src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwNone_GeomW4P1_sse2_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_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_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_sse2_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_sse2_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          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->ic->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 114     ewtab            = fr->ic->tabq_coul_FDV0;
 115     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 116     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 117
 118     /* Setup water-specific parameters */
 119     inr              = nlist->iinr[0];
 120     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 121     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 122     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 148                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 149
 150         fix1             = _mm_setzero_pd();
 151         fiy1             = _mm_setzero_pd();
 152         fiz1             = _mm_setzero_pd();
 153         fix2             = _mm_setzero_pd();
 154         fiy2             = _mm_setzero_pd();
 155         fiz2             = _mm_setzero_pd();
 156         fix3             = _mm_setzero_pd();
 157         fiy3             = _mm_setzero_pd();
 158         fiz3             = _mm_setzero_pd();
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx10             = _mm_sub_pd(ix1,jx0);
 179             dy10             = _mm_sub_pd(iy1,jy0);
 180             dz10             = _mm_sub_pd(iz1,jz0);
 181             dx20             = _mm_sub_pd(ix2,jx0);
 182             dy20             = _mm_sub_pd(iy2,jy0);
 183             dz20             = _mm_sub_pd(iz2,jz0);
 184             dx30             = _mm_sub_pd(ix3,jx0);
 185             dy30             = _mm_sub_pd(iy3,jy0);
 186             dz30             = _mm_sub_pd(iz3,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 190             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 191             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 192
 193             rinv10           = sse2_invsqrt_d(rsq10);
 194             rinv20           = sse2_invsqrt_d(rsq20);
 195             rinv30           = sse2_invsqrt_d(rsq30);
 196
 197             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 198             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 199             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 203
 204             fjx0             = _mm_setzero_pd();
 205             fjy0             = _mm_setzero_pd();
 206             fjz0             = _mm_setzero_pd();
 207
 208             /**************************
 209              * CALCULATE INTERACTIONS *
 210              **************************/
 211
 212             r10              = _mm_mul_pd(rsq10,rinv10);
 213
 214             /* Compute parameters for interactions between i and j atoms */
 215             qq10             = _mm_mul_pd(iq1,jq0);
 216
 217             /* EWALD ELECTROSTATICS */
 218
 219             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 220             ewrt             = _mm_mul_pd(r10,ewtabscale);
 221             ewitab           = _mm_cvttpd_epi32(ewrt);
 222             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 223             ewitab           = _mm_slli_epi32(ewitab,2);
 224             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 225             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 226             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 227             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 228             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 229             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 230             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 231             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 232             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 233             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 234
 235             /* Update potential sum for this i atom from the interaction with this j atom. */
 236             velecsum         = _mm_add_pd(velecsum,velec);
 237
 238             fscal            = felec;
 239
 240             /* Calculate temporary vectorial force */
 241             tx               = _mm_mul_pd(fscal,dx10);
 242             ty               = _mm_mul_pd(fscal,dy10);
 243             tz               = _mm_mul_pd(fscal,dz10);
 244
 245             /* Update vectorial force */
 246             fix1             = _mm_add_pd(fix1,tx);
 247             fiy1             = _mm_add_pd(fiy1,ty);
 248             fiz1             = _mm_add_pd(fiz1,tz);
 249
 250             fjx0             = _mm_add_pd(fjx0,tx);
 251             fjy0             = _mm_add_pd(fjy0,ty);
 252             fjz0             = _mm_add_pd(fjz0,tz);
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             r20              = _mm_mul_pd(rsq20,rinv20);
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq20             = _mm_mul_pd(iq2,jq0);
 262
 263             /* EWALD ELECTROSTATICS */
