src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwNone_GeomP1P1_sse2_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     __m128i          vfitab;
  87     __m128i          ifour       = _mm_set1_epi32(4);
  88     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  89     real             *vftab;
  90     __m128d          dummy_mask,cutoff_mask;
  91     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  92     __m128d          one     = _mm_set1_pd(1.0);
  93     __m128d          two     = _mm_set1_pd(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_pd(fr->ic->epsfac);
 106     charge           = mdatoms->chargeA;
 107
 108     vftab            = kernel_data->table_elec->data;
 109     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     /* Start outer loop over neighborlists */
 120     for(iidx=0; iidx<nri; iidx++)
 121     {
 122         /* Load shift vector for this list */
 123         i_shift_offset   = DIM*shiftidx[iidx];
 124
 125         /* Load limits for loop over neighbors */
 126         j_index_start    = jindex[iidx];
 127         j_index_end      = jindex[iidx+1];
 128
 129         /* Get outer coordinate index */
 130         inr              = iinr[iidx];
 131         i_coord_offset   = DIM*inr;
 132
 133         /* Load i particle coords and add shift vector */
 134         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 135
 136         fix0             = _mm_setzero_pd();
 137         fiy0             = _mm_setzero_pd();
 138         fiz0             = _mm_setzero_pd();
 139
 140         /* Load parameters for i particles */
 141         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 142
 143         /* Reset potential sums */
 144         velecsum         = _mm_setzero_pd();
 145
 146         /* Start inner kernel loop */
 147         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 148         {
 149
 150             /* Get j neighbor index, and coordinate index */
 151             jnrA             = jjnr[jidx];
 152             jnrB             = jjnr[jidx+1];
 153             j_coord_offsetA  = DIM*jnrA;
 154             j_coord_offsetB  = DIM*jnrB;
 155
 156             /* load j atom coordinates */
 157             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 158                                               &jx0,&jy0,&jz0);
 159
 160             /* Calculate displacement vector */
 161             dx00             = _mm_sub_pd(ix0,jx0);
 162             dy00             = _mm_sub_pd(iy0,jy0);
 163             dz00             = _mm_sub_pd(iz0,jz0);
 164
 165             /* Calculate squared distance and things based on it */
 166             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 167
 168             rinv00           = sse2_invsqrt_d(rsq00);
 169
 170             /* Load parameters for j particles */
 171             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 172
 173             /**************************
 174              * CALCULATE INTERACTIONS *
 175              **************************/
 176
 177             r00              = _mm_mul_pd(rsq00,rinv00);
 178
 179             /* Compute parameters for interactions between i and j atoms */
 180             qq00             = _mm_mul_pd(iq0,jq0);
 181
 182             /* Calculate table index by multiplying r with table scale and truncate to integer */
 183             rt               = _mm_mul_pd(r00,vftabscale);
 184             vfitab           = _mm_cvttpd_epi32(rt);
 185             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 186             vfitab           = _mm_slli_epi32(vfitab,2);
 187
 188             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 189             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 190             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 191             GMX_MM_TRANSPOSE2_PD(Y,F);
 192             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 193             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 194             GMX_MM_TRANSPOSE2_PD(G,H);
 195             Heps             = _mm_mul_pd(vfeps,H);
 196             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 197             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 198             velec            = _mm_mul_pd(qq00,VV);
 199             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 200             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 201
 202             /* Update potential sum for this i atom from the interaction with this j atom. */
 203             velecsum         = _mm_add_pd(velecsum,velec);
 204
 205             fscal            = felec;
 206
 207             /* Calculate temporary vectorial force */
 208             tx               = _mm_mul_pd(fscal,dx00);
 209             ty               = _mm_mul_pd(fscal,dy00);
 210             tz               = _mm_mul_pd(fscal,dz00);
 211
 212             /* Update vectorial force */
 213             fix0             = _mm_add_pd(fix0,tx);
 214             fiy0             = _mm_add_pd(fiy0,ty);
 215             fiz0             = _mm_add_pd(fiz0,tz);
 216
 217             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 218
 219             /* Inner loop uses 43 flops */
 220         }
 221
 222         if(jidx<j_index_end)
 223         {
 224
 225             jnrA             = jjnr[jidx];
 226             j_coord_offsetA  = DIM*jnrA;
 227
 228             /* load j atom coordinates */
 229             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 230                                               &jx0,&jy0,&jz0);
 231
 232             /* Calculate displacement vector */
 233             dx00             = _mm_sub_pd(ix0,jx0);
 234             dy00             = _mm_sub_pd(iy0,jy0);
 235             dz00             = _mm_sub_pd(iz0,jz0);
 236
 237             /* Calculate squared distance and things based on it */
 238             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 239
 240             rinv00           = sse2_invsqrt_d(rsq00);
 241
 242             /* Load parameters for j particles */
 243             jq0              = _mm_load_sd(charge+jnrA+0);
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             r00              = _mm_mul_pd(rsq00,rinv00);
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm_mul_pd(iq0,jq0);
 253
 254             /* Calculate table index by multiplying r with table scale and truncate to integer */
