src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSw_VdwNone_GeomW4P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_ElecEwSw_VdwNone_GeomW4P1_VF_sse4_1_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSw_VdwNone_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              vdwioffset1;
  80     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  81     int              vdwioffset2;
  82     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  83     int              vdwioffset3;
  84     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  88     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128i          ewitab;
  93     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  96     real             rswitch_scalar,d_scalar;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->ic->epsfac);
 113     charge           = mdatoms->chargeA;
 114
 115     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 116     ewtab            = fr->ic->tabq_coul_FDV0;
 117     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 118     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 123     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 124     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->ic->rcoulomb;
 128     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 130
 131     rswitch_scalar   = fr->ic->rcoulomb_switch;
 132     rswitch          = _mm_set1_pd(rswitch_scalar);
 133     /* Setup switch parameters */
 134     d_scalar         = rcutoff_scalar-rswitch_scalar;
 135     d                = _mm_set1_pd(d_scalar);
 136     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 137     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 140     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 167                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 168
 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
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx10             = _mm_sub_pd(ix1,jx0);
 198             dy10             = _mm_sub_pd(iy1,jy0);
 199             dz10             = _mm_sub_pd(iz1,jz0);
 200             dx20             = _mm_sub_pd(ix2,jx0);
 201             dy20             = _mm_sub_pd(iy2,jy0);
 202             dz20             = _mm_sub_pd(iz2,jz0);
 203             dx30             = _mm_sub_pd(ix3,jx0);
 204             dy30             = _mm_sub_pd(iy3,jy0);
 205             dz30             = _mm_sub_pd(iz3,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 209             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 210             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 211
 212             rinv10           = sse41_invsqrt_d(rsq10);
 213             rinv20           = sse41_invsqrt_d(rsq20);
 214             rinv30           = sse41_invsqrt_d(rsq30);
 215
 216             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 217             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 218             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 222
 223             fjx0             = _mm_setzero_pd();
 224             fjy0             = _mm_setzero_pd();
 225             fjz0             = _mm_setzero_pd();
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             if (gmx_mm_any_lt(rsq10,rcutoff2))
 232             {
 233
 234             r10              = _mm_mul_pd(rsq10,rinv10);
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq10             = _mm_mul_pd(iq1,jq0);
 238
 239             /* EWALD ELECTROSTATICS */
 240
 241             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 242             ewrt             = _mm_mul_pd(r10,ewtabscale);
 243             ewitab           = _mm_cvttpd_epi32(ewrt);
 244             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 245             ewitab           = _mm_slli_epi32(ewitab,2);
 246             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 247             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 248             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 249             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 250             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 251             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 252             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 253             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 254             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 255             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 256
 257             d                = _mm_sub_pd(r10,rswitch);
 258             d                = _mm_max_pd(d,_mm_setzero_pd());
 259             d2               = _mm_mul_pd(d,d);
 260             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 261
 262             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 263
 264             /* Evaluate switch function */
 265             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 266             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 267             velec            = _mm_mul_pd(velec,sw);
 268             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 269
 270             /* Update potential sum for this i atom from the interaction with this j atom. */
 271             velec            = _mm_and_pd(velec,cutoff_mask);
 272             velecsum         = _mm_add_pd(velecsum,velec);
 273
 274             fscal            = felec;
 275
 276             fscal            = _mm_and_pd(fscal,cutoff_mask);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_pd(fscal,dx10);
 280             ty               = _mm_mul_pd(fscal,dy10);
 281             tz               = _mm_mul_pd(fscal,dz10);
 282
 283             /* Update vectorial force */
 284             fix1             = _mm_add_pd(fix1,tx);
 285             fiy1             = _mm_add_pd(fiy1,ty);
 286             fiz1             = _mm_add_pd(fiz1,tz);
 287
 288             fjx0             = _mm_add_pd(fjx0,tx);
 289             fjy0             = _mm_add_pd(fjy0,ty);