 264
 265             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 266             ewrt             = _mm_mul_pd(r20,ewtabscale);
 267             ewitab           = _mm_cvttpd_epi32(ewrt);
 268             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 269             ewitab           = _mm_slli_epi32(ewitab,2);
 270             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 271             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 272             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 273             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 274             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 275             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 276             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 277             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 278             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 279             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_pd(velecsum,velec);
 283
 284             fscal            = felec;
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_pd(fscal,dx20);
 288             ty               = _mm_mul_pd(fscal,dy20);
 289             tz               = _mm_mul_pd(fscal,dz20);
 290
 291             /* Update vectorial force */
 292             fix2             = _mm_add_pd(fix2,tx);
 293             fiy2             = _mm_add_pd(fiy2,ty);
 294             fiz2             = _mm_add_pd(fiz2,tz);
 295
 296             fjx0             = _mm_add_pd(fjx0,tx);
 297             fjy0             = _mm_add_pd(fjy0,ty);
 298             fjz0             = _mm_add_pd(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r30              = _mm_mul_pd(rsq30,rinv30);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq30             = _mm_mul_pd(iq3,jq0);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _mm_mul_pd(r30,ewtabscale);
 313             ewitab           = _mm_cvttpd_epi32(ewrt);
 314             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 315             ewitab           = _mm_slli_epi32(ewitab,2);
 316             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 317             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 318             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 319             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 320             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 321             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 322             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 323             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 324             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 325             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_pd(fscal,dx30);
 334             ty               = _mm_mul_pd(fscal,dy30);
 335             tz               = _mm_mul_pd(fscal,dz30);
 336
 337             /* Update vectorial force */
 338             fix3             = _mm_add_pd(fix3,tx);
 339             fiy3             = _mm_add_pd(fiy3,ty);
 340             fiz3             = _mm_add_pd(fiz3,tz);
 341
 342             fjx0             = _mm_add_pd(fjx0,tx);
 343             fjy0             = _mm_add_pd(fjy0,ty);
 344             fjz0             = _mm_add_pd(fjz0,tz);
 345
 346             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 347
 348             /* Inner loop uses 126 flops */
 349         }
 350
 351         if(jidx<j_index_end)
 352         {
 353
 354             jnrA             = jjnr[jidx];
 355             j_coord_offsetA  = DIM*jnrA;
 356
 357             /* load j atom coordinates */
 358             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 359                                               &jx0,&jy0,&jz0);
 360
 361             /* Calculate displacement vector */
 362             dx10             = _mm_sub_pd(ix1,jx0);
 363             dy10             = _mm_sub_pd(iy1,jy0);
 364             dz10             = _mm_sub_pd(iz1,jz0);
 365             dx20             = _mm_sub_pd(ix2,jx0);
 366             dy20             = _mm_sub_pd(iy2,jy0);
 367             dz20             = _mm_sub_pd(iz2,jz0);
 368             dx30             = _mm_sub_pd(ix3,jx0);
 369             dy30             = _mm_sub_pd(iy3,jy0);
 370             dz30             = _mm_sub_pd(iz3,jz0);
 371
 372             /* Calculate squared distance and things based on it */
 373             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 374             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 375             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 376
 377             rinv10           = sse2_invsqrt_d(rsq10);
 378             rinv20           = sse2_invsqrt_d(rsq20);
 379             rinv30           = sse2_invsqrt_d(rsq30);
 380
 381             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 382             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 383             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 384
 385             /* Load parameters for j particles */
 386             jq0              = _mm_load_sd(charge+jnrA+0);
 387
