src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwNone_GeomW4P1_sse4_1_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_sse4_1_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset1;
  80     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  81     int              vdwioffset2;
  82     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  83     int              vdwioffset3;
  84     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  88     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128i          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_pd(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113
 114     vftab            = kernel_data->table_elec->data;
 115     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 120     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 121     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 147                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 148
 149         fix1             = _mm_setzero_pd();
 150         fiy1             = _mm_setzero_pd();
 151         fiz1             = _mm_setzero_pd();
 152         fix2             = _mm_setzero_pd();
 153         fiy2             = _mm_setzero_pd();
 154         fiz2             = _mm_setzero_pd();
 155         fix3             = _mm_setzero_pd();
 156         fiy3             = _mm_setzero_pd();
 157         fiz3             = _mm_setzero_pd();
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_pd();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171
 172             /* load j atom coordinates */
 173             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx10             = _mm_sub_pd(ix1,jx0);
 178             dy10             = _mm_sub_pd(iy1,jy0);
 179             dz10             = _mm_sub_pd(iz1,jz0);
 180             dx20             = _mm_sub_pd(ix2,jx0);
 181             dy20             = _mm_sub_pd(iy2,jy0);
 182             dz20             = _mm_sub_pd(iz2,jz0);
 183             dx30             = _mm_sub_pd(ix3,jx0);
 184             dy30             = _mm_sub_pd(iy3,jy0);
 185             dz30             = _mm_sub_pd(iz3,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 189             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 190             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 191
 192             rinv10           = sse41_invsqrt_d(rsq10);
 193             rinv20           = sse41_invsqrt_d(rsq20);
 194             rinv30           = sse41_invsqrt_d(rsq30);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 198
 199             fjx0             = _mm_setzero_pd();
 200             fjy0             = _mm_setzero_pd();
 201             fjz0             = _mm_setzero_pd();
 202
 203             /**************************
 204              * CALCULATE INTERACTIONS *
 205              **************************/
 206
 207             r10              = _mm_mul_pd(rsq10,rinv10);
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq10             = _mm_mul_pd(iq1,jq0);
 211
 212             /* Calculate table index by multiplying r with table scale and truncate to integer */
 213             rt               = _mm_mul_pd(r10,vftabscale);
 214             vfitab           = _mm_cvttpd_epi32(rt);
 215             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 216             vfitab           = _mm_slli_epi32(vfitab,2);
 217
 218             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 219             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 220             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 221             GMX_MM_TRANSPOSE2_PD(Y,F);
 222             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 223             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 224             GMX_MM_TRANSPOSE2_PD(G,H);
 225             Heps             = _mm_mul_pd(vfeps,H);
 226             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 227             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 228             velec            = _mm_mul_pd(qq10,VV);
 229             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 230             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velecsum         = _mm_add_pd(velecsum,velec);
 234
 235             fscal            = felec;
 236
 237             /* Calculate temporary vectorial force */
 238             tx               = _mm_mul_pd(fscal,dx10);
 239             ty               = _mm_mul_pd(fscal,dy10);
 240             tz               = _mm_mul_pd(fscal,dz10);
 241
 242             /* Update vectorial force */
 243             fix1             = _mm_add_pd(fix1,tx);
 244             fiy1             = _mm_add_pd(fiy1,ty);
 245             fiz1             = _mm_add_pd(fiz1,tz);
 246
 247             fjx0             = _mm_add_pd(fjx0,tx);
 248             fjy0             = _mm_add_pd(fjy0,ty);
 249             fjz0             = _mm_add_pd(fjz0,tz);
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             r20              = _mm_mul_pd(rsq20,rinv20);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq20             = _mm_mul_pd(iq2,jq0);
 259
 260             /* Calculate table index by multiplying r with table scale and truncate to integer */
 261             rt               = _mm_mul_pd(r20,vftabscale);
 262             vfitab           = _mm_cvttpd_epi32(rt);