 290             fjz0             = _mm_add_pd(fjz0,tz);
 291
 292             }
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq20,rcutoff2))
 299             {
 300
 301             r20              = _mm_mul_pd(rsq20,rinv20);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq20             = _mm_mul_pd(iq2,jq0);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _mm_mul_pd(r20,ewtabscale);
 310             ewitab           = _mm_cvttpd_epi32(ewrt);
 311             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 312             ewitab           = _mm_slli_epi32(ewitab,2);
 313             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 314             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 315             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 316             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 317             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 318             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 319             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 320             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 321             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 322             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 323
 324             d                = _mm_sub_pd(r20,rswitch);
 325             d                = _mm_max_pd(d,_mm_setzero_pd());
 326             d2               = _mm_mul_pd(d,d);
 327             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 328
 329             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 330
 331             /* Evaluate switch function */
 332             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 333             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 334             velec            = _mm_mul_pd(velec,sw);
 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             r30              = _mm_mul_pd(rsq30,rinv30);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq30             = _mm_mul_pd(iq3,jq0);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 376             ewrt             = _mm_mul_pd(r30,ewtabscale);
 377             ewitab           = _mm_cvttpd_epi32(ewrt);
 378             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 379             ewitab           = _mm_slli_epi32(ewitab,2);
 380             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 381             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 382             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 383             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 384             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 385             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 386             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 387             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 388             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 389             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 390
 391             d                = _mm_sub_pd(r30,rswitch);
 392             d                = _mm_max_pd(d,_mm_setzero_pd());
 393             d2               = _mm_mul_pd(d,d);
 394             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 395
 396             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 397
 398             /* Evaluate switch function */
 399             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 400             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 401             velec            = _mm_mul_pd(velec,sw);
 402             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_and_pd(velec,cutoff_mask);
 406             velecsum         = _mm_add_pd(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             fscal            = _mm_and_pd(fscal,cutoff_mask);
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm_mul_pd(fscal,dx30);
 414             ty               = _mm_mul_pd(fscal,dy30);
 415             tz               = _mm_mul_pd(fscal,dz30);
 416
 417             /* Update vectorial force */
 418             fix3             = _mm_add_pd(fix3,tx);
 419             fiy3             = _mm_add_pd(fiy3,ty);
 420             fiz3             = _mm_add_pd(fiz3,tz);
 421
 422             fjx0             = _mm_add_pd(fjx0,tx);
 423             fjy0             = _mm_add_pd(fjy0,ty);
 424             fjz0             = _mm_add_pd(fjz0,tz);
 425
 426             }
 427
 428             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 429
 430             /* Inner loop uses 198 flops */
 431         }
 432
 433         if(jidx<j_index_end)
 434         {
 435
 436             jnrA             = jjnr[jidx];
 437             j_coord_offsetA  = DIM*jnrA;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 441                                               &jx0,&jy0,&jz0);
 442
 443             /* Calculate displacement vector */
 444             dx10             = _mm_sub_pd(ix1,jx0);
 445             dy10             = _mm_sub_pd(iy1,jy0);
 446             dz10             = _mm_sub_pd(iz1,jz0);
 447             dx20             = _mm_sub_pd(ix2,jx0);
 448             dy20             = _mm_sub_pd(iy2,jy0);
 449             dz20             = _mm_sub_pd(iz2,jz0);
 450             dx30             = _mm_sub_pd(ix3,jx0);
 451             dy30             = _mm_sub_pd(iy3,jy0);
 452             dz30             = _mm_sub_pd(iz3,jz0);
 453
 454             /* Calculate squared distance and things based on it */
 455             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 456             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 457             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 458
 459             rinv10           = sse41_invsqrt_d(rsq10);
 460             rinv20           = sse41_invsqrt_d(rsq20);
 461             rinv30           = sse41_invsqrt_d(rsq30);
 462