 388             fjx0             = _mm_setzero_pd();
 389             fjy0             = _mm_setzero_pd();
 390             fjz0             = _mm_setzero_pd();
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             r10              = _mm_mul_pd(rsq10,rinv10);
 397
 398             /* Compute parameters for interactions between i and j atoms */
 399             qq10             = _mm_mul_pd(iq1,jq0);
 400
 401             /* EWALD ELECTROSTATICS */
 402
 403             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 404             ewrt             = _mm_mul_pd(r10,ewtabscale);
 405             ewitab           = _mm_cvttpd_epi32(ewrt);
 406             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 407             ewitab           = _mm_slli_epi32(ewitab,2);
 408             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 409             ewtabD           = _mm_setzero_pd();
 410             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 411             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 412             ewtabFn          = _mm_setzero_pd();
 413             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 414             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 415             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 416             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 417             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 421             velecsum         = _mm_add_pd(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm_mul_pd(fscal,dx10);
 429             ty               = _mm_mul_pd(fscal,dy10);
 430             tz               = _mm_mul_pd(fscal,dz10);
 431
 432             /* Update vectorial force */
 433             fix1             = _mm_add_pd(fix1,tx);
 434             fiy1             = _mm_add_pd(fiy1,ty);
 435             fiz1             = _mm_add_pd(fiz1,tz);
 436
 437             fjx0             = _mm_add_pd(fjx0,tx);
 438             fjy0             = _mm_add_pd(fjy0,ty);
 439             fjz0             = _mm_add_pd(fjz0,tz);
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             r20              = _mm_mul_pd(rsq20,rinv20);
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             qq20             = _mm_mul_pd(iq2,jq0);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 453             ewrt             = _mm_mul_pd(r20,ewtabscale);
 454             ewitab           = _mm_cvttpd_epi32(ewrt);
 455             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 456             ewitab           = _mm_slli_epi32(ewitab,2);
 457             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 458             ewtabD           = _mm_setzero_pd();
 459             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 460             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 461             ewtabFn          = _mm_setzero_pd();
 462             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 463             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 464             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 465             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 466             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 467
 468             /* Update potential sum for this i atom from the interaction with this j atom. */
 469             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 470             velecsum         = _mm_add_pd(velecsum,velec);
 471
 472             fscal            = felec;
 473
 474             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 475
 476             /* Calculate temporary vectorial force */
 477             tx               = _mm_mul_pd(fscal,dx20);
 478             ty               = _mm_mul_pd(fscal,dy20);
 479             tz               = _mm_mul_pd(fscal,dz20);
 480
 481             /* Update vectorial force */
 482             fix2             = _mm_add_pd(fix2,tx);
 483             fiy2             = _mm_add_pd(fiy2,ty);
 484             fiz2             = _mm_add_pd(fiz2,tz);
 485
 486             fjx0             = _mm_add_pd(fjx0,tx);
 487             fjy0             = _mm_add_pd(fjy0,ty);
 488             fjz0             = _mm_add_pd(fjz0,tz);
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             r30              = _mm_mul_pd(rsq30,rinv30);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq30             = _mm_mul_pd(iq3,jq0);
 498
 499             /* EWALD ELECTROSTATICS */
 500
 501             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 502             ewrt             = _mm_mul_pd(r30,ewtabscale);
 503             ewitab           = _mm_cvttpd_epi32(ewrt);
 504             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 505             ewitab           = _mm_slli_epi32(ewitab,2);
 506             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 507             ewtabD           = _mm_setzero_pd();
 508             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 509             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 510             ewtabFn          = _mm_setzero_pd();
