src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_sse4_1_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d          dummy_mask,cutoff_mask;
 102     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 103     __m128d          one     = _mm_set1_pd(1.0);
 104     __m128d          two     = _mm_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_pd(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118     krf              = _mm_set1_pd(fr->ic->k_rf);
 119     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 120     crf              = _mm_set1_pd(fr->ic->c_rf);
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 128     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 129     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 130     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 131
 132     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 133     rcutoff_scalar   = fr->ic->rcoulomb;
 134     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 135     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 136
 137     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 138     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169         fix1             = _mm_setzero_pd();
 170         fiy1             = _mm_setzero_pd();
 171         fiz1             = _mm_setzero_pd();
 172         fix2             = _mm_setzero_pd();
 173         fiy2             = _mm_setzero_pd();
 174         fiz2             = _mm_setzero_pd();
 175         fix3             = _mm_setzero_pd();
 176         fiy3             = _mm_setzero_pd();
 177         fiz3             = _mm_setzero_pd();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_pd();
 181         vvdwsum          = _mm_setzero_pd();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_pd(ix0,jx0);
 199             dy00             = _mm_sub_pd(iy0,jy0);
 200             dz00             = _mm_sub_pd(iz0,jz0);
 201             dx10             = _mm_sub_pd(ix1,jx0);
 202             dy10             = _mm_sub_pd(iy1,jy0);
 203             dz10             = _mm_sub_pd(iz1,jz0);
 204             dx20             = _mm_sub_pd(ix2,jx0);
 205             dy20             = _mm_sub_pd(iy2,jy0);
 206             dz20             = _mm_sub_pd(iz2,jz0);
 207             dx30             = _mm_sub_pd(ix3,jx0);
 208             dy30             = _mm_sub_pd(iy3,jy0);
 209             dz30             = _mm_sub_pd(iz3,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 215             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 216
 217             rinv10           = sse41_invsqrt_d(rsq10);
 218             rinv20           = sse41_invsqrt_d(rsq20);
 219             rinv30           = sse41_invsqrt_d(rsq30);
 220
 221             rinvsq00         = sse41_inv_d(rsq00);
 222             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 223             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 224             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             if (gmx_mm_any_lt(rsq00,rcutoff2))
 240             {
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 245
 246             /* LENNARD-JONES DISPERSION/REPULSION */
 247
 248             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 249             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 250             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 251             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 252                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 253             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 254
 255             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 259             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 260
 261             fscal            = fvdw;
 262
 263             fscal            = _mm_and_pd(fscal,cutoff_mask);
 264
 265             /* Calculate temporary vectorial force */
 266             tx               = _mm_mul_pd(fscal,dx00);
 267             ty               = _mm_mul_pd(fscal,dy00);
 268             tz               = _mm_mul_pd(fscal,dz00);
 269
 270             /* Update vectorial force */
 271             fix0             = _mm_add_pd(fix0,tx);
 272             fiy0             = _mm_add_pd(fiy0,ty);
 273             fiz0             = _mm_add_pd(fiz0,tz);
 274
 275             fjx0             = _mm_add_pd(fjx0,tx);
 276             fjy0             = _mm_add_pd(fjy0,ty);
 277             fjz0             = _mm_add_pd(fjz0,tz);
 278
 279             }
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             if (gmx_mm_any_lt(rsq10,rcutoff2))
 286             {
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq10             = _mm_mul_pd(iq1,jq0);
 290
 291             /* REACTION-FIELD ELECTROSTATICS */
 292             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 293             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 294
 295             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velec            = _mm_and_pd(velec,cutoff_mask);