 463             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 464             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 465             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 466
 467             /* Load parameters for j particles */
 468             jq0              = _mm_load_sd(charge+jnrA+0);
 469
 470             fjx0             = _mm_setzero_pd();
 471             fjy0             = _mm_setzero_pd();
 472             fjz0             = _mm_setzero_pd();
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             if (gmx_mm_any_lt(rsq10,rcutoff2))
 479             {
 480
 481             r10              = _mm_mul_pd(rsq10,rinv10);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq10             = _mm_mul_pd(iq1,jq0);
 485
 486             /* EWALD ELECTROSTATICS */
 487
 488             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 489             ewrt             = _mm_mul_pd(r10,ewtabscale);
 490             ewitab           = _mm_cvttpd_epi32(ewrt);
 491             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 492             ewitab           = _mm_slli_epi32(ewitab,2);
 493             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 494             ewtabD           = _mm_setzero_pd();
 495             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 496             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 497             ewtabFn          = _mm_setzero_pd();
 498             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 499             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 500             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 501             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 502             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 503
 504             d                = _mm_sub_pd(r10,rswitch);
 505             d                = _mm_max_pd(d,_mm_setzero_pd());
 506             d2               = _mm_mul_pd(d,d);
 507             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 508
 509             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 510
 511             /* Evaluate switch function */
 512             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 513             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 514             velec            = _mm_mul_pd(velec,sw);
 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             r20              = _mm_mul_pd(rsq20,rinv20);
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq20             = _mm_mul_pd(iq2,jq0);
 555
 556             /* EWALD ELECTROSTATICS */
 557
 558             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 559             ewrt             = _mm_mul_pd(r20,ewtabscale);
 560             ewitab           = _mm_cvttpd_epi32(ewrt);
 561             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 562             ewitab           = _mm_slli_epi32(ewitab,2);
 563             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 564             ewtabD           = _mm_setzero_pd();
 565             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 566             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 567             ewtabFn          = _mm_setzero_pd();
 568             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 569             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 570             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 571             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 572             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 573
 574             d                = _mm_sub_pd(r20,rswitch);
 575             d                = _mm_max_pd(d,_mm_setzero_pd());
 576             d2               = _mm_mul_pd(d,d);
 577             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 578
 579             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 580
 581             /* Evaluate switch function */
 582             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 583             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 584             velec            = _mm_mul_pd(velec,sw);
 585             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velec            = _mm_and_pd(velec,cutoff_mask);
 589             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 590             velecsum         = _mm_add_pd(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             fscal            = _mm_and_pd(fscal,cutoff_mask);
 595
 596             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm_mul_pd(fscal,dx20);
 600             ty               = _mm_mul_pd(fscal,dy20);
 601             tz               = _mm_mul_pd(fscal,dz20);
 602
 603             /* Update vectorial force */
 604             fix2             = _mm_add_pd(fix2,tx);
 605             fiy2             = _mm_add_pd(fiy2,ty);
 606             fiz2             = _mm_add_pd(fiz2,tz);
 607
 608             fjx0             = _mm_add_pd(fjx0,tx);
 609             fjy0             = _mm_add_pd(fjy0,ty);
 610             fjz0             = _mm_add_pd(fjz0,tz);
 611
 612             }
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             if (gmx_mm_any_lt(rsq30,rcutoff2))
 619             {
 620
 621             r30              = _mm_mul_pd(rsq30,rinv30);
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq30             = _mm_mul_pd(iq3,jq0);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm_mul_pd(r30,ewtabscale);
 630             ewitab           = _mm_cvttpd_epi32(ewrt);
 631             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 632             ewitab           = _mm_slli_epi32(ewitab,2);
 633             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 634             ewtabD           = _mm_setzero_pd();
 635             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 636             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 637             ewtabFn          = _mm_setzero_pd();