 511             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 512             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 513             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 514             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 515             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 519             velecsum         = _mm_add_pd(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 524
 525             /* Calculate temporary vectorial force */
 526             tx               = _mm_mul_pd(fscal,dx30);
 527             ty               = _mm_mul_pd(fscal,dy30);
 528             tz               = _mm_mul_pd(fscal,dz30);
 529
 530             /* Update vectorial force */
 531             fix3             = _mm_add_pd(fix3,tx);
 532             fiy3             = _mm_add_pd(fiy3,ty);
 533             fiz3             = _mm_add_pd(fiz3,tz);
 534
 535             fjx0             = _mm_add_pd(fjx0,tx);
 536             fjy0             = _mm_add_pd(fjy0,ty);
 537             fjz0             = _mm_add_pd(fjz0,tz);
 538
 539             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 540
 541             /* Inner loop uses 126 flops */
 542         }
 543
 544         /* End of innermost loop */
 545
 546         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 547                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 548
 549         ggid                        = gid[iidx];
 550         /* Update potential energies */
 551         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 552
 553         /* Increment number of inner iterations */
 554         inneriter                  += j_index_end - j_index_start;
 555
 556         /* Outer loop uses 19 flops */
 557     }
 558
 559     /* Increment number of outer iterations */
 560     outeriter        += nri;
 561
 562     /* Update outer/inner flops */
 563
 564     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*126);
 565 }
 566 /*
 567  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_double
 568  * Electrostatics interaction: Ewald
 569  * VdW interaction:            None
 570  * Geometry:                   Water4-Particle
 571  * Calculate force/pot:        Force
 572  */
 573 void
 574 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_double
 575                     (t_nblist                    * gmx_restrict       nlist,
 576                      rvec                        * gmx_restrict          xx,
 577                      rvec                        * gmx_restrict          ff,
 578                      struct t_forcerec           * gmx_restrict          fr,
 579                      t_mdatoms                   * gmx_restrict     mdatoms,
 580                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 581                      t_nrnb                      * gmx_restrict        nrnb)
 582 {
 583     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 584      * just 0 for non-waters.
 585      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 586      * jnr indices corresponding to data put in the four positions in the SIMD register.
 587      */
 588     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 589     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 590     int              jnrA,jnrB;
 591     int              j_coord_offsetA,j_coord_offsetB;
 592     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 593     real             rcutoff_scalar;
 594     real             *shiftvec,*fshift,*x,*f;
 595     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 596     int              vdwioffset1;
 597     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 598     int              vdwioffset2;
 599     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 600     int              vdwioffset3;
 601     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 602     int              vdwjidx0A,vdwjidx0B;
 603     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 604     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 605     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 606     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 607     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 608     real             *charge;
 609     __m128i          ewitab;
 610     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 611     real             *ewtab;
 612     __m128d          dummy_mask,cutoff_mask;
 613     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 614     __m128d          one     = _mm_set1_pd(1.0);
 615     __m128d          two     = _mm_set1_pd(2.0);
 616     x                = xx[0];
 617     f                = ff[0];
 618
 619     nri              = nlist->nri;
 620     iinr             = nlist->iinr;
 621     jindex           = nlist->jindex;
 622     jjnr             = nlist->jjnr;
 623     shiftidx         = nlist->shift;
 624     gid              = nlist->gid;
 625     shiftvec         = fr->shift_vec[0];
 626     fshift           = fr->fshift[0];
 627     facel            = _mm_set1_pd(fr->ic->epsfac);
 628     charge           = mdatoms->chargeA;
 629
 630     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 631     ewtab            = fr->ic->tabq_coul_F;
 632     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 633     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 634
