src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSw_VdwNone_GeomW4P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset1;
  81     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  82     int              vdwioffset2;
  83     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  84     int              vdwioffset3;
  85     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128i          ewitab;
  94     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  96     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  97     real             rswitch_scalar,d_scalar;
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 124     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 125     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 126
 127     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 130     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 131
 132     rswitch_scalar   = fr->rcoulomb_switch;
 133     rswitch          = _mm_set1_pd(rswitch_scalar);
 134     /* Setup switch parameters */
 135     d_scalar         = rcutoff_scalar-rswitch_scalar;
 136     d                = _mm_set1_pd(d_scalar);
 137     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 138     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 141     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 142     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 168                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 169
 170         fix1             = _mm_setzero_pd();
 171         fiy1             = _mm_setzero_pd();
 172         fiz1             = _mm_setzero_pd();
 173         fix2             = _mm_setzero_pd();
 174         fiy2             = _mm_setzero_pd();
 175         fiz2             = _mm_setzero_pd();
 176         fix3             = _mm_setzero_pd();
 177         fiy3             = _mm_setzero_pd();
 178         fiz3             = _mm_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _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             dx10             = _mm_sub_pd(ix1,jx0);
 199             dy10             = _mm_sub_pd(iy1,jy0);
 200             dz10             = _mm_sub_pd(iz1,jz0);
 201             dx20             = _mm_sub_pd(ix2,jx0);
 202             dy20             = _mm_sub_pd(iy2,jy0);
 203             dz20             = _mm_sub_pd(iz2,jz0);
 204             dx30             = _mm_sub_pd(ix3,jx0);
 205             dy30             = _mm_sub_pd(iy3,jy0);
 206             dz30             = _mm_sub_pd(iz3,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 211             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 212
 213             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 214             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 215             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 216
 217             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 218             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 219             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 223
 224             fjx0             = _mm_setzero_pd();
 225             fjy0             = _mm_setzero_pd();
 226             fjz0             = _mm_setzero_pd();
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             if (gmx_mm_any_lt(rsq10,rcutoff2))
 233             {
 234
 235             r10              = _mm_mul_pd(rsq10,rinv10);
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq10             = _mm_mul_pd(iq1,jq0);
 239
 240             /* EWALD ELECTROSTATICS */
 241
 242             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 243             ewrt             = _mm_mul_pd(r10,ewtabscale);
 244             ewitab           = _mm_cvttpd_epi32(ewrt);
 245             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 246             ewitab           = _mm_slli_epi32(ewitab,2);
 247             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 248             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 249             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 250             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 251             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 252             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 253             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 254             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 255             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 256             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 257
 258             d                = _mm_sub_pd(r10,rswitch);
 259             d                = _mm_max_pd(d,_mm_setzero_pd());
 260             d2               = _mm_mul_pd(d,d);
 261             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)))))));
 262
 263             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 264
 265             /* Evaluate switch function */
 266             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 267             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 268             velec            = _mm_mul_pd(velec,sw);
 269             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 270
 271             /* Update potential sum for this i atom from the interaction with this j atom. */
 272             velec            = _mm_and_pd(velec,cutoff_mask);
 273             velecsum         = _mm_add_pd(velecsum,velec);
 274
 275             fscal            = felec;
 276
 277             fscal            = _mm_and_pd(fscal,cutoff_mask);
 278
 279             /* Calculate temporary vectorial force */
 280             tx               = _mm_mul_pd(fscal,dx10);
 281             ty               = _mm_mul_pd(fscal,dy10);
 282             tz               = _mm_mul_pd(fscal,dz10);
 283
 284             /* Update vectorial force */
 285             fix1             = _mm_add_pd(fix1,tx);
 286             fiy1             = _mm_add_pd(fiy1,ty);
 287             fiz1             = _mm_add_pd(fiz1,tz);
 288
 289             fjx0             = _mm_add_pd(fjx0,tx);