 638             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 639             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 640             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 641             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 642             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 643
 644             d                = _mm_sub_pd(r30,rswitch);
 645             d                = _mm_max_pd(d,_mm_setzero_pd());
 646             d2               = _mm_mul_pd(d,d);
 647             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 648
 649             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 650
 651             /* Evaluate switch function */
 652             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 653             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 654             velec            = _mm_mul_pd(velec,sw);
 655             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 656
 657             /* Update potential sum for this i atom from the interaction with this j atom. */
 658             velec            = _mm_and_pd(velec,cutoff_mask);
 659             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 660             velecsum         = _mm_add_pd(velecsum,velec);
 661
 662             fscal            = felec;
 663
 664             fscal            = _mm_and_pd(fscal,cutoff_mask);
 665
 666             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm_mul_pd(fscal,dx30);
 670             ty               = _mm_mul_pd(fscal,dy30);
 671             tz               = _mm_mul_pd(fscal,dz30);
 672
 673             /* Update vectorial force */
 674             fix3             = _mm_add_pd(fix3,tx);
 675             fiy3             = _mm_add_pd(fiy3,ty);
 676             fiz3             = _mm_add_pd(fiz3,tz);
 677
 678             fjx0             = _mm_add_pd(fjx0,tx);
 679             fjy0             = _mm_add_pd(fjy0,ty);
 680             fjz0             = _mm_add_pd(fjz0,tz);
 681
 682             }
 683
 684             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 685
 686             /* Inner loop uses 198 flops */
 687         }
 688
 689         /* End of innermost loop */
 690
 691         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 692                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 693
 694         ggid                        = gid[iidx];
 695         /* Update potential energies */
 696         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 697
 698         /* Increment number of inner iterations */
 699         inneriter                  += j_index_end - j_index_start;
 700
 701         /* Outer loop uses 19 flops */
 702     }
 703
 704     /* Increment number of outer iterations */
 705     outeriter        += nri;
 706
 707     /* Update outer/inner flops */
 708
 709     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*198);
 710 }
 711 /*
 712  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse4_1_double
 713  * Electrostatics interaction: Ewald
 714  * VdW interaction:            None
 715  * Geometry:                   Water4-Particle
 716  * Calculate force/pot:        Force
 717  */
 718 void
 719 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse4_1_double
 720                     (t_nblist                    * gmx_restrict       nlist,
 721                      rvec                        * gmx_restrict          xx,
 722                      rvec                        * gmx_restrict          ff,
 723                      struct t_forcerec           * gmx_restrict          fr,
 724                      t_mdatoms                   * gmx_restrict     mdatoms,
 725                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 726                      t_nrnb                      * gmx_restrict        nrnb)
 727 {
 728     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 729      * just 0 for non-waters.
 730      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 731      * jnr indices corresponding to data put in the four positions in the SIMD register.
 732      */
 733     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 734     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 735     int              jnrA,jnrB;
 736     int              j_coord_offsetA,j_coord_offsetB;
 737     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 738     real             rcutoff_scalar;
 739     real             *shiftvec,*fshift,*x,*f;
 740     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 741     int              vdwioffset1;
 742     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 743     int              vdwioffset2;
 744     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 745     int              vdwioffset3;
 746     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 747     int              vdwjidx0A,vdwjidx0B;
 748     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 749     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 750     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 751     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 752     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 753     real             *charge;
 754     __m128i          ewitab;
 755     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 756     real             *ewtab;
 757     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 758     real             rswitch_scalar,d_scalar;
 759     __m128d          dummy_mask,cutoff_mask;
 760     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 761     __m128d          one     = _mm_set1_pd(1.0);
 762     __m128d          two     = _mm_set1_pd(2.0);
 763     x                = xx[0];
 764     f                = ff[0];
 765
 766     nri              = nlist->nri;
 767     iinr             = nlist->iinr;
 768     jindex           = nlist->jindex;
 769     jjnr             = nlist->jjnr;
 770     shiftidx         = nlist->shift;
 771     gid              = nlist->gid;
 772     shiftvec         = fr->shift_vec[0];
