src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwNone_GeomW3P1_sse4_1_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 sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          ewitab;
  96     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  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     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 124     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 125     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 151                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 152
 153         fix0             = _mm_setzero_pd();
 154         fiy0             = _mm_setzero_pd();
 155         fiz0             = _mm_setzero_pd();
 156         fix1             = _mm_setzero_pd();
 157         fiy1             = _mm_setzero_pd();
 158         fiz1             = _mm_setzero_pd();
 159         fix2             = _mm_setzero_pd();
 160         fiy2             = _mm_setzero_pd();
 161         fiz2             = _mm_setzero_pd();
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184             dx10             = _mm_sub_pd(ix1,jx0);
 185             dy10             = _mm_sub_pd(iy1,jy0);
 186             dz10             = _mm_sub_pd(iz1,jz0);
 187             dx20             = _mm_sub_pd(ix2,jx0);
 188             dy20             = _mm_sub_pd(iy2,jy0);
 189             dz20             = _mm_sub_pd(iz2,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 193             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 194             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 195
 196             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 197             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 198             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 199
 200             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 201             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 202             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 206
 207             fjx0             = _mm_setzero_pd();
 208             fjy0             = _mm_setzero_pd();
 209             fjz0             = _mm_setzero_pd();
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             r00              = _mm_mul_pd(rsq00,rinv00);
 216
 217             /* Compute parameters for interactions between i and j atoms */
 218             qq00             = _mm_mul_pd(iq0,jq0);
 219
 220             /* EWALD ELECTROSTATICS */
 221
 222             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 223             ewrt             = _mm_mul_pd(r00,ewtabscale);
 224             ewitab           = _mm_cvttpd_epi32(ewrt);
 225             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 226             ewitab           = _mm_slli_epi32(ewitab,2);
 227             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 228             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 229             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 230             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 231             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 232             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 233             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 234             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 235             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 236             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 237
 238             /* Update potential sum for this i atom from the interaction with this j atom. */
 239             velecsum         = _mm_add_pd(velecsum,velec);
 240
 241             fscal            = felec;
 242
 243             /* Calculate temporary vectorial force */
 244             tx               = _mm_mul_pd(fscal,dx00);
 245             ty               = _mm_mul_pd(fscal,dy00);
 246             tz               = _mm_mul_pd(fscal,dz00);
 247
 248             /* Update vectorial force */
 249             fix0             = _mm_add_pd(fix0,tx);
 250             fiy0             = _mm_add_pd(fiy0,ty);
 251             fiz0             = _mm_add_pd(fiz0,tz);
 252
 253             fjx0             = _mm_add_pd(fjx0,tx);
 254             fjy0             = _mm_add_pd(fjy0,ty);
 255             fjz0             = _mm_add_pd(fjz0,tz);
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             r10              = _mm_mul_pd(rsq10,rinv10);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             qq10             = _mm_mul_pd(iq1,jq0);
 265
 266             /* EWALD ELECTROSTATICS */
 267
 268             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 269             ewrt             = _mm_mul_pd(r10,ewtabscale);
 270             ewitab           = _mm_cvttpd_epi32(ewrt);
 271             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 272             ewitab           = _mm_slli_epi32(ewitab,2);
 273             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 274             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 275             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 276             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 277             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 278             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 279             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 280             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 281             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 282             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velecsum         = _mm_add_pd(velecsum,velec);
 286
 287             fscal            = felec;
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm_mul_pd(fscal,dx10);
 291             ty               = _mm_mul_pd(fscal,dy10);
 292             tz               = _mm_mul_pd(fscal,dz10);
 293
 294             /* Update vectorial force */
 295             fix1             = _mm_add_pd(fix1,tx);
 296             fiy1             = _mm_add_pd(fiy1,ty);
