src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  86     real             rswitch_scalar,d_scalar;
  87     __m128d          dummy_mask,cutoff_mask;
  88     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  89     __m128d          one     = _mm_set1_pd(1.0);
  90     __m128d          two     = _mm_set1_pd(2.0);
  91     x                = xx[0];
  92     f                = ff[0];
  93
  94     nri              = nlist->nri;
  95     iinr             = nlist->iinr;
  96     jindex           = nlist->jindex;
  97     jjnr             = nlist->jjnr;
  98     shiftidx         = nlist->shift;
  99     gid              = nlist->gid;
 100     shiftvec         = fr->shift_vec[0];
 101     fshift           = fr->fshift[0];
 102     facel            = _mm_set1_pd(fr->epsfac);
 103     charge           = mdatoms->chargeA;
 104     krf              = _mm_set1_pd(fr->ic->k_rf);
 105     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 106     crf              = _mm_set1_pd(fr->ic->c_rf);
 107     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     /* Setup water-specific parameters */
 112     inr              = nlist->iinr[0];
 113     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 114     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 115     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 116     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 117
 118     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 119     rcutoff_scalar   = fr->rcoulomb;
 120     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 121     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 122
 123     rswitch_scalar   = fr->rvdw_switch;
 124     rswitch          = _mm_set1_pd(rswitch_scalar);
 125     /* Setup switch parameters */
 126     d_scalar         = rcutoff_scalar-rswitch_scalar;
 127     d                = _mm_set1_pd(d_scalar);
 128     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 129     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 130     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 131     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 132     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 133     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 160
 161         fix0             = _mm_setzero_pd();
 162         fiy0             = _mm_setzero_pd();
 163         fiz0             = _mm_setzero_pd();
 164         fix1             = _mm_setzero_pd();
 165         fiy1             = _mm_setzero_pd();
 166         fiz1             = _mm_setzero_pd();
 167         fix2             = _mm_setzero_pd();
 168         fiy2             = _mm_setzero_pd();
 169         fiz2             = _mm_setzero_pd();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_pd();
 173         vvdwsum          = _mm_setzero_pd();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_pd(ix0,jx0);
 191             dy00             = _mm_sub_pd(iy0,jy0);
 192             dz00             = _mm_sub_pd(iz0,jz0);
 193             dx10             = _mm_sub_pd(ix1,jx0);
 194             dy10             = _mm_sub_pd(iy1,jy0);
 195             dz10             = _mm_sub_pd(iz1,jz0);
 196             dx20             = _mm_sub_pd(ix2,jx0);
 197             dy20             = _mm_sub_pd(iy2,jy0);
 198             dz20             = _mm_sub_pd(iz2,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 204
 205             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 206             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 207             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 208
 209             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 210             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 211             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 215             vdwjidx0A        = 2*vdwtype[jnrA+0];
 216             vdwjidx0B        = 2*vdwtype[jnrB+0];
 217
 218             fjx0             = _mm_setzero_pd();
 219             fjy0             = _mm_setzero_pd();
 220             fjz0             = _mm_setzero_pd();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             if (gmx_mm_any_lt(rsq00,rcutoff2))
 227             {
 228
 229             r00              = _mm_mul_pd(rsq00,rinv00);
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm_mul_pd(iq0,jq0);
 233             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 235
 236             /* REACTION-FIELD ELECTROSTATICS */
 237             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 238             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 239
 240             /* LENNARD-JONES DISPERSION/REPULSION */
 241
 242             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 243             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 244             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 245             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 246             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 247
 248             d                = _mm_sub_pd(r00,rswitch);
 249             d                = _mm_max_pd(d,_mm_setzero_pd());
 250             d2               = _mm_mul_pd(d,d);
 251             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 252
 253             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 254
 255             /* Evaluate switch function */
 256             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 257             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 258             vvdw             = _mm_mul_pd(vvdw,sw);
 259             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velec            = _mm_and_pd(velec,cutoff_mask);
 263             velecsum         = _mm_add_pd(velecsum,velec);
 264             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 265             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_pd(felec,fvdw);
 268
 269             fscal            = _mm_and_pd(fscal,cutoff_mask);
 270
 271             /* Update vectorial force */
 272             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 273             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 274             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 275
 276             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 277             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 278             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 279
 280             }
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             if (gmx_mm_any_lt(rsq10,rcutoff2))
 287             {
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq10             = _mm_mul_pd(iq1,jq0);
 291
 292             /* REACTION-FIELD ELECTROSTATICS */
 293             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 294             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 295
 296             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_and_pd(velec,cutoff_mask);
 300             velecsum         = _mm_add_pd(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             fscal            = _mm_and_pd(fscal,cutoff_mask);
 305
 306             /* Update vectorial force */
 307             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 308             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 309             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 310
