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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_sse4_1_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset1;
  81     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  82     int              vdwioffset2;
  83     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  84     int              vdwioffset3;
  85     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114
 115     vftab            = kernel_data->table_elec->data;
 116     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 117
 118     /* Setup water-specific parameters */
 119     inr              = nlist->iinr[0];
 120     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 121     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 122     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 148                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 149
 150         fix1             = _mm_setzero_pd();
 151         fiy1             = _mm_setzero_pd();
 152         fiz1             = _mm_setzero_pd();
 153         fix2             = _mm_setzero_pd();
 154         fiy2             = _mm_setzero_pd();
 155         fiz2             = _mm_setzero_pd();
 156         fix3             = _mm_setzero_pd();
 157         fiy3             = _mm_setzero_pd();
 158         fiz3             = _mm_setzero_pd();
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx10             = _mm_sub_pd(ix1,jx0);
 179             dy10             = _mm_sub_pd(iy1,jy0);
 180             dz10             = _mm_sub_pd(iz1,jz0);
 181             dx20             = _mm_sub_pd(ix2,jx0);
 182             dy20             = _mm_sub_pd(iy2,jy0);
 183             dz20             = _mm_sub_pd(iz2,jz0);
 184             dx30             = _mm_sub_pd(ix3,jx0);
 185             dy30             = _mm_sub_pd(iy3,jy0);
 186             dz30             = _mm_sub_pd(iz3,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 190             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 191             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 192
 193             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 194             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 195             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 196
 197             /* Load parameters for j particles */
 198             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 199
 200             fjx0             = _mm_setzero_pd();
 201             fjy0             = _mm_setzero_pd();
 202             fjz0             = _mm_setzero_pd();
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             r10              = _mm_mul_pd(rsq10,rinv10);
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq10             = _mm_mul_pd(iq1,jq0);
 212
 213             /* Calculate table index by multiplying r with table scale and truncate to integer */
 214             rt               = _mm_mul_pd(r10,vftabscale);
 215             vfitab           = _mm_cvttpd_epi32(rt);
 216             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 217             vfitab           = _mm_slli_epi32(vfitab,2);
 218
 219             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 220             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 221             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 222             GMX_MM_TRANSPOSE2_PD(Y,F);
 223             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 224             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 225             GMX_MM_TRANSPOSE2_PD(G,H);
 226             Heps             = _mm_mul_pd(vfeps,H);
 227             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 228             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 229             velec            = _mm_mul_pd(qq10,VV);
 230             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 231             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             velecsum         = _mm_add_pd(velecsum,velec);
 235
 236             fscal            = felec;
 237
 238             /* Calculate temporary vectorial force */
 239             tx               = _mm_mul_pd(fscal,dx10);
 240             ty               = _mm_mul_pd(fscal,dy10);
 241             tz               = _mm_mul_pd(fscal,dz10);
 242
 243             /* Update vectorial force */
 244             fix1             = _mm_add_pd(fix1,tx);
 245             fiy1             = _mm_add_pd(fiy1,ty);
 246             fiz1             = _mm_add_pd(fiz1,tz);
 247
 248             fjx0             = _mm_add_pd(fjx0,tx);
 249             fjy0             = _mm_add_pd(fjy0,ty);
 250             fjz0             = _mm_add_pd(fjz0,tz);
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             r20              = _mm_mul_pd(rsq20,rinv20);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq20             = _mm_mul_pd(iq2,jq0);
 260
 261             /* Calculate table index by multiplying r with table scale and truncate to integer */
 262             rt               = _mm_mul_pd(r20,vftabscale);
 263             vfitab           = _mm_cvttpd_epi32(rt);
 264             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 265             vfitab           = _mm_slli_epi32(vfitab,2);
 266
 267             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 268             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 269             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 270             GMX_MM_TRANSPOSE2_PD(Y,F);
 271             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 272             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 273             GMX_MM_TRANSPOSE2_PD(G,H);
 274             Heps             = _mm_mul_pd(vfeps,H);
 275             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 276             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 277             velec            = _mm_mul_pd(qq20,VV);
 278             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 279             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_pd(velecsum,velec);
 283
 284             fscal            = felec;
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_pd(fscal,dx20);
 288             ty               = _mm_mul_pd(fscal,dy20);