 299             velecsum         = _mm_add_pd(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303             fscal            = _mm_and_pd(fscal,cutoff_mask);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_pd(fscal,dx10);
 307             ty               = _mm_mul_pd(fscal,dy10);
 308             tz               = _mm_mul_pd(fscal,dz10);
 309
 310             /* Update vectorial force */
 311             fix1             = _mm_add_pd(fix1,tx);
 312             fiy1             = _mm_add_pd(fiy1,ty);
 313             fiz1             = _mm_add_pd(fiz1,tz);
 314
 315             fjx0             = _mm_add_pd(fjx0,tx);
 316             fjy0             = _mm_add_pd(fjy0,ty);
 317             fjz0             = _mm_add_pd(fjz0,tz);
 318
 319             }
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             if (gmx_mm_any_lt(rsq20,rcutoff2))
 326             {
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq20             = _mm_mul_pd(iq2,jq0);
 330
 331             /* REACTION-FIELD ELECTROSTATICS */
 332             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 333             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 334
 335             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _mm_and_pd(velec,cutoff_mask);
 339             velecsum         = _mm_add_pd(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             fscal            = _mm_and_pd(fscal,cutoff_mask);
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_pd(fscal,dx20);
 347             ty               = _mm_mul_pd(fscal,dy20);
 348             tz               = _mm_mul_pd(fscal,dz20);
 349
 350             /* Update vectorial force */
 351             fix2             = _mm_add_pd(fix2,tx);
 352             fiy2             = _mm_add_pd(fiy2,ty);
 353             fiz2             = _mm_add_pd(fiz2,tz);
 354
 355             fjx0             = _mm_add_pd(fjx0,tx);
 356             fjy0             = _mm_add_pd(fjy0,ty);
 357             fjz0             = _mm_add_pd(fjz0,tz);
 358
 359             }
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             if (gmx_mm_any_lt(rsq30,rcutoff2))
 366             {
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq30             = _mm_mul_pd(iq3,jq0);
 370
 371             /* REACTION-FIELD ELECTROSTATICS */
 372             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 373             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 374
 375             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm_and_pd(velec,cutoff_mask);
 379             velecsum         = _mm_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm_and_pd(fscal,cutoff_mask);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm_mul_pd(fscal,dx30);
 387             ty               = _mm_mul_pd(fscal,dy30);
 388             tz               = _mm_mul_pd(fscal,dz30);
 389
 390             /* Update vectorial force */
 391             fix3             = _mm_add_pd(fix3,tx);
 392             fiy3             = _mm_add_pd(fiy3,ty);
 393             fiz3             = _mm_add_pd(fiz3,tz);
 394
 395             fjx0             = _mm_add_pd(fjx0,tx);
 396             fjy0             = _mm_add_pd(fjy0,ty);
 397             fjz0             = _mm_add_pd(fjz0,tz);
 398
 399             }
 400
 401             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 402
 403             /* Inner loop uses 152 flops */
 404         }
 405
 406         if(jidx<j_index_end)
 407         {
 408
 409             jnrA             = jjnr[jidx];
 410             j_coord_offsetA  = DIM*jnrA;
 411
 412             /* load j atom coordinates */
 413             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 414                                               &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _mm_sub_pd(ix0,jx0);
 418             dy00             = _mm_sub_pd(iy0,jy0);
 419             dz00             = _mm_sub_pd(iz0,jz0);
 420             dx10             = _mm_sub_pd(ix1,jx0);
 421             dy10             = _mm_sub_pd(iy1,jy0);
 422             dz10             = _mm_sub_pd(iz1,jz0);
 423             dx20             = _mm_sub_pd(ix2,jx0);
 424             dy20             = _mm_sub_pd(iy2,jy0);
 425             dz20             = _mm_sub_pd(iz2,jz0);
 426             dx30             = _mm_sub_pd(ix3,jx0);
 427             dy30             = _mm_sub_pd(iy3,jy0);
 428             dz30             = _mm_sub_pd(iz3,jz0);
 429
 430             /* Calculate squared distance and things based on it */
 431             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 432             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 433             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 434             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 435
 436             rinv10           = sse41_invsqrt_d(rsq10);
 437             rinv20           = sse41_invsqrt_d(rsq20);
 438             rinv30           = sse41_invsqrt_d(rsq30);
 439
 440             rinvsq00         = sse41_inv_d(rsq00);
 441             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 442             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 443             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 444
