src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_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              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 105     real             rswitch_scalar,d_scalar;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 141
 142     rswitch_scalar   = fr->rvdw_switch;
 143     rswitch          = _mm_set1_pd(rswitch_scalar);
 144     /* Setup switch parameters */
 145     d_scalar         = rcutoff_scalar-rswitch_scalar;
 146     d                = _mm_set1_pd(d_scalar);
 147     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 148     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 151     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 179
 180         fix0             = _mm_setzero_pd();
 181         fiy0             = _mm_setzero_pd();
 182         fiz0             = _mm_setzero_pd();
 183         fix1             = _mm_setzero_pd();
 184         fiy1             = _mm_setzero_pd();
 185         fiz1             = _mm_setzero_pd();
 186         fix2             = _mm_setzero_pd();
 187         fiy2             = _mm_setzero_pd();
 188         fiz2             = _mm_setzero_pd();
 189         fix3             = _mm_setzero_pd();
 190         fiy3             = _mm_setzero_pd();
 191         fiz3             = _mm_setzero_pd();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm_setzero_pd();
 195         vvdwsum          = _mm_setzero_pd();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206
 207             /* load j atom coordinates */
 208             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 209                                               &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_pd(ix0,jx0);
 213             dy00             = _mm_sub_pd(iy0,jy0);
 214             dz00             = _mm_sub_pd(iz0,jz0);
 215             dx10             = _mm_sub_pd(ix1,jx0);
 216             dy10             = _mm_sub_pd(iy1,jy0);
 217             dz10             = _mm_sub_pd(iz1,jz0);
 218             dx20             = _mm_sub_pd(ix2,jx0);
 219             dy20             = _mm_sub_pd(iy2,jy0);
 220             dz20             = _mm_sub_pd(iz2,jz0);
 221             dx30             = _mm_sub_pd(ix3,jx0);
 222             dy30             = _mm_sub_pd(iy3,jy0);
 223             dz30             = _mm_sub_pd(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 229             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 230
 231             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 232             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 233             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 234             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 235
 236             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 237             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 238             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 239             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 243             vdwjidx0A        = 2*vdwtype[jnrA+0];
 244             vdwjidx0B        = 2*vdwtype[jnrB+0];
 245
 246             fjx0             = _mm_setzero_pd();
 247             fjy0             = _mm_setzero_pd();
 248             fjz0             = _mm_setzero_pd();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             if (gmx_mm_any_lt(rsq00,rcutoff2))
 255             {
 256
 257             r00              = _mm_mul_pd(rsq00,rinv00);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 261                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 262
 263             /* LENNARD-JONES DISPERSION/REPULSION */
 264
 265             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 266             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 267             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 268             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 269             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 270
 271             d                = _mm_sub_pd(r00,rswitch);
 272             d                = _mm_max_pd(d,_mm_setzero_pd());
 273             d2               = _mm_mul_pd(d,d);
 274             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 275
 276             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 277
 278             /* Evaluate switch function */
 279             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 280             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 281             vvdw             = _mm_mul_pd(vvdw,sw);
 282             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 286             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 287
 288             fscal            = fvdw;
 289
 290             fscal            = _mm_and_pd(fscal,cutoff_mask);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 294             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 295             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 296
 297             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 298             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 299             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 300
 301             }
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm_any_lt(rsq10,rcutoff2))
 308             {
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm_mul_pd(iq1,jq0);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 315             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 316
 317             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_and_pd(velec,cutoff_mask);
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm_and_pd(fscal,cutoff_mask);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 329             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 330             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 331
 332             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 333             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 334             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 335
 336             }
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             if (gmx_mm_any_lt(rsq20,rcutoff2))
 343             {
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq20             = _mm_mul_pd(iq2,jq0);
 347
 348             /* REACTION-FIELD ELECTROSTATICS */
 349             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 350             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 351
 352             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm_and_pd(velec,cutoff_mask);
