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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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_ElecCSTab_VdwNone_GeomW4P1_VF_sse4_1_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_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              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116
 117     vftab            = kernel_data->table_elec->data;
 118     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 123     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 124     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 150                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 151
 152         fix1             = _mm_setzero_pd();
 153         fiy1             = _mm_setzero_pd();
 154         fiz1             = _mm_setzero_pd();
 155         fix2             = _mm_setzero_pd();
 156         fiy2             = _mm_setzero_pd();
 157         fiz2             = _mm_setzero_pd();
 158         fix3             = _mm_setzero_pd();
 159         fiy3             = _mm_setzero_pd();
 160         fiz3             = _mm_setzero_pd();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174
 175             /* load j atom coordinates */
 176             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx10             = _mm_sub_pd(ix1,jx0);
 181             dy10             = _mm_sub_pd(iy1,jy0);
 182             dz10             = _mm_sub_pd(iz1,jz0);
 183             dx20             = _mm_sub_pd(ix2,jx0);
 184             dy20             = _mm_sub_pd(iy2,jy0);
 185             dz20             = _mm_sub_pd(iz2,jz0);
 186             dx30             = _mm_sub_pd(ix3,jx0);
 187             dy30             = _mm_sub_pd(iy3,jy0);
 188             dz30             = _mm_sub_pd(iz3,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 192             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 193             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 194
 195             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 196             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 197             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 201
 202             fjx0             = _mm_setzero_pd();
 203             fjy0             = _mm_setzero_pd();
 204             fjz0             = _mm_setzero_pd();
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             r10              = _mm_mul_pd(rsq10,rinv10);
 211
 212             /* Compute parameters for interactions between i and j atoms */
 213             qq10             = _mm_mul_pd(iq1,jq0);
 214
 215             /* Calculate table index by multiplying r with table scale and truncate to integer */
 216             rt               = _mm_mul_pd(r10,vftabscale);
 217             vfitab           = _mm_cvttpd_epi32(rt);
 218             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 219             vfitab           = _mm_slli_epi32(vfitab,2);
 220
 221             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 222             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 223             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 224             GMX_MM_TRANSPOSE2_PD(Y,F);
 225             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 226             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 227             GMX_MM_TRANSPOSE2_PD(G,H);
 228             Heps             = _mm_mul_pd(vfeps,H);
 229             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 230             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 231             velec            = _mm_mul_pd(qq10,VV);
 232             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 233             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 234
 235             /* Update potential sum for this i atom from the interaction with this j atom. */
 236             velecsum         = _mm_add_pd(velecsum,velec);
 237
 238             fscal            = felec;
 239
 240             /* Calculate temporary vectorial force */
 241             tx               = _mm_mul_pd(fscal,dx10);
 242             ty               = _mm_mul_pd(fscal,dy10);
 243             tz               = _mm_mul_pd(fscal,dz10);
 244
 245             /* Update vectorial force */
 246             fix1             = _mm_add_pd(fix1,tx);
 247             fiy1             = _mm_add_pd(fiy1,ty);
 248             fiz1             = _mm_add_pd(fiz1,tz);
 249
 250             fjx0             = _mm_add_pd(fjx0,tx);
 251             fjy0             = _mm_add_pd(fjy0,ty);
 252             fjz0             = _mm_add_pd(fjz0,tz);
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             r20              = _mm_mul_pd(rsq20,rinv20);
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq20             = _mm_mul_pd(iq2,jq0);
 262
 263             /* Calculate table index by multiplying r with table scale and truncate to integer */
 264             rt               = _mm_mul_pd(r20,vftabscale);
 265             vfitab           = _mm_cvttpd_epi32(rt);
 266             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 267             vfitab           = _mm_slli_epi32(vfitab,2);
 268
 269             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 270             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 271             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 272             GMX_MM_TRANSPOSE2_PD(Y,F);
 273             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 274             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 275             GMX_MM_TRANSPOSE2_PD(G,H);
 276             Heps             = _mm_mul_pd(vfeps,H);
 277             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 278             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 279             velec            = _mm_mul_pd(qq20,VV);
 280             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 281             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm_add_pd(velecsum,velec);
 285
 286             fscal            = felec;
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm_mul_pd(fscal,dx20);
 290             ty               = _mm_mul_pd(fscal,dy20);
 291             tz               = _mm_mul_pd(fscal,dz20);