 297             fiz1             = _mm_add_pd(fiz1,tz);
 298
 299             fjx0             = _mm_add_pd(fjx0,tx);
 300             fjy0             = _mm_add_pd(fjy0,ty);
 301             fjz0             = _mm_add_pd(fjz0,tz);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             r20              = _mm_mul_pd(rsq20,rinv20);
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq20             = _mm_mul_pd(iq2,jq0);
 311
 312             /* EWALD ELECTROSTATICS */
 313
 314             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 315             ewrt             = _mm_mul_pd(r20,ewtabscale);
 316             ewitab           = _mm_cvttpd_epi32(ewrt);
 317             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 318             ewitab           = _mm_slli_epi32(ewitab,2);
 319             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 320             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 321             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 322             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 323             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 324             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 325             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 326             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 327             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 328             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm_add_pd(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm_mul_pd(fscal,dx20);
 337             ty               = _mm_mul_pd(fscal,dy20);
 338             tz               = _mm_mul_pd(fscal,dz20);
 339
 340             /* Update vectorial force */
 341             fix2             = _mm_add_pd(fix2,tx);
 342             fiy2             = _mm_add_pd(fiy2,ty);
 343             fiz2             = _mm_add_pd(fiz2,tz);
 344
 345             fjx0             = _mm_add_pd(fjx0,tx);
 346             fjy0             = _mm_add_pd(fjy0,ty);
 347             fjz0             = _mm_add_pd(fjz0,tz);
 348
 349             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 350
 351             /* Inner loop uses 126 flops */
 352         }
 353
 354         if(jidx<j_index_end)
 355         {
 356
 357             jnrA             = jjnr[jidx];
 358             j_coord_offsetA  = DIM*jnrA;
 359
 360             /* load j atom coordinates */
 361             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 362                                               &jx0,&jy0,&jz0);
 363
 364             /* Calculate displacement vector */
 365             dx00             = _mm_sub_pd(ix0,jx0);
 366             dy00             = _mm_sub_pd(iy0,jy0);
 367             dz00             = _mm_sub_pd(iz0,jz0);
 368             dx10             = _mm_sub_pd(ix1,jx0);
 369             dy10             = _mm_sub_pd(iy1,jy0);
 370             dz10             = _mm_sub_pd(iz1,jz0);
 371             dx20             = _mm_sub_pd(ix2,jx0);
 372             dy20             = _mm_sub_pd(iy2,jy0);
 373             dz20             = _mm_sub_pd(iz2,jz0);
 374
 375             /* Calculate squared distance and things based on it */
 376             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 377             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 378             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 379
 380             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 381             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 382             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 383
 384             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 385             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 386             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 387
 388             /* Load parameters for j particles */
 389             jq0              = _mm_load_sd(charge+jnrA+0);
 390
 391             fjx0             = _mm_setzero_pd();
 392             fjy0             = _mm_setzero_pd();
 393             fjz0             = _mm_setzero_pd();
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             r00              = _mm_mul_pd(rsq00,rinv00);
 400
 401             /* Compute parameters for interactions between i and j atoms */
 402             qq00             = _mm_mul_pd(iq0,jq0);
 403
 404             /* EWALD ELECTROSTATICS */
 405
 406             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 407             ewrt             = _mm_mul_pd(r00,ewtabscale);
 408             ewitab           = _mm_cvttpd_epi32(ewrt);
 409             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 410             ewitab           = _mm_slli_epi32(ewitab,2);
 411             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 412             ewtabD           = _mm_setzero_pd();
 413             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 414             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 415             ewtabFn          = _mm_setzero_pd();
 416             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 417             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 418             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 419             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 420             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 424             velecsum         = _mm_add_pd(velecsum,velec);
 425
 426             fscal            = felec;
 427
 428             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 429
 430             /* Calculate temporary vectorial force */
 431             tx               = _mm_mul_pd(fscal,dx00);
 432             ty               = _mm_mul_pd(fscal,dy00);
 433             tz               = _mm_mul_pd(fscal,dz00);
 434
 435             /* Update vectorial force */
 436             fix0             = _mm_add_pd(fix0,tx);
 437             fiy0             = _mm_add_pd(fiy0,ty);
 438             fiz0             = _mm_add_pd(fiz0,tz);
 439