 356             velecsum         = _mm_add_pd(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             fscal            = _mm_and_pd(fscal,cutoff_mask);
 361
 362             /* Update vectorial force */
 363             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 364             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 365             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 366
 367             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 368             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 369             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 370
 371             }
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             if (gmx_mm_any_lt(rsq30,rcutoff2))
 378             {
 379
 380             /* Compute parameters for interactions between i and j atoms */
 381             qq30             = _mm_mul_pd(iq3,jq0);
 382
 383             /* REACTION-FIELD ELECTROSTATICS */
 384             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 385             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 386
 387             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velec            = _mm_and_pd(velec,cutoff_mask);
 391             velecsum         = _mm_add_pd(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             fscal            = _mm_and_pd(fscal,cutoff_mask);
 396
 397             /* Update vectorial force */
 398             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 399             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 400             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 401
 402             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 403             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 404             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 405
 406             }
 407
 408             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 409
 410             /* Inner loop uses 182 flops */
 411         }
 412
 413         if(jidx<j_index_end)
 414         {
 415
 416             jnrA             = jjnr[jidx];
 417             j_coord_offsetA  = DIM*jnrA;
 418
 419             /* load j atom coordinates */
 420             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 421                                               &jx0,&jy0,&jz0);
 422
 423             /* Calculate displacement vector */
 424             dx00             = _mm_sub_pd(ix0,jx0);
 425             dy00             = _mm_sub_pd(iy0,jy0);
 426             dz00             = _mm_sub_pd(iz0,jz0);
 427             dx10             = _mm_sub_pd(ix1,jx0);
 428             dy10             = _mm_sub_pd(iy1,jy0);
 429             dz10             = _mm_sub_pd(iz1,jz0);
 430             dx20             = _mm_sub_pd(ix2,jx0);
 431             dy20             = _mm_sub_pd(iy2,jy0);
 432             dz20             = _mm_sub_pd(iz2,jz0);
 433             dx30             = _mm_sub_pd(ix3,jx0);
 434             dy30             = _mm_sub_pd(iy3,jy0);
 435             dz30             = _mm_sub_pd(iz3,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 439             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 440             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 441             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 442
 443             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 444             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 445             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 446             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 447
 448             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 449             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 450             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 451             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 452
 453             /* Load parameters for j particles */
 454             jq0              = _mm_load_sd(charge+jnrA+0);
 455             vdwjidx0A        = 2*vdwtype[jnrA+0];
 456
 457             fjx0             = _mm_setzero_pd();
 458             fjy0             = _mm_setzero_pd();
 459             fjz0             = _mm_setzero_pd();
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm_any_lt(rsq00,rcutoff2))
 466             {
 467
 468             r00              = _mm_mul_pd(rsq00,rinv00);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 472
 473             /* LENNARD-JONES DISPERSION/REPULSION */
 474
 475             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 476             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 477             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 478             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 479             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 480
 481             d                = _mm_sub_pd(r00,rswitch);
 482             d                = _mm_max_pd(d,_mm_setzero_pd());
 483             d2               = _mm_mul_pd(d,d);
 484             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 485
 486             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 487
 488             /* Evaluate switch function */
 489             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 490             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 491             vvdw             = _mm_mul_pd(vvdw,sw);
 492             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 496             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 497             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 498
 499             fscal            = fvdw;
 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             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 507             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 508             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 509
 510             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 511             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 512             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 513
 514             }
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             if (gmx_mm_any_lt(rsq10,rcutoff2))
 521             {
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq10             = _mm_mul_pd(iq1,jq0);
 525
 526             /* REACTION-FIELD ELECTROSTATICS */
 527             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 528             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 529
