src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwNone_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_ElecEw_VdwNone_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_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          ewitab;
  87     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  88     real             *ewtab;
  89     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->ic->epsfac);
 105     charge           = mdatoms->chargeA;
 106
 107     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 108     ewtab            = fr->ic->tabq_coul_FDV0;
 109     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 110     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 111
 112     /* Avoid stupid compiler warnings */
 113     jnrA = jnrB = 0;
 114     j_coord_offsetA = 0;
 115     j_coord_offsetB = 0;
 116
 117     outeriter        = 0;
 118     inneriter        = 0;
 119
 120     /* Start outer loop over neighborlists */
 121     for(iidx=0; iidx<nri; iidx++)
 122     {
 123         /* Load shift vector for this list */
 124         i_shift_offset   = DIM*shiftidx[iidx];
 125
 126         /* Load limits for loop over neighbors */
 127         j_index_start    = jindex[iidx];
 128         j_index_end      = jindex[iidx+1];
 129
 130         /* Get outer coordinate index */
 131         inr              = iinr[iidx];
 132         i_coord_offset   = DIM*inr;
 133
 134         /* Load i particle coords and add shift vector */
 135         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 136
 137         fix0             = _mm_setzero_pd();
 138         fiy0             = _mm_setzero_pd();
 139         fiz0             = _mm_setzero_pd();
 140
 141         /* Load parameters for i particles */
 142         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 143
 144         /* Reset potential sums */
 145         velecsum         = _mm_setzero_pd();
 146
 147         /* Start inner kernel loop */
 148         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 149         {
 150
 151             /* Get j neighbor index, and coordinate index */
 152             jnrA             = jjnr[jidx];
 153             jnrB             = jjnr[jidx+1];
 154             j_coord_offsetA  = DIM*jnrA;
 155             j_coord_offsetB  = DIM*jnrB;
 156
 157             /* load j atom coordinates */
 158             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 159                                               &jx0,&jy0,&jz0);
 160
 161             /* Calculate displacement vector */
 162             dx00             = _mm_sub_pd(ix0,jx0);
 163             dy00             = _mm_sub_pd(iy0,jy0);
 164             dz00             = _mm_sub_pd(iz0,jz0);
 165
 166             /* Calculate squared distance and things based on it */
 167             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 168
 169             rinv00           = sse2_invsqrt_d(rsq00);
 170
 171             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 172
 173             /* Load parameters for j particles */
 174             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 175
 176             /**************************
 177              * CALCULATE INTERACTIONS *
 178              **************************/
 179
 180             r00              = _mm_mul_pd(rsq00,rinv00);
 181
 182             /* Compute parameters for interactions between i and j atoms */
 183             qq00             = _mm_mul_pd(iq0,jq0);
 184
 185             /* EWALD ELECTROSTATICS */
 186
 187             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 188             ewrt             = _mm_mul_pd(r00,ewtabscale);
 189             ewitab           = _mm_cvttpd_epi32(ewrt);
 190             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 191             ewitab           = _mm_slli_epi32(ewitab,2);
 192             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 193             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 194             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 195             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 196             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 197             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 198             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 199             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 200             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 201             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 202
 203             /* Update potential sum for this i atom from the interaction with this j atom. */
 204             velecsum         = _mm_add_pd(velecsum,velec);
 205
 206             fscal            = felec;
 207
 208             /* Calculate temporary vectorial force */
 209             tx               = _mm_mul_pd(fscal,dx00);
 210             ty               = _mm_mul_pd(fscal,dy00);
 211             tz               = _mm_mul_pd(fscal,dz00);
 212
 213             /* Update vectorial force */
 214             fix0             = _mm_add_pd(fix0,tx);
 215             fiy0             = _mm_add_pd(fiy0,ty);
 216             fiz0             = _mm_add_pd(fiz0,tz);
 217
 218             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 219
 220             /* Inner loop uses 41 flops */
 221         }
 222
 223         if(jidx<j_index_end)
 224         {
 225
 226             jnrA             = jjnr[jidx];
 227             j_coord_offsetA  = DIM*jnrA;
 228
 229             /* load j atom coordinates */
 230             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 231                                               &jx0,&jy0,&jz0);
 232
 233             /* Calculate displacement vector */
 234             dx00             = _mm_sub_pd(ix0,jx0);
 235             dy00             = _mm_sub_pd(iy0,jy0);
 236             dz00             = _mm_sub_pd(iz0,jz0);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 240
 241             rinv00           = sse2_invsqrt_d(rsq00);
 242
 243             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 244
 245             /* Load parameters for j particles */
 246             jq0              = _mm_load_sd(charge+jnrA+0);
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm_mul_pd(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq00             = _mm_mul_pd(iq0,jq0);
 256
 257             /* EWALD ELECTROSTATICS */
 258
 259             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 260             ewrt             = _mm_mul_pd(r00,ewtabscale);