 773     fshift           = fr->fshift[0];
 774     facel            = _mm_set1_pd(fr->ic->epsfac);
 775     charge           = mdatoms->chargeA;
 776
 777     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 778     ewtab            = fr->ic->tabq_coul_FDV0;
 779     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 780     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 781
 782     /* Setup water-specific parameters */
 783     inr              = nlist->iinr[0];
 784     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 785     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 786     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 787
 788     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 789     rcutoff_scalar   = fr->ic->rcoulomb;
 790     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 791     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 792
 793     rswitch_scalar   = fr->ic->rcoulomb_switch;
 794     rswitch          = _mm_set1_pd(rswitch_scalar);
 795     /* Setup switch parameters */
 796     d_scalar         = rcutoff_scalar-rswitch_scalar;
 797     d                = _mm_set1_pd(d_scalar);
 798     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 799     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 800     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 801     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 802     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 803     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 804
 805     /* Avoid stupid compiler warnings */
 806     jnrA = jnrB = 0;
 807     j_coord_offsetA = 0;
 808     j_coord_offsetB = 0;
 809
 810     outeriter        = 0;
 811     inneriter        = 0;
 812
 813     /* Start outer loop over neighborlists */
 814     for(iidx=0; iidx<nri; iidx++)
 815     {
 816         /* Load shift vector for this list */
 817         i_shift_offset   = DIM*shiftidx[iidx];
 818
 819         /* Load limits for loop over neighbors */
 820         j_index_start    = jindex[iidx];
 821         j_index_end      = jindex[iidx+1];
 822
 823         /* Get outer coordinate index */
 824         inr              = iinr[iidx];
 825         i_coord_offset   = DIM*inr;
 826
 827         /* Load i particle coords and add shift vector */
 828         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 829                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 830
 831         fix1             = _mm_setzero_pd();
 832         fiy1             = _mm_setzero_pd();
 833         fiz1             = _mm_setzero_pd();
 834         fix2             = _mm_setzero_pd();
 835         fiy2             = _mm_setzero_pd();
 836         fiz2             = _mm_setzero_pd();
 837         fix3             = _mm_setzero_pd();
 838         fiy3             = _mm_setzero_pd();
 839         fiz3             = _mm_setzero_pd();
 840
 841         /* Start inner kernel loop */
 842         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 843         {
 844
 845             /* Get j neighbor index, and coordinate index */
 846             jnrA             = jjnr[jidx];
 847             jnrB             = jjnr[jidx+1];
 848             j_coord_offsetA  = DIM*jnrA;
 849             j_coord_offsetB  = DIM*jnrB;
 850
 851             /* load j atom coordinates */
 852             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 853                                               &jx0,&jy0,&jz0);
 854
 855             /* Calculate displacement vector */
 856             dx10             = _mm_sub_pd(ix1,jx0);
 857             dy10             = _mm_sub_pd(iy1,jy0);
 858             dz10             = _mm_sub_pd(iz1,jz0);
 859             dx20             = _mm_sub_pd(ix2,jx0);
 860             dy20             = _mm_sub_pd(iy2,jy0);
 861             dz20             = _mm_sub_pd(iz2,jz0);
 862             dx30             = _mm_sub_pd(ix3,jx0);
 863             dy30             = _mm_sub_pd(iy3,jy0);
 864             dz30             = _mm_sub_pd(iz3,jz0);
 865
 866             /* Calculate squared distance and things based on it */
 867             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 868             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 869             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 870
 871             rinv10           = sse41_invsqrt_d(rsq10);
 872             rinv20           = sse41_invsqrt_d(rsq20);
 873             rinv30           = sse41_invsqrt_d(rsq30);
 874
 875             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 876             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 877             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 878
 879             /* Load parameters for j particles */
 880             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 881
 882             fjx0             = _mm_setzero_pd();
 883             fjy0             = _mm_setzero_pd();
 884             fjz0             = _mm_setzero_pd();
 885
 886             /**************************
 887              * CALCULATE INTERACTIONS *
 888              **************************/
 889
 890             if (gmx_mm_any_lt(rsq10,rcutoff2))
 891             {
 892
 893             r10              = _mm_mul_pd(rsq10,rinv10);
 894
 895             /* Compute parameters for interactions between i and j atoms */
 896             qq10             = _mm_mul_pd(iq1,jq0);
 897
 898             /* EWALD ELECTROSTATICS */
 899
 900             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 901             ewrt             = _mm_mul_pd(r10,ewtabscale);
 902             ewitab           = _mm_cvttpd_epi32(ewrt);
 903             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 904             ewitab           = _mm_slli_epi32(ewitab,2);