 289             tz               = _mm_mul_pd(fscal,dz20);
 290
 291             /* Update vectorial force */
 292             fix2             = _mm_add_pd(fix2,tx);
 293             fiy2             = _mm_add_pd(fiy2,ty);
 294             fiz2             = _mm_add_pd(fiz2,tz);
 295
 296             fjx0             = _mm_add_pd(fjx0,tx);
 297             fjy0             = _mm_add_pd(fjy0,ty);
 298             fjz0             = _mm_add_pd(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r30              = _mm_mul_pd(rsq30,rinv30);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq30             = _mm_mul_pd(iq3,jq0);
 308
 309             /* Calculate table index by multiplying r with table scale and truncate to integer */
 310             rt               = _mm_mul_pd(r30,vftabscale);
 311             vfitab           = _mm_cvttpd_epi32(rt);
 312             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 313             vfitab           = _mm_slli_epi32(vfitab,2);
 314
 315             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 316             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 317             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 318             GMX_MM_TRANSPOSE2_PD(Y,F);
 319             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 320             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 321             GMX_MM_TRANSPOSE2_PD(G,H);
 322             Heps             = _mm_mul_pd(vfeps,H);
 323             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 324             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 325             velec            = _mm_mul_pd(qq30,VV);
 326             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 327             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velecsum         = _mm_add_pd(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm_mul_pd(fscal,dx30);
 336             ty               = _mm_mul_pd(fscal,dy30);
 337             tz               = _mm_mul_pd(fscal,dz30);
 338
 339             /* Update vectorial force */
 340             fix3             = _mm_add_pd(fix3,tx);
 341             fiy3             = _mm_add_pd(fiy3,ty);
 342             fiz3             = _mm_add_pd(fiz3,tz);
 343
 344             fjx0             = _mm_add_pd(fjx0,tx);
 345             fjy0             = _mm_add_pd(fjy0,ty);
 346             fjz0             = _mm_add_pd(fjz0,tz);
 347
 348             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 349
 350             /* Inner loop uses 132 flops */
 351         }
 352
 353         if(jidx<j_index_end)
 354         {
 355
 356             jnrA             = jjnr[jidx];
 357             j_coord_offsetA  = DIM*jnrA;
 358
 359             /* load j atom coordinates */
 360             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 361                                               &jx0,&jy0,&jz0);
 362
 363             /* Calculate displacement vector */
 364             dx10             = _mm_sub_pd(ix1,jx0);
 365             dy10             = _mm_sub_pd(iy1,jy0);
 366             dz10             = _mm_sub_pd(iz1,jz0);
 367             dx20             = _mm_sub_pd(ix2,jx0);
 368             dy20             = _mm_sub_pd(iy2,jy0);
 369             dz20             = _mm_sub_pd(iz2,jz0);
 370             dx30             = _mm_sub_pd(ix3,jx0);
 371             dy30             = _mm_sub_pd(iy3,jy0);
 372             dz30             = _mm_sub_pd(iz3,jz0);
 373
 374             /* Calculate squared distance and things based on it */
 375             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 376             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 377             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 378
 379             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 380             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 381             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 382
 383             /* Load parameters for j particles */
 384             jq0              = _mm_load_sd(charge+jnrA+0);
 385
 386             fjx0             = _mm_setzero_pd();
 387             fjy0             = _mm_setzero_pd();
 388             fjz0             = _mm_setzero_pd();
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             r10              = _mm_mul_pd(rsq10,rinv10);
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq10             = _mm_mul_pd(iq1,jq0);
 398
 399             /* Calculate table index by multiplying r with table scale and truncate to integer */
 400             rt               = _mm_mul_pd(r10,vftabscale);
 401             vfitab           = _mm_cvttpd_epi32(rt);
 402             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 403             vfitab           = _mm_slli_epi32(vfitab,2);
 404
 405             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 406             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 407             F                = _mm_setzero_pd();
 408             GMX_MM_TRANSPOSE2_PD(Y,F);
 409             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 410             H                = _mm_setzero_pd();
 411             GMX_MM_TRANSPOSE2_PD(G,H);
 412             Heps             = _mm_mul_pd(vfeps,H);
 413             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 414             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 415             velec            = _mm_mul_pd(qq10,VV);
 416             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 417             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 421             velecsum         = _mm_add_pd(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm_mul_pd(fscal,dx10);
 429             ty               = _mm_mul_pd(fscal,dy10);
 430             tz               = _mm_mul_pd(fscal,dz10);
 431
 432             /* Update vectorial force */
 433             fix1             = _mm_add_pd(fix1,tx);
 434             fiy1             = _mm_add_pd(fiy1,ty);
 435             fiz1             = _mm_add_pd(fiz1,tz);
 436
 437             fjx0             = _mm_add_pd(fjx0,tx);
 438             fjy0             = _mm_add_pd(fjy0,ty);
 439             fjz0             = _mm_add_pd(fjz0,tz);