 292
 293             /* Update vectorial force */
 294             fix2             = _mm_add_pd(fix2,tx);
 295             fiy2             = _mm_add_pd(fiy2,ty);
 296             fiz2             = _mm_add_pd(fiz2,tz);
 297
 298             fjx0             = _mm_add_pd(fjx0,tx);
 299             fjy0             = _mm_add_pd(fjy0,ty);
 300             fjz0             = _mm_add_pd(fjz0,tz);
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             r30              = _mm_mul_pd(rsq30,rinv30);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq30             = _mm_mul_pd(iq3,jq0);
 310
 311             /* Calculate table index by multiplying r with table scale and truncate to integer */
 312             rt               = _mm_mul_pd(r30,vftabscale);
 313             vfitab           = _mm_cvttpd_epi32(rt);
 314             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 315             vfitab           = _mm_slli_epi32(vfitab,2);
 316
 317             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 318             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 319             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 320             GMX_MM_TRANSPOSE2_PD(Y,F);
 321             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 322             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 323             GMX_MM_TRANSPOSE2_PD(G,H);
 324             Heps             = _mm_mul_pd(vfeps,H);
 325             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 326             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 327             velec            = _mm_mul_pd(qq30,VV);
 328             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 329             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velecsum         = _mm_add_pd(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             /* Calculate temporary vectorial force */
 337             tx               = _mm_mul_pd(fscal,dx30);
 338             ty               = _mm_mul_pd(fscal,dy30);
 339             tz               = _mm_mul_pd(fscal,dz30);
 340
 341             /* Update vectorial force */
 342             fix3             = _mm_add_pd(fix3,tx);
 343             fiy3             = _mm_add_pd(fiy3,ty);
 344             fiz3             = _mm_add_pd(fiz3,tz);
 345
 346             fjx0             = _mm_add_pd(fjx0,tx);
 347             fjy0             = _mm_add_pd(fjy0,ty);
 348             fjz0             = _mm_add_pd(fjz0,tz);
 349
 350             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 351
 352             /* Inner loop uses 132 flops */
 353         }
 354
 355         if(jidx<j_index_end)
 356         {
 357
 358             jnrA             = jjnr[jidx];
 359             j_coord_offsetA  = DIM*jnrA;
 360
 361             /* load j atom coordinates */
 362             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 363                                               &jx0,&jy0,&jz0);
 364
 365             /* Calculate displacement vector */
 366             dx10             = _mm_sub_pd(ix1,jx0);
 367             dy10             = _mm_sub_pd(iy1,jy0);
 368             dz10             = _mm_sub_pd(iz1,jz0);
 369             dx20             = _mm_sub_pd(ix2,jx0);
 370             dy20             = _mm_sub_pd(iy2,jy0);
 371             dz20             = _mm_sub_pd(iz2,jz0);
 372             dx30             = _mm_sub_pd(ix3,jx0);
 373             dy30             = _mm_sub_pd(iy3,jy0);
 374             dz30             = _mm_sub_pd(iz3,jz0);
 375
 376             /* Calculate squared distance and things based on it */
 377             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 378             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 379             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 380
 381             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 382             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 383             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 384
 385             /* Load parameters for j particles */
 386             jq0              = _mm_load_sd(charge+jnrA+0);
 387
 388             fjx0             = _mm_setzero_pd();
 389             fjy0             = _mm_setzero_pd();
 390             fjz0             = _mm_setzero_pd();
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             r10              = _mm_mul_pd(rsq10,rinv10);
 397
 398             /* Compute parameters for interactions between i and j atoms */
 399             qq10             = _mm_mul_pd(iq1,jq0);
 400
 401             /* Calculate table index by multiplying r with table scale and truncate to integer */
 402             rt               = _mm_mul_pd(r10,vftabscale);
 403             vfitab           = _mm_cvttpd_epi32(rt);
 404             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 405             vfitab           = _mm_slli_epi32(vfitab,2);
 406
 407             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 408             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 409             F                = _mm_setzero_pd();
 410             GMX_MM_TRANSPOSE2_PD(Y,F);
 411             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 412             H                = _mm_setzero_pd();
 413             GMX_MM_TRANSPOSE2_PD(G,H);
 414             Heps             = _mm_mul_pd(vfeps,H);
 415             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 416             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 417             velec            = _mm_mul_pd(qq10,VV);
 418             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 419             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 420
 421             /* Update potential sum for this i atom from the interaction with this j atom. */
 422             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 423             velecsum         = _mm_add_pd(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm_mul_pd(fscal,dx10);
 431             ty               = _mm_mul_pd(fscal,dy10);
 432             tz               = _mm_mul_pd(fscal,dz10);
 433
 434             /* Update vectorial force */
 435             fix1             = _mm_add_pd(fix1,tx);