 905             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 906             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 907             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 908             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 909             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 910             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 911             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 912             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 913             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 914             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 915
 916             d                = _mm_sub_pd(r10,rswitch);
 917             d                = _mm_max_pd(d,_mm_setzero_pd());
 918             d2               = _mm_mul_pd(d,d);
 919             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 920
 921             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 922
 923             /* Evaluate switch function */
 924             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 925             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 926             cutoff_mask      = _mm_cmplt_pd(rsq10,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,dx10);
 934             ty               = _mm_mul_pd(fscal,dy10);
 935             tz               = _mm_mul_pd(fscal,dz10);
 936
 937             /* Update vectorial force */
 938             fix1             = _mm_add_pd(fix1,tx);
 939             fiy1             = _mm_add_pd(fiy1,ty);
 940             fiz1             = _mm_add_pd(fiz1,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(rsq20,rcutoff2))
 953             {
 954
 955             r20              = _mm_mul_pd(rsq20,rinv20);
 956
 957             /* Compute parameters for interactions between i and j atoms */
 958             qq20             = _mm_mul_pd(iq2,jq0);
 959
 960             /* EWALD ELECTROSTATICS */
 961
 962             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 963             ewrt             = _mm_mul_pd(r20,ewtabscale);
 964             ewitab           = _mm_cvttpd_epi32(ewrt);
 965             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 966             ewitab           = _mm_slli_epi32(ewitab,2);
 967             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 968             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 969             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 970             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 971             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 972             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 973             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 974             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 975             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 976             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 977
 978             d                = _mm_sub_pd(r20,rswitch);
 979             d                = _mm_max_pd(d,_mm_setzero_pd());
 980             d2               = _mm_mul_pd(d,d);
 981             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 982
 983             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 984
 985             /* Evaluate switch function */
 986             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 987             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 988             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 989
 990             fscal            = felec;
 991
 992             fscal            = _mm_and_pd(fscal,cutoff_mask);
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm_mul_pd(fscal,dx20);
 996             ty               = _mm_mul_pd(fscal,dy20);
 997             tz               = _mm_mul_pd(fscal,dz20);
 998
 999             /* Update vectorial force */
1000             fix2             = _mm_add_pd(fix2,tx);
1001             fiy2             = _mm_add_pd(fiy2,ty);
1002             fiz2             = _mm_add_pd(fiz2,tz);
1003
1004             fjx0             = _mm_add_pd(fjx0,tx);
1005             fjy0             = _mm_add_pd(fjy0,ty);
1006             fjz0             = _mm_add_pd(fjz0,tz);
1007
1008             }
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             if (gmx_mm_any_lt(rsq30,rcutoff2))
1015             {
1016
1017             r30              = _mm_mul_pd(rsq30,rinv30);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq30             = _mm_mul_pd(iq3,jq0);
1021
1022             /* EWALD ELECTROSTATICS */
1023
1024             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1025             ewrt             = _mm_mul_pd(r30,ewtabscale);
1026             ewitab           = _mm_cvttpd_epi32(ewrt);
1027             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1028             ewitab           = _mm_slli_epi32(ewitab,2);
1029             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1030             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1031             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1032             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1033             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1034             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1035             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1036             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1037             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1038             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1039
1040             d                = _mm_sub_pd(r30,rswitch);
1041             d                = _mm_max_pd(d,_mm_setzero_pd());
1042             d2               = _mm_mul_pd(d,d);
1043             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1044
1045             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1046
1047             /* Evaluate switch function */
1048             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1049             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1050             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm_and_pd(fscal,cutoff_mask);