 635     /* Setup water-specific parameters */
 636     inr              = nlist->iinr[0];
 637     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 638     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 639     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 640
 641     /* Avoid stupid compiler warnings */
 642     jnrA = jnrB = 0;
 643     j_coord_offsetA = 0;
 644     j_coord_offsetB = 0;
 645
 646     outeriter        = 0;
 647     inneriter        = 0;
 648
 649     /* Start outer loop over neighborlists */
 650     for(iidx=0; iidx<nri; iidx++)
 651     {
 652         /* Load shift vector for this list */
 653         i_shift_offset   = DIM*shiftidx[iidx];
 654
 655         /* Load limits for loop over neighbors */
 656         j_index_start    = jindex[iidx];
 657         j_index_end      = jindex[iidx+1];
 658
 659         /* Get outer coordinate index */
 660         inr              = iinr[iidx];
 661         i_coord_offset   = DIM*inr;
 662
 663         /* Load i particle coords and add shift vector */
 664         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 665                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 666
 667         fix1             = _mm_setzero_pd();
 668         fiy1             = _mm_setzero_pd();
 669         fiz1             = _mm_setzero_pd();
 670         fix2             = _mm_setzero_pd();
 671         fiy2             = _mm_setzero_pd();
 672         fiz2             = _mm_setzero_pd();
 673         fix3             = _mm_setzero_pd();
 674         fiy3             = _mm_setzero_pd();
 675         fiz3             = _mm_setzero_pd();
 676
 677         /* Start inner kernel loop */
 678         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 679         {
 680
 681             /* Get j neighbor index, and coordinate index */
 682             jnrA             = jjnr[jidx];
 683             jnrB             = jjnr[jidx+1];
 684             j_coord_offsetA  = DIM*jnrA;
 685             j_coord_offsetB  = DIM*jnrB;
 686
 687             /* load j atom coordinates */
 688             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 689                                               &jx0,&jy0,&jz0);
 690
 691             /* Calculate displacement vector */
 692             dx10             = _mm_sub_pd(ix1,jx0);
 693             dy10             = _mm_sub_pd(iy1,jy0);
 694             dz10             = _mm_sub_pd(iz1,jz0);
 695             dx20             = _mm_sub_pd(ix2,jx0);
 696             dy20             = _mm_sub_pd(iy2,jy0);
 697             dz20             = _mm_sub_pd(iz2,jz0);
 698             dx30             = _mm_sub_pd(ix3,jx0);
 699             dy30             = _mm_sub_pd(iy3,jy0);
 700             dz30             = _mm_sub_pd(iz3,jz0);
 701
 702             /* Calculate squared distance and things based on it */
 703             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 704             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 705             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 706
 707             rinv10           = sse2_invsqrt_d(rsq10);
 708             rinv20           = sse2_invsqrt_d(rsq20);
 709             rinv30           = sse2_invsqrt_d(rsq30);
 710
 711             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 712             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 713             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 714
 715             /* Load parameters for j particles */
 716             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 717
 718             fjx0             = _mm_setzero_pd();
 719             fjy0             = _mm_setzero_pd();
 720             fjz0             = _mm_setzero_pd();
 721
 722             /**************************
 723              * CALCULATE INTERACTIONS *
 724              **************************/
 725
 726             r10              = _mm_mul_pd(rsq10,rinv10);
 727
 728             /* Compute parameters for interactions between i and j atoms */
 729             qq10             = _mm_mul_pd(iq1,jq0);
 730
 731             /* EWALD ELECTROSTATICS */
 732
 733             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 734             ewrt             = _mm_mul_pd(r10,ewtabscale);
 735             ewitab           = _mm_cvttpd_epi32(ewrt);
 736             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 737             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 738                                          &ewtabF,&ewtabFn);
 739             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 740             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 741
 742             fscal            = felec;
 743
 744             /* Calculate temporary vectorial force */
 745             tx               = _mm_mul_pd(fscal,dx10);
 746             ty               = _mm_mul_pd(fscal,dy10);
 747             tz               = _mm_mul_pd(fscal,dz10);
 748
 749             /* Update vectorial force */
 750             fix1             = _mm_add_pd(fix1,tx);
 751             fiy1             = _mm_add_pd(fiy1,ty);
 752             fiz1             = _mm_add_pd(fiz1,tz);
 753