 436             fiy1             = _mm_add_pd(fiy1,ty);
 437             fiz1             = _mm_add_pd(fiz1,tz);
 438
 439             fjx0             = _mm_add_pd(fjx0,tx);
 440             fjy0             = _mm_add_pd(fjy0,ty);
 441             fjz0             = _mm_add_pd(fjz0,tz);
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             r20              = _mm_mul_pd(rsq20,rinv20);
 448
 449             /* Compute parameters for interactions between i and j atoms */
 450             qq20             = _mm_mul_pd(iq2,jq0);
 451
 452             /* Calculate table index by multiplying r with table scale and truncate to integer */
 453             rt               = _mm_mul_pd(r20,vftabscale);
 454             vfitab           = _mm_cvttpd_epi32(rt);
 455             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 456             vfitab           = _mm_slli_epi32(vfitab,2);
 457
 458             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 459             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 460             F                = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(Y,F);
 462             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 463             H                = _mm_setzero_pd();
 464             GMX_MM_TRANSPOSE2_PD(G,H);
 465             Heps             = _mm_mul_pd(vfeps,H);
 466             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 467             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 468             velec            = _mm_mul_pd(qq20,VV);
 469             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 470             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 474             velecsum         = _mm_add_pd(velecsum,velec);
 475
 476             fscal            = felec;
 477
 478             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm_mul_pd(fscal,dx20);
 482             ty               = _mm_mul_pd(fscal,dy20);
 483             tz               = _mm_mul_pd(fscal,dz20);
 484
 485             /* Update vectorial force */
 486             fix2             = _mm_add_pd(fix2,tx);
 487             fiy2             = _mm_add_pd(fiy2,ty);
 488             fiz2             = _mm_add_pd(fiz2,tz);
 489
 490             fjx0             = _mm_add_pd(fjx0,tx);
 491             fjy0             = _mm_add_pd(fjy0,ty);
 492             fjz0             = _mm_add_pd(fjz0,tz);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r30              = _mm_mul_pd(rsq30,rinv30);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq30             = _mm_mul_pd(iq3,jq0);
 502
 503             /* Calculate table index by multiplying r with table scale and truncate to integer */
 504             rt               = _mm_mul_pd(r30,vftabscale);
 505             vfitab           = _mm_cvttpd_epi32(rt);
 506             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 507             vfitab           = _mm_slli_epi32(vfitab,2);
 508
 509             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 510             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 511             F                = _mm_setzero_pd();
 512             GMX_MM_TRANSPOSE2_PD(Y,F);
 513             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 514             H                = _mm_setzero_pd();
 515             GMX_MM_TRANSPOSE2_PD(G,H);
 516             Heps             = _mm_mul_pd(vfeps,H);
 517             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 518             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 519             velec            = _mm_mul_pd(qq30,VV);
 520             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 521             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 525             velecsum         = _mm_add_pd(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx30);
 533             ty               = _mm_mul_pd(fscal,dy30);
 534             tz               = _mm_mul_pd(fscal,dz30);
 535
 536             /* Update vectorial force */
 537             fix3             = _mm_add_pd(fix3,tx);
 538             fiy3             = _mm_add_pd(fiy3,ty);
 539             fiz3             = _mm_add_pd(fiz3,tz);
 540
 541             fjx0             = _mm_add_pd(fjx0,tx);
 542             fjy0             = _mm_add_pd(fjy0,ty);
 543             fjz0             = _mm_add_pd(fjz0,tz);
 544
 545             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 546
 547             /* Inner loop uses 132 flops */
 548         }
 549
 550         /* End of innermost loop */
 551
 552         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 553                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 554
 555         ggid                        = gid[iidx];
 556         /* Update potential energies */
 557         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 558
 559         /* Increment number of inner iterations */
 560         inneriter                  += j_index_end - j_index_start;
 561
 562         /* Outer loop uses 19 flops */
 563     }
 564
 565     /* Increment number of outer iterations */
 566     outeriter        += nri;
 567
 568     /* Update outer/inner flops */
 569
 570     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*132);
 571 }
 572 /*
 573  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_double
 574  * Electrostatics interaction: CubicSplineTable
 575  * VdW interaction:            None
 576  * Geometry:                   Water4-Particle
 577  * Calculate force/pot:        Force
 578  */
 579 void
 580 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_double
 581                     (t_nblist                    * gmx_restrict       nlist,
 582                      rvec                        * gmx_restrict          xx,
 583                      rvec                        * gmx_restrict          ff,
 584                      t_forcerec                  * gmx_restrict          fr,
 585                      t_mdatoms                   * gmx_restrict     mdatoms,
 586                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 587                      t_nrnb                      * gmx_restrict        nrnb)
 588 {
 589     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 590      * just 0 for non-waters.