 445             /* Load parameters for j particles */
 446             jq0              = _mm_load_sd(charge+jnrA+0);
 447             vdwjidx0A        = 2*vdwtype[jnrA+0];
 448
 449             fjx0             = _mm_setzero_pd();
 450             fjy0             = _mm_setzero_pd();
 451             fjz0             = _mm_setzero_pd();
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             if (gmx_mm_any_lt(rsq00,rcutoff2))
 458             {
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 462
 463             /* LENNARD-JONES DISPERSION/REPULSION */
 464
 465             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 466             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 467             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 468             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 469                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 470             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 471
 472             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 476             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 477             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 478
 479             fscal            = fvdw;
 480
 481             fscal            = _mm_and_pd(fscal,cutoff_mask);
 482
 483             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm_mul_pd(fscal,dx00);
 487             ty               = _mm_mul_pd(fscal,dy00);
 488             tz               = _mm_mul_pd(fscal,dz00);
 489
 490             /* Update vectorial force */
 491             fix0             = _mm_add_pd(fix0,tx);
 492             fiy0             = _mm_add_pd(fiy0,ty);
 493             fiz0             = _mm_add_pd(fiz0,tz);
 494
 495             fjx0             = _mm_add_pd(fjx0,tx);
 496             fjy0             = _mm_add_pd(fjy0,ty);
 497             fjz0             = _mm_add_pd(fjz0,tz);
 498
 499             }
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             if (gmx_mm_any_lt(rsq10,rcutoff2))
 506             {
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq10             = _mm_mul_pd(iq1,jq0);
 510
 511             /* REACTION-FIELD ELECTROSTATICS */
 512             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 513             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 514
 515             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velec            = _mm_and_pd(velec,cutoff_mask);
 519             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 520             velecsum         = _mm_add_pd(velecsum,velec);
 521
 522             fscal            = felec;
 523
 524             fscal            = _mm_and_pd(fscal,cutoff_mask);
 525
 526             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_pd(fscal,dx10);
 530             ty               = _mm_mul_pd(fscal,dy10);
 531             tz               = _mm_mul_pd(fscal,dz10);
 532
 533             /* Update vectorial force */
 534             fix1             = _mm_add_pd(fix1,tx);
 535             fiy1             = _mm_add_pd(fiy1,ty);
 536             fiz1             = _mm_add_pd(fiz1,tz);
 537
 538             fjx0             = _mm_add_pd(fjx0,tx);
 539             fjy0             = _mm_add_pd(fjy0,ty);
 540             fjz0             = _mm_add_pd(fjz0,tz);
 541
 542             }
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             if (gmx_mm_any_lt(rsq20,rcutoff2))
 549             {
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq20             = _mm_mul_pd(iq2,jq0);
 553
 554             /* REACTION-FIELD ELECTROSTATICS */
 555             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 556             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 557
 558             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm_and_pd(velec,cutoff_mask);
 562             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 563             velecsum         = _mm_add_pd(velecsum,velec);
 564
 565             fscal            = felec;
 566
 567             fscal            = _mm_and_pd(fscal,cutoff_mask);
 568
 569             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm_mul_pd(fscal,dx20);
 573             ty               = _mm_mul_pd(fscal,dy20);
 574             tz               = _mm_mul_pd(fscal,dz20);
 575
 576             /* Update vectorial force */
 577             fix2             = _mm_add_pd(fix2,tx);
 578             fiy2             = _mm_add_pd(fiy2,ty);
 579             fiz2             = _mm_add_pd(fiz2,tz);
 580
 581             fjx0             = _mm_add_pd(fjx0,tx);
 582             fjy0             = _mm_add_pd(fjy0,ty);
 583             fjz0             = _mm_add_pd(fjz0,tz);
 584
 585             }
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             if (gmx_mm_any_lt(rsq30,rcutoff2))
 592             {
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq30             = _mm_mul_pd(iq3,jq0);
 596
 597             /* REACTION-FIELD ELECTROSTATICS */