 311             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 312             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 313             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 314
 315             }
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             if (gmx_mm_any_lt(rsq20,rcutoff2))
 322             {
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq20             = _mm_mul_pd(iq2,jq0);
 326
 327             /* REACTION-FIELD ELECTROSTATICS */
 328             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 329             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 330
 331             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velec            = _mm_and_pd(velec,cutoff_mask);
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             fscal            = _mm_and_pd(fscal,cutoff_mask);
 340
 341             /* Update vectorial force */
 342             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 343             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 344             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 345
 346             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 347             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 348             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 349
 350             }
 351
 352             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 353
 354             /* Inner loop uses 154 flops */
 355         }
 356
 357         if(jidx<j_index_end)
 358         {
 359
 360             jnrA             = jjnr[jidx];
 361             j_coord_offsetA  = DIM*jnrA;
 362
 363             /* load j atom coordinates */
 364             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 365                                               &jx0,&jy0,&jz0);
 366
 367             /* Calculate displacement vector */
 368             dx00             = _mm_sub_pd(ix0,jx0);
 369             dy00             = _mm_sub_pd(iy0,jy0);
 370             dz00             = _mm_sub_pd(iz0,jz0);
 371             dx10             = _mm_sub_pd(ix1,jx0);
 372             dy10             = _mm_sub_pd(iy1,jy0);
 373             dz10             = _mm_sub_pd(iz1,jz0);
 374             dx20             = _mm_sub_pd(ix2,jx0);
 375             dy20             = _mm_sub_pd(iy2,jy0);
 376             dz20             = _mm_sub_pd(iz2,jz0);
 377
 378             /* Calculate squared distance and things based on it */
 379             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 380             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 381             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 382
 383             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 384             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 385             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 386
 387             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 388             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 389             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 390
 391             /* Load parameters for j particles */
 392             jq0              = _mm_load_sd(charge+jnrA+0);
 393             vdwjidx0A        = 2*vdwtype[jnrA+0];
 394
 395             fjx0             = _mm_setzero_pd();
 396             fjy0             = _mm_setzero_pd();
 397             fjz0             = _mm_setzero_pd();
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             if (gmx_mm_any_lt(rsq00,rcutoff2))
 404             {
 405
 406             r00              = _mm_mul_pd(rsq00,rinv00);
 407
 408             /* Compute parameters for interactions between i and j atoms */
 409             qq00             = _mm_mul_pd(iq0,jq0);
 410             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 411
 412             /* REACTION-FIELD ELECTROSTATICS */
 413             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 414             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 415
 416             /* LENNARD-JONES DISPERSION/REPULSION */
 417
 418             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 419             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 420             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 421             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 422             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 423
 424             d                = _mm_sub_pd(r00,rswitch);
 425             d                = _mm_max_pd(d,_mm_setzero_pd());
 426             d2               = _mm_mul_pd(d,d);
 427             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 428
 429             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 430
 431             /* Evaluate switch function */
 432             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 433             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 434             vvdw             = _mm_mul_pd(vvdw,sw);
 435             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velec            = _mm_and_pd(velec,cutoff_mask);
 439             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 440             velecsum         = _mm_add_pd(velecsum,velec);
 441             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 442             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 443             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 444
 445             fscal            = _mm_add_pd(felec,fvdw);
 446
 447             fscal            = _mm_and_pd(fscal,cutoff_mask);
 448
 449             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 450
 451             /* Update vectorial force */
 452             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 453             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 454             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 455
 456             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 457             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 458             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 459
 460             }
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             if (gmx_mm_any_lt(rsq10,rcutoff2))
 467             {
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq10             = _mm_mul_pd(iq1,jq0);
 471
 472             /* REACTION-FIELD ELECTROSTATICS */
 473             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 474             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 475
 476             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velec            = _mm_and_pd(velec,cutoff_mask);
 480             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 481             velecsum         = _mm_add_pd(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485             fscal            = _mm_and_pd(fscal,cutoff_mask);
 486
 487             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 488
 489             /* Update vectorial force */
 490             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 491             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 492             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 493
 494             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 495             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 496             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 497
 498             }
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             if (gmx_mm_any_lt(rsq20,rcutoff2))