 754             fjx0             = _mm_add_pd(fjx0,tx);
 755             fjy0             = _mm_add_pd(fjy0,ty);
 756             fjz0             = _mm_add_pd(fjz0,tz);
 757
 758             /**************************
 759              * CALCULATE INTERACTIONS *
 760              **************************/
 761
 762             r20              = _mm_mul_pd(rsq20,rinv20);
 763
 764             /* Compute parameters for interactions between i and j atoms */
 765             qq20             = _mm_mul_pd(iq2,jq0);
 766
 767             /* EWALD ELECTROSTATICS */
 768
 769             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 770             ewrt             = _mm_mul_pd(r20,ewtabscale);
 771             ewitab           = _mm_cvttpd_epi32(ewrt);
 772             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 773             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 774                                          &ewtabF,&ewtabFn);
 775             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 776             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 777
 778             fscal            = felec;
 779
 780             /* Calculate temporary vectorial force */
 781             tx               = _mm_mul_pd(fscal,dx20);
 782             ty               = _mm_mul_pd(fscal,dy20);
 783             tz               = _mm_mul_pd(fscal,dz20);
 784
 785             /* Update vectorial force */
 786             fix2             = _mm_add_pd(fix2,tx);
 787             fiy2             = _mm_add_pd(fiy2,ty);
 788             fiz2             = _mm_add_pd(fiz2,tz);
 789
 790             fjx0             = _mm_add_pd(fjx0,tx);
 791             fjy0             = _mm_add_pd(fjy0,ty);
 792             fjz0             = _mm_add_pd(fjz0,tz);
 793
 794             /**************************
 795              * CALCULATE INTERACTIONS *
 796              **************************/
 797
 798             r30              = _mm_mul_pd(rsq30,rinv30);
 799
 800             /* Compute parameters for interactions between i and j atoms */
 801             qq30             = _mm_mul_pd(iq3,jq0);
 802
 803             /* EWALD ELECTROSTATICS */
 804
 805             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 806             ewrt             = _mm_mul_pd(r30,ewtabscale);
 807             ewitab           = _mm_cvttpd_epi32(ewrt);
 808             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 809             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 810                                          &ewtabF,&ewtabFn);
 811             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 812             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 813
 814             fscal            = felec;
 815
 816             /* Calculate temporary vectorial force */
 817             tx               = _mm_mul_pd(fscal,dx30);
 818             ty               = _mm_mul_pd(fscal,dy30);
 819             tz               = _mm_mul_pd(fscal,dz30);
 820
 821             /* Update vectorial force */
 822             fix3             = _mm_add_pd(fix3,tx);
 823             fiy3             = _mm_add_pd(fiy3,ty);
 824             fiz3             = _mm_add_pd(fiz3,tz);
 825
 826             fjx0             = _mm_add_pd(fjx0,tx);
 827             fjy0             = _mm_add_pd(fjy0,ty);
 828             fjz0             = _mm_add_pd(fjz0,tz);
 829
 830             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 831
 832             /* Inner loop uses 111 flops */
 833         }
 834
 835         if(jidx<j_index_end)
 836         {
 837
 838             jnrA             = jjnr[jidx];
 839             j_coord_offsetA  = DIM*jnrA;
 840
 841             /* load j atom coordinates */
 842             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 843                                               &jx0,&jy0,&jz0);
 844
 845             /* Calculate displacement vector */
 846             dx10             = _mm_sub_pd(ix1,jx0);
 847             dy10             = _mm_sub_pd(iy1,jy0);
 848             dz10             = _mm_sub_pd(iz1,jz0);
 849             dx20             = _mm_sub_pd(ix2,jx0);
 850             dy20             = _mm_sub_pd(iy2,jy0);
 851             dz20             = _mm_sub_pd(iz2,jz0);
 852             dx30             = _mm_sub_pd(ix3,jx0);
 853             dy30             = _mm_sub_pd(iy3,jy0);
 854             dz30             = _mm_sub_pd(iz3,jz0);
 855
 856             /* Calculate squared distance and things based on it */
 857             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 858             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 859             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 860
 861             rinv10           = sse2_invsqrt_d(rsq10);
 862             rinv20           = sse2_invsqrt_d(rsq20);
 863             rinv30           = sse2_invsqrt_d(rsq30);
 864
 865             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 866             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 867             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 868
 869             /* Load parameters for j particles */
 870             jq0              = _mm_load_sd(charge+jnrA+0);