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             r20              = _mm_mul_pd(rsq20,rinv20);
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             qq20             = _mm_mul_pd(iq2,jq0);
 449
 450             /* Calculate table index by multiplying r with table scale and truncate to integer */
 451             rt               = _mm_mul_pd(r20,vftabscale);
 452             vfitab           = _mm_cvttpd_epi32(rt);
 453             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 454             vfitab           = _mm_slli_epi32(vfitab,2);
 455
 456             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 457             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 458             F                = _mm_setzero_pd();
 459             GMX_MM_TRANSPOSE2_PD(Y,F);
 460             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 461             H                = _mm_setzero_pd();
 462             GMX_MM_TRANSPOSE2_PD(G,H);
 463             Heps             = _mm_mul_pd(vfeps,H);
 464             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 465             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 466             velec            = _mm_mul_pd(qq20,VV);
 467             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 468             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 472             velecsum         = _mm_add_pd(velecsum,velec);
 473
 474             fscal            = felec;
 475
 476             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 477
 478             /* Calculate temporary vectorial force */
 479             tx               = _mm_mul_pd(fscal,dx20);
 480             ty               = _mm_mul_pd(fscal,dy20);
 481             tz               = _mm_mul_pd(fscal,dz20);
 482
 483             /* Update vectorial force */
 484             fix2             = _mm_add_pd(fix2,tx);
 485             fiy2             = _mm_add_pd(fiy2,ty);
 486             fiz2             = _mm_add_pd(fiz2,tz);
 487
 488             fjx0             = _mm_add_pd(fjx0,tx);
 489             fjy0             = _mm_add_pd(fjy0,ty);
 490             fjz0             = _mm_add_pd(fjz0,tz);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             r30              = _mm_mul_pd(rsq30,rinv30);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq30             = _mm_mul_pd(iq3,jq0);
 500
 501             /* Calculate table index by multiplying r with table scale and truncate to integer */
 502             rt               = _mm_mul_pd(r30,vftabscale);
 503             vfitab           = _mm_cvttpd_epi32(rt);
 504             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 505             vfitab           = _mm_slli_epi32(vfitab,2);
 506
 507             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 508             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 509             F                = _mm_setzero_pd();
 510             GMX_MM_TRANSPOSE2_PD(Y,F);
 511             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 512             H                = _mm_setzero_pd();
 513             GMX_MM_TRANSPOSE2_PD(G,H);
 514             Heps             = _mm_mul_pd(vfeps,H);
 515             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 516             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 517             velec            = _mm_mul_pd(qq30,VV);
 518             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 519             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 520
 521             /* Update potential sum for this i atom from the interaction with this j atom. */
 522             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 523             velecsum         = _mm_add_pd(velecsum,velec);
 524
 525             fscal            = felec;
 526
 527             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_pd(fscal,dx30);
 531             ty               = _mm_mul_pd(fscal,dy30);
 532             tz               = _mm_mul_pd(fscal,dz30);
 533
 534             /* Update vectorial force */
 535             fix3             = _mm_add_pd(fix3,tx);
 536             fiy3             = _mm_add_pd(fiy3,ty);
 537             fiz3             = _mm_add_pd(fiz3,tz);
 538
 539             fjx0             = _mm_add_pd(fjx0,tx);
 540             fjy0             = _mm_add_pd(fjy0,ty);
 541             fjz0             = _mm_add_pd(fjz0,tz);
 542
 543             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 544
 545             /* Inner loop uses 132 flops */
 546         }
 547
 548         /* End of innermost loop */
 549
 550         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 551                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 552
 553         ggid                        = gid[iidx];
 554         /* Update potential energies */
 555         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 556
 557         /* Increment number of inner iterations */
 558         inneriter                  += j_index_end - j_index_start;
 559
 560         /* Outer loop uses 19 flops */
 561     }
 562
 563     /* Increment number of outer iterations */
 564     outeriter        += nri;
 565
 566     /* Update outer/inner flops */
 567
 568     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*132);
 569 }
 570 /*
 571  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_double
 572  * Electrostatics interaction: CubicSplineTable
 573  * VdW interaction:            None
 574  * Geometry:                   Water4-Particle
 575  * Calculate force/pot:        Force
 576  */
 577 void
 578 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_double
 579                     (t_nblist                    * gmx_restrict       nlist,
 580                      rvec                        * gmx_restrict          xx,
 581                      rvec                        * gmx_restrict          ff,
 582                      t_forcerec                  * gmx_restrict          fr,
 583                      t_mdatoms                   * gmx_restrict     mdatoms,
 584                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 585                      t_nrnb                      * gmx_restrict        nrnb)
 586 {
 587     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 588      * just 0 for non-waters.
