src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwNone_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015, 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/math/vec.h"
  46 #include "gromacs/legacyheaders/nrnb.h"
  47
  48 #include "gromacs/simd/math_x86_sse4_1_double.h"
  49 #include "kernelutil_x86_sse4_1_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse4_1_double
  53  * Electrostatics interaction: CubicSplineTable
  54  * VdW interaction:            None
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse4_1_double
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB;
  76     int              j_coord_offsetA,j_coord_offsetB;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  81     int              vdwioffset0;
  82     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  83     int              vdwjidx0A,vdwjidx0B;
  84     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  85     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     vftab            = kernel_data->table_elec->data;
 111     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 137
 138         fix0             = _mm_setzero_pd();
 139         fiy0             = _mm_setzero_pd();
 140         fiz0             = _mm_setzero_pd();
 141
 142         /* Load parameters for i particles */
 143         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 144
 145         /* Reset potential sums */
 146         velecsum         = _mm_setzero_pd();
 147
 148         /* Start inner kernel loop */
 149         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 150         {
 151
 152             /* Get j neighbor index, and coordinate index */
 153             jnrA             = jjnr[jidx];
 154             jnrB             = jjnr[jidx+1];
 155             j_coord_offsetA  = DIM*jnrA;
 156             j_coord_offsetB  = DIM*jnrB;
 157
 158             /* load j atom coordinates */
 159             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 160                                               &jx0,&jy0,&jz0);
 161
 162             /* Calculate displacement vector */
 163             dx00             = _mm_sub_pd(ix0,jx0);
 164             dy00             = _mm_sub_pd(iy0,jy0);
 165             dz00             = _mm_sub_pd(iz0,jz0);
 166
 167             /* Calculate squared distance and things based on it */
 168             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 169
 170             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 171
 172             /* Load parameters for j particles */
 173             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 174
 175             /**************************
 176              * CALCULATE INTERACTIONS *
 177              **************************/
 178
 179             r00              = _mm_mul_pd(rsq00,rinv00);
 180
 181             /* Compute parameters for interactions between i and j atoms */
 182             qq00             = _mm_mul_pd(iq0,jq0);
 183
 184             /* Calculate table index by multiplying r with table scale and truncate to integer */
 185             rt               = _mm_mul_pd(r00,vftabscale);
 186             vfitab           = _mm_cvttpd_epi32(rt);
 187             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 188             vfitab           = _mm_slli_epi32(vfitab,2);
 189
 190             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 191             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 192             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 193             GMX_MM_TRANSPOSE2_PD(Y,F);
 194             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 195             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 196             GMX_MM_TRANSPOSE2_PD(G,H);
 197             Heps             = _mm_mul_pd(vfeps,H);
 198             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 199             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 200             velec            = _mm_mul_pd(qq00,VV);
 201             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 202             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 203
 204             /* Update potential sum for this i atom from the interaction with this j atom. */
 205             velecsum         = _mm_add_pd(velecsum,velec);
 206
 207             fscal            = felec;
 208
 209             /* Calculate temporary vectorial force */
 210             tx               = _mm_mul_pd(fscal,dx00);
 211             ty               = _mm_mul_pd(fscal,dy00);
 212             tz               = _mm_mul_pd(fscal,dz00);
 213
 214             /* Update vectorial force */
 215             fix0             = _mm_add_pd(fix0,tx);
 216             fiy0             = _mm_add_pd(fiy0,ty);
 217             fiz0             = _mm_add_pd(fiz0,tz);
 218
 219             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 220
 221             /* Inner loop uses 43 flops */
 222         }
 223
 224         if(jidx<j_index_end)
 225         {
 226
 227             jnrA             = jjnr[jidx];
 228             j_coord_offsetA  = DIM*jnrA;
 229
 230             /* load j atom coordinates */
 231             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 232                                               &jx0,&jy0,&jz0);
 233
 234             /* Calculate displacement vector */
 235             dx00             = _mm_sub_pd(ix0,jx0);
 236             dy00             = _mm_sub_pd(iy0,jy0);
 237             dz00             = _mm_sub_pd(iz0,jz0);
 238
 239             /* Calculate squared distance and things based on it */
 240             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 241
 242             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 243
 244             /* Load parameters for j particles */
 245             jq0              = _mm_load_sd(charge+jnrA+0);
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             r00              = _mm_mul_pd(rsq00,rinv00);
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             qq00             = _mm_mul_pd(iq0,jq0);
 255
 256             /* Calculate table index by multiplying r with table scale and truncate to integer */
 257             rt               = _mm_mul_pd(r00,vftabscale);
 258             vfitab           = _mm_cvttpd_epi32(rt);