 261             ewitab           = _mm_cvttpd_epi32(ewrt);
 262             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 263             ewitab           = _mm_slli_epi32(ewitab,2);
 264             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 265             ewtabD           = _mm_setzero_pd();
 266             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 267             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 268             ewtabFn          = _mm_setzero_pd();
 269             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 270             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 271             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 272             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 273             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 277             velecsum         = _mm_add_pd(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm_mul_pd(fscal,dx00);
 285             ty               = _mm_mul_pd(fscal,dy00);
 286             tz               = _mm_mul_pd(fscal,dz00);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_add_pd(fix0,tx);
 290             fiy0             = _mm_add_pd(fiy0,ty);
 291             fiz0             = _mm_add_pd(fiz0,tz);
 292
 293             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 294
 295             /* Inner loop uses 41 flops */
 296         }
 297
 298         /* End of innermost loop */
 299
 300         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 301                                               f+i_coord_offset,fshift+i_shift_offset);
 302
 303         ggid                        = gid[iidx];
 304         /* Update potential energies */
 305         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 306
 307         /* Increment number of inner iterations */
 308         inneriter                  += j_index_end - j_index_start;
 309
 310         /* Outer loop uses 8 flops */
 311     }
 312
 313     /* Increment number of outer iterations */
 314     outeriter        += nri;
 315
 316     /* Update outer/inner flops */
 317
 318     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*41);
 319 }
 320 /*
 321  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse2_double
 322  * Electrostatics interaction: Ewald
 323  * VdW interaction:            None
 324  * Geometry:                   Particle-Particle
 325  * Calculate force/pot:        Force
 326  */
 327 void
 328 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse2_double
 329                     (t_nblist                    * gmx_restrict       nlist,
 330                      rvec                        * gmx_restrict          xx,
 331                      rvec                        * gmx_restrict          ff,
 332                      struct t_forcerec           * gmx_restrict          fr,
 333                      t_mdatoms                   * gmx_restrict     mdatoms,
 334                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 335                      t_nrnb                      * gmx_restrict        nrnb)
 336 {
 337     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 338      * just 0 for non-waters.
 339      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 340      * jnr indices corresponding to data put in the four positions in the SIMD register.
 341      */
 342     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 343     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 344     int              jnrA,jnrB;
 345     int              j_coord_offsetA,j_coord_offsetB;
 346     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 347     real             rcutoff_scalar;
 348     real             *shiftvec,*fshift,*x,*f;
 349     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 350     int              vdwioffset0;
 351     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 352     int              vdwjidx0A,vdwjidx0B;
 353     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 354     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 355     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 356     real             *charge;
 357     __m128i          ewitab;
 358     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 359     real             *ewtab;
 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     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 379     ewtab            = fr->ic->tabq_coul_F;
 380     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 381     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_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           = sse2_invsqrt_d(rsq00);
 438
 439             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 440
 441             /* Load parameters for j particles */
 442             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             r00              = _mm_mul_pd(rsq00,rinv00);
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq00             = _mm_mul_pd(iq0,jq0);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 456             ewrt             = _mm_mul_pd(r00,ewtabscale);
 457             ewitab           = _mm_cvttpd_epi32(ewrt);
 458             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 459             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 460                                          &ewtabF,&ewtabFn);
 461             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 462             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 463
 464             fscal            = felec;
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm_mul_pd(fscal,dx00);
 468             ty               = _mm_mul_pd(fscal,dy00);
 469             tz               = _mm_mul_pd(fscal,dz00);
 470
 471             /* Update vectorial force */
 472             fix0             = _mm_add_pd(fix0,tx);
 473             fiy0             = _mm_add_pd(fiy0,ty);
 474             fiz0             = _mm_add_pd(fiz0,tz);
 475
 476             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 477
 478             /* Inner loop uses 36 flops */
 479         }
 480
 481         if(jidx<j_index_end)
 482         {
 483
 484             jnrA             = jjnr[jidx];
 485             j_coord_offsetA  = DIM*jnrA;
 486
 487             /* load j atom coordinates */
 488             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 489                                               &jx0,&jy0,&jz0