 263             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 264             vfitab           = _mm_slli_epi32(vfitab,2);
 265
 266             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 267             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 269             GMX_MM_TRANSPOSE2_PD(Y,F);
 270             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 271             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 272             GMX_MM_TRANSPOSE2_PD(G,H);
 273             Heps             = _mm_mul_pd(vfeps,H);
 274             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 275             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 276             velec            = _mm_mul_pd(qq20,VV);
 277             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 278             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm_add_pd(velecsum,velec);
 282
 283             fscal            = felec;
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_pd(fscal,dx20);
 287             ty               = _mm_mul_pd(fscal,dy20);
 288             tz               = _mm_mul_pd(fscal,dz20);
 289
 290             /* Update vectorial force */
 291             fix2             = _mm_add_pd(fix2,tx);
 292             fiy2             = _mm_add_pd(fiy2,ty);
 293             fiz2             = _mm_add_pd(fiz2,tz);
 294
 295             fjx0             = _mm_add_pd(fjx0,tx);
 296             fjy0             = _mm_add_pd(fjy0,ty);
 297             fjz0             = _mm_add_pd(fjz0,tz);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r30              = _mm_mul_pd(rsq30,rinv30);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq30             = _mm_mul_pd(iq3,jq0);
 307
 308             /* Calculate table index by multiplying r with table scale and truncate to integer */
 309             rt               = _mm_mul_pd(r30,vftabscale);
 310             vfitab           = _mm_cvttpd_epi32(rt);
 311             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 312             vfitab           = _mm_slli_epi32(vfitab,2);
 313
 314             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 315             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 316             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 317             GMX_MM_TRANSPOSE2_PD(Y,F);
 318             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 319             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 320             GMX_MM_TRANSPOSE2_PD(G,H);
 321             Heps             = _mm_mul_pd(vfeps,H);
 322             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 323             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 324             velec            = _mm_mul_pd(qq30,VV);
 325             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 326             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_pd(fscal,dx30);
 335             ty               = _mm_mul_pd(fscal,dy30);
 336             tz               = _mm_mul_pd(fscal,dz30);
 337
 338             /* Update vectorial force */
 339             fix3             = _mm_add_pd(fix3,tx);
 340             fiy3             = _mm_add_pd(fiy3,ty);
 341             fiz3             = _mm_add_pd(fiz3,tz);
 342
 343             fjx0             = _mm_add_pd(fjx0,tx);
 344             fjy0             = _mm_add_pd(fjy0,ty);
 345             fjz0             = _mm_add_pd(fjz0,tz);
 346
 347             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 348
 349             /* Inner loop uses 132 flops */
 350         }
 351
 352         if(jidx<j_index_end)
 353         {
 354
 355             jnrA             = jjnr[jidx];
 356             j_coord_offsetA  = DIM*jnrA;
 357
 358             /* load j atom coordinates */
 359             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 360                                               &jx0,&jy0,&jz0);
 361
 362             /* Calculate displacement vector */
 363             dx10             = _mm_sub_pd(ix1,jx0);
 364             dy10             = _mm_sub_pd(iy1,jy0);
 365             dz10             = _mm_sub_pd(iz1,jz0);
 366             dx20             = _mm_sub_pd(ix2,jx0);
 367             dy20             = _mm_sub_pd(iy2,jy0);
 368             dz20             = _mm_sub_pd(iz2,jz0);
 369             dx30             = _mm_sub_pd(ix3,jx0);
 370             dy30             = _mm_sub_pd(iy3,jy0);
 371             dz30             = _mm_sub_pd(iz3,jz0);
 372
 373             /* Calculate squared distance and things based on it */
 374             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 375             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 376             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 377
 378             rinv10           = sse41_invsqrt_d(rsq10);
 379             rinv20           = sse41_invsqrt_d(rsq20);
 380             rinv30           = sse41_invsqrt_d(rsq30);
 381
 382             /* Load parameters for j particles */
 383             jq0              = _mm_load_sd(charge+jnrA+0);
 384
 385             fjx0             = _mm_setzero_pd();
 386             fjy0             = _mm_setzero_pd();
 387             fjz0             = _mm_setzero_pd();
 388
 389             /**************************
 390              * CALCULATE INTERACTIONS *