 259             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 260             vfitab           = _mm_slli_epi32(vfitab,2);
 261
 262             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 263             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 264             F                = _mm_setzero_pd();
 265             GMX_MM_TRANSPOSE2_PD(Y,F);
 266             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 267             H                = _mm_setzero_pd();
 268             GMX_MM_TRANSPOSE2_PD(G,H);
 269             Heps             = _mm_mul_pd(vfeps,H);
 270             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 271             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 272             velec            = _mm_mul_pd(qq00,VV);
 273             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 274             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 278             velecsum         = _mm_add_pd(velecsum,velec);
 279
 280             fscal            = felec;
 281
 282             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_pd(fscal,dx00);
 286             ty               = _mm_mul_pd(fscal,dy00);
 287             tz               = _mm_mul_pd(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_add_pd(fix0,tx);
 291             fiy0             = _mm_add_pd(fiy0,ty);
 292             fiz0             = _mm_add_pd(fiz0,tz);
 293
 294             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 295
 296             /* Inner loop uses 43 flops */
 297         }
 298
 299         /* End of innermost loop */
 300
 301         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 302                                               f+i_coord_offset,fshift+i_shift_offset);
 303
 304         ggid                        = gid[iidx];
 305         /* Update potential energies */
 306         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 307
 308         /* Increment number of inner iterations */
 309         inneriter                  += j_index_end - j_index_start;
 310
 311         /* Outer loop uses 8 flops */
 312     }
 313
 314     /* Increment number of outer iterations */
 315     outeriter        += nri;
 316
 317     /* Update outer/inner flops */
 318
 319     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*43);
 320 }
 321 /*
 322  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse4_1_double
 323  * Electrostatics interaction: CubicSplineTable
 324  * VdW interaction:            None
 325  * Geometry:                   Particle-Particle
 326  * Calculate force/pot:        Force
 327  */
 328 void
 329 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse4_1_double
 330                     (t_nblist                    * gmx_restrict       nlist,
 331                      rvec                        * gmx_restrict          xx,
 332                      rvec                        * gmx_restrict          ff,
 333                      t_forcerec                  * gmx_restrict          fr,
 334                      t_mdatoms                   * gmx_restrict     mdatoms,
 335                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 336                      t_nrnb                      * gmx_restrict        nrnb)
 337 {
 338     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 339      * just 0 for non-waters.
 340      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 341      * jnr indices corresponding to data put in the four positions in the SIMD register.
 342      */
 343     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 344     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 345     int              jnrA,jnrB;
 346     int              j_coord_offsetA,j_coord_offsetB;
 347     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 348     real             rcutoff_scalar;
 349     real             *shiftvec,*fshift,*x,*f;
 350     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 351     int              vdwioffset0;
 352     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 353     int              vdwjidx0A,vdwjidx0B;
 354     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 355     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 356     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 357     real             *charge;
 358     __m128i          vfitab;
 359     __m128i          ifour       = _mm_set1_epi32(4);
 360     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 361     real             *vftab;
 362     __m128d          dummy_mask,cutoff_mask;
 363     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 364     __m128d          one     = _mm_set1_pd(1.0);
 365     __m128d          two     = _mm_set1_pd(2.0);
 366     x                = xx[0];
 367     f                = ff[0];
 368
 369     nri              = nlist->nri;
 370     iinr             = nlist->iinr;
 371     jindex           = nlist->jindex;
 372     jjnr             = nlist->jjnr;
 373     shiftidx         = nlist->shift;
 374     gid              = nlist->gid;
 375     shiftvec         = fr->shift_vec[0];
 376     fshift           = fr->fshift[0];
 377     facel            = _mm_set1_pd(fr->epsfac);
 378     charge           = mdatoms->chargeA;
 379
 380     vftab            = kernel_data->table_elec->data;
 381     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 382
 383     /* Avoid stupid compiler warnings */
 384     jnrA = jnrB = 0;
 385     j_coord_offsetA = 0;
 386     j_coord_offsetB = 0;
 387
 388     outeriter        = 0;
 389     inneriter        = 0;
 390
 391     /* Start outer loop over neighborlists */
 392     for(iidx=0; iidx<nri; iidx++)
 393     {
 394         /* Load shift vector for this list */
 395         i_shift_offset   = DIM*shiftidx[iidx];
 396
 397         /* Load limits for loop over neighbors */
 398         j_index_start    = jindex[iidx];
 399         j_index_end      = jindex[iidx+1];
 400
 401         /* Get outer coordinate index */
 402         inr              = iinr[iidx];
 403         i_coord_offset   = DIM*inr;
 404
 405         /* Load i particle coords and add shift vector */
 406         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 407
 408         fix0             = _mm_setzero_pd();
 409         fiy0             = _mm_setzero_pd();
 410         fiz0             = _mm_setzero_pd();