 290             fjy0             = _mm_add_pd(fjy0,ty);
 291             fjz0             = _mm_add_pd(fjz0,tz);
 292
 293             }
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             if (gmx_mm_any_lt(rsq20,rcutoff2))
 300             {
 301
 302             r20              = _mm_mul_pd(rsq20,rinv20);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq20             = _mm_mul_pd(iq2,jq0);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm_mul_pd(r20,ewtabscale);
 311             ewitab           = _mm_cvttpd_epi32(ewrt);
 312             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 313             ewitab           = _mm_slli_epi32(ewitab,2);
 314             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 315             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 316             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 317             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 318             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 319             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 320             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 321             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 322             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 323             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 324
 325             d                = _mm_sub_pd(r20,rswitch);
 326             d                = _mm_max_pd(d,_mm_setzero_pd());
 327             d2               = _mm_mul_pd(d,d);
 328             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)))))));
 329
 330             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 331
 332             /* Evaluate switch function */
 333             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 334             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 335             velec            = _mm_mul_pd(velec,sw);
 336             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_and_pd(velec,cutoff_mask);
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm_and_pd(fscal,cutoff_mask);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_pd(fscal,dx20);
 348             ty               = _mm_mul_pd(fscal,dy20);
 349             tz               = _mm_mul_pd(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_pd(fix2,tx);
 353             fiy2             = _mm_add_pd(fiy2,ty);
 354             fiz2             = _mm_add_pd(fiz2,tz);
 355
 356             fjx0             = _mm_add_pd(fjx0,tx);
 357             fjy0             = _mm_add_pd(fjy0,ty);
 358             fjz0             = _mm_add_pd(fjz0,tz);
 359
 360             }
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm_any_lt(rsq30,rcutoff2))
 367             {
 368
 369             r30              = _mm_mul_pd(rsq30,rinv30);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq30             = _mm_mul_pd(iq3,jq0);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 377             ewrt             = _mm_mul_pd(r30,ewtabscale);
 378             ewitab           = _mm_cvttpd_epi32(ewrt);
 379             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 380             ewitab           = _mm_slli_epi32(ewitab,2);
 381             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 382             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 383             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 384             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 385             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 386             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 387             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 388             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 389             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 390             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 391
 392             d                = _mm_sub_pd(r30,rswitch);
 393             d                = _mm_max_pd(d,_mm_setzero_pd());
 394             d2               = _mm_mul_pd(d,d);
 395             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)))))));
 396
 397             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 398
 399             /* Evaluate switch function */
 400             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 401             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 402             velec            = _mm_mul_pd(velec,sw);
 403             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velec            = _mm_and_pd(velec,cutoff_mask);
 407             velecsum         = _mm_add_pd(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             fscal            = _mm_and_pd(fscal,cutoff_mask);
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm_mul_pd(fscal,dx30);
 415             ty               = _mm_mul_pd(fscal,dy30);
 416             tz               = _mm_mul_pd(fscal,dz30);
 417
 418             /* Update vectorial force */
 419             fix3             = _mm_add_pd(fix3,tx);
 420             fiy3             = _mm_add_pd(fiy3,ty);
 421             fiz3             = _mm_add_pd(fiz3,tz);
 422
 423             fjx0             = _mm_add_pd(fjx0,tx);
 424             fjy0             = _mm_add_pd(fjy0,ty);
 425             fjz0             = _mm_add_pd(fjz0,tz);
 426
 427             }
 428
 429             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 430
 431             /* Inner loop uses 198 flops */
 432         }
 433
 434         if(jidx<j_index_end)
 435         {
 436
 437             jnrA             = jjnr[jidx];
 438             j_coord_offsetA  = DIM*jnrA;
 439
 440             /* load j atom coordinates */
 441             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 442                                               &jx0,&jy0,&jz0);
 443
 444             /* Calculate displacement vector */
 445             dx10             = _mm_sub_pd(ix1,jx0);
 446             dy10             = _mm_sub_pd(iy1,jy0);