 440             fjx0             = _mm_add_pd(fjx0,tx);
 441             fjy0             = _mm_add_pd(fjy0,ty);
 442             fjz0             = _mm_add_pd(fjz0,tz);
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             r10              = _mm_mul_pd(rsq10,rinv10);
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq10             = _mm_mul_pd(iq1,jq0);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 456             ewrt             = _mm_mul_pd(r10,ewtabscale);
 457             ewitab           = _mm_cvttpd_epi32(ewrt);
 458             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 459             ewitab           = _mm_slli_epi32(ewitab,2);
 460             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 461             ewtabD           = _mm_setzero_pd();
 462             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 463             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 464             ewtabFn          = _mm_setzero_pd();
 465             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 466             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 467             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 468             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 469             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 473             velecsum         = _mm_add_pd(velecsum,velec);
 474
 475             fscal            = felec;
 476
 477             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 478
 479             /* Calculate temporary vectorial force */
 480             tx               = _mm_mul_pd(fscal,dx10);
 481             ty               = _mm_mul_pd(fscal,dy10);
 482             tz               = _mm_mul_pd(fscal,dz10);
 483
 484             /* Update vectorial force */
 485             fix1             = _mm_add_pd(fix1,tx);
 486             fiy1             = _mm_add_pd(fiy1,ty);
 487             fiz1             = _mm_add_pd(fiz1,tz);
 488
 489             fjx0             = _mm_add_pd(fjx0,tx);
 490             fjy0             = _mm_add_pd(fjy0,ty);
 491             fjz0             = _mm_add_pd(fjz0,tz);
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             r20              = _mm_mul_pd(rsq20,rinv20);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq20             = _mm_mul_pd(iq2,jq0);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 505             ewrt             = _mm_mul_pd(r20,ewtabscale);
 506             ewitab           = _mm_cvttpd_epi32(ewrt);
 507             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 508             ewitab           = _mm_slli_epi32(ewitab,2);
 509             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 510             ewtabD           = _mm_setzero_pd();
 511             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 512             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 513             ewtabFn          = _mm_setzero_pd();
 514             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 515             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 516             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 517             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 518             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 522             velecsum         = _mm_add_pd(velecsum,velec);
 523
 524             fscal            = felec;
 525
 526             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_pd(fscal,dx20);
 530             ty               = _mm_mul_pd(fscal,dy20);
 531             tz               = _mm_mul_pd(fscal,dz20);
 532
 533             /* Update vectorial force */
 534             fix2             = _mm_add_pd(fix2,tx);
 535             fiy2             = _mm_add_pd(fiy2,ty);
 536             fiz2             = _mm_add_pd(fiz2,tz);
 537
 538             fjx0             = _mm_add_pd(fjx0,tx);
 539             fjy0             = _mm_add_pd(fjy0,ty);
 540             fjz0             = _mm_add_pd(fjz0,tz);
 541
 542             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 543
 544             /* Inner loop uses 126 flops */
 545         }
 546
 547         /* End of innermost loop */
 548
 549         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 550                                               f+i_coord_offset,fshift+i_shift_offset);
 551
 552         ggid                        = gid[iidx];
 553         /* Update potential energies */
 554         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 555
 556         /* Increment number of inner iterations */
 557         inneriter                  += j_index_end - j_index_start;
 558
 559         /* Outer loop uses 19 flops */
 560     }
 561
 562     /* Increment number of outer iterations */
 563     outeriter        += nri;
 564
 565     /* Update outer/inner flops */
 566
 567     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*126);
 568 }
 569 /*
 570  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse4_1_double
 571  * Electrostatics interaction: Ewald
 572  * VdW interaction:            None
 573  * Geometry:                   Water3-Particle
 574  * Calculate force/pot:        Force
 575  */
 576 void
 577 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse4_1_double
 578                     (t_nblist                    * gmx_restrict       nlist,
 579                      rvec                        * gmx_restrict          xx,
 580                      rvec                        * gmx_restrict          ff,
 581                      t_forcerec                  * gmx_restrict          fr,
 582                      t_mdatoms                   * gmx_restrict     mdatoms,
 583                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 584                      t_nrnb                      * gmx_restrict        nrnb)
 585 {
 586     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 587      * just 0 for non-waters.
 588      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 589      * jnr indices corresponding to data put in the four positions in the SIMD register.