 530             cutoff_mask      = _mm_cmplt_pd(rsq10,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             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 545             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 546             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 547
 548             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 549             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 550             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 551
 552             }
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm_any_lt(rsq20,rcutoff2))
 559             {
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq20             = _mm_mul_pd(iq2,jq0);
 563
 564             /* REACTION-FIELD ELECTROSTATICS */
 565             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 566             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 567
 568             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_and_pd(velec,cutoff_mask);
 572             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 573             velecsum         = _mm_add_pd(velecsum,velec);
 574
 575             fscal            = felec;
 576
 577             fscal            = _mm_and_pd(fscal,cutoff_mask);
 578
 579             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 580
 581             /* Update vectorial force */
 582             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 583             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 584             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 585
 586             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 587             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 588             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 589
 590             }
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             if (gmx_mm_any_lt(rsq30,rcutoff2))
 597             {
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             qq30             = _mm_mul_pd(iq3,jq0);
 601
 602             /* REACTION-FIELD ELECTROSTATICS */
 603             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 604             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 605
 606             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velec            = _mm_and_pd(velec,cutoff_mask);
 610             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 611             velecsum         = _mm_add_pd(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm_and_pd(fscal,cutoff_mask);
 616
 617             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 618
 619             /* Update vectorial force */
 620             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 621             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 622             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 623
 624             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 625             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 626             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 627
 628             }
 629
 630             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 631
 632             /* Inner loop uses 182 flops */
 633         }
 634
 635         /* End of innermost loop */
 636
 637         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 638                                               f+i_coord_offset,fshift+i_shift_offset);
 639
 640         ggid                        = gid[iidx];
 641         /* Update potential energies */
 642         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 643         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 644
 645         /* Increment number of inner iterations */
 646         inneriter                  += j_index_end - j_index_start;
 647
 648         /* Outer loop uses 26 flops */
 649     }
 650
 651     /* Increment number of outer iterations */
 652     outeriter        += nri;
 653
 654     /* Update outer/inner flops */
 655
 656     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 657 }
 658 /*
 659  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_128_fma_double
 660  * Electrostatics interaction: ReactionField
 661  * VdW interaction:            LennardJones
 662  * Geometry:                   Water4-Particle
 663  * Calculate force/pot:        Force
 664  */
 665 void
 666 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_128_fma_double
 667                     (t_nblist                    * gmx_restrict       nlist,
 668                      rvec                        * gmx_restrict          xx,
 669                      rvec                        * gmx_restrict          ff,
 670                      t_forcerec                  * gmx_restrict          fr,
 671                      t_mdatoms                   * gmx_restrict     mdatoms,
 672                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 673                      t_nrnb                      * gmx_restrict        nrnb)
 674 {
 675     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 676      * just 0 for non-waters.
 677      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 678      * jnr indices corresponding to data put in the four positions in the SIMD register.
 679      */
 680     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 681     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 682     int              jnrA,jnrB;
 683     int              j_coord_offsetA,j_coord_offsetB;
 684     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 685     real             rcutoff_scalar;
 686     real             *shiftvec,*fshift,*x,*f;
 687     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 688     int              vdwioffset0;
 689     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 690     int              vdwioffset1;
 691     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 692     int              vdwioffset2;
 693     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 694     int              vdwioffset3;
 695     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 696     int              vdwjidx0A,vdwjidx0B;
 697     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 698     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 699     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 700     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 701     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 702     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 703     real             *charge;
 704     int              nvdwtype;
 705     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 706     int              *vdwtype;
 707     real             *vdwparam;