 447             dz10             = _mm_sub_pd(iz1,jz0);
 448             dx20             = _mm_sub_pd(ix2,jx0);
 449             dy20             = _mm_sub_pd(iy2,jy0);
 450             dz20             = _mm_sub_pd(iz2,jz0);
 451             dx30             = _mm_sub_pd(ix3,jx0);
 452             dy30             = _mm_sub_pd(iy3,jy0);
 453             dz30             = _mm_sub_pd(iz3,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 457             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 458             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 459
 460             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 461             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 462             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 463
 464             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 465             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 466             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 467
 468             /* Load parameters for j particles */
 469             jq0              = _mm_load_sd(charge+jnrA+0);
 470
 471             fjx0             = _mm_setzero_pd();
 472             fjy0             = _mm_setzero_pd();
 473             fjz0             = _mm_setzero_pd();
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             if (gmx_mm_any_lt(rsq10,rcutoff2))
 480             {
 481
 482             r10              = _mm_mul_pd(rsq10,rinv10);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq10             = _mm_mul_pd(iq1,jq0);
 486
 487             /* EWALD ELECTROSTATICS */
 488
 489             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 490             ewrt             = _mm_mul_pd(r10,ewtabscale);
 491             ewitab           = _mm_cvttpd_epi32(ewrt);
 492             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 493             ewitab           = _mm_slli_epi32(ewitab,2);
 494             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 495             ewtabD           = _mm_setzero_pd();
 496             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 497             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 498             ewtabFn          = _mm_setzero_pd();
 499             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 500             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 501             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 502             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 503             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 504
 505             d                = _mm_sub_pd(r10,rswitch);
 506             d                = _mm_max_pd(d,_mm_setzero_pd());
 507             d2               = _mm_mul_pd(d,d);
 508             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)))))));
 509
 510             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 511
 512             /* Evaluate switch function */
 513             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 514             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 515             velec            = _mm_mul_pd(velec,sw);
 516             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _mm_and_pd(velec,cutoff_mask);
 520             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 521             velecsum         = _mm_add_pd(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             fscal            = _mm_and_pd(fscal,cutoff_mask);
 526
 527             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_pd(fscal,dx10);
 531             ty               = _mm_mul_pd(fscal,dy10);
 532             tz               = _mm_mul_pd(fscal,dz10);
 533
 534             /* Update vectorial force */
 535             fix1             = _mm_add_pd(fix1,tx);
 536             fiy1             = _mm_add_pd(fiy1,ty);
 537             fiz1             = _mm_add_pd(fiz1,tz);
 538
 539             fjx0             = _mm_add_pd(fjx0,tx);
 540             fjy0             = _mm_add_pd(fjy0,ty);
 541             fjz0             = _mm_add_pd(fjz0,tz);
 542
 543             }
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq20,rcutoff2))
 550             {
 551
 552             r20              = _mm_mul_pd(rsq20,rinv20);
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq20             = _mm_mul_pd(iq2,jq0);
 556
 557             /* EWALD ELECTROSTATICS */
 558
 559             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 560             ewrt             = _mm_mul_pd(r20,ewtabscale);
 561             ewitab           = _mm_cvttpd_epi32(ewrt);
 562             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 563             ewitab           = _mm_slli_epi32(ewitab,2);
 564             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 565             ewtabD           = _mm_setzero_pd();
 566             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 567             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 568             ewtabFn          = _mm_setzero_pd();
 569             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 570             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 571             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 572             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 573             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 574
 575             d                = _mm_sub_pd(r20,rswitch);
 576             d                = _mm_max_pd(d,_mm_setzero_pd());
 577             d2               = _mm_mul_pd(d,d);
 578             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)))))));
 579
 580             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 581
 582             /* Evaluate switch function */
 583             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 584             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 585             velec            = _mm_mul_pd(velec,sw);