 591      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 592      * jnr indices corresponding to data put in the four positions in the SIMD register.
 593      */
 594     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 595     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 596     int              jnrA,jnrB;
 597     int              j_coord_offsetA,j_coord_offsetB;
 598     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 599     real             rcutoff_scalar;
 600     real             *shiftvec,*fshift,*x,*f;
 601     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 602     int              vdwioffset1;
 603     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 604     int              vdwioffset2;
 605     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 606     int              vdwioffset3;
 607     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 608     int              vdwjidx0A,vdwjidx0B;
 609     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 610     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 611     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 612     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 613     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 614     real             *charge;
 615     __m128i          vfitab;
 616     __m128i          ifour       = _mm_set1_epi32(4);
 617     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 618     real             *vftab;
 619     __m128d          dummy_mask,cutoff_mask;
 620     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 621     __m128d          one     = _mm_set1_pd(1.0);
 622     __m128d          two     = _mm_set1_pd(2.0);
 623     x                = xx[0];
 624     f                = ff[0];
 625
 626     nri              = nlist->nri;
 627     iinr             = nlist->iinr;
 628     jindex           = nlist->jindex;
 629     jjnr             = nlist->jjnr;
 630     shiftidx         = nlist->shift;
 631     gid              = nlist->gid;
 632     shiftvec         = fr->shift_vec[0];
 633     fshift           = fr->fshift[0];
 634     facel            = _mm_set1_pd(fr->epsfac);
 635     charge           = mdatoms->chargeA;
 636
 637     vftab            = kernel_data->table_elec->data;
 638     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 639
 640     /* Setup water-specific parameters */
 641     inr              = nlist->iinr[0];
 642     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 643     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 644     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 645
 646     /* Avoid stupid compiler warnings */
 647     jnrA = jnrB = 0;
 648     j_coord_offsetA = 0;
 649     j_coord_offsetB = 0;
 650
 651     outeriter        = 0;
 652     inneriter        = 0;
 653
 654     /* Start outer loop over neighborlists */
 655     for(iidx=0; iidx<nri; iidx++)
 656     {
 657         /* Load shift vector for this list */
 658         i_shift_offset   = DIM*shiftidx[iidx];
 659
 660         /* Load limits for loop over neighbors */
 661         j_index_start    = jindex[iidx];
 662         j_index_end      = jindex[iidx+1];
 663
 664         /* Get outer coordinate index */
 665         inr              = iinr[iidx];
 666         i_coord_offset   = DIM*inr;
 667
 668         /* Load i particle coords and add shift vector */
 669         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 670                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 671
 672         fix1             = _mm_setzero_pd();
 673         fiy1             = _mm_setzero_pd();
 674         fiz1             = _mm_setzero_pd();
 675         fix2             = _mm_setzero_pd();
 676         fiy2             = _mm_setzero_pd();
 677         fiz2             = _mm_setzero_pd();
 678         fix3             = _mm_setzero_pd();
 679         fiy3             = _mm_setzero_pd();
 680         fiz3             = _mm_setzero_pd();
 681
 682         /* Start inner kernel loop */
 683         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 684         {
 685
 686             /* Get j neighbor index, and coordinate index */
 687             jnrA             = jjnr[jidx];
 688             jnrB             = jjnr[jidx+1];
 689             j_coord_offsetA  = DIM*jnrA;
 690             j_coord_offsetB  = DIM*jnrB;
 691
 692             /* load j atom coordinates */
 693             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 694                                               &jx0,&jy0,&jz0);
 695
 696             /* Calculate displacement vector */
 697             dx10             = _mm_sub_pd(ix1,jx0);
 698             dy10             = _mm_sub_pd(iy1,jy0);
 699             dz10             = _mm_sub_pd(iz1,jz0);
 700             dx20             = _mm_sub_pd(ix2,jx0);
 701             dy20             = _mm_sub_pd(iy2,jy0);
 702             dz20             = _mm_sub_pd(iz2,jz0);
 703             dx30             = _mm_sub_pd(ix3,jx0);
 704             dy30             = _mm_sub_pd(iy3,jy0);
 705             dz30             = _mm_sub_pd(iz3,jz0);
 706
 707             /* Calculate squared distance and things based on it */