 391              **************************/
 392
 393             r10              = _mm_mul_pd(rsq10,rinv10);
 394
 395             /* Compute parameters for interactions between i and j atoms */
 396             qq10             = _mm_mul_pd(iq1,jq0);
 397
 398             /* Calculate table index by multiplying r with table scale and truncate to integer */
 399             rt               = _mm_mul_pd(r10,vftabscale);
 400             vfitab           = _mm_cvttpd_epi32(rt);
 401             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 402             vfitab           = _mm_slli_epi32(vfitab,2);
 403
 404             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 405             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 406             F                = _mm_setzero_pd();
 407             GMX_MM_TRANSPOSE2_PD(Y,F);
 408             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 409             H                = _mm_setzero_pd();
 410             GMX_MM_TRANSPOSE2_PD(G,H);
 411             Heps             = _mm_mul_pd(vfeps,H);
 412             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 413             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 414             velec            = _mm_mul_pd(qq10,VV);
 415             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 416             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 420             velecsum         = _mm_add_pd(velecsum,velec);
 421
 422             fscal            = felec;
 423
 424             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm_mul_pd(fscal,dx10);
 428             ty               = _mm_mul_pd(fscal,dy10);
 429             tz               = _mm_mul_pd(fscal,dz10);
 430
 431             /* Update vectorial force */
 432             fix1             = _mm_add_pd(fix1,tx);
 433             fiy1             = _mm_add_pd(fiy1,ty);
 434             fiz1             = _mm_add_pd(fiz1,tz);
 435
 436             fjx0             = _mm_add_pd(fjx0,tx);
 437             fjy0             = _mm_add_pd(fjy0,ty);
 438             fjz0             = _mm_add_pd(fjz0,tz);
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r20              = _mm_mul_pd(rsq20,rinv20);
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             qq20             = _mm_mul_pd(iq2,jq0);
 448
 449             /* Calculate table index by multiplying r with table scale and truncate to integer */
 450             rt               = _mm_mul_pd(r20,vftabscale);
 451             vfitab           = _mm_cvttpd_epi32(rt);
 452             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 453             vfitab           = _mm_slli_epi32(vfitab,2);
 454
 455             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 456             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 457             F                = _mm_setzero_pd();
 458             GMX_MM_TRANSPOSE2_PD(Y,F);
 459             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 460             H                = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(G,H);
 462             Heps             = _mm_mul_pd(vfeps,H);
 463             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 464             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 465             velec            = _mm_mul_pd(qq20,VV);
 466             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 467             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 468
 469             /* Update potential sum for this i atom from the interaction with this j atom. */
 470             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 471             velecsum         = _mm_add_pd(velecsum,velec);
 472
 473             fscal            = felec;
 474
 475             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 476
 477             /* Calculate temporary vectorial force */
 478             tx               = _mm_mul_pd(fscal,dx20);
 479             ty               = _mm_mul_pd(fscal,dy20);
 480             tz               = _mm_mul_pd(fscal,dz20);
 481
 482             /* Update vectorial force */
 483             fix2             = _mm_add_pd(fix2,tx);
 484             fiy2             = _mm_add_pd(fiy2,ty);
 485             fiz2             = _mm_add_pd(fiz2,tz);
 486
 487             fjx0             = _mm_add_pd(fjx0,tx);
 488             fjy0             = _mm_add_pd(fjy0,ty);
 489             fjz0             = _mm_add_pd(fjz0,tz);
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             r30              = _mm_mul_pd(rsq30,rinv30);
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq30             = _mm_mul_pd(iq3,jq0);
 499
 500             /* Calculate table index by multiplying r with table scale and truncate to integer */
 501             rt               = _mm_mul_pd(r30,vftabscale);
 502             vfitab           = _mm_cvttpd_epi32(rt);
 503             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 504             vfitab           = _mm_slli_epi32(vfitab,2);
 505
 506             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 507             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 508             F                = _mm_setzero_pd();
 509             GMX_MM_TRANSPOSE2_PD(Y,F);
 510             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 511             H                = _mm_setzero_pd();