 590      */
 591     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 592     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 593     int              jnrA,jnrB;
 594     int              j_coord_offsetA,j_coord_offsetB;
 595     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 596     real             rcutoff_scalar;
 597     real             *shiftvec,*fshift,*x,*f;
 598     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 599     int              vdwioffset0;
 600     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 601     int              vdwioffset1;
 602     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 603     int              vdwioffset2;
 604     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 605     int              vdwjidx0A,vdwjidx0B;
 606     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 607     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 608     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 609     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 610     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 611     real             *charge;
 612     __m128i          ewitab;
 613     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 614     real             *ewtab;
 615     __m128d          dummy_mask,cutoff_mask;
 616     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 617     __m128d          one     = _mm_set1_pd(1.0);
 618     __m128d          two     = _mm_set1_pd(2.0);
 619     x                = xx[0];
 620     f                = ff[0];
 621
 622     nri              = nlist->nri;
 623     iinr             = nlist->iinr;
 624     jindex           = nlist->jindex;
 625     jjnr             = nlist->jjnr;
 626     shiftidx         = nlist->shift;
 627     gid              = nlist->gid;
 628     shiftvec         = fr->shift_vec[0];
 629     fshift           = fr->fshift[0];
 630     facel            = _mm_set1_pd(fr->epsfac);
 631     charge           = mdatoms->chargeA;
 632
 633     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 634     ewtab            = fr->ic->tabq_coul_F;
 635     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 636     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 637
 638     /* Setup water-specific parameters */
 639     inr              = nlist->iinr[0];
 640     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 641     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 642     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 643
 644     /* Avoid stupid compiler warnings */
 645     jnrA = jnrB = 0;
 646     j_coord_offsetA = 0;
 647     j_coord_offsetB = 0;
 648
 649     outeriter        = 0;
 650     inneriter        = 0;
 651
 652     /* Start outer loop over neighborlists */
 653     for(iidx=0; iidx<nri; iidx++)
 654     {
 655         /* Load shift vector for this list */
 656         i_shift_offset   = DIM*shiftidx[iidx];
 657
 658         /* Load limits for loop over neighbors */
 659         j_index_start    = jindex[iidx];
 660         j_index_end      = jindex[iidx+1];
 661
 662         /* Get outer coordinate index */
 663         inr              = iinr[iidx];
 664         i_coord_offset   = DIM*inr;
 665
 666         /* Load i particle coords and add shift vector */
 667         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 668                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 669
 670         fix0             = _mm_setzero_pd();
 671         fiy0             = _mm_setzero_pd();
 672         fiz0             = _mm_setzero_pd();
 673         fix1             = _mm_setzero_pd();
 674         fiy1             = _mm_setzero_pd();
 675         fiz1             = _mm_setzero_pd();
 676         fix2             = _mm_setzero_pd();
 677         fiy2             = _mm_setzero_pd();
 678         fiz2             = _mm_setzero_pd();
 679
 680         /* Start inner kernel loop */
 681         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 682         {
 683
 684             /* Get j neighbor index, and coordinate index */
 685             jnrA             = jjnr[jidx];
 686             jnrB             = jjnr[jidx+1];
 687             j_coord_offsetA  = DIM*jnrA;
 688             j_coord_offsetB  = DIM*jnrB;
 689
 690             /* load j atom coordinates */
 691             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 692                                               &jx0,&jy0,&jz0);
 693
 694             /* Calculate displacement vector */
 695             dx00             = _mm_sub_pd(ix0,jx0);
 696             dy00             = _mm_sub_pd(iy0,jy0);
 697             dz00             = _mm_sub_pd(iz0,jz0);
 698             dx10             = _mm_sub_pd(ix1,jx0);
 699             dy10             = _mm_sub_pd(iy1,jy0);
 700             dz10             = _mm_sub_pd(iz1,jz0);
 701             dx20             = _mm_sub_pd(ix2,jx0);
 702             dy20             = _mm_sub_pd(iy2,jy0);
 703             dz20             = _mm_sub_pd(iz2,jz0);
 704
 705             /* Calculate squared distance and things based on it */
 706             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 707             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 708             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 709
 710             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 711             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 712             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 713
 714             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 715             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 716             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 717
 718             /* Load parameters for j particles */
 719             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 720
 721             fjx0             = _mm_setzero_pd();
 722             fjy0             = _mm_setzero_pd();