 708     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 709     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 710     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 711     real             rswitch_scalar,d_scalar;
 712     __m128d          dummy_mask,cutoff_mask;
 713     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 714     __m128d          one     = _mm_set1_pd(1.0);
 715     __m128d          two     = _mm_set1_pd(2.0);
 716     x                = xx[0];
 717     f                = ff[0];
 718
 719     nri              = nlist->nri;
 720     iinr             = nlist->iinr;
 721     jindex           = nlist->jindex;
 722     jjnr             = nlist->jjnr;
 723     shiftidx         = nlist->shift;
 724     gid              = nlist->gid;
 725     shiftvec         = fr->shift_vec[0];
 726     fshift           = fr->fshift[0];
 727     facel            = _mm_set1_pd(fr->epsfac);
 728     charge           = mdatoms->chargeA;
 729     krf              = _mm_set1_pd(fr->ic->k_rf);
 730     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 731     crf              = _mm_set1_pd(fr->ic->c_rf);
 732     nvdwtype         = fr->ntype;
 733     vdwparam         = fr->nbfp;
 734     vdwtype          = mdatoms->typeA;
 735
 736     /* Setup water-specific parameters */
 737     inr              = nlist->iinr[0];
 738     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 739     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 740     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 741     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 742
 743     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 744     rcutoff_scalar   = fr->rcoulomb;
 745     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 746     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 747
 748     rswitch_scalar   = fr->rvdw_switch;
 749     rswitch          = _mm_set1_pd(rswitch_scalar);
 750     /* Setup switch parameters */
 751     d_scalar         = rcutoff_scalar-rswitch_scalar;
 752     d                = _mm_set1_pd(d_scalar);
 753     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 754     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 755     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 756     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 757     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 758     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 759
 760     /* Avoid stupid compiler warnings */
 761     jnrA = jnrB = 0;
 762     j_coord_offsetA = 0;
 763     j_coord_offsetB = 0;
 764
 765     outeriter        = 0;
 766     inneriter        = 0;
 767
 768     /* Start outer loop over neighborlists */
 769     for(iidx=0; iidx<nri; iidx++)
 770     {
 771         /* Load shift vector for this list */
 772         i_shift_offset   = DIM*shiftidx[iidx];
 773
 774         /* Load limits for loop over neighbors */
 775         j_index_start    = jindex[iidx];
 776         j_index_end      = jindex[iidx+1];
 777
 778         /* Get outer coordinate index */
 779         inr              = iinr[iidx];
 780         i_coord_offset   = DIM*inr;
 781
 782         /* Load i particle coords and add shift vector */
 783         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 784                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 785
 786         fix0             = _mm_setzero_pd();
 787         fiy0             = _mm_setzero_pd();
 788         fiz0             = _mm_setzero_pd();
 789         fix1             = _mm_setzero_pd();
 790         fiy1             = _mm_setzero_pd();
 791         fiz1             = _mm_setzero_pd();
 792         fix2             = _mm_setzero_pd();
 793         fiy2             = _mm_setzero_pd();
 794         fiz2             = _mm_setzero_pd();
 795         fix3             = _mm_setzero_pd();
 796         fiy3             = _mm_setzero_pd();
 797         fiz3             = _mm_setzero_pd();
 798
 799         /* Start inner kernel loop */
 800         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 801         {
 802
 803             /* Get j neighbor index, and coordinate index */
 804             jnrA             = jjnr[jidx];
 805             jnrB             = jjnr[jidx+1];
 806             j_coord_offsetA  = DIM*jnrA;
 807             j_coord_offsetB  = DIM*jnrB;
 808
 809             /* load j atom coordinates */
 810             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 811                                               &jx0,&jy0,&jz0);
 812
 813             /* Calculate displacement vector */
 814             dx00             = _mm_sub_pd(ix0,jx0);
 815             dy00             = _mm_sub_pd(iy0,jy0);
 816             dz00             = _mm_sub_pd(iz0,jz0);
 817             dx10             = _mm_sub_pd(ix1,jx0);
 818             dy10             = _mm_sub_pd(iy1,jy0);
 819             dz10             = _mm_sub_pd(iz1,jz0);
 820             dx20             = _mm_sub_pd(ix2,jx0);
 821             dy20             = _mm_sub_pd(iy2,jy0);
 822             dz20             = _mm_sub_pd(iz2,jz0);
 823             dx30             = _mm_sub_pd(ix3,jx0);
 824             dy30             = _mm_sub_pd(iy3,jy0);
 825             dz30             = _mm_sub_pd(iz3,jz0);
 826
 827             /* Calculate squared distance and things based on it */
 828             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 829             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 830             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 831             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 832
 833             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 834             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 835             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 836             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 837
 838             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 839             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 840             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 841             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 842
 843             /* Load parameters for j particles */
 844             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 845             vdwjidx0A        = 2*vdwtype[jnrA+0];
 846             vdwjidx0B        = 2*vdwtype[jnrB+0];
 847
 848             fjx0             = _mm_setzero_pd();
 849             fjy0             = _mm_setzero_pd();
 850             fjz0             = _mm_setzero_pd();
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             if (gmx_mm_any_lt(rsq00,rcutoff2))
 857             {
 858