 512             GMX_MM_TRANSPOSE2_PD(G,H);
 513             Heps             = _mm_mul_pd(vfeps,H);
 514             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 515             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 516             velec            = _mm_mul_pd(qq30,VV);
 517             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 518             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 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,dx30);
 530             ty               = _mm_mul_pd(fscal,dy30);
 531             tz               = _mm_mul_pd(fscal,dz30);
 532
 533             /* Update vectorial force */
 534             fix3             = _mm_add_pd(fix3,tx);
 535             fiy3             = _mm_add_pd(fiy3,ty);
 536             fiz3             = _mm_add_pd(fiz3,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 132 flops */
 545         }
 546
 547         /* End of innermost loop */
 548
 549         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 550                                               f+i_coord_offset+DIM,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_W4_VF,outeriter*19 + inneriter*132);
 568 }
 569 /*
 570  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_double
 571  * Electrostatics interaction: CubicSplineTable
 572  * VdW interaction:            None
 573  * Geometry:                   Water4-Particle
 574  * Calculate force/pot:        Force
 575  */
 576 void
 577 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_double
 578                     (t_nblist                    * gmx_restrict       nlist,
 579                      rvec                        * gmx_restrict          xx,
 580                      rvec                        * gmx_restrict          ff,
 581                      struct 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              vdwioffset1;
 600     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 601     int              vdwioffset2;
 602     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 603     int              vdwioffset3;
 604     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 605     int              vdwjidx0A,vdwjidx0B;
 606     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 607     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 608     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 609     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 610     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 611     real             *charge;
 612     __m128i          vfitab;
 613     __m128i          ifour       = _mm_set1_epi32(4);
 614     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 615     real             *vftab;
 616     __m128d          dummy_mask,cutoff_mask;
 617     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 618     __m128d          one     = _mm_set1_pd(1.0);
 619     __m128d          two     = _mm_set1_pd(2.0);
 620     x                = xx[0];
 621     f                = ff[0];
 622
 623     nri              = nlist->nri;
 624     iinr             = nlist->iinr;
 625     jindex           = nlist->jindex;
 626     jjnr             = nlist->jjnr;
 627     shiftidx         = nlist->shift;
 628     gid              = nlist->gid;
 629     shiftvec         = fr->shift_vec[0];
 630     fshift           = fr->fshift[0];
 631     facel            = _mm_set1_pd(fr->ic->epsfac);
 632     charge           = mdatoms->chargeA;
 633
 634     vftab            = kernel_data->table_elec->data;
 635     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 636
 637     /* Setup water-specific parameters */
 638     inr              = nlist->iinr[0];
 639     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 640     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 641     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 642
 643     /* Avoid stupid compiler warnings */
 644     jnrA = jnrB = 0;
 645     j_coord_offsetA = 0;
 646     j_coord_offsetB = 0;
 647
 648     outeriter        = 0;
 649     inneriter        = 0;
 650
 651     /* Start outer loop over neighborlists */
 652     for(iidx=0; iidx<nri; iidx++)
 653     {
 654         /* Load shift vector for this list */
 655         i_shift_offset   = DIM*shiftidx[iidx];
 656
 657         /* Load limits for loop over neighbors */
 658         j_index_start    = jindex[iidx];
 659         j_index_end      = jindex[iidx+1];
 660
 661         /* Get outer coordinate index */
 662         inr              = iinr[iidx];
 663         i_coord_offset   = DIM*inr;
 664
 665         /* Load i particle coords and add shift vector */
 666         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 667                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 668
 669         fix1             = _mm_setzero_pd();
 670         fiy1             = _mm_setzero_pd();
 671         fiz1             = _mm_setzero_pd();
 672         fix2             = _mm_setzero_pd();
 673         fiy2             = _mm_setzero_pd();
 674         fiz2             = _mm_setzero_pd();
 675         fix3             = _mm_setzero_pd();
 676         fiy3             = _mm_setzero_pd();
 677         fiz3             = _mm_setzero_pd();
 678
 679         /* Start inner kernel loop */
 680         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 681         {