1055
1056             /* Calculate temporary vectorial force */
1057             tx               = _mm_mul_pd(fscal,dx30);
1058             ty               = _mm_mul_pd(fscal,dy30);
1059             tz               = _mm_mul_pd(fscal,dz30);
1060
1061             /* Update vectorial force */
1062             fix3             = _mm_add_pd(fix3,tx);
1063             fiy3             = _mm_add_pd(fiy3,ty);
1064             fiz3             = _mm_add_pd(fiz3,tz);
1065
1066             fjx0             = _mm_add_pd(fjx0,tx);
1067             fjy0             = _mm_add_pd(fjy0,ty);
1068             fjz0             = _mm_add_pd(fjz0,tz);
1069
1070             }
1071
1072             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1073
1074             /* Inner loop uses 189 flops */
1075         }
1076
1077         if(jidx<j_index_end)
1078         {
1079
1080             jnrA             = jjnr[jidx];
1081             j_coord_offsetA  = DIM*jnrA;
1082
1083             /* load j atom coordinates */
1084             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1085                                               &jx0,&jy0,&jz0);
1086
1087             /* Calculate displacement vector */
1088             dx10             = _mm_sub_pd(ix1,jx0);
1089             dy10             = _mm_sub_pd(iy1,jy0);
1090             dz10             = _mm_sub_pd(iz1,jz0);
1091             dx20             = _mm_sub_pd(ix2,jx0);
1092             dy20             = _mm_sub_pd(iy2,jy0);
1093             dz20             = _mm_sub_pd(iz2,jz0);
1094             dx30             = _mm_sub_pd(ix3,jx0);
1095             dy30             = _mm_sub_pd(iy3,jy0);
1096             dz30             = _mm_sub_pd(iz3,jz0);
1097
1098             /* Calculate squared distance and things based on it */
1099             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1100             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1101             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1102
1103             rinv10           = sse41_invsqrt_d(rsq10);
1104             rinv20           = sse41_invsqrt_d(rsq20);
1105             rinv30           = sse41_invsqrt_d(rsq30);
1106
1107             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1108             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1109             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1110
1111             /* Load parameters for j particles */
1112             jq0              = _mm_load_sd(charge+jnrA+0);
1113
1114             fjx0             = _mm_setzero_pd();
1115             fjy0             = _mm_setzero_pd();
1116             fjz0             = _mm_setzero_pd();
1117
1118             /**************************
1119              * CALCULATE INTERACTIONS *
1120              **************************/
1121
1122             if (gmx_mm_any_lt(rsq10,rcutoff2))
1123             {
1124
1125             r10              = _mm_mul_pd(rsq10,rinv10);
1126
1127             /* Compute parameters for interactions between i and j atoms */
1128             qq10             = _mm_mul_pd(iq1,jq0);
1129
1130             /* EWALD ELECTROSTATICS */
1131
1132             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1133             ewrt             = _mm_mul_pd(r10,ewtabscale);
1134             ewitab           = _mm_cvttpd_epi32(ewrt);
1135             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1136             ewitab           = _mm_slli_epi32(ewitab,2);
1137             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1138             ewtabD           = _mm_setzero_pd();
1139             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1140             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1141             ewtabFn          = _mm_setzero_pd();
1142             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1143             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1144             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1145             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1146             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1147
1148             d                = _mm_sub_pd(r10,rswitch);
1149             d                = _mm_max_pd(d,_mm_setzero_pd());
1150             d2               = _mm_mul_pd(d,d);
1151             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1152
1153             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1154
1155             /* Evaluate switch function */
1156             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1157             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1158             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1159
1160             fscal            = felec;
1161
1162             fscal            = _mm_and_pd(fscal,cutoff_mask);
1163
1164             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1165
1166             /* Calculate temporary vectorial force */
1167             tx               = _mm_mul_pd(fscal,dx10);
1168             ty               = _mm_mul_pd(fscal,dy10);
1169             tz               = _mm_mul_pd(fscal,dz10);
1170
1171             /* Update vectorial force */
1172             fix1             = _mm_add_pd(fix1,tx);
1173             fiy1             = _mm_add_pd(fiy1,ty);
1174             fiz1             = _mm_add_pd(fiz1,tz);
1175
1176             fjx0             = _mm_add_pd(fjx0,tx);
1177             fjy0             = _mm_add_pd(fjy0,ty);
1178             fjz0             = _mm_add_pd(fjz0,tz);
1179
1180             }
1181
1182             /**************************
1183              * CALCULATE INTERACTIONS *
1184              **************************/
1185
1186             if (gmx_mm_any_lt(rsq20,rcutoff2))
1187             {
1188
1189             r20              = _mm_mul_pd(rsq20,rinv20);
1190
1191             /* Compute parameters for interactions between i and j atoms */
1192             qq20             = _mm_mul_pd(iq2,jq0);