 589      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 590      * jnr indices corresponding to data put in the four positions in the SIMD register.
 591      */
 592     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 593     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 594     int              jnrA,jnrB;
 595     int              j_coord_offsetA,j_coord_offsetB;
 596     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 597     real             rcutoff_scalar;
 598     real             *shiftvec,*fshift,*x,*f;
 599     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 600     int              vdwioffset1;
 601     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 602     int              vdwioffset2;
 603     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 604     int              vdwioffset3;
 605     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 606     int              vdwjidx0A,vdwjidx0B;
 607     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 608     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 609     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 610     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 611     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 612     real             *charge;
 613     __m128i          vfitab;
 614     __m128i          ifour       = _mm_set1_epi32(4);
 615     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 616     real             *vftab;
 617     __m128d          dummy_mask,cutoff_mask;
 618     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 619     __m128d          one     = _mm_set1_pd(1.0);
 620     __m128d          two     = _mm_set1_pd(2.0);
 621     x                = xx[0];
 622     f                = ff[0];
 623
 624     nri              = nlist->nri;
 625     iinr             = nlist->iinr;
 626     jindex           = nlist->jindex;
 627     jjnr             = nlist->jjnr;
 628     shiftidx         = nlist->shift;
 629     gid              = nlist->gid;
 630     shiftvec         = fr->shift_vec[0];
 631     fshift           = fr->fshift[0];
 632     facel            = _mm_set1_pd(fr->epsfac);
 633     charge           = mdatoms->chargeA;
 634
 635     vftab            = kernel_data->table_elec->data;
 636     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 637
 638     /* Setup water-specific parameters */
 639     inr              = nlist->iinr[0];
 640     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 641     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 642     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 643
 644     /* Avoid stupid compiler warnings */
 645     jnrA = jnrB = 0;
 646     j_coord_offsetA = 0;
 647     j_coord_offsetB = 0;
 648
 649     outeriter        = 0;
 650     inneriter        = 0;
 651
 652     /* Start outer loop over neighborlists */
 653     for(iidx=0; iidx<nri; iidx++)
 654     {
 655         /* Load shift vector for this list */
 656         i_shift_offset   = DIM*shiftidx[iidx];
 657
 658         /* Load limits for loop over neighbors */
 659         j_index_start    = jindex[iidx];
 660         j_index_end      = jindex[iidx+1];
 661
 662         /* Get outer coordinate index */
 663         inr              = iinr[iidx];
 664         i_coord_offset   = DIM*inr;
 665
 666         /* Load i particle coords and add shift vector */
 667         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 668                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 669
 670         fix1             = _mm_setzero_pd();
 671         fiy1             = _mm_setzero_pd();
 672         fiz1             = _mm_setzero_pd();
 673         fix2             = _mm_setzero_pd();
 674         fiy2             = _mm_setzero_pd();
 675         fiz2             = _mm_setzero_pd();
 676         fix3             = _mm_setzero_pd();
 677         fiy3             = _mm_setzero_pd();
 678         fiz3             = _mm_setzero_pd();
 679
 680         /* Start inner kernel loop */
 681         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 682         {
 683
 684             /* Get j neighbor index, and coordinate index */
 685             jnrA             = jjnr[jidx];
 686             jnrB             = jjnr[jidx+1];
 687             j_coord_offsetA  = DIM*jnrA;
 688             j_coord_offsetB  = DIM*jnrB;
 689
 690             /* load j atom coordinates */
 691             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 692                                               &jx0,&jy0,&jz0);
 693
 694             /* Calculate displacement vector */
 695             dx10             = _mm_sub_pd(ix1,jx0);
 696             dy10             = _mm_sub_pd(iy1,jy0);
 697             dz10             = _mm_sub_pd(iz1,jz0);
 698             dx20             = _mm_sub_pd(ix2,jx0);
 699             dy20             = _mm_sub_pd(iy2,jy0);
 700             dz20             = _mm_sub_pd(iz2,jz0);
 701             dx30             = _mm_sub_pd(ix3,jx0);
 702             dy30             = _mm_sub_pd(iy3,jy0);
 703             dz30             = _mm_sub_pd(iz3,jz0);
 704