 255             rt               = _mm_mul_pd(r00,vftabscale);
 256             vfitab           = _mm_cvttpd_epi32(rt);
 257             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 258             vfitab           = _mm_slli_epi32(vfitab,2);
 259
 260             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 261             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 262             F                = _mm_setzero_pd();
 263             GMX_MM_TRANSPOSE2_PD(Y,F);
 264             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 265             H                = _mm_setzero_pd();
 266             GMX_MM_TRANSPOSE2_PD(G,H);
 267             Heps             = _mm_mul_pd(vfeps,H);
 268             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 269             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 270             velec            = _mm_mul_pd(qq00,VV);
 271             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 272             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 276             velecsum         = _mm_add_pd(velecsum,velec);
 277
 278             fscal            = felec;
 279
 280             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_pd(fscal,dx00);
 284             ty               = _mm_mul_pd(fscal,dy00);
 285             tz               = _mm_mul_pd(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_pd(fix0,tx);
 289             fiy0             = _mm_add_pd(fiy0,ty);
 290             fiz0             = _mm_add_pd(fiz0,tz);
 291
 292             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 293
 294             /* Inner loop uses 43 flops */
 295         }
 296
 297         /* End of innermost loop */
 298
 299         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 300                                               f+i_coord_offset,fshift+i_shift_offset);
 301
 302         ggid                        = gid[iidx];
 303         /* Update potential energies */
 304         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 305
 306         /* Increment number of inner iterations */
 307         inneriter                  += j_index_end - j_index_start;
 308
 309         /* Outer loop uses 8 flops */
 310     }
 311
 312     /* Increment number of outer iterations */
 313     outeriter        += nri;
 314
 315     /* Update outer/inner flops */
 316
 317     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*43);
 318 }
 319 /*
 320  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_double
 321  * Electrostatics interaction: CubicSplineTable
 322  * VdW interaction:            None
 323  * Geometry:                   Particle-Particle
 324  * Calculate force/pot:        Force
 325  */
 326 void
 327 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_double
 328                     (t_nblist                    * gmx_restrict       nlist,
 329                      rvec                        * gmx_restrict          xx,
 330                      rvec                        * gmx_restrict          ff,
 331                      struct t_forcerec           * gmx_restrict          fr,
 332                      t_mdatoms                   * gmx_restrict     mdatoms,
 333                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 334                      t_nrnb                      * gmx_restrict        nrnb)
 335 {
 336     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 337      * just 0 for non-waters.
 338      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 339      * jnr indices corresponding to data put in the four positions in the SIMD register.
 340      */
 341     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 342     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 343     int              jnrA,jnrB;
 344     int              j_coord_offsetA,j_coord_offsetB;
 345     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 346     real             rcutoff_scalar;
 347     real             *shiftvec,*fshift,*x,*f;
 348     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 349     int              vdwioffset0;
 350     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 351     int              vdwjidx0A,vdwjidx0B;
 352     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 353     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 354     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 355     real             *charge;
 356     __m128i          vfitab;
 357     __m128i          ifour       = _mm_set1_epi32(4);
 358     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 359     real             *vftab;
 360     __m128d          dummy_mask,cutoff_mask;
 361     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 362     __m128d          one     = _mm_set1_pd(1.0);
 363     __m128d          two     = _mm_set1_pd(2.0);
 364     x                = xx[0];
 365     f                = ff[0];
 366
 367     nri              = nlist->nri;
 368     iinr             = nlist->iinr;
 369     jindex           = nlist->jindex;
 370     jjnr             = nlist->jjnr;
 371     shiftidx         = nlist->shift;
 372     gid              = nlist->gid;
 373     shiftvec         = fr->shift_vec[0];
 374     fshift           = fr->fshift[0];
 375     facel            = _mm_set1_pd(fr->ic->epsfac);
 376     charge           = mdatoms->chargeA;
 377
 378     vftab            = kernel_data->table_elec->data;
 379     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 380
 381     /* Avoid stupid compiler warnings */
 382     jnrA = jnrB = 0;
 383     j_coord_offsetA = 0;
 384     j_coord_offsetB = 0;
 385
 386     outeriter        = 0;
 387     inneriter        = 0;
 388
 389     /* Start outer loop over neighborlists */
 390     for(iidx=0; iidx<nri; iidx++)
 391     {
 392         /* Load shift vector for this list */
 393         i_shift_offset   = DIM*shiftidx[iidx];
 394
 395         /* Load limits for loop over neighbors */
 396         j_index_start    = jindex[iidx];
 397         j_index_end      = jindex[iidx+1];
 398
 399         /* Get outer coordinate index */
 400         inr              = iinr[iidx];
 401         i_coord_offset   = DIM*inr;
 402
 403         /* Load i particle coords and add shift vector */
 404         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 405
 406         fix0             = _mm_setzero_pd();
 407         fiy0             = _mm_setzero_pd();
 408         fiz0             = _mm_setzero_pd();