 708             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 709             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 710             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 711
 712             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 713             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 714             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 715
 716             /* Load parameters for j particles */
 717             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 718
 719             fjx0             = _mm_setzero_pd();
 720             fjy0             = _mm_setzero_pd();
 721             fjz0             = _mm_setzero_pd();
 722
 723             /**************************
 724              * CALCULATE INTERACTIONS *
 725              **************************/
 726
 727             r10              = _mm_mul_pd(rsq10,rinv10);
 728
 729             /* Compute parameters for interactions between i and j atoms */
 730             qq10             = _mm_mul_pd(iq1,jq0);
 731
 732             /* Calculate table index by multiplying r with table scale and truncate to integer */
 733             rt               = _mm_mul_pd(r10,vftabscale);
 734             vfitab           = _mm_cvttpd_epi32(rt);
 735             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 736             vfitab           = _mm_slli_epi32(vfitab,2);
 737
 738             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 739             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 740             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 741             GMX_MM_TRANSPOSE2_PD(Y,F);
 742             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 743             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 744             GMX_MM_TRANSPOSE2_PD(G,H);
 745             Heps             = _mm_mul_pd(vfeps,H);
 746             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 747             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 748             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 749
 750             fscal            = felec;
 751
 752             /* Calculate temporary vectorial force */
 753             tx               = _mm_mul_pd(fscal,dx10);
 754             ty               = _mm_mul_pd(fscal,dy10);
 755             tz               = _mm_mul_pd(fscal,dz10);
 756
 757             /* Update vectorial force */
 758             fix1             = _mm_add_pd(fix1,tx);
 759             fiy1             = _mm_add_pd(fiy1,ty);
 760             fiz1             = _mm_add_pd(fiz1,tz);
 761
 762             fjx0             = _mm_add_pd(fjx0,tx);
 763             fjy0             = _mm_add_pd(fjy0,ty);
 764             fjz0             = _mm_add_pd(fjz0,tz);
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             r20              = _mm_mul_pd(rsq20,rinv20);
 771
 772             /* Compute parameters for interactions between i and j atoms */
 773             qq20             = _mm_mul_pd(iq2,jq0);
 774
 775             /* Calculate table index by multiplying r with table scale and truncate to integer */
 776             rt               = _mm_mul_pd(r20,vftabscale);
 777             vfitab           = _mm_cvttpd_epi32(rt);
 778             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 779             vfitab           = _mm_slli_epi32(vfitab,2);
 780
 781             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 782             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 783             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 784             GMX_MM_TRANSPOSE2_PD(Y,F);
 785             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 786             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 787             GMX_MM_TRANSPOSE2_PD(G,H);
 788             Heps             = _mm_mul_pd(vfeps,H);
 789             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 790             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 791             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 792
 793             fscal            = felec;
 794
 795             /* Calculate temporary vectorial force */
 796             tx               = _mm_mul_pd(fscal,dx20);
 797             ty               = _mm_mul_pd(fscal,dy20);
 798             tz               = _mm_mul_pd(fscal,dz20);
 799
 800             /* Update vectorial force */
 801             fix2             = _mm_add_pd(fix2,tx);
 802             fiy2             = _mm_add_pd(fiy2,ty);
 803             fiz2             = _mm_add_pd(fiz2,tz);
 804
 805             fjx0             = _mm_add_pd(fjx0,tx);
 806             fjy0             = _mm_add_pd(fjy0,ty);
 807             fjz0             = _mm_add_pd(fjz0,tz);
 808
 809             /**************************
 810              * CALCULATE INTERACTIONS *
 811              **************************/
 812
 813             r30              = _mm_mul_pd(rsq30,rinv30);
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             qq30             = _mm_mul_pd(iq3,jq0);
 817