 586             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             velec            = _mm_and_pd(velec,cutoff_mask);
 590             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 591             velecsum         = _mm_add_pd(velecsum,velec);
 592
 593             fscal            = felec;
 594
 595             fscal            = _mm_and_pd(fscal,cutoff_mask);
 596
 597             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm_mul_pd(fscal,dx20);
 601             ty               = _mm_mul_pd(fscal,dy20);
 602             tz               = _mm_mul_pd(fscal,dz20);
 603
 604             /* Update vectorial force */
 605             fix2             = _mm_add_pd(fix2,tx);
 606             fiy2             = _mm_add_pd(fiy2,ty);
 607             fiz2             = _mm_add_pd(fiz2,tz);
 608
 609             fjx0             = _mm_add_pd(fjx0,tx);
 610             fjy0             = _mm_add_pd(fjy0,ty);
 611             fjz0             = _mm_add_pd(fjz0,tz);
 612
 613             }
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm_any_lt(rsq30,rcutoff2))
 620             {
 621
 622             r30              = _mm_mul_pd(rsq30,rinv30);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq30             = _mm_mul_pd(iq3,jq0);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 630             ewrt             = _mm_mul_pd(r30,ewtabscale);
 631             ewitab           = _mm_cvttpd_epi32(ewrt);
 632             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 633             ewitab           = _mm_slli_epi32(ewitab,2);
 634             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 635             ewtabD           = _mm_setzero_pd();
 636             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 637             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 638             ewtabFn          = _mm_setzero_pd();
 639             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 640             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 641             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 642             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 643             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 644
 645             d                = _mm_sub_pd(r30,rswitch);
 646             d                = _mm_max_pd(d,_mm_setzero_pd());
 647             d2               = _mm_mul_pd(d,d);
 648             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)))))));
 649
 650             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 651
 652             /* Evaluate switch function */
 653             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 654             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 655             velec            = _mm_mul_pd(velec,sw);
 656             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 657
 658             /* Update potential sum for this i atom from the interaction with this j atom. */
 659             velec            = _mm_and_pd(velec,cutoff_mask);
 660             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 661             velecsum         = _mm_add_pd(velecsum,velec);
 662
 663             fscal            = felec;
 664
 665             fscal            = _mm_and_pd(fscal,cutoff_mask);
 666
 667             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 668
 669             /* Calculate temporary vectorial force */
 670             tx               = _mm_mul_pd(fscal,dx30);
 671             ty               = _mm_mul_pd(fscal,dy30);
 672             tz               = _mm_mul_pd(fscal,dz30);
 673
 674             /* Update vectorial force */
 675             fix3             = _mm_add_pd(fix3,tx);
 676             fiy3             = _mm_add_pd(fiy3,ty);
 677             fiz3             = _mm_add_pd(fiz3,tz);
 678
 679             fjx0             = _mm_add_pd(fjx0,tx);
 680             fjy0             = _mm_add_pd(fjy0,ty);
 681             fjz0             = _mm_add_pd(fjz0,tz);
 682
 683             }
 684
 685             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 686
 687             /* Inner loop uses 198 flops */
 688         }
 689
 690         /* End of innermost loop */
 691
 692         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 693                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 694
 695         ggid                        = gid[iidx];
 696         /* Update potential energies */
 697         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 698
 699         /* Increment number of inner iterations */
 700         inneriter                  += j_index_end - j_index_start;
 701
 702         /* Outer loop uses 19 flops */
 703     }
 704
 705     /* Increment number of outer iterations */
 706     outeriter        += nri;
 707
 708     /* Update outer/inner flops */
 709
 710     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*198);
 711 }
 712 /*
 713  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse2_double
 714  * Electrostatics interaction: Ewald
 715  * VdW interaction:            None
 716  * Geometry:                   Water4-Particle
 717  * Calculate force/pot:        Force
 718  */
 719 void
 720 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse2_double
 721                     (t_nblist                    * gmx_restrict       nlist,
 722                      rvec                        * gmx_restrict          xx,
 723                      rvec                        * gmx_restrict          ff,
 724                      t_forcerec                  * gmx_restrict          fr,
 725                      t_mdatoms                   * gmx_restrict     mdatoms,
 726                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 727                      t_nrnb                      * gmx_restrict        nrnb)
 728 {
 729     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 730      * just 0 for non-waters.
 731      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 732      * jnr indices corresponding to data put in the four positions in the SIMD register.