 505             {
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             qq20             = _mm_mul_pd(iq2,jq0);
 509
 510             /* REACTION-FIELD ELECTROSTATICS */
 511             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 512             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 513
 514             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm_and_pd(velec,cutoff_mask);
 518             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 519             velecsum         = _mm_add_pd(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             fscal            = _mm_and_pd(fscal,cutoff_mask);
 524
 525             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 526
 527             /* Update vectorial force */
 528             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 529             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 530             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 531
 532             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 533             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 534             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 535
 536             }
 537
 538             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 539
 540             /* Inner loop uses 154 flops */
 541         }
 542
 543         /* End of innermost loop */
 544
 545         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 546                                               f+i_coord_offset,fshift+i_shift_offset);
 547
 548         ggid                        = gid[iidx];
 549         /* Update potential energies */
 550         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 551         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 552
 553         /* Increment number of inner iterations */
 554         inneriter                  += j_index_end - j_index_start;
 555
 556         /* Outer loop uses 20 flops */
 557     }
 558
 559     /* Increment number of outer iterations */
 560     outeriter        += nri;
 561
 562     /* Update outer/inner flops */
 563
 564     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 565 }
 566 /*
 567  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 568  * Electrostatics interaction: ReactionField
 569  * VdW interaction:            LennardJones
 570  * Geometry:                   Water3-Particle
 571  * Calculate force/pot:        Force
 572  */
 573 void
 574 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 575                     (t_nblist * gmx_restrict                nlist,
 576                      rvec * gmx_restrict                    xx,
 577                      rvec * gmx_restrict                    ff,
 578                      t_forcerec * gmx_restrict              fr,
 579                      t_mdatoms * gmx_restrict               mdatoms,
 580                      nb_kernel_data_t * gmx_restrict        kernel_data,
 581                      t_nrnb * gmx_restrict                  nrnb)
 582 {
 583     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 584      * just 0 for non-waters.
 585      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 586      * jnr indices corresponding to data put in the four positions in the SIMD register.
 587      */
 588     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 589     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 590     int              jnrA,jnrB;
 591     int              j_coord_offsetA,j_coord_offsetB;
 592     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 593     real             rcutoff_scalar;
 594     real             *shiftvec,*fshift,*x,*f;
 595     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 596     int              vdwioffset0;
 597     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 598     int              vdwioffset1;
 599     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 600     int              vdwioffset2;
 601     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 602     int              vdwjidx0A,vdwjidx0B;
 603     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 604     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 605     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 606     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 607     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 608     real             *charge;
 609     int              nvdwtype;
 610     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 611     int              *vdwtype;
 612     real             *vdwparam;
 613     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 614     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 615     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 616     real             rswitch_scalar,d_scalar;
 617     __m128d          dummy_mask,cutoff_mask;
 618     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 619     __m128d          one     = _mm_set1_pd(1.0);
 620     __m128d          two     = _mm_set1_pd(2.0);
 621     x                = xx[0];
 622     f                = ff[0];
 623
 624     nri              = nlist->nri;
 625     iinr             = nlist->iinr;
 626     jindex           = nlist->jindex;
 627     jjnr             = nlist->jjnr;
 628     shiftidx         = nlist->shift;
 629     gid              = nlist->gid;
 630     shiftvec         = fr->shift_vec[0];
 631     fshift           = fr->fshift[0];
 632     facel            = _mm_set1_pd(fr->epsfac);
 633     charge           = mdatoms->chargeA;
 634     krf              = _mm_set1_pd(fr->ic->k_rf);
 635     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 636     crf              = _mm_set1_pd(fr->ic->c_rf);
 637     nvdwtype         = fr->ntype;
 638     vdwparam         = fr->nbfp;
 639     vdwtype          = mdatoms->typeA;
 640
 641     /* Setup water-specific parameters */
 642     inr              = nlist->iinr[0];
 643     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 644     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 645     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 646     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 647
 648     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 649     rcutoff_scalar   = fr->rcoulomb;
 650     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 651     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 652
 653     rswitch_scalar   = fr->rvdw_switch;
 654     rswitch          = _mm_set1_pd(rswitch_scalar);
 655     /* Setup switch parameters */
 656     d_scalar         = rcutoff_scalar-rswitch_scalar;
 657     d                = _mm_set1_pd(d_scalar);
 658     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 659     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 660     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 661     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 662     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 663     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 664
 665     /* Avoid stupid compiler warnings */
 666     jnrA = jnrB = 0;
 667     j_coord_offsetA = 0;
 668     j_coord_offsetB = 0;
 669
 670     outeriter        = 0;
 671     inneriter        = 0;
 672
 673     /* Start outer loop over neighborlists */
 674     for(iidx=0; iidx<nri; iidx++)
 675     {
 676         /* Load shift vector for this list */
 677         i_shift_offset   = DIM*shiftidx[iidx];
 678
 679         /* Load limits for loop over neighbors */