 598             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 599             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 600
 601             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 602
 603             /* Update potential sum for this i atom from the interaction with this j atom. */
 604             velec            = _mm_and_pd(velec,cutoff_mask);
 605             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 606             velecsum         = _mm_add_pd(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             fscal            = _mm_and_pd(fscal,cutoff_mask);
 611
 612             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 613
 614             /* Calculate temporary vectorial force */
 615             tx               = _mm_mul_pd(fscal,dx30);
 616             ty               = _mm_mul_pd(fscal,dy30);
 617             tz               = _mm_mul_pd(fscal,dz30);
 618
 619             /* Update vectorial force */
 620             fix3             = _mm_add_pd(fix3,tx);
 621             fiy3             = _mm_add_pd(fiy3,ty);
 622             fiz3             = _mm_add_pd(fiz3,tz);
 623
 624             fjx0             = _mm_add_pd(fjx0,tx);
 625             fjy0             = _mm_add_pd(fjy0,ty);
 626             fjz0             = _mm_add_pd(fjz0,tz);
 627
 628             }
 629
 630             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 631
 632             /* Inner loop uses 152 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*152);
 657 }
 658 /*
 659  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_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_VdwLJSh_GeomW4P1_F_sse4_1_double
 667                     (t_nblist                    * gmx_restrict       nlist,
 668                      rvec                        * gmx_restrict          xx,
 669                      rvec                        * gmx_restrict          ff,
 670                      struct 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          dummy_mask,cutoff_mask;
 711     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 712     __m128d          one     = _mm_set1_pd(1.0);
 713     __m128d          two     = _mm_set1_pd(2.0);
 714     x                = xx[0];
 715     f                = ff[0];
 716
 717     nri              = nlist->nri;
 718     iinr             = nlist->iinr;
 719     jindex           = nlist->jindex;
 720     jjnr             = nlist->jjnr;
 721     shiftidx         = nlist->shift;
 722     gid              = nlist->gid;
 723     shiftvec         = fr->shift_vec[0];
 724     fshift           = fr->fshift[0];
 725     facel            = _mm_set1_pd(fr->ic->epsfac);
 726     charge           = mdatoms->chargeA;
 727     krf              = _mm_set1_pd(fr->ic->k_rf);
 728     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 729     crf              = _mm_set1_pd(fr->ic->c_rf);
 730     nvdwtype         = fr->ntype;
 731     vdwparam         = fr->nbfp;
 732     vdwtype          = mdatoms->typeA;
 733
 734     /* Setup water-specific parameters */
 735     inr              = nlist->iinr[0];
 736     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 737     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 738     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 739     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 740
 741     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 742     rcutoff_scalar   = fr->ic->rcoulomb;
 743     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 744     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 745
 746     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 747     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 748
 749     /* Avoid stupid compiler warnings */
 750     jnrA = jnrB = 0;
 751     j_coord_offsetA = 0;
 752     j_coord_offsetB = 0;
 753
 754     outeriter        = 0;
 755     inneriter        = 0;
 756
 757     /* Start outer loop over neighborlists */
 758     for(iidx=0; iidx<nri; iidx++)
 759     {
 760         /* Load shift vector for this list */
 761         i_shift_offset   = DIM*shiftidx[iidx];
 762
 763         /* Load limits for loop over neighbors */
 764         j_index_start    = jindex[iidx];
 765         j_index_end      = jindex[iidx+1];
 766
 767         /* Get outer coordinate index */
 768         inr              = iinr[iidx];
 769         i_coord_offset   = DIM*inr;
 770
 771         /* Load i particle coords and add shift vector */
 772         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 773                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 774
 775         fix0             = _mm_setzero_pd();
 776         fiy0             = _mm_setzero_pd();
 777         fiz0             = _mm_setzero_pd();
 778         fix1             = _mm_setzero_pd();
 779         fiy1             = _mm_setzero_pd();
 780         fiz1             = _mm_setzero_pd();
 781         fix2             = _mm_setzero_pd();
 782         fiy2             = _mm_setzero_pd();
 783         fiz2             = _mm_setzero_pd();
 784         fix3             = _mm_setzero_pd();
 785         fiy3             = _mm_setzero_pd();