 733      */
 734     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 735     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 736     int              jnrA,jnrB;
 737     int              j_coord_offsetA,j_coord_offsetB;
 738     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 739     real             rcutoff_scalar;
 740     real             *shiftvec,*fshift,*x,*f;
 741     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 742     int              vdwioffset1;
 743     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 744     int              vdwioffset2;
 745     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 746     int              vdwioffset3;
 747     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 748     int              vdwjidx0A,vdwjidx0B;
 749     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 750     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 751     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 752     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 753     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 754     real             *charge;
 755     __m128i          ewitab;
 756     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 757     real             *ewtab;
 758     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 759     real             rswitch_scalar,d_scalar;
 760     __m128d          dummy_mask,cutoff_mask;
 761     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 762     __m128d          one     = _mm_set1_pd(1.0);
 763     __m128d          two     = _mm_set1_pd(2.0);
 764     x                = xx[0];
 765     f                = ff[0];
 766
 767     nri              = nlist->nri;
 768     iinr             = nlist->iinr;
 769     jindex           = nlist->jindex;
 770     jjnr             = nlist->jjnr;
 771     shiftidx         = nlist->shift;
 772     gid              = nlist->gid;
 773     shiftvec         = fr->shift_vec[0];
 774     fshift           = fr->fshift[0];
 775     facel            = _mm_set1_pd(fr->epsfac);
 776     charge           = mdatoms->chargeA;
 777
 778     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 779     ewtab            = fr->ic->tabq_coul_FDV0;
 780     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 781     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 782
 783     /* Setup water-specific parameters */
 784     inr              = nlist->iinr[0];
 785     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 786     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 787     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 788
 789     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 790     rcutoff_scalar   = fr->rcoulomb;
 791     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 792     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 793
 794     rswitch_scalar   = fr->rcoulomb_switch;
 795     rswitch          = _mm_set1_pd(rswitch_scalar);
 796     /* Setup switch parameters */
 797     d_scalar         = rcutoff_scalar-rswitch_scalar;
 798     d                = _mm_set1_pd(d_scalar);
 799     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 800     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 801     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 802     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 803     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 804     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 805
 806     /* Avoid stupid compiler warnings */
 807     jnrA = jnrB = 0;
 808     j_coord_offsetA = 0;
 809     j_coord_offsetB = 0;
 810
 811     outeriter        = 0;
 812     inneriter        = 0;
 813
 814     /* Start outer loop over neighborlists */
 815     for(iidx=0; iidx<nri; iidx++)
 816     {
 817         /* Load shift vector for this list */
 818         i_shift_offset   = DIM*shiftidx[iidx];
 819
 820         /* Load limits for loop over neighbors */
 821         j_index_start    = jindex[iidx];
 822         j_index_end      = jindex[iidx+1];
 823
 824         /* Get outer coordinate index */
 825         inr              = iinr[iidx];
 826         i_coord_offset   = DIM*inr;
 827
 828         /* Load i particle coords and add shift vector */
 829         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 830                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 831
 832         fix1             = _mm_setzero_pd();
 833         fiy1             = _mm_setzero_pd();
 834         fiz1             = _mm_setzero_pd();
 835         fix2             = _mm_setzero_pd();
 836         fiy2             = _mm_setzero_pd();
 837         fiz2             = _mm_setzero_pd();
 838         fix3             = _mm_setzero_pd();
 839         fiy3             = _mm_setzero_pd();
 840         fiz3             = _mm_setzero_pd();
 841
 842         /* Start inner kernel loop */
 843         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 844         {
 845
 846             /* Get j neighbor index, and coordinate index */
 847             jnrA             = jjnr[jidx];
 848             jnrB             = jjnr[jidx+1];
 849             j_coord_offsetA  = DIM*jnrA;
 850             j_coord_offsetB  = DIM*jnrB;
 851
 852             /* load j atom coordinates */
 853             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 854                                               &jx0,&jy0,&jz0);
 855
 856             /* Calculate displacement vector */
 857             dx10             = _mm_sub_pd(ix1,jx0);
 858             dy10             = _mm_sub_pd(iy1,jy0);
 859             dz10             = _mm_sub_pd(iz1,jz0);
 860             dx20             = _mm_sub_pd(ix2,jx0);
 861             dy20             = _mm_sub_pd(iy2,jy0);
 862             dz20             = _mm_sub_pd(iz2,jz0);
 863             dx30             = _mm_sub_pd(ix3,jx0);
 864             dy30             = _mm_sub_pd(iy3,jy0);
 865             dz30             = _mm_sub_pd(iz3,jz0);
 866
 867             /* Calculate squared distance and things based on it */
 868             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 869             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 870             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 871
 872             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 873             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 874             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 875