 705             /* Calculate squared distance and things based on it */
 706             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 707             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 708             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 709
 710             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 711             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 712             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 713
 714             /* Load parameters for j particles */
 715             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 716
 717             fjx0             = _mm_setzero_pd();
 718             fjy0             = _mm_setzero_pd();
 719             fjz0             = _mm_setzero_pd();
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             r10              = _mm_mul_pd(rsq10,rinv10);
 726
 727             /* Compute parameters for interactions between i and j atoms */
 728             qq10             = _mm_mul_pd(iq1,jq0);
 729
 730             /* Calculate table index by multiplying r with table scale and truncate to integer */
 731             rt               = _mm_mul_pd(r10,vftabscale);
 732             vfitab           = _mm_cvttpd_epi32(rt);
 733             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 734             vfitab           = _mm_slli_epi32(vfitab,2);
 735
 736             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 737             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 738             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 739             GMX_MM_TRANSPOSE2_PD(Y,F);
 740             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 741             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 742             GMX_MM_TRANSPOSE2_PD(G,H);
 743             Heps             = _mm_mul_pd(vfeps,H);
 744             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 745             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 746             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 747
 748             fscal            = felec;
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm_mul_pd(fscal,dx10);
 752             ty               = _mm_mul_pd(fscal,dy10);
 753             tz               = _mm_mul_pd(fscal,dz10);
 754
 755             /* Update vectorial force */
 756             fix1             = _mm_add_pd(fix1,tx);
 757             fiy1             = _mm_add_pd(fiy1,ty);
 758             fiz1             = _mm_add_pd(fiz1,tz);
 759
 760             fjx0             = _mm_add_pd(fjx0,tx);
 761             fjy0             = _mm_add_pd(fjy0,ty);
 762             fjz0             = _mm_add_pd(fjz0,tz);
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             r20              = _mm_mul_pd(rsq20,rinv20);
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq20             = _mm_mul_pd(iq2,jq0);
 772
 773             /* Calculate table index by multiplying r with table scale and truncate to integer */
 774             rt               = _mm_mul_pd(r20,vftabscale);
 775             vfitab           = _mm_cvttpd_epi32(rt);
 776             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 777             vfitab           = _mm_slli_epi32(vfitab,2);
 778
 779             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 780             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 781             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 782             GMX_MM_TRANSPOSE2_PD(Y,F);
 783             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 784             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 785             GMX_MM_TRANSPOSE2_PD(G,H);
 786             Heps             = _mm_mul_pd(vfeps,H);
 787             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 788             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 789             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 790
 791             fscal            = felec;
 792
 793             /* Calculate temporary vectorial force */
 794             tx               = _mm_mul_pd(fscal,dx20);
 795             ty               = _mm_mul_pd(fscal,dy20);
 796             tz               = _mm_mul_pd(fscal,dz20);
 797
 798             /* Update vectorial force */
 799             fix2             = _mm_add_pd(fix2,tx);
 800             fiy2             = _mm_add_pd(fiy2,ty);
 801             fiz2             = _mm_add_pd(fiz2,tz);
 802
 803             fjx0             = _mm_add_pd(fjx0,tx);
 804             fjy0             = _mm_add_pd(fjy0,ty);
 805             fjz0             = _mm_add_pd(fjz0,tz);
 806
 807             /**************************
 808              * CALCULATE INTERACTIONS *
 809              **************************/
 810
 811             r30              = _mm_mul_pd(rsq30,rinv30);
 812
 813             /* Compute parameters for interactions between i and j atoms */
 814             qq30             = _mm_mul_pd(iq3,jq0);
 815
 816             /* Calculate table index by multiplying r with table scale and truncate to integer */