 786         fiz3             = _mm_setzero_pd();
 787
 788         /* Start inner kernel loop */
 789         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 790         {
 791
 792             /* Get j neighbor index, and coordinate index */
 793             jnrA             = jjnr[jidx];
 794             jnrB             = jjnr[jidx+1];
 795             j_coord_offsetA  = DIM*jnrA;
 796             j_coord_offsetB  = DIM*jnrB;
 797
 798             /* load j atom coordinates */
 799             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 800                                               &jx0,&jy0,&jz0);
 801
 802             /* Calculate displacement vector */
 803             dx00             = _mm_sub_pd(ix0,jx0);
 804             dy00             = _mm_sub_pd(iy0,jy0);
 805             dz00             = _mm_sub_pd(iz0,jz0);
 806             dx10             = _mm_sub_pd(ix1,jx0);
 807             dy10             = _mm_sub_pd(iy1,jy0);
 808             dz10             = _mm_sub_pd(iz1,jz0);
 809             dx20             = _mm_sub_pd(ix2,jx0);
 810             dy20             = _mm_sub_pd(iy2,jy0);
 811             dz20             = _mm_sub_pd(iz2,jz0);
 812             dx30             = _mm_sub_pd(ix3,jx0);
 813             dy30             = _mm_sub_pd(iy3,jy0);
 814             dz30             = _mm_sub_pd(iz3,jz0);
 815
 816             /* Calculate squared distance and things based on it */
 817             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 818             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 819             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 820             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 821
 822             rinv10           = sse41_invsqrt_d(rsq10);
 823             rinv20           = sse41_invsqrt_d(rsq20);
 824             rinv30           = sse41_invsqrt_d(rsq30);
 825
 826             rinvsq00         = sse41_inv_d(rsq00);
 827             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 828             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 829             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 830
 831             /* Load parameters for j particles */
 832             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 833             vdwjidx0A        = 2*vdwtype[jnrA+0];
 834             vdwjidx0B        = 2*vdwtype[jnrB+0];
 835
 836             fjx0             = _mm_setzero_pd();
 837             fjy0             = _mm_setzero_pd();
 838             fjz0             = _mm_setzero_pd();
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             if (gmx_mm_any_lt(rsq00,rcutoff2))
 845             {
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 849                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 850
 851             /* LENNARD-JONES DISPERSION/REPULSION */
 852
 853             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 854             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 855
 856             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 857
 858             fscal            = fvdw;
 859
 860             fscal            = _mm_and_pd(fscal,cutoff_mask);
 861
 862             /* Calculate temporary vectorial force */
 863             tx               = _mm_mul_pd(fscal,dx00);
 864             ty               = _mm_mul_pd(fscal,dy00);
 865             tz               = _mm_mul_pd(fscal,dz00);
 866
 867             /* Update vectorial force */
 868             fix0             = _mm_add_pd(fix0,tx);
 869             fiy0             = _mm_add_pd(fiy0,ty);
 870             fiz0             = _mm_add_pd(fiz0,tz);
 871
 872             fjx0             = _mm_add_pd(fjx0,tx);
 873             fjy0             = _mm_add_pd(fjy0,ty);
 874             fjz0             = _mm_add_pd(fjz0,tz);
 875
 876             }
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             if (gmx_mm_any_lt(rsq10,rcutoff2))
 883             {
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq10             = _mm_mul_pd(iq1,jq0);
 887
 888             /* REACTION-FIELD ELECTROSTATICS */
 889             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 890
 891             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 892
 893             fscal            = felec;
 894
 895             fscal            = _mm_and_pd(fscal,cutoff_mask);
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm_mul_pd(fscal,dx10);
 899             ty               = _mm_mul_pd(fscal,dy10);
 900             tz               = _mm_mul_pd(fscal,dz10);
 901
 902             /* Update vectorial force */
 903             fix1             = _mm_add_pd(fix1,tx);
 904             fiy1             = _mm_add_pd(fiy1,ty);
 905             fiz1             = _mm_add_pd(fiz1,tz);
 906
 907             fjx0             = _mm_add_pd(fjx0,tx);
 908             fjy0             = _mm_add_pd(fjy0,ty);
 909             fjz0             = _mm_add_pd(fjz0,tz);
 910
 911             }
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             if (gmx_mm_any_lt(rsq20,rcutoff2))
 918             {