 680         j_index_start    = jindex[iidx];
 681         j_index_end      = jindex[iidx+1];
 682
 683         /* Get outer coordinate index */
 684         inr              = iinr[iidx];
 685         i_coord_offset   = DIM*inr;
 686
 687         /* Load i particle coords and add shift vector */
 688         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 689                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 690
 691         fix0             = _mm_setzero_pd();
 692         fiy0             = _mm_setzero_pd();
 693         fiz0             = _mm_setzero_pd();
 694         fix1             = _mm_setzero_pd();
 695         fiy1             = _mm_setzero_pd();
 696         fiz1             = _mm_setzero_pd();
 697         fix2             = _mm_setzero_pd();
 698         fiy2             = _mm_setzero_pd();
 699         fiz2             = _mm_setzero_pd();
 700
 701         /* Start inner kernel loop */
 702         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 703         {
 704
 705             /* Get j neighbor index, and coordinate index */
 706             jnrA             = jjnr[jidx];
 707             jnrB             = jjnr[jidx+1];
 708             j_coord_offsetA  = DIM*jnrA;
 709             j_coord_offsetB  = DIM*jnrB;
 710
 711             /* load j atom coordinates */
 712             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 713                                               &jx0,&jy0,&jz0);
 714
 715             /* Calculate displacement vector */
 716             dx00             = _mm_sub_pd(ix0,jx0);
 717             dy00             = _mm_sub_pd(iy0,jy0);
 718             dz00             = _mm_sub_pd(iz0,jz0);
 719             dx10             = _mm_sub_pd(ix1,jx0);
 720             dy10             = _mm_sub_pd(iy1,jy0);
 721             dz10             = _mm_sub_pd(iz1,jz0);
 722             dx20             = _mm_sub_pd(ix2,jx0);
 723             dy20             = _mm_sub_pd(iy2,jy0);
 724             dz20             = _mm_sub_pd(iz2,jz0);
 725
 726             /* Calculate squared distance and things based on it */
 727             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 728             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 729             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 730
 731             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 732             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 733             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 734
 735             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 736             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 737             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 738
 739             /* Load parameters for j particles */
 740             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 741             vdwjidx0A        = 2*vdwtype[jnrA+0];
 742             vdwjidx0B        = 2*vdwtype[jnrB+0];
 743
 744             fjx0             = _mm_setzero_pd();
 745             fjy0             = _mm_setzero_pd();
 746             fjz0             = _mm_setzero_pd();
 747
 748             /**************************
 749              * CALCULATE INTERACTIONS *
 750              **************************/
 751
 752             if (gmx_mm_any_lt(rsq00,rcutoff2))
 753             {
 754
 755             r00              = _mm_mul_pd(rsq00,rinv00);
 756
 757             /* Compute parameters for interactions between i and j atoms */
 758             qq00             = _mm_mul_pd(iq0,jq0);
 759             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 760                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 761
 762             /* REACTION-FIELD ELECTROSTATICS */
 763             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 764
 765             /* LENNARD-JONES DISPERSION/REPULSION */
 766
 767             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 768             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 769             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 770             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 771             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 772
 773             d                = _mm_sub_pd(r00,rswitch);
 774             d                = _mm_max_pd(d,_mm_setzero_pd());
 775             d2               = _mm_mul_pd(d,d);
 776             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 777
 778             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 779
 780             /* Evaluate switch function */
 781             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 782             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 783             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 784
 785             fscal            = _mm_add_pd(felec,fvdw);
 786
 787             fscal            = _mm_and_pd(fscal,cutoff_mask);
 788
 789             /* Update vectorial force */
 790             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 791             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 792             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 793
 794             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 795             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 796             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 797
 798             }
 799
 800             /**************************
 801              * CALCULATE INTERACTIONS *
 802              **************************/
 803
 804             if (gmx_mm_any_lt(rsq10,rcutoff2))
 805             {
 806
 807             /* Compute parameters for interactions between i and j atoms */
 808             qq10             = _mm_mul_pd(iq1,jq0);
 809
 810             /* REACTION-FIELD ELECTROSTATICS */
 811             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 812
 813             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 814
 815             fscal            = felec;
 816
 817             fscal            = _mm_and_pd(fscal,cutoff_mask);
 818
 819             /* Update vectorial force */
 820             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 821             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 822             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 823
 824             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 825             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 826             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 827
 828             }
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             if (gmx_mm_any_lt(rsq20,rcutoff2))
 835             {
 836
 837             /* Compute parameters for interactions between i and j atoms */
 838             qq20             = _mm_mul_pd(iq2,jq0);
 839
 840             /* REACTION-FIELD ELECTROSTATICS */
 841             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 842
 843             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 844
 845             fscal            = felec;
 846
 847             fscal            = _mm_and_pd(fscal,cutoff_mask);
 848
 849             /* Update vectorial force */
 850             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 851             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 852             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 853