 817             rt               = _mm_mul_pd(r30,vftabscale);
 818             vfitab           = _mm_cvttpd_epi32(rt);
 819             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 820             vfitab           = _mm_slli_epi32(vfitab,2);
 821
 822             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 823             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 824             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 825             GMX_MM_TRANSPOSE2_PD(Y,F);
 826             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 827             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 828             GMX_MM_TRANSPOSE2_PD(G,H);
 829             Heps             = _mm_mul_pd(vfeps,H);
 830             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 831             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 832             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 833
 834             fscal            = felec;
 835
 836             /* Calculate temporary vectorial force */
 837             tx               = _mm_mul_pd(fscal,dx30);
 838             ty               = _mm_mul_pd(fscal,dy30);
 839             tz               = _mm_mul_pd(fscal,dz30);
 840
 841             /* Update vectorial force */
 842             fix3             = _mm_add_pd(fix3,tx);
 843             fiy3             = _mm_add_pd(fiy3,ty);
 844             fiz3             = _mm_add_pd(fiz3,tz);
 845
 846             fjx0             = _mm_add_pd(fjx0,tx);
 847             fjy0             = _mm_add_pd(fjy0,ty);
 848             fjz0             = _mm_add_pd(fjz0,tz);
 849
 850             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 851
 852             /* Inner loop uses 120 flops */
 853         }
 854
 855         if(jidx<j_index_end)
 856         {
 857
 858             jnrA             = jjnr[jidx];
 859             j_coord_offsetA  = DIM*jnrA;
 860
 861             /* load j atom coordinates */
 862             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 863                                               &jx0,&jy0,&jz0);
 864
 865             /* Calculate displacement vector */
 866             dx10             = _mm_sub_pd(ix1,jx0);
 867             dy10             = _mm_sub_pd(iy1,jy0);
 868             dz10             = _mm_sub_pd(iz1,jz0);
 869             dx20             = _mm_sub_pd(ix2,jx0);
 870             dy20             = _mm_sub_pd(iy2,jy0);
 871             dz20             = _mm_sub_pd(iz2,jz0);
 872             dx30             = _mm_sub_pd(ix3,jx0);
 873             dy30             = _mm_sub_pd(iy3,jy0);
 874             dz30             = _mm_sub_pd(iz3,jz0);
 875
 876             /* Calculate squared distance and things based on it */
 877             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 878             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 879             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 880
 881             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 882             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 883             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 884
 885             /* Load parameters for j particles */
 886             jq0              = _mm_load_sd(charge+jnrA+0);
 887
 888             fjx0             = _mm_setzero_pd();
 889             fjy0             = _mm_setzero_pd();
 890             fjz0             = _mm_setzero_pd();
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             r10              = _mm_mul_pd(rsq10,rinv10);
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq10             = _mm_mul_pd(iq1,jq0);
 900
 901             /* Calculate table index by multiplying r with table scale and truncate to integer */
 902             rt               = _mm_mul_pd(r10,vftabscale);
 903             vfitab           = _mm_cvttpd_epi32(rt);
 904             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 905             vfitab           = _mm_slli_epi32(vfitab,2);
 906
 907             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 908             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 909             F                = _mm_setzero_pd();
 910             GMX_MM_TRANSPOSE2_PD(Y,F);
 911             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 912             H                = _mm_setzero_pd();
 913             GMX_MM_TRANSPOSE2_PD(G,H);
 914             Heps             = _mm_mul_pd(vfeps,H);
 915             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 916             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 917             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 918
 919             fscal            = felec;
 920
 921             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 922
 923             /* Calculate temporary vectorial force */
 924             tx               = _mm_mul_pd(fscal,dx10);
 925             ty               = _mm_mul_pd(fscal,dy10);
 926             tz               = _mm_mul_pd(fscal,dz10);
 927
 928             /* Update vectorial force */