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             qq20             = _mm_mul_pd(iq2,jq0);
 922
 923             /* REACTION-FIELD ELECTROSTATICS */
 924             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 925
 926             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 927
 928             fscal            = felec;
 929
 930             fscal            = _mm_and_pd(fscal,cutoff_mask);
 931
 932             /* Calculate temporary vectorial force */
 933             tx               = _mm_mul_pd(fscal,dx20);
 934             ty               = _mm_mul_pd(fscal,dy20);
 935             tz               = _mm_mul_pd(fscal,dz20);
 936
 937             /* Update vectorial force */
 938             fix2             = _mm_add_pd(fix2,tx);
 939             fiy2             = _mm_add_pd(fiy2,ty);
 940             fiz2             = _mm_add_pd(fiz2,tz);
 941
 942             fjx0             = _mm_add_pd(fjx0,tx);
 943             fjy0             = _mm_add_pd(fjy0,ty);
 944             fjz0             = _mm_add_pd(fjz0,tz);
 945
 946             }
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             if (gmx_mm_any_lt(rsq30,rcutoff2))
 953             {
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq30             = _mm_mul_pd(iq3,jq0);
 957
 958             /* REACTION-FIELD ELECTROSTATICS */
 959             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 960
 961             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 962
 963             fscal            = felec;
 964
 965             fscal            = _mm_and_pd(fscal,cutoff_mask);
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm_mul_pd(fscal,dx30);
 969             ty               = _mm_mul_pd(fscal,dy30);
 970             tz               = _mm_mul_pd(fscal,dz30);
 971
 972             /* Update vectorial force */
 973             fix3             = _mm_add_pd(fix3,tx);
 974             fiy3             = _mm_add_pd(fiy3,ty);
 975             fiz3             = _mm_add_pd(fiz3,tz);
 976
 977             fjx0             = _mm_add_pd(fjx0,tx);
 978             fjy0             = _mm_add_pd(fjy0,ty);
 979             fjz0             = _mm_add_pd(fjz0,tz);
 980
 981             }
 982
 983             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 984
 985             /* Inner loop uses 123 flops */
 986         }
 987
 988         if(jidx<j_index_end)
 989         {
 990
 991             jnrA             = jjnr[jidx];
 992             j_coord_offsetA  = DIM*jnrA;
 993
 994             /* load j atom coordinates */
 995             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 996                                               &jx0,&jy0,&jz0);
 997
 998             /* Calculate displacement vector */
 999             dx00             = _mm_sub_pd(ix0,jx0);
1000             dy00             = _mm_sub_pd(iy0,jy0);
1001             dz00             = _mm_sub_pd(iz0,jz0);
1002             dx10             = _mm_sub_pd(ix1,jx0);
1003             dy10             = _mm_sub_pd(iy1,jy0);
1004             dz10             = _mm_sub_pd(iz1,jz0);
1005             dx20             = _mm_sub_pd(ix2,jx0);
1006             dy20             = _mm_sub_pd(iy2,jy0);
1007             dz20             = _mm_sub_pd(iz2,jz0);
1008             dx30             = _mm_sub_pd(ix3,jx0);
1009             dy30             = _mm_sub_pd(iy3,jy0);
1010             dz30             = _mm_sub_pd(iz3,jz0);
1011
1012             /* Calculate squared distance and things based on it */
1013             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1014             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1015             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1016             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1017
1018             rinv10           = sse41_invsqrt_d(rsq10);
1019             rinv20           = sse41_invsqrt_d(rsq20);
1020             rinv30           = sse41_invsqrt_d(rsq30);
1021
1022             rinvsq00         = sse41_inv_d(rsq00);
1023             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1024             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1025             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1026
1027             /* Load parameters for j particles */
1028             jq0              = _mm_load_sd(charge+jnrA+0);
1029             vdwjidx0A        = 2*vdwtype[jnrA+0];
1030
1031             fjx0             = _mm_setzero_pd();
1032             fjy0             = _mm_setzero_pd();
1033             fjz0             = _mm_setzero_pd();
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             if (gmx_mm_any_lt(rsq00,rcutoff2))
1040             {
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1044
1045             /* LENNARD-JONES DISPERSION/REPULSION */
1046
1047             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1048             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1049
1050             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1051
1052             fscal            = fvdw;
1053