 854             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 855             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 856             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 857
 858             }
 859
 860             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 861
 862             /* Inner loop uses 133 flops */
 863         }
 864
 865         if(jidx<j_index_end)
 866         {
 867
 868             jnrA             = jjnr[jidx];
 869             j_coord_offsetA  = DIM*jnrA;
 870
 871             /* load j atom coordinates */
 872             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 873                                               &jx0,&jy0,&jz0);
 874
 875             /* Calculate displacement vector */
 876             dx00             = _mm_sub_pd(ix0,jx0);
 877             dy00             = _mm_sub_pd(iy0,jy0);
 878             dz00             = _mm_sub_pd(iz0,jz0);
 879             dx10             = _mm_sub_pd(ix1,jx0);
 880             dy10             = _mm_sub_pd(iy1,jy0);
 881             dz10             = _mm_sub_pd(iz1,jz0);
 882             dx20             = _mm_sub_pd(ix2,jx0);
 883             dy20             = _mm_sub_pd(iy2,jy0);
 884             dz20             = _mm_sub_pd(iz2,jz0);
 885
 886             /* Calculate squared distance and things based on it */
 887             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 888             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 889             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 890
 891             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 892             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 893             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 894
 895             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 896             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 897             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 898
 899             /* Load parameters for j particles */
 900             jq0              = _mm_load_sd(charge+jnrA+0);
 901             vdwjidx0A        = 2*vdwtype[jnrA+0];
 902
 903             fjx0             = _mm_setzero_pd();
 904             fjy0             = _mm_setzero_pd();
 905             fjz0             = _mm_setzero_pd();
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             if (gmx_mm_any_lt(rsq00,rcutoff2))
 912             {
 913
 914             r00              = _mm_mul_pd(rsq00,rinv00);
 915
 916             /* Compute parameters for interactions between i and j atoms */
 917             qq00             = _mm_mul_pd(iq0,jq0);
 918             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 919
 920             /* REACTION-FIELD ELECTROSTATICS */
 921             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 922
 923             /* LENNARD-JONES DISPERSION/REPULSION */
 924
 925             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 926             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 927             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 928             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 929             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 930
 931             d                = _mm_sub_pd(r00,rswitch);
 932             d                = _mm_max_pd(d,_mm_setzero_pd());
 933             d2               = _mm_mul_pd(d,d);
 934             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 935
 936             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 937
 938             /* Evaluate switch function */
 939             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 940             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 941             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 942
 943             fscal            = _mm_add_pd(felec,fvdw);
 944
 945             fscal            = _mm_and_pd(fscal,cutoff_mask);
 946
 947             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 948
 949             /* Update vectorial force */
 950             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 951             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 952             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 953
 954             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 955             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 956             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 957
 958             }
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             if (gmx_mm_any_lt(rsq10,rcutoff2))
 965             {
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq10             = _mm_mul_pd(iq1,jq0);
 969
 970             /* REACTION-FIELD ELECTROSTATICS */
 971             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 972
 973             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 974
 975             fscal            = felec;
 976
 977             fscal            = _mm_and_pd(fscal,cutoff_mask);
 978
 979             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 980
 981             /* Update vectorial force */
 982             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 983             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 984             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 985
 986             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 987             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 988             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 989
 990             }
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             if (gmx_mm_any_lt(rsq20,rcutoff2))
 997             {
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             qq20             = _mm_mul_pd(iq2,jq0);
1001
1002             /* REACTION-FIELD ELECTROSTATICS */
1003             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1004
1005             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1006
1007             fscal            = felec;
1008
1009             fscal            = _mm_and_pd(fscal,cutoff_mask);
1010
1011             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1012
1013             /* Update vectorial force */
1014             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1015             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1016             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1017
1018             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1019             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1020             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1021
1022             }
1023
1024             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1025
1026             /* Inner loop uses 133 flops */
1027         }
1028
1029         /* End of innermost loop */
1030
1031         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1032                                               f+i_coord_offset,fshift+i_shift_offset);
1033
1034         /* Increment number of inner iterations */
1035         inneriter                  += j_index_end - j_index_start;
1036
1037         /* Outer loop uses 18 flops */
1038     }
1039
1040     /* Increment number of outer iterations */
1041     outeriter        += nri;
1042
1043     /* Update outer/inner flops */
1044
1045     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*133);
1046 }