 876             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 877             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 878             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 879
 880             /* Load parameters for j particles */
 881             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 882
 883             fjx0             = _mm_setzero_pd();
 884             fjy0             = _mm_setzero_pd();
 885             fjz0             = _mm_setzero_pd();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             if (gmx_mm_any_lt(rsq10,rcutoff2))
 892             {
 893
 894             r10              = _mm_mul_pd(rsq10,rinv10);
 895
 896             /* Compute parameters for interactions between i and j atoms */
 897             qq10             = _mm_mul_pd(iq1,jq0);
 898
 899             /* EWALD ELECTROSTATICS */
 900
 901             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 902             ewrt             = _mm_mul_pd(r10,ewtabscale);
 903             ewitab           = _mm_cvttpd_epi32(ewrt);
 904             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 905             ewitab           = _mm_slli_epi32(ewitab,2);
 906             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 907             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 908             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 909             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 910             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 911             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 912             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 913             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 914             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 915             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 916
 917             d                = _mm_sub_pd(r10,rswitch);
 918             d                = _mm_max_pd(d,_mm_setzero_pd());
 919             d2               = _mm_mul_pd(d,d);
 920             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)))))));
 921
 922             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 923
 924             /* Evaluate switch function */
 925             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 926             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 927             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm_and_pd(fscal,cutoff_mask);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm_mul_pd(fscal,dx10);
 935             ty               = _mm_mul_pd(fscal,dy10);
 936             tz               = _mm_mul_pd(fscal,dz10);
 937
 938             /* Update vectorial force */
 939             fix1             = _mm_add_pd(fix1,tx);
 940             fiy1             = _mm_add_pd(fiy1,ty);
 941             fiz1             = _mm_add_pd(fiz1,tz);
 942
 943             fjx0             = _mm_add_pd(fjx0,tx);
 944             fjy0             = _mm_add_pd(fjy0,ty);
 945             fjz0             = _mm_add_pd(fjz0,tz);
 946
 947             }
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             if (gmx_mm_any_lt(rsq20,rcutoff2))
 954             {
 955
 956             r20              = _mm_mul_pd(rsq20,rinv20);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq20             = _mm_mul_pd(iq2,jq0);
 960
 961             /* EWALD ELECTROSTATICS */
 962
 963             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 964             ewrt             = _mm_mul_pd(r20,ewtabscale);
 965             ewitab           = _mm_cvttpd_epi32(ewrt);
 966             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 967             ewitab           = _mm_slli_epi32(ewitab,2);
 968             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 969             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 970             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 971             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 972             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 973             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 974             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 975             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 976             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 977             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 978
 979             d                = _mm_sub_pd(r20,rswitch);
 980             d                = _mm_max_pd(d,_mm_setzero_pd());
 981             d2               = _mm_mul_pd(d,d);
 982             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)))))));
 983
 984             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 985
 986             /* Evaluate switch function */
 987             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 988             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 989             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm_and_pd(fscal,cutoff_mask);
 994
 995             /* Calculate temporary vectorial force */
 996             tx               = _mm_mul_pd(fscal,dx20);
 997             ty               = _mm_mul_pd(fscal,dy20);
 998             tz               = _mm_mul_pd(fscal,dz20);
 999
1000             /* Update vectorial force */
1001             fix2             = _mm_add_pd(fix2,tx);
1002             fiy2             = _mm_add_pd(fiy2,ty);
1003             fiz2             = _mm_add_pd(fiz2,tz);
1004
1005             fjx0             = _mm_add_pd(fjx0,tx);
1006             fjy0             = _mm_add_pd(fjy0,ty);
1007             fjz0             = _mm_add_pd(fjz0,tz);
1008
1009             }
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             if (gmx_mm_any_lt(rsq30,rcutoff2))
1016             {
1017
1018             r30              = _mm_mul_pd(rsq30,rinv30);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq30             = _mm_mul_pd(iq3,jq0);
1022
1023             /* EWALD ELECTROSTATICS */
1024
1025             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026             ewrt             = _mm_mul_pd(r30,ewtabscale);
1027             ewitab           = _mm_cvttpd_epi32(ewrt);
1028             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1029             ewitab           = _mm_slli_epi32(ewitab,2);