 859             r00              = _mm_mul_pd(rsq00,rinv00);
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 863                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 864
 865             /* LENNARD-JONES DISPERSION/REPULSION */
 866
 867             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 868             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 869             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 870             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 871             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 872
 873             d                = _mm_sub_pd(r00,rswitch);
 874             d                = _mm_max_pd(d,_mm_setzero_pd());
 875             d2               = _mm_mul_pd(d,d);
 876             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 877
 878             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 879
 880             /* Evaluate switch function */
 881             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 882             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 883             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 884
 885             fscal            = fvdw;
 886
 887             fscal            = _mm_and_pd(fscal,cutoff_mask);
 888
 889             /* Update vectorial force */
 890             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 891             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 892             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 893
 894             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 895             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 896             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 897
 898             }
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             if (gmx_mm_any_lt(rsq10,rcutoff2))
 905             {
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq10             = _mm_mul_pd(iq1,jq0);
 909
 910             /* REACTION-FIELD ELECTROSTATICS */
 911             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 912
 913             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 914
 915             fscal            = felec;
 916
 917             fscal            = _mm_and_pd(fscal,cutoff_mask);
 918
 919             /* Update vectorial force */
 920             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 921             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 922             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 923
 924             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 925             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 926             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 927
 928             }
 929
 930             /**************************
 931              * CALCULATE INTERACTIONS *
 932              **************************/
 933
 934             if (gmx_mm_any_lt(rsq20,rcutoff2))
 935             {
 936
 937             /* Compute parameters for interactions between i and j atoms */
 938             qq20             = _mm_mul_pd(iq2,jq0);
 939
 940             /* REACTION-FIELD ELECTROSTATICS */
 941             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 942
 943             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 944
 945             fscal            = felec;
 946
 947             fscal            = _mm_and_pd(fscal,cutoff_mask);
 948
 949             /* Update vectorial force */
 950             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 951             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 952             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 953
 954             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 955             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 956             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 957
 958             }
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             if (gmx_mm_any_lt(rsq30,rcutoff2))
 965             {
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq30             = _mm_mul_pd(iq3,jq0);
 969
 970             /* REACTION-FIELD ELECTROSTATICS */
 971             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 972
 973             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 974
 975             fscal            = felec;
 976
 977             fscal            = _mm_and_pd(fscal,cutoff_mask);
 978
 979             /* Update vectorial force */
 980             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 981             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 982             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 983
 984             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 985             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 986             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 987
 988             }
 989
 990             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 991
 992             /* Inner loop uses 161 flops */
 993         }
 994
 995         if(jidx<j_index_end)
 996         {
 997
 998             jnrA             = jjnr[jidx];
 999             j_coord_offsetA  = DIM*jnrA;
1000
1001             /* load j atom coordinates */
1002             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1003                                               &jx0,&jy0,&jz0);
1004
1005             /* Calculate displacement vector */
1006             dx00             = _mm_sub_pd(ix0,jx0);
1007             dy00             = _mm_sub_pd(iy0,jy0);
1008             dz00             = _mm_sub_pd(iz0,jz0);
1009             dx10             = _mm_sub_pd(ix1,jx0);
1010             dy10             = _mm_sub_pd(iy1,jy0);
1011             dz10             = _mm_sub_pd(iz1,jz0);
1012             dx20             = _mm_sub_pd(ix2,jx0);
1013             dy20             = _mm_sub_pd(iy2,jy0);
1014             dz20             = _mm_sub_pd(iz2,jz0);
1015             dx30             = _mm_sub_pd(ix3,jx0);
1016             dy30             = _mm_sub_pd(iy3,jy0);
1017             dz30             = _mm_sub_pd(iz3,jz0);
1018
1019             /* Calculate squared distance and things based on it */
1020             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1021             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1022             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1023             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1024