 929             fix1             = _mm_add_pd(fix1,tx);
 930             fiy1             = _mm_add_pd(fiy1,ty);
 931             fiz1             = _mm_add_pd(fiz1,tz);
 932
 933             fjx0             = _mm_add_pd(fjx0,tx);
 934             fjy0             = _mm_add_pd(fjy0,ty);
 935             fjz0             = _mm_add_pd(fjz0,tz);
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             r20              = _mm_mul_pd(rsq20,rinv20);
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             qq20             = _mm_mul_pd(iq2,jq0);
 945
 946             /* Calculate table index by multiplying r with table scale and truncate to integer */
 947             rt               = _mm_mul_pd(r20,vftabscale);
 948             vfitab           = _mm_cvttpd_epi32(rt);
 949             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 950             vfitab           = _mm_slli_epi32(vfitab,2);
 951
 952             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 953             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 954             F                = _mm_setzero_pd();
 955             GMX_MM_TRANSPOSE2_PD(Y,F);
 956             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 957             H                = _mm_setzero_pd();
 958             GMX_MM_TRANSPOSE2_PD(G,H);
 959             Heps             = _mm_mul_pd(vfeps,H);
 960             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 961             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 962             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 963
 964             fscal            = felec;
 965
 966             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 967
 968             /* Calculate temporary vectorial force */
 969             tx               = _mm_mul_pd(fscal,dx20);
 970             ty               = _mm_mul_pd(fscal,dy20);
 971             tz               = _mm_mul_pd(fscal,dz20);
 972
 973             /* Update vectorial force */
 974             fix2             = _mm_add_pd(fix2,tx);
 975             fiy2             = _mm_add_pd(fiy2,ty);
 976             fiz2             = _mm_add_pd(fiz2,tz);
 977
 978             fjx0             = _mm_add_pd(fjx0,tx);
 979             fjy0             = _mm_add_pd(fjy0,ty);
 980             fjz0             = _mm_add_pd(fjz0,tz);
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             r30              = _mm_mul_pd(rsq30,rinv30);
 987
 988             /* Compute parameters for interactions between i and j atoms */
 989             qq30             = _mm_mul_pd(iq3,jq0);
 990
 991             /* Calculate table index by multiplying r with table scale and truncate to integer */
 992             rt               = _mm_mul_pd(r30,vftabscale);
 993             vfitab           = _mm_cvttpd_epi32(rt);
 994             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 995             vfitab           = _mm_slli_epi32(vfitab,2);
 996
 997             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 998             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 999             F                = _mm_setzero_pd();
1000             GMX_MM_TRANSPOSE2_PD(Y,F);
1001             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1002             H                = _mm_setzero_pd();
1003             GMX_MM_TRANSPOSE2_PD(G,H);
1004             Heps             = _mm_mul_pd(vfeps,H);
1005             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1006             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1007             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1008
1009             fscal            = felec;
1010
1011             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1012
1013             /* Calculate temporary vectorial force */
1014             tx               = _mm_mul_pd(fscal,dx30);
1015             ty               = _mm_mul_pd(fscal,dy30);
1016             tz               = _mm_mul_pd(fscal,dz30);
1017
1018             /* Update vectorial force */
1019             fix3             = _mm_add_pd(fix3,tx);
1020             fiy3             = _mm_add_pd(fiy3,ty);
1021             fiz3             = _mm_add_pd(fiz3,tz);
1022
1023             fjx0             = _mm_add_pd(fjx0,tx);
1024             fjy0             = _mm_add_pd(fjy0,ty);
1025             fjz0             = _mm_add_pd(fjz0,tz);
1026
1027             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1028
1029             /* Inner loop uses 120 flops */
1030         }
1031
1032         /* End of innermost loop */
1033
1034         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1035                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1036
1037         /* Increment number of inner iterations */
1038         inneriter                  += j_index_end - j_index_start;
1039
1040         /* Outer loop uses 18 flops */
1041     }
1042
1043     /* Increment number of outer iterations */
1044     outeriter        += nri;
1045
1046     /* Update outer/inner flops */
1047
1048     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1049 }