1054             fscal            = _mm_and_pd(fscal,cutoff_mask);
1055
1056             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1057
1058             /* Calculate temporary vectorial force */
1059             tx               = _mm_mul_pd(fscal,dx00);
1060             ty               = _mm_mul_pd(fscal,dy00);
1061             tz               = _mm_mul_pd(fscal,dz00);
1062
1063             /* Update vectorial force */
1064             fix0             = _mm_add_pd(fix0,tx);
1065             fiy0             = _mm_add_pd(fiy0,ty);
1066             fiz0             = _mm_add_pd(fiz0,tz);
1067
1068             fjx0             = _mm_add_pd(fjx0,tx);
1069             fjy0             = _mm_add_pd(fjy0,ty);
1070             fjz0             = _mm_add_pd(fjz0,tz);
1071
1072             }
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             if (gmx_mm_any_lt(rsq10,rcutoff2))
1079             {
1080
1081             /* Compute parameters for interactions between i and j atoms */
1082             qq10             = _mm_mul_pd(iq1,jq0);
1083
1084             /* REACTION-FIELD ELECTROSTATICS */
1085             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1086
1087             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm_and_pd(fscal,cutoff_mask);
1092
1093             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm_mul_pd(fscal,dx10);
1097             ty               = _mm_mul_pd(fscal,dy10);
1098             tz               = _mm_mul_pd(fscal,dz10);
1099
1100             /* Update vectorial force */
1101             fix1             = _mm_add_pd(fix1,tx);
1102             fiy1             = _mm_add_pd(fiy1,ty);
1103             fiz1             = _mm_add_pd(fiz1,tz);
1104
1105             fjx0             = _mm_add_pd(fjx0,tx);
1106             fjy0             = _mm_add_pd(fjy0,ty);
1107             fjz0             = _mm_add_pd(fjz0,tz);
1108
1109             }
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             if (gmx_mm_any_lt(rsq20,rcutoff2))
1116             {
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq20             = _mm_mul_pd(iq2,jq0);
1120
1121             /* REACTION-FIELD ELECTROSTATICS */
1122             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1123
1124             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1125
1126             fscal            = felec;
1127
1128             fscal            = _mm_and_pd(fscal,cutoff_mask);
1129
1130             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1131
1132             /* Calculate temporary vectorial force */
1133             tx               = _mm_mul_pd(fscal,dx20);
1134             ty               = _mm_mul_pd(fscal,dy20);
1135             tz               = _mm_mul_pd(fscal,dz20);
1136
1137             /* Update vectorial force */
1138             fix2             = _mm_add_pd(fix2,tx);
1139             fiy2             = _mm_add_pd(fiy2,ty);
1140             fiz2             = _mm_add_pd(fiz2,tz);
1141
1142             fjx0             = _mm_add_pd(fjx0,tx);
1143             fjy0             = _mm_add_pd(fjy0,ty);
1144             fjz0             = _mm_add_pd(fjz0,tz);
1145
1146             }
1147
1148             /**************************
1149              * CALCULATE INTERACTIONS *
1150              **************************/
1151
1152             if (gmx_mm_any_lt(rsq30,rcutoff2))
1153             {
1154
1155             /* Compute parameters for interactions between i and j atoms */
1156             qq30             = _mm_mul_pd(iq3,jq0);
1157
1158             /* REACTION-FIELD ELECTROSTATICS */
1159             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1160
1161             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_and_pd(fscal,cutoff_mask);
1166
1167             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm_mul_pd(fscal,dx30);
1171             ty               = _mm_mul_pd(fscal,dy30);
1172             tz               = _mm_mul_pd(fscal,dz30);
1173
1174             /* Update vectorial force */
1175             fix3             = _mm_add_pd(fix3,tx);
1176             fiy3             = _mm_add_pd(fiy3,ty);
1177             fiz3             = _mm_add_pd(fiz3,tz);
1178
1179             fjx0             = _mm_add_pd(fjx0,tx);
1180             fjy0             = _mm_add_pd(fjy0,ty);
1181             fjz0             = _mm_add_pd(fjz0,tz);
1182
1183             }
1184
1185             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1186
1187             /* Inner loop uses 123 flops */
1188         }
1189
1190         /* End of innermost loop */
1191
1192         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1193                                               f+i_coord_offset,fshift+i_shift_offset);
1194
1195         /* Increment number of inner iterations */
1196         inneriter                  += j_index_end - j_index_start;
1197
1198         /* Outer loop uses 24 flops */
1199     }
1200
1201     /* Increment number of outer iterations */
1202     outeriter        += nri;
1203
1204     /* Update outer/inner flops */
1205
1206     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1207 }