 723             fjz0             = _mm_setzero_pd();
 724
 725             /**************************
 726              * CALCULATE INTERACTIONS *
 727              **************************/
 728
 729             r00              = _mm_mul_pd(rsq00,rinv00);
 730
 731             /* Compute parameters for interactions between i and j atoms */
 732             qq00             = _mm_mul_pd(iq0,jq0);
 733
 734             /* EWALD ELECTROSTATICS */
 735
 736             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 737             ewrt             = _mm_mul_pd(r00,ewtabscale);
 738             ewitab           = _mm_cvttpd_epi32(ewrt);
 739             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 740             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 741                                          &ewtabF,&ewtabFn);
 742             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 743             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 744
 745             fscal            = felec;
 746
 747             /* Calculate temporary vectorial force */
 748             tx               = _mm_mul_pd(fscal,dx00);
 749             ty               = _mm_mul_pd(fscal,dy00);
 750             tz               = _mm_mul_pd(fscal,dz00);
 751
 752             /* Update vectorial force */
 753             fix0             = _mm_add_pd(fix0,tx);
 754             fiy0             = _mm_add_pd(fiy0,ty);
 755             fiz0             = _mm_add_pd(fiz0,tz);
 756
 757             fjx0             = _mm_add_pd(fjx0,tx);
 758             fjy0             = _mm_add_pd(fjy0,ty);
 759             fjz0             = _mm_add_pd(fjz0,tz);
 760
 761             /**************************
 762              * CALCULATE INTERACTIONS *
 763              **************************/
 764
 765             r10              = _mm_mul_pd(rsq10,rinv10);
 766
 767             /* Compute parameters for interactions between i and j atoms */
 768             qq10             = _mm_mul_pd(iq1,jq0);
 769
 770             /* EWALD ELECTROSTATICS */
 771
 772             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 773             ewrt             = _mm_mul_pd(r10,ewtabscale);
 774             ewitab           = _mm_cvttpd_epi32(ewrt);
 775             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 776             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 777                                          &ewtabF,&ewtabFn);
 778             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 779             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 780
 781             fscal            = felec;
 782
 783             /* Calculate temporary vectorial force */
 784             tx               = _mm_mul_pd(fscal,dx10);
 785             ty               = _mm_mul_pd(fscal,dy10);
 786             tz               = _mm_mul_pd(fscal,dz10);
 787
 788             /* Update vectorial force */
 789             fix1             = _mm_add_pd(fix1,tx);
 790             fiy1             = _mm_add_pd(fiy1,ty);
 791             fiz1             = _mm_add_pd(fiz1,tz);
 792
 793             fjx0             = _mm_add_pd(fjx0,tx);
 794             fjy0             = _mm_add_pd(fjy0,ty);
 795             fjz0             = _mm_add_pd(fjz0,tz);
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             r20              = _mm_mul_pd(rsq20,rinv20);
 802
 803             /* Compute parameters for interactions between i and j atoms */
 804             qq20             = _mm_mul_pd(iq2,jq0);
 805
 806             /* EWALD ELECTROSTATICS */
 807
 808             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 809             ewrt             = _mm_mul_pd(r20,ewtabscale);
 810             ewitab           = _mm_cvttpd_epi32(ewrt);
 811             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 812             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 813                                          &ewtabF,&ewtabFn);
 814             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 815             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 816
 817             fscal            = felec;
 818
 819             /* Calculate temporary vectorial force */
 820             tx               = _mm_mul_pd(fscal,dx20);
 821             ty               = _mm_mul_pd(fscal,dy20);
 822             tz               = _mm_mul_pd(fscal,dz20);
 823
 824             /* Update vectorial force */
 825             fix2             = _mm_add_pd(fix2,tx);
 826             fiy2             = _mm_add_pd(fiy2,ty);
 827             fiz2             = _mm_add_pd(fiz2,tz);
 828
 829             fjx0             = _mm_add_pd(fjx0,tx);
 830             fjy0             = _mm_add_pd(fjy0,ty);
 831             fjz0             = _mm_add_pd(fjz0,tz);
 832
 833             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 834
 835             /* Inner loop uses 111 flops */
 836         }
 837
 838         if(jidx<j_index_end)
 839         {
 840
 841             jnrA             = jjnr[jidx];
 842             j_coord_offsetA  = DIM*jnrA;
 843
 844             /* load j atom coordinates */
 845             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 846                                               &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm_sub_pd(ix0,jx0);
 850             dy00             = _mm_sub_pd(iy0,jy0);
 851             dz00             = _mm_sub_pd(iz0,jz0);
 852             dx10             = _mm_sub_pd(ix1,jx0);
 853             dy10             = _mm_sub_pd(iy1,jy0);
 854             dz10             = _mm_sub_pd(iz1,jz0);
 855             dx20             = _mm_sub_pd(ix2,jx0);
 856             dy20             = _mm_sub_pd(iy2,jy0);
 857             dz20             = _mm_sub_pd(iz2,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 861             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 862             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 863