1193
1194             /* EWALD ELECTROSTATICS */
1195
1196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1197             ewrt             = _mm_mul_pd(r20,ewtabscale);
1198             ewitab           = _mm_cvttpd_epi32(ewrt);
1199             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1200             ewitab           = _mm_slli_epi32(ewitab,2);
1201             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1202             ewtabD           = _mm_setzero_pd();
1203             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1204             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1205             ewtabFn          = _mm_setzero_pd();
1206             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1207             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1208             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1209             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1210             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1211
1212             d                = _mm_sub_pd(r20,rswitch);
1213             d                = _mm_max_pd(d,_mm_setzero_pd());
1214             d2               = _mm_mul_pd(d,d);
1215             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1216
1217             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1218
1219             /* Evaluate switch function */
1220             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1221             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1222             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1223
1224             fscal            = felec;
1225
1226             fscal            = _mm_and_pd(fscal,cutoff_mask);
1227
1228             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm_mul_pd(fscal,dx20);
1232             ty               = _mm_mul_pd(fscal,dy20);
1233             tz               = _mm_mul_pd(fscal,dz20);
1234
1235             /* Update vectorial force */
1236             fix2             = _mm_add_pd(fix2,tx);
1237             fiy2             = _mm_add_pd(fiy2,ty);
1238             fiz2             = _mm_add_pd(fiz2,tz);
1239
1240             fjx0             = _mm_add_pd(fjx0,tx);
1241             fjy0             = _mm_add_pd(fjy0,ty);
1242             fjz0             = _mm_add_pd(fjz0,tz);
1243
1244             }
1245
1246             /**************************
1247              * CALCULATE INTERACTIONS *
1248              **************************/
1249
1250             if (gmx_mm_any_lt(rsq30,rcutoff2))
1251             {
1252
1253             r30              = _mm_mul_pd(rsq30,rinv30);
1254
1255             /* Compute parameters for interactions between i and j atoms */
1256             qq30             = _mm_mul_pd(iq3,jq0);
1257
1258             /* EWALD ELECTROSTATICS */
1259
1260             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1261             ewrt             = _mm_mul_pd(r30,ewtabscale);
1262             ewitab           = _mm_cvttpd_epi32(ewrt);
1263             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1264             ewitab           = _mm_slli_epi32(ewitab,2);
1265             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1266             ewtabD           = _mm_setzero_pd();
1267             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1268             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1269             ewtabFn          = _mm_setzero_pd();
1270             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1271             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1272             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1273             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1274             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1275
1276             d                = _mm_sub_pd(r30,rswitch);
1277             d                = _mm_max_pd(d,_mm_setzero_pd());
1278             d2               = _mm_mul_pd(d,d);
1279             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1280
1281             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1282
1283             /* Evaluate switch function */
1284             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1285             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1286             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1287
1288             fscal            = felec;
1289
1290             fscal            = _mm_and_pd(fscal,cutoff_mask);
1291
1292             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1293
1294             /* Calculate temporary vectorial force */
1295             tx               = _mm_mul_pd(fscal,dx30);
1296             ty               = _mm_mul_pd(fscal,dy30);
1297             tz               = _mm_mul_pd(fscal,dz30);
1298
1299             /* Update vectorial force */
1300             fix3             = _mm_add_pd(fix3,tx);
1301             fiy3             = _mm_add_pd(fiy3,ty);
1302             fiz3             = _mm_add_pd(fiz3,tz);
1303
1304             fjx0             = _mm_add_pd(fjx0,tx);
1305             fjy0             = _mm_add_pd(fjy0,ty);
1306             fjz0             = _mm_add_pd(fjz0,tz);
1307
1308             }
1309
1310             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1311
1312             /* Inner loop uses 189 flops */
1313         }
1314
1315         /* End of innermost loop */
1316
1317         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1318                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1319
1320         /* Increment number of inner iterations */
1321         inneriter                  += j_index_end - j_index_start;
1322
1323         /* Outer loop uses 18 flops */
1324     }
1325
1326     /* Increment number of outer iterations */
1327     outeriter        += nri;
1328
1329     /* Update outer/inner flops */
1330
1331     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*189);
1332 }