 409
 410         /* Load parameters for i particles */
 411         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 412
 413         /* Start inner kernel loop */
 414         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 415         {
 416
 417             /* Get j neighbor index, and coordinate index */
 418             jnrA             = jjnr[jidx];
 419             jnrB             = jjnr[jidx+1];
 420             j_coord_offsetA  = DIM*jnrA;
 421             j_coord_offsetB  = DIM*jnrB;
 422
 423             /* load j atom coordinates */
 424             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 425                                               &jx0,&jy0,&jz0);
 426
 427             /* Calculate displacement vector */
 428             dx00             = _mm_sub_pd(ix0,jx0);
 429             dy00             = _mm_sub_pd(iy0,jy0);
 430             dz00             = _mm_sub_pd(iz0,jz0);
 431
 432             /* Calculate squared distance and things based on it */
 433             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 434
 435             rinv00           = sse2_invsqrt_d(rsq00);
 436
 437             /* Load parameters for j particles */
 438             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r00              = _mm_mul_pd(rsq00,rinv00);
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             qq00             = _mm_mul_pd(iq0,jq0);
 448
 449             /* Calculate table index by multiplying r with table scale and truncate to integer */
 450             rt               = _mm_mul_pd(r00,vftabscale);
 451             vfitab           = _mm_cvttpd_epi32(rt);
 452             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 453             vfitab           = _mm_slli_epi32(vfitab,2);
 454
 455             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 456             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 457             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 458             GMX_MM_TRANSPOSE2_PD(Y,F);
 459             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 460             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 461             GMX_MM_TRANSPOSE2_PD(G,H);
 462             Heps             = _mm_mul_pd(vfeps,H);
 463             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 464             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 465             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 466
 467             fscal            = felec;
 468
 469             /* Calculate temporary vectorial force */
 470             tx               = _mm_mul_pd(fscal,dx00);
 471             ty               = _mm_mul_pd(fscal,dy00);
 472             tz               = _mm_mul_pd(fscal,dz00);
 473
 474             /* Update vectorial force */
 475             fix0             = _mm_add_pd(fix0,tx);
 476             fiy0             = _mm_add_pd(fiy0,ty);
 477             fiz0             = _mm_add_pd(fiz0,tz);
 478
 479             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 480
 481             /* Inner loop uses 39 flops */
 482         }
 483
 484         if(jidx<j_index_end)
 485         {
 486
 487             jnrA             = jjnr[jidx];
 488             j_coord_offsetA  = DIM*jnrA;
 489
 490             /* load j atom coordinates */
 491             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 492                                               &jx0,&jy0,&jz0);
 493
 494             /* Calculate displacement vector */
 495             dx00             = _mm_sub_pd(ix0,jx0);
 496             dy00             = _mm_sub_pd(iy0,jy0);
 497             dz00             = _mm_sub_pd(iz0,jz0);
 498
 499             /* Calculate squared distance and things based on it */
 500             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 501
 502             rinv00           = sse2_invsqrt_d(rsq00);
 503
 504             /* Load parameters for j particles */
 505             jq0              = _mm_load_sd(charge+jnrA+0);
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             r00              = _mm_mul_pd(rsq00,rinv00);
 512
 513             /* Compute parameters for interactions between i and j atoms */
 514             qq00             = _mm_mul_pd(iq0,jq0);
 515
 516             /* Calculate table index by multiplying r with table scale and truncate to integer */
 517             rt               = _mm_mul_pd(r00,vftabscale);
 518             vfitab           = _mm_cvttpd_epi32(rt);
 519             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 520             vfitab           = _mm_slli_epi32(vfitab,2);
 521
 522             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 523             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 524             F                = _mm_setzero_pd();
 525             GMX_MM_TRANSPOSE2_PD(Y,F);
 526             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 527             H                = _mm_setzero_pd();
 528             GMX_MM_TRANSPOSE2_PD(G,H);
 529             Heps             = _mm_mul_pd(vfeps,H);
 530             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 531             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 532             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm_mul_pd(fscal,dx00);
 540             ty               = _mm_mul_pd(fscal,dy00);
 541             tz               = _mm_mul_pd(fscal,dz00);
 542
 543             /* Update vectorial force */
 544             fix0             = _mm_add_pd(fix0,tx);
 545             fiy0             = _mm_add_pd(fiy0,ty);
 546             fiz0             = _mm_add_pd(fiz0,tz);
 547
 548             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 549
 550             /* Inner loop uses 39 flops */
 551         }
 552
 553         /* End of innermost loop */
 554
 555         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 556                                               f+i_coord_offset,fshift+i_shift_offset);
 557
 558         /* Increment number of inner iterations */
 559         inneriter                  += j_index_end - j_index_start;
 560
 561         /* Outer loop uses 7 flops */
 562     }
 563
 564     /* Increment number of outer iterations */
 565     outeriter        += nri;
 566
 567     /* Update outer/inner flops */
 568
 569     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 570 }