 682
 683             /* Get j neighbor index, and coordinate index */
 684             jnrA             = jjnr[jidx];
 685             jnrB             = jjnr[jidx+1];
 686             j_coord_offsetA  = DIM*jnrA;
 687             j_coord_offsetB  = DIM*jnrB;
 688
 689             /* load j atom coordinates */
 690             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 691                                               &jx0,&jy0,&jz0);
 692
 693             /* Calculate displacement vector */
 694             dx10             = _mm_sub_pd(ix1,jx0);
 695             dy10             = _mm_sub_pd(iy1,jy0);
 696             dz10             = _mm_sub_pd(iz1,jz0);
 697             dx20             = _mm_sub_pd(ix2,jx0);
 698             dy20             = _mm_sub_pd(iy2,jy0);
 699             dz20             = _mm_sub_pd(iz2,jz0);
 700             dx30             = _mm_sub_pd(ix3,jx0);
 701             dy30             = _mm_sub_pd(iy3,jy0);
 702             dz30             = _mm_sub_pd(iz3,jz0);
 703
 704             /* Calculate squared distance and things based on it */
 705             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 706             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 707             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 708
 709             rinv10           = sse41_invsqrt_d(rsq10);
 710             rinv20           = sse41_invsqrt_d(rsq20);
 711             rinv30           = sse41_invsqrt_d(rsq30);
 712
 713             /* Load parameters for j particles */
 714             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 715
 716             fjx0             = _mm_setzero_pd();
 717             fjy0             = _mm_setzero_pd();
 718             fjz0             = _mm_setzero_pd();
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             r10              = _mm_mul_pd(rsq10,rinv10);
 725
 726             /* Compute parameters for interactions between i and j atoms */
 727             qq10             = _mm_mul_pd(iq1,jq0);
 728
 729             /* Calculate table index by multiplying r with table scale and truncate to integer */
 730             rt               = _mm_mul_pd(r10,vftabscale);
 731             vfitab           = _mm_cvttpd_epi32(rt);
 732             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 733             vfitab           = _mm_slli_epi32(vfitab,2);
 734
 735             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 736             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 737             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 738             GMX_MM_TRANSPOSE2_PD(Y,F);
 739             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 740             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 741             GMX_MM_TRANSPOSE2_PD(G,H);
 742             Heps             = _mm_mul_pd(vfeps,H);
 743             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 744             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 745             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 746
 747             fscal            = felec;
 748
 749             /* Calculate temporary vectorial force */
 750             tx               = _mm_mul_pd(fscal,dx10);
 751             ty               = _mm_mul_pd(fscal,dy10);
 752             tz               = _mm_mul_pd(fscal,dz10);
 753
 754             /* Update vectorial force */
 755             fix1             = _mm_add_pd(fix1,tx);
 756             fiy1             = _mm_add_pd(fiy1,ty);
 757             fiz1             = _mm_add_pd(fiz1,tz);
 758
 759             fjx0             = _mm_add_pd(fjx0,tx);
 760             fjy0             = _mm_add_pd(fjy0,ty);
 761             fjz0             = _mm_add_pd(fjz0,tz);
 762
 763             /**************************
 764              * CALCULATE INTERACTIONS *
 765              **************************/
 766
 767             r20              = _mm_mul_pd(rsq20,rinv20);
 768
 769             /* Compute parameters for interactions between i and j atoms */
 770             qq20             = _mm_mul_pd(iq2,jq0);
 771
 772             /* Calculate table index by multiplying r with table scale and truncate to integer */
 773             rt               = _mm_mul_pd(r20,vftabscale);
 774             vfitab           = _mm_cvttpd_epi32(rt);
 775             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 776             vfitab           = _mm_slli_epi32(vfitab,2);
 777
 778             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 779             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 780             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 781             GMX_MM_TRANSPOSE2_PD(Y,F);
 782             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 783             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 784             GMX_MM_TRANSPOSE2_PD(G,H);
 785             Heps             = _mm_mul_pd(vfeps,H);
 786             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 787             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 788             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 789
 790             fscal            = felec;
 791
 792             /* Calculate temporary vectorial force */
 793             tx               = _mm_mul_pd(fscal,dx20);