 864             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 865             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 866             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 867
 868             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 869             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 870             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 871
 872             /* Load parameters for j particles */
 873             jq0              = _mm_load_sd(charge+jnrA+0);
 874
 875             fjx0             = _mm_setzero_pd();
 876             fjy0             = _mm_setzero_pd();
 877             fjz0             = _mm_setzero_pd();
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             r00              = _mm_mul_pd(rsq00,rinv00);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq00             = _mm_mul_pd(iq0,jq0);
 887
 888             /* EWALD ELECTROSTATICS */
 889
 890             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 891             ewrt             = _mm_mul_pd(r00,ewtabscale);
 892             ewitab           = _mm_cvttpd_epi32(ewrt);
 893             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 894             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 895             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 896             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 897
 898             fscal            = felec;
 899
 900             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 901
 902             /* Calculate temporary vectorial force */
 903             tx               = _mm_mul_pd(fscal,dx00);
 904             ty               = _mm_mul_pd(fscal,dy00);
 905             tz               = _mm_mul_pd(fscal,dz00);
 906
 907             /* Update vectorial force */
 908             fix0             = _mm_add_pd(fix0,tx);
 909             fiy0             = _mm_add_pd(fiy0,ty);
 910             fiz0             = _mm_add_pd(fiz0,tz);
 911
 912             fjx0             = _mm_add_pd(fjx0,tx);
 913             fjy0             = _mm_add_pd(fjy0,ty);
 914             fjz0             = _mm_add_pd(fjz0,tz);
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             r10              = _mm_mul_pd(rsq10,rinv10);
 921
 922             /* Compute parameters for interactions between i and j atoms */
 923             qq10             = _mm_mul_pd(iq1,jq0);
 924
 925             /* EWALD ELECTROSTATICS */
 926
 927             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 928             ewrt             = _mm_mul_pd(r10,ewtabscale);
 929             ewitab           = _mm_cvttpd_epi32(ewrt);
 930             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 931             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 932             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 933             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 934
 935             fscal            = felec;
 936
 937             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm_mul_pd(fscal,dx10);
 941             ty               = _mm_mul_pd(fscal,dy10);
 942             tz               = _mm_mul_pd(fscal,dz10);
 943
 944             /* Update vectorial force */
 945             fix1             = _mm_add_pd(fix1,tx);
 946             fiy1             = _mm_add_pd(fiy1,ty);
 947             fiz1             = _mm_add_pd(fiz1,tz);
 948
 949             fjx0             = _mm_add_pd(fjx0,tx);
 950             fjy0             = _mm_add_pd(fjy0,ty);
 951             fjz0             = _mm_add_pd(fjz0,tz);
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             r20              = _mm_mul_pd(rsq20,rinv20);
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq20             = _mm_mul_pd(iq2,jq0);
 961
 962             /* EWALD ELECTROSTATICS */
 963
 964             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 965             ewrt             = _mm_mul_pd(r20,ewtabscale);
 966             ewitab           = _mm_cvttpd_epi32(ewrt);
 967             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 968             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 969             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 970             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 971
 972             fscal            = felec;
 973
 974             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 975
 976             /* Calculate temporary vectorial force */
 977             tx               = _mm_mul_pd(fscal,dx20);
 978             ty               = _mm_mul_pd(fscal,dy20);
 979             tz               = _mm_mul_pd(fscal,dz20);
 980
 981             /* Update vectorial force */
 982             fix2             = _mm_add_pd(fix2,tx);
 983             fiy2             = _mm_add_pd(fiy2,ty);
 984             fiz2             = _mm_add_pd(fiz2,tz);
 985
 986             fjx0             = _mm_add_pd(fjx0,tx);
 987             fjy0             = _mm_add_pd(fjy0,ty);
 988             fjz0             = _mm_add_pd(fjz0,tz);
 989
 990             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 991
 992             /* Inner loop uses 111 flops */
 993         }
 994
 995         /* End of innermost loop */
 996
 997         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 998                                               f+i_coord_offset,fshift+i_shift_offset);
 999
1000         /* Increment number of inner iterations */
1001         inneriter                  += j_index_end - j_index_start;
1002
1003         /* Outer loop uses 18 flops */
1004     }
1005
1006     /* Increment number of outer iterations */
1007     outeriter        += nri;
1008
1009     /* Update outer/inner flops */
1010
1011     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*111);
1012 }