 818             /* Calculate table index by multiplying r with table scale and truncate to integer */
 819             rt               = _mm_mul_pd(r30,vftabscale);
 820             vfitab           = _mm_cvttpd_epi32(rt);
 821             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 822             vfitab           = _mm_slli_epi32(vfitab,2);
 823
 824             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 825             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 826             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 827             GMX_MM_TRANSPOSE2_PD(Y,F);
 828             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 829             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 830             GMX_MM_TRANSPOSE2_PD(G,H);
 831             Heps             = _mm_mul_pd(vfeps,H);
 832             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 833             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 834             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 835
 836             fscal            = felec;
 837
 838             /* Calculate temporary vectorial force */
 839             tx               = _mm_mul_pd(fscal,dx30);
 840             ty               = _mm_mul_pd(fscal,dy30);
 841             tz               = _mm_mul_pd(fscal,dz30);
 842
 843             /* Update vectorial force */
 844             fix3             = _mm_add_pd(fix3,tx);
 845             fiy3             = _mm_add_pd(fiy3,ty);
 846             fiz3             = _mm_add_pd(fiz3,tz);
 847
 848             fjx0             = _mm_add_pd(fjx0,tx);
 849             fjy0             = _mm_add_pd(fjy0,ty);
 850             fjz0             = _mm_add_pd(fjz0,tz);
 851
 852             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 853
 854             /* Inner loop uses 120 flops */
 855         }
 856
 857         if(jidx<j_index_end)
 858         {
 859
 860             jnrA             = jjnr[jidx];
 861             j_coord_offsetA  = DIM*jnrA;
 862
 863             /* load j atom coordinates */
 864             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 865                                               &jx0,&jy0,&jz0);
 866
 867             /* Calculate displacement vector */
 868             dx10             = _mm_sub_pd(ix1,jx0);
 869             dy10             = _mm_sub_pd(iy1,jy0);
 870             dz10             = _mm_sub_pd(iz1,jz0);
 871             dx20             = _mm_sub_pd(ix2,jx0);
 872             dy20             = _mm_sub_pd(iy2,jy0);
 873             dz20             = _mm_sub_pd(iz2,jz0);
 874             dx30             = _mm_sub_pd(ix3,jx0);
 875             dy30             = _mm_sub_pd(iy3,jy0);
 876             dz30             = _mm_sub_pd(iz3,jz0);
 877
 878             /* Calculate squared distance and things based on it */
 879             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 880             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 881             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 882
 883             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 884             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 885             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 886
 887             /* Load parameters for j particles */
 888             jq0              = _mm_load_sd(charge+jnrA+0);
 889
 890             fjx0             = _mm_setzero_pd();
 891             fjy0             = _mm_setzero_pd();
 892             fjz0             = _mm_setzero_pd();
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             r10              = _mm_mul_pd(rsq10,rinv10);
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq10             = _mm_mul_pd(iq1,jq0);
 902
 903             /* Calculate table index by multiplying r with table scale and truncate to integer */
 904             rt               = _mm_mul_pd(r10,vftabscale);
 905             vfitab           = _mm_cvttpd_epi32(rt);
 906             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 907             vfitab           = _mm_slli_epi32(vfitab,2);
 908
 909             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 910             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 911             F                = _mm_setzero_pd();
 912             GMX_MM_TRANSPOSE2_PD(Y,F);
 913             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 914             H                = _mm_setzero_pd();
 915             GMX_MM_TRANSPOSE2_PD(G,H);
 916             Heps             = _mm_mul_pd(vfeps,H);
 917             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 918             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 919             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 920
 921             fscal            = felec;
 922
 923             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 924
 925             /* Calculate temporary vectorial force */
 926             tx               = _mm_mul_pd(fscal,dx10);
 927             ty               = _mm_mul_pd(fscal,dy10);
 928             tz               = _mm_mul_pd(fscal,dz10);