1025             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1026             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1027             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1028             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1029
1030             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1031             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1032             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1033             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1034
1035             /* Load parameters for j particles */
1036             jq0              = _mm_load_sd(charge+jnrA+0);
1037             vdwjidx0A        = 2*vdwtype[jnrA+0];
1038
1039             fjx0             = _mm_setzero_pd();
1040             fjy0             = _mm_setzero_pd();
1041             fjz0             = _mm_setzero_pd();
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             if (gmx_mm_any_lt(rsq00,rcutoff2))
1048             {
1049
1050             r00              = _mm_mul_pd(rsq00,rinv00);
1051
1052             /* Compute parameters for interactions between i and j atoms */
1053             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1054
1055             /* LENNARD-JONES DISPERSION/REPULSION */
1056
1057             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1058             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1059             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1060             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1061             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1062
1063             d                = _mm_sub_pd(r00,rswitch);
1064             d                = _mm_max_pd(d,_mm_setzero_pd());
1065             d2               = _mm_mul_pd(d,d);
1066             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1067
1068             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1069
1070             /* Evaluate switch function */
1071             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1072             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1073             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1074
1075             fscal            = fvdw;
1076
1077             fscal            = _mm_and_pd(fscal,cutoff_mask);
1078
1079             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1080
1081             /* Update vectorial force */
1082             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1083             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1084             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1085
1086             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1087             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1088             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1089
1090             }
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm_any_lt(rsq10,rcutoff2))
1097             {
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq10             = _mm_mul_pd(iq1,jq0);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1104
1105             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_and_pd(fscal,cutoff_mask);
1110
1111             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1112
1113             /* Update vectorial force */
1114             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1115             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1116             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1117
1118             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1119             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1120             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1121
1122             }
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             if (gmx_mm_any_lt(rsq20,rcutoff2))
1129             {
1130
1131             /* Compute parameters for interactions between i and j atoms */
1132             qq20             = _mm_mul_pd(iq2,jq0);
1133
1134             /* REACTION-FIELD ELECTROSTATICS */
1135             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1136
1137             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1138
1139             fscal            = felec;
1140
1141             fscal            = _mm_and_pd(fscal,cutoff_mask);
1142
1143             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1144
1145             /* Update vectorial force */
1146             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1147             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1148             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1149
1150             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1151             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1152             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1153
1154             }
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             if (gmx_mm_any_lt(rsq30,rcutoff2))
1161             {
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq30             = _mm_mul_pd(iq3,jq0);
1165
1166             /* REACTION-FIELD ELECTROSTATICS */
1167             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
1168
1169             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1170
1171             fscal            = felec;
1172
1173             fscal            = _mm_and_pd(fscal,cutoff_mask);
1174
1175             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1176
1177             /* Update vectorial force */
1178             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1179             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1180             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1181
1182             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1183             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1184             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1185
1186             }
1187
1188             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1189
1190             /* Inner loop uses 161 flops */
1191         }
1192
1193         /* End of innermost loop */
1194
1195         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1196                                               f+i_coord_offset,fshift+i_shift_offset);
1197
1198         /* Increment number of inner iterations */
1199         inneriter                  += j_index_end - j_index_start;
1200
1201         /* Outer loop uses 24 flops */
1202     }
1203
1204     /* Increment number of outer iterations */
1205     outeriter        += nri;
1206
1207     /* Update outer/inner flops */
1208
1209     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*161);
1210 }