(rsq00,rcutoff2))
 769             {
 770
 771             r00              = _mm_mul_pd(rsq00,rinv00);
 772
 773             /* Compute parameters for interactions between i and j atoms */
 774             qq00             = _mm_mul_pd(iq0,jq0);
 775             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 776                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 777
 778             /* REACTION-FIELD ELECTROSTATICS */
 779             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 780
 781             /* LENNARD-JONES DISPERSION/REPULSION */
 782
 783             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 784             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 785             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 786             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 787             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 788
 789             d                = _mm_sub_pd(r00,rswitch);
 790             d                = _mm_max_pd(d,_mm_setzero_pd());
 791             d2               = _mm_mul_pd(d,d);
 792             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 793
 794             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 795
 796             /* Evaluate switch function */
 797             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 798             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 799             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 800
 801             fscal            = _mm_add_pd(felec,fvdw);
 802
 803             fscal            = _mm_and_pd(fscal,cutoff_mask);
 804
 805             /* Update vectorial force */
 806             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 807             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 808             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 809
 810             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 811             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 812             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 813
 814             }
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             if (gmx_mm_any_lt(rsq10,rcutoff2))
 821             {
 822
 823             /* Compute parameters for interactions between i and j atoms */
 824             qq10             = _mm_mul_pd(iq1,jq0);
 825
 826             /* REACTION-FIELD ELECTROSTATICS */
 827             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 828
 829             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 830
 831             fscal            = felec;
 832
 833             fscal            = _mm_and_pd(fscal,cutoff_mask);
 834
 835             /* Update vectorial force */
 836             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 837             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 838             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 839
 840             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 841             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 842             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 843
 844             }
 845
 846             /**************************
 847              * CALCULATE INTERACTIONS *
 848              **************************/
 849
 850             if (gmx_mm_any_lt(rsq20,rcutoff2))
 851             {
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq20             = _mm_mul_pd(iq2,jq0);
 855
 856             /* REACTION-FIELD ELECTROSTATICS */
 857             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 858
 859             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 860
 861             fscal            = felec;
 862
 863             fscal            = _mm_and_pd(fscal,cutoff_mask);
 864
 865             /* Update vectorial force */
 866             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 867             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 868             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 869
 870             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 871             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 872             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 873
 874             }
 875
 876             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 877
 878             /* Inner loop uses 133 flops */
 879         }
 880
 881         if(jidx<j_index_end)
 882         {
 883
 884             jnrA             = jjnr[jidx];
 885             j_coord_offsetA  = DIM*jnrA;
 886
 887             /* load j atom coordinates */
 888             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 889                                               &jx0,&jy0,&jz0);
 890
 891             /* Calculate displacement vector */
 892             dx00             = _mm_sub_pd(ix0,jx0);
 893             dy00             = _mm_sub_pd(iy0,jy0);
 894             dz00             = _mm_sub_pd(iz0,jz0);
 895             dx10             = _mm_sub_pd(ix1,jx0);
 896             dy10             = _mm_sub_pd(iy1,jy0);
 897             dz10             = _mm_sub_pd(iz1,jz0);
 898             dx20             = _mm_sub_pd(ix2,jx0);
 899             dy20             = _mm_sub_pd(iy2,jy0);
 900             dz20             = _mm_sub_pd(iz2,jz0);
 901
 902             /* Calculate squared distance and things based on it */
 903             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 904             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 905             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 906
 907             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 908             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 909             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 910
 911             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 912             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 913             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 914
 915             /* Load parameters for j particles */
 916             jq0              = _mm_load_sd(charge+jnrA+0);
 917             vdwjidx0A        = 2*vdwtype[jnrA+0];
 918
 919             fjx0             = _mm_setzero_pd();
 920             fjy0             = _mm_setzero_pd();
 921             fjz0             = _mm_setzero_pd();
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             if (gmx_mm_any_lt(rsq00,rcutoff2))
 928             {
 929
 930             r00              = _mm_mul_pd(rsq00,rinv00);
 931
 932             /* Compute parameters for interactions between i and j atoms */
 933             qq00             = _mm_mul_pd(iq0,jq0);
 934             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 935
 936             /* REACTION-FIELD ELECTROSTATICS */
 937             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 938
 939             /* LENNARD-JONES DISPERSION/REPULSION */
 940
 941             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 942             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 943             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 944             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 945             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 946
 947             d                = _mm_sub_pd(r00,rswitch);
 948             d                = _mm_max_pd(d,_mm_setzero_pd());
 949             d2               = _mm_mul_pd(d,d);
 950             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 951
 952             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 953
 954             /* Evaluate switch function */
 955             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 956             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 957             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 958
 959             fscal            = _mm_add_pd(felec,fvdw);
 960
 961             fscal            = _mm_and_pd(fscal,cutoff_mask);
 962
 963             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 964
 965             /* Update vectorial force */
 966             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 967             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 968             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 969
 970             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 971             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 972             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 973
 974             }
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             if (gmx_mm_any_lt(rsq10,rcutoff2))
 981             {
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq10             = _mm_mul_pd(iq1,jq0);
 985
 986             /* REACTION-FIELD ELECTROSTATICS */
 987             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 988
 989             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm_and_pd(fscal,cutoff_mask);
 994
 995             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 996
 997             /* Update vectorial force */
 998             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 999             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1000             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1001
1002             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1003             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1004             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1005
1006             }
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             if (gmx_mm_any_lt(rsq20,rcutoff2))
1013             {
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq20             = _mm_mul_pd(iq2,jq0);
1017
1018             /* REACTION-FIELD ELECTROSTATICS */
1019             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1020
1021             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1022
1023             fscal            = felec;
1024
1025             fscal            = _mm_and_pd(fscal,cutoff_mask);
1026
1027             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1028
1029             /* Update vectorial force */
1030             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1031             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1032             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1033
1034             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1035             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1036             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1037
1038             }
1039
1040             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1041
1042             /* Inner loop uses 133 flops */
1043         }
1044
1045         /* End of innermost loop */
1046
1047         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1048                                               f+i_coord_offset,fshift+i_shift_offset);
1049
1050         /* Increment number of inner iterations */
1051         inneriter                  += j_index_end - j_index_start;
1052
1053         /* Outer loop uses 18 flops */
1054     }
1055
1056     /* Increment number of outer iterations */
1057     outeriter        += nri;
1058
1059     /* Update outer/inner flops */
1060
1061     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*133);
1062 }