 411
 412         /* Load parameters for i particles */
 413         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 414
 415         /* Start inner kernel loop */
 416         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 417         {
 418
 419             /* Get j neighbor index, and coordinate index */
 420             jnrA             = jjnr[jidx];
 421             jnrB             = jjnr[jidx+1];
 422             j_coord_offsetA  = DIM*jnrA;
 423             j_coord_offsetB  = DIM*jnrB;
 424
 425             /* load j atom coordinates */
 426             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                               &jx0,&jy0,&jz0);
 428
 429             /* Calculate displacement vector */
 430             dx00             = _mm_sub_pd(ix0,jx0);
 431             dy00             = _mm_sub_pd(iy0,jy0);
 432             dz00             = _mm_sub_pd(iz0,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 436
 437             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 438
 439             /* Load parameters for j particles */
 440             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             r00              = _mm_mul_pd(rsq00,rinv00);
 447
 448             /* Compute parameters for interactions between i and j atoms */
 449             qq00             = _mm_mul_pd(iq0,jq0);
 450
 451             /* Calculate table index by multiplying r with table scale and truncate to integer */
 452             rt               = _mm_mul_pd(r00,vftabscale);
 453             vfitab           = _mm_cvttpd_epi32(rt);
 454             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 455             vfitab           = _mm_slli_epi32(vfitab,2);
 456
 457             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 458             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 459             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 460             GMX_MM_TRANSPOSE2_PD(Y,F);
 461             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 462             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 463             GMX_MM_TRANSPOSE2_PD(G,H);
 464             Heps             = _mm_mul_pd(vfeps,H);
 465             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 466             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 467             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 468
 469             fscal            = felec;
 470
 471             /* Calculate temporary vectorial force */
 472             tx               = _mm_mul_pd(fscal,dx00);
 473             ty               = _mm_mul_pd(fscal,dy00);
 474             tz               = _mm_mul_pd(fscal,dz00);
 475
 476             /* Update vectorial force */
 477             fix0             = _mm_add_pd(fix0,tx);
 478             fiy0             = _mm_add_pd(fiy0,ty);
 479             fiz0             = _mm_add_pd(fiz0,tz);
 480
 481             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 482
 483             /* Inner loop uses 39 flops */
 484         }
 485
 486         if(jidx<j_index_end)
 487         {
 488
 489             jnrA             = jjnr[jidx];
 490             j_coord_offsetA  = DIM*jnrA;
 491
 492             /* load j atom coordinates */
 493             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 494                                               &jx0,&jy0,&jz0);
 495
 496             /* Calculate displacement vector */
 497             dx00             = _mm_sub_pd(ix0,jx0);
 498             dy00             = _mm_sub_pd(iy0,jy0);
 499             dz00             = _mm_sub_pd(iz0,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 503
 504             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 505
 506             /* Load parameters for j particles */
 507             jq0              = _mm_load_sd(charge+jnrA+0);
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             r00              = _mm_mul_pd(rsq00,rinv00);
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             qq00             = _mm_mul_pd(iq0,jq0);
 517
 518             /* Calculate table index by multiplying r with table scale and truncate to integer */
 519             rt               = _mm_mul_pd(r00,vftabscale);
 520             vfitab           = _mm_cvttpd_epi32(rt);
 521             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 522             vfitab           = _mm_slli_epi32(vfitab,2);
 523
 524             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 525             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 526             F                = _mm_setzero_pd();
 527             GMX_MM_TRANSPOSE2_PD(Y,F);
 528             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 529             H                = _mm_setzero_pd();
 530             GMX_MM_TRANSPOSE2_PD(G,H);
 531             Heps             = _mm_mul_pd(vfeps,H);
 532             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 533             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 534             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 539
 540             /* Calculate temporary vectorial force */
 541             tx               = _mm_mul_pd(fscal,dx00);
 542             ty               = _mm_mul_pd(fscal,dy00);
 543             tz               = _mm_mul_pd(fscal,dz00);
 544
 545             /* Update vectorial force */
 546             fix0             = _mm_add_pd(fix0,tx);
 547             fiy0             = _mm_add_pd(fiy0,ty);
 548             fiz0             = _mm_add_pd(fiz0,tz);
 549
 550             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 551
 552             /* Inner loop uses 39 flops */
 553         }
 554
 555         /* End of innermost loop */
 556
 557         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 558                                               f+i_coord_offset,fshift+i_shift_offset);
 559
 560         /* Increment number of inner iterations */
 561         inneriter                  += j_index_end - j_index_start;
 562
 563         /* Outer loop uses 7 flops */
 564     }
 565
 566     /* Increment number of outer iterations */
 567     outeriter        += nri;
 568
 569     /* Update outer/inner flops */
 570
 571     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 572 }