 929
 930             /* Update vectorial force */
 931             fix1             = _mm_add_pd(fix1,tx);
 932             fiy1             = _mm_add_pd(fiy1,ty);
 933             fiz1             = _mm_add_pd(fiz1,tz);
 934
 935             fjx0             = _mm_add_pd(fjx0,tx);
 936             fjy0             = _mm_add_pd(fjy0,ty);
 937             fjz0             = _mm_add_pd(fjz0,tz);
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             r20              = _mm_mul_pd(rsq20,rinv20);
 944
 945             /* Compute parameters for interactions between i and j atoms */
 946             qq20             = _mm_mul_pd(iq2,jq0);
 947
 948             /* Calculate table index by multiplying r with table scale and truncate to integer */
 949             rt               = _mm_mul_pd(r20,vftabscale);
 950             vfitab           = _mm_cvttpd_epi32(rt);
 951             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 952             vfitab           = _mm_slli_epi32(vfitab,2);
 953
 954             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 955             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 956             F                = _mm_setzero_pd();
 957             GMX_MM_TRANSPOSE2_PD(Y,F);
 958             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 959             H                = _mm_setzero_pd();
 960             GMX_MM_TRANSPOSE2_PD(G,H);
 961             Heps             = _mm_mul_pd(vfeps,H);
 962             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 963             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 964             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 965
 966             fscal            = felec;
 967
 968             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 969
 970             /* Calculate temporary vectorial force */
 971             tx               = _mm_mul_pd(fscal,dx20);
 972             ty               = _mm_mul_pd(fscal,dy20);
 973             tz               = _mm_mul_pd(fscal,dz20);
 974
 975             /* Update vectorial force */
 976             fix2             = _mm_add_pd(fix2,tx);
 977             fiy2             = _mm_add_pd(fiy2,ty);
 978             fiz2             = _mm_add_pd(fiz2,tz);
 979
 980             fjx0             = _mm_add_pd(fjx0,tx);
 981             fjy0             = _mm_add_pd(fjy0,ty);
 982             fjz0             = _mm_add_pd(fjz0,tz);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r30              = _mm_mul_pd(rsq30,rinv30);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq30             = _mm_mul_pd(iq3,jq0);
 992
 993             /* Calculate table index by multiplying r with table scale and truncate to integer */
 994             rt               = _mm_mul_pd(r30,vftabscale);
 995             vfitab           = _mm_cvttpd_epi32(rt);
 996             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 997             vfitab           = _mm_slli_epi32(vfitab,2);
 998
 999             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1000             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1001             F                = _mm_setzero_pd();
1002             GMX_MM_TRANSPOSE2_PD(Y,F);
1003             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1004             H                = _mm_setzero_pd();
1005             GMX_MM_TRANSPOSE2_PD(G,H);
1006             Heps             = _mm_mul_pd(vfeps,H);
1007             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1008             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1009             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1010
1011             fscal            = felec;
1012
1013             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1014
1015             /* Calculate temporary vectorial force */
1016             tx               = _mm_mul_pd(fscal,dx30);
1017             ty               = _mm_mul_pd(fscal,dy30);
1018             tz               = _mm_mul_pd(fscal,dz30);
1019
1020             /* Update vectorial force */
1021             fix3             = _mm_add_pd(fix3,tx);
1022             fiy3             = _mm_add_pd(fiy3,ty);
1023             fiz3             = _mm_add_pd(fiz3,tz);
1024
1025             fjx0             = _mm_add_pd(fjx0,tx);
1026             fjy0             = _mm_add_pd(fjy0,ty);
1027             fjz0             = _mm_add_pd(fjz0,tz);
1028
1029             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1030
1031             /* Inner loop uses 120 flops */
1032         }
1033
1034         /* End of innermost loop */
1035
1036         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1037                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1038
1039         /* Increment number of inner iterations */
1040         inneriter                  += j_index_end - j_index_start;
1041
1042         /* Outer loop uses 18 flops */
1043     }
1044
1045     /* Increment number of outer iterations */
1046     outeriter        += nri;
1047
1048     /* Update outer/inner flops */
1049
1050     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1051 }