 871
 872             fjx0             = _mm_setzero_pd();
 873             fjy0             = _mm_setzero_pd();
 874             fjz0             = _mm_setzero_pd();
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             r10              = _mm_mul_pd(rsq10,rinv10);
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq10             = _mm_mul_pd(iq1,jq0);
 884
 885             /* EWALD ELECTROSTATICS */
 886
 887             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 888             ewrt             = _mm_mul_pd(r10,ewtabscale);
 889             ewitab           = _mm_cvttpd_epi32(ewrt);
 890             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 891             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 892             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 893             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 894
 895             fscal            = felec;
 896
 897             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 898
 899             /* Calculate temporary vectorial force */
 900             tx               = _mm_mul_pd(fscal,dx10);
 901             ty               = _mm_mul_pd(fscal,dy10);
 902             tz               = _mm_mul_pd(fscal,dz10);
 903
 904             /* Update vectorial force */
 905             fix1             = _mm_add_pd(fix1,tx);
 906             fiy1             = _mm_add_pd(fiy1,ty);
 907             fiz1             = _mm_add_pd(fiz1,tz);
 908
 909             fjx0             = _mm_add_pd(fjx0,tx);
 910             fjy0             = _mm_add_pd(fjy0,ty);
 911             fjz0             = _mm_add_pd(fjz0,tz);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r20              = _mm_mul_pd(rsq20,rinv20);
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq20             = _mm_mul_pd(iq2,jq0);
 921
 922             /* EWALD ELECTROSTATICS */
 923
 924             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 925             ewrt             = _mm_mul_pd(r20,ewtabscale);
 926             ewitab           = _mm_cvttpd_epi32(ewrt);
 927             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 928             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 929             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 930             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 931
 932             fscal            = felec;
 933
 934             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm_mul_pd(fscal,dx20);
 938             ty               = _mm_mul_pd(fscal,dy20);
 939             tz               = _mm_mul_pd(fscal,dz20);
 940
 941             /* Update vectorial force */
 942             fix2             = _mm_add_pd(fix2,tx);
 943             fiy2             = _mm_add_pd(fiy2,ty);
 944             fiz2             = _mm_add_pd(fiz2,tz);
 945
 946             fjx0             = _mm_add_pd(fjx0,tx);
 947             fjy0             = _mm_add_pd(fjy0,ty);
 948             fjz0             = _mm_add_pd(fjz0,tz);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             r30              = _mm_mul_pd(rsq30,rinv30);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq30             = _mm_mul_pd(iq3,jq0);
 958
 959             /* EWALD ELECTROSTATICS */
 960
 961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 962             ewrt             = _mm_mul_pd(r30,ewtabscale);
 963             ewitab           = _mm_cvttpd_epi32(ewrt);
 964             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 965             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 966             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 967             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 968
 969             fscal            = felec;
 970
 971             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 972
 973             /* Calculate temporary vectorial force */
 974             tx               = _mm_mul_pd(fscal,dx30);
 975             ty               = _mm_mul_pd(fscal,dy30);
 976             tz               = _mm_mul_pd(fscal,dz30);
 977
 978             /* Update vectorial force */
 979             fix3             = _mm_add_pd(fix3,tx);
 980             fiy3             = _mm_add_pd(fiy3,ty);
 981             fiz3             = _mm_add_pd(fiz3,tz);
 982
 983             fjx0             = _mm_add_pd(fjx0,tx);
 984             fjy0             = _mm_add_pd(fjy0,ty);
 985             fjz0             = _mm_add_pd(fjz0,tz);
 986
 987             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 988
 989             /* Inner loop uses 111 flops */
 990         }
 991
 992         /* End of innermost loop */
 993
 994         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 995                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 996
 997         /* Increment number of inner iterations */
 998         inneriter                  += j_index_end - j_index_start;
 999
1000         /* Outer loop uses 18 flops */
1001     }
1002
1003     /* Increment number of outer iterations */
1004     outeriter        += nri;
1005
1006     /* Update outer/inner flops */
1007
1008     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*111);
1009 }