 794             ty               = _mm_mul_pd(fscal,dy20);
 795             tz               = _mm_mul_pd(fscal,dz20);
 796
 797             /* Update vectorial force */
 798             fix2             = _mm_add_pd(fix2,tx);
 799             fiy2             = _mm_add_pd(fiy2,ty);
 800             fiz2             = _mm_add_pd(fiz2,tz);
 801
 802             fjx0             = _mm_add_pd(fjx0,tx);
 803             fjy0             = _mm_add_pd(fjy0,ty);
 804             fjz0             = _mm_add_pd(fjz0,tz);
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             r30              = _mm_mul_pd(rsq30,rinv30);
 811
 812             /* Compute parameters for interactions between i and j atoms */
 813             qq30             = _mm_mul_pd(iq3,jq0);
 814
 815             /* Calculate table index by multiplying r with table scale and truncate to integer */
 816             rt               = _mm_mul_pd(r30,vftabscale);
 817             vfitab           = _mm_cvttpd_epi32(rt);
 818             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 819             vfitab           = _mm_slli_epi32(vfitab,2);
 820
 821             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 822             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 823             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 824             GMX_MM_TRANSPOSE2_PD(Y,F);
 825             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 826             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 827             GMX_MM_TRANSPOSE2_PD(G,H);
 828             Heps             = _mm_mul_pd(vfeps,H);
 829             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 830             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 831             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 832
 833             fscal            = felec;
 834
 835             /* Calculate temporary vectorial force */
 836             tx               = _mm_mul_pd(fscal,dx30);
 837             ty               = _mm_mul_pd(fscal,dy30);
 838             tz               = _mm_mul_pd(fscal,dz30);
 839
 840             /* Update vectorial force */
 841             fix3             = _mm_add_pd(fix3,tx);
 842             fiy3             = _mm_add_pd(fiy3,ty);
 843             fiz3             = _mm_add_pd(fiz3,tz);
 844
 845             fjx0             = _mm_add_pd(fjx0,tx);
 846             fjy0             = _mm_add_pd(fjy0,ty);
 847             fjz0             = _mm_add_pd(fjz0,tz);
 848
 849             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 850
 851             /* Inner loop uses 120 flops */
 852         }
 853
 854         if(jidx<j_index_end)
 855         {
 856
 857             jnrA             = jjnr[jidx];
 858             j_coord_offsetA  = DIM*jnrA;
 859
 860             /* load j atom coordinates */
 861             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 862                                               &jx0,&jy0,&jz0);
 863
 864             /* Calculate displacement vector */
 865             dx10             = _mm_sub_pd(ix1,jx0);
 866             dy10             = _mm_sub_pd(iy1,jy0);
 867             dz10             = _mm_sub_pd(iz1,jz0);
 868             dx20             = _mm_sub_pd(ix2,jx0);
 869             dy20             = _mm_sub_pd(iy2,jy0);
 870             dz20             = _mm_sub_pd(iz2,jz0);
 871             dx30             = _mm_sub_pd(ix3,jx0);
 872             dy30             = _mm_sub_pd(iy3,jy0);
 873             dz30             = _mm_sub_pd(iz3,jz0);
 874
 875             /* Calculate squared distance and things based on it */
 876             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 877             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 878             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 879
 880             rinv10           = sse41_invsqrt_d(rsq10);
 881             rinv20           = sse41_invsqrt_d(rsq20);
 882             rinv30           = sse41_invsqrt_d(rsq30);
 883
 884             /* Load parameters for j particles */
 885             jq0              = _mm_load_sd(charge+jnrA+0);
 886
 887             fjx0             = _mm_setzero_pd();
 888             fjy0             = _mm_setzero_pd();
 889             fjz0             = _mm_setzero_pd();
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             r10              = _mm_mul_pd(rsq10,rinv10);
 896
 897             /* Compute parameters for interactions between i and j atoms */
 898             qq10             = _mm_mul_pd(iq1,jq0);
 899
 900             /* Calculate table index by multiplying r with table scale and truncate to integer */
 901             rt               = _mm_mul_pd(r10,vftabscale);
 902             vfitab           = _mm_cvttpd_epi32(rt);
 903             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 904             vfitab           = _mm_slli_epi32(vfitab,2);
 905
 906             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 907             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 908             F                = _mm_setzero_pd();
 909             GMX_MM_TRANSPOSE2_PD(Y,F);
 910             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 911             H                = _mm_setzero_pd();
 912             GMX_MM_TRANSPOSE2_PD(G,H);