1030             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1031             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1032             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1033             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1034             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1035             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1036             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1037             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1038             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1039             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1040
1041             d                = _mm_sub_pd(r30,rswitch);
1042             d                = _mm_max_pd(d,_mm_setzero_pd());
1043             d2               = _mm_mul_pd(d,d);
1044             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)))))));
1045
1046             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1047
1048             /* Evaluate switch function */
1049             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1050             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1051             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1052
1053             fscal            = felec;
1054
1055             fscal            = _mm_and_pd(fscal,cutoff_mask);
1056
1057             /* Calculate temporary vectorial force */
1058             tx               = _mm_mul_pd(fscal,dx30);
1059             ty               = _mm_mul_pd(fscal,dy30);
1060             tz               = _mm_mul_pd(fscal,dz30);
1061
1062             /* Update vectorial force */
1063             fix3             = _mm_add_pd(fix3,tx);
1064             fiy3             = _mm_add_pd(fiy3,ty);
1065             fiz3             = _mm_add_pd(fiz3,tz);
1066
1067             fjx0             = _mm_add_pd(fjx0,tx);
1068             fjy0             = _mm_add_pd(fjy0,ty);
1069             fjz0             = _mm_add_pd(fjz0,tz);
1070
1071             }
1072
1073             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1074
1075             /* Inner loop uses 189 flops */
1076         }
1077
1078         if(jidx<j_index_end)
1079         {
1080
1081             jnrA             = jjnr[jidx];
1082             j_coord_offsetA  = DIM*jnrA;
1083
1084             /* load j atom coordinates */
1085             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1086                                               &jx0,&jy0,&jz0);
1087
1088             /* Calculate displacement vector */
1089             dx10             = _mm_sub_pd(ix1,jx0);
1090             dy10             = _mm_sub_pd(iy1,jy0);
1091             dz10             = _mm_sub_pd(iz1,jz0);
1092             dx20             = _mm_sub_pd(ix2,jx0);
1093             dy20             = _mm_sub_pd(iy2,jy0);
1094             dz20             = _mm_sub_pd(iz2,jz0);
1095             dx30             = _mm_sub_pd(ix3,jx0);
1096             dy30             = _mm_sub_pd(iy3,jy0);
1097             dz30             = _mm_sub_pd(iz3,jz0);
1098
1099             /* Calculate squared distance and things based on it */
1100             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1101             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1102             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1103
1104             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1105             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1106             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1107
1108             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1109             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1110             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1111
1112             /* Load parameters for j particles */
1113             jq0              = _mm_load_sd(charge+jnrA+0);
1114
1115             fjx0             = _mm_setzero_pd();
1116             fjy0             = _mm_setzero_pd();
1117             fjz0             = _mm_setzero_pd();
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             if (gmx_mm_any_lt(rsq10,rcutoff2))
1124             {
1125
1126             r10              = _mm_mul_pd(rsq10,rinv10);
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             qq10             = _mm_mul_pd(iq1,jq0);
1130
1131             /* EWALD ELECTROSTATICS */
1132
1133             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1134             ewrt             = _mm_mul_pd(r10,ewtabscale);
1135             ewitab           = _mm_cvttpd_epi32(ewrt);
1136             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1137             ewitab           = _mm_slli_epi32(ewitab,2);
1138             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1139             ewtabD           = _mm_setzero_pd();
1140             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1141             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1142             ewtabFn          = _mm_setzero_pd();
1143             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1144             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1145             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1146             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1147             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1148
1149             d                = _mm_sub_pd(r10,rswitch);
1150             d                = _mm_max_pd(d,_mm_setzero_pd());
1151             d2               = _mm_mul_pd(d,d);
1152             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)))))));
1153
1154             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1155
1156             /* Evaluate switch function */
1157             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1158             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1159             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_and_pd(fscal,cutoff_mask);
1164
1165             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm_mul_pd(fscal,dx10);
1169             ty               = _mm_mul_pd(fscal,dy10);
1170             tz               = _mm_mul_pd(fscal,dz10);
1171
1172             /* Update vectorial force */
1173             fix1             = _mm_add_pd(fix1,tx);
1174             fiy1             = _mm_add_pd(fiy1,ty);
1175             fiz1             = _mm_add_pd(fiz1,tz);
1176
1177             fjx0             = _mm_add_pd(fjx0,tx);
1178             fjy0             = _mm_add_pd(fjy0,ty);
1179             fjz0             = _mm_add_pd(fjz0,tz);
1180
1181             }
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             if (gmx_mm_any_lt(rsq20,rcutoff2))
1188             {
1189
1190             r20              = _mm_mul_pd(rsq20,rinv20);
1191
1192             /* Compute parameters for interactions between i and j atoms */
1193             qq20             = _mm_mul_pd(iq2,jq0);
1194
1195             /* EWALD ELECTROSTATICS */
1196
1197             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1198             ewrt             = _mm_mul_pd(r20,ewtabscale);
1199             ewitab           = _mm_cvttpd_epi32(ewrt);
1200             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1201             ewitab           = _mm_slli_epi32(ewitab,2);
1202             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1203             ewtabD           = _mm_setzero_pd();
1204             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1205             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1206             ewtabFn          = _mm_setzero_pd();
1207             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1208             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1209             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1210             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1211             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1212
1213             d                = _mm_sub_pd(r20,rswitch);
1214             d                = _mm_max_pd(d,_mm_setzero_pd());
1215             d2               = _mm_mul_pd(d,d);
1216             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)))))));
1217
1218             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1219
1220             /* Evaluate switch function */
1221             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1222             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1223             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1224
1225             fscal            = felec;
1226
1227             fscal            = _mm_and_pd(fscal,cutoff_mask);
1228
1229             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1230
1231             /* Calculate temporary vectorial force */
1232             tx               = _mm_mul_pd(fscal,dx20);
1233             ty               = _mm_mul_pd(fscal,dy20);
1234             tz               = _mm_mul_pd(fscal,dz20);
1235
1236             /* Update vectorial force */
1237             fix2             = _mm_add_pd(fix2,tx);
1238             fiy2             = _mm_add_pd(fiy2,ty);
1239             fiz2             = _mm_add_pd(fiz2,tz);
1240
1241             fjx0             = _mm_add_pd(fjx0,tx);
1242             fjy0             = _mm_add_pd(fjy0,ty);
1243             fjz0             = _mm_add_pd(fjz0,tz);
1244
1245             }
1246
1247             /**************************
1248              * CALCULATE INTERACTIONS *
1249              **************************/
1250
1251             if (gmx_mm_any_lt(rsq30,rcutoff2))
1252             {
1253
1254             r30              = _mm_mul_pd(rsq30,rinv30);
1255
1256             /* Compute parameters for interactions between i and j atoms */
1257             qq30             = _mm_mul_pd(iq3,jq0);
1258
1259             /* EWALD ELECTROSTATICS */
1260
1261             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1262             ewrt             = _mm_mul_pd(r30,ewtabscale);
1263             ewitab           = _mm_cvttpd_epi32(ewrt);
1264             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1265             ewitab           = _mm_slli_epi32(ewitab,2);
1266             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1267             ewtabD           = _mm_setzero_pd();
1268             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1269             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1270             ewtabFn          = _mm_setzero_pd();
1271             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1272             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1273             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1274             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1275             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1276
1277             d                = _mm_sub_pd(r30,rswitch);
1278             d                = _mm_max_pd(d,_mm_setzero_pd());
1279             d2               = _mm_mul_pd(d,d);
1280             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)))))));
1281
1282             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1283
1284             /* Evaluate switch function */
1285             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1286             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1287             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1288
1289             fscal            = felec;
1290
1291             fscal            = _mm_and_pd(fscal,cutoff_mask);
1292
1293             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1294
1295             /* Calculate temporary vectorial force */
1296             tx               = _mm_mul_pd(fscal,dx30);
1297             ty               = _mm_mul_pd(fscal,dy30);
1298             tz               = _mm_mul_pd(fscal,dz30);
1299
1300             /* Update vectorial force */
1301             fix3             = _mm_add_pd(fix3,tx);
1302             fiy3             = _mm_add_pd(fiy3,ty);
1303             fiz3             = _mm_add_pd(fiz3,tz);
1304
1305             fjx0             = _mm_add_pd(fjx0,tx);
1306             fjy0             = _mm_add_pd(fjy0,ty);
1307             fjz0             = _mm_add_pd(fjz0,tz);
1308
1309             }
1310
1311             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1312
1313             /* Inner loop uses 189 flops */
1314         }
1315
1316         /* End of innermost loop */
1317
1318         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1319                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1320
1321         /* Increment number of inner iterations */
1322         inneriter                  += j_index_end - j_index_start;
1323
1324         /* Outer loop uses 18 flops */
1325     }
1326
1327     /* Increment number of outer iterations */
1328     outeriter        += nri;
1329
1330     /* Update outer/inner flops */
1331
1332     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*189);
1333 }