 913             Heps             = _mm_mul_pd(vfeps,H);
 914             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 915             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 916             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 917
 918             fscal            = felec;
 919
 920             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm_mul_pd(fscal,dx10);
 924             ty               = _mm_mul_pd(fscal,dy10);
 925             tz               = _mm_mul_pd(fscal,dz10);
 926
 927             /* Update vectorial force */
 928             fix1             = _mm_add_pd(fix1,tx);
 929             fiy1             = _mm_add_pd(fiy1,ty);
 930             fiz1             = _mm_add_pd(fiz1,tz);
 931
 932             fjx0             = _mm_add_pd(fjx0,tx);
 933             fjy0             = _mm_add_pd(fjy0,ty);
 934             fjz0             = _mm_add_pd(fjz0,tz);
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             r20              = _mm_mul_pd(rsq20,rinv20);
 941
 942             /* Compute parameters for interactions between i and j atoms */
 943             qq20             = _mm_mul_pd(iq2,jq0);
 944
 945             /* Calculate table index by multiplying r with table scale and truncate to integer */
 946             rt               = _mm_mul_pd(r20,vftabscale);
 947             vfitab           = _mm_cvttpd_epi32(rt);
 948             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 949             vfitab           = _mm_slli_epi32(vfitab,2);
 950
 951             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 952             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 953             F                = _mm_setzero_pd();
 954             GMX_MM_TRANSPOSE2_PD(Y,F);
 955             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 956             H                = _mm_setzero_pd();
 957             GMX_MM_TRANSPOSE2_PD(G,H);
 958             Heps             = _mm_mul_pd(vfeps,H);
 959             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 960             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 961             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 962
 963             fscal            = felec;
 964
 965             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm_mul_pd(fscal,dx20);
 969             ty               = _mm_mul_pd(fscal,dy20);
 970             tz               = _mm_mul_pd(fscal,dz20);
 971
 972             /* Update vectorial force */
 973             fix2             = _mm_add_pd(fix2,tx);
 974             fiy2             = _mm_add_pd(fiy2,ty);
 975             fiz2             = _mm_add_pd(fiz2,tz);
 976
 977             fjx0             = _mm_add_pd(fjx0,tx);
 978             fjy0             = _mm_add_pd(fjy0,ty);
 979             fjz0             = _mm_add_pd(fjz0,tz);
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             r30              = _mm_mul_pd(rsq30,rinv30);
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq30             = _mm_mul_pd(iq3,jq0);
 989
 990             /* Calculate table index by multiplying r with table scale and truncate to integer */
 991             rt               = _mm_mul_pd(r30,vftabscale);
 992             vfitab           = _mm_cvttpd_epi32(rt);
 993             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 994             vfitab           = _mm_slli_epi32(vfitab,2);
 995
 996             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 997             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 998             F                = _mm_setzero_pd();
 999             GMX_MM_TRANSPOSE2_PD(Y,F);
1000             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1001             H                = _mm_setzero_pd();
1002             GMX_MM_TRANSPOSE2_PD(G,H);
1003             Heps             = _mm_mul_pd(vfeps,H);
1004             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1005             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1006             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm_mul_pd(fscal,dx30);
1014             ty               = _mm_mul_pd(fscal,dy30);
1015             tz               = _mm_mul_pd(fscal,dz30);
1016
1017             /* Update vectorial force */
1018             fix3             = _mm_add_pd(fix3,tx);
1019             fiy3             = _mm_add_pd(fiy3,ty);
1020             fiz3             = _mm_add_pd(fiz3,tz);
1021
1022             fjx0             = _mm_add_pd(fjx0,tx);
1023             fjy0             = _mm_add_pd(fjy0,ty);
1024             fjz0             = _mm_add_pd(fjz0,tz);
1025
1026             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1027
1028             /* Inner loop uses 120 flops */
1029         }
1030
1031         /* End of innermost loop */
1032
1033         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1034                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1035
1036         /* Increment number of inner iterations */
1037         inneriter                  += j_index_end - j_index_start;
1038
1039         /* Outer loop uses 18 flops */
1040     }
1041
1042     /* Increment number of outer iterations */
1043     outeriter        += nri;
1044
1045     /* Update outer/inner flops */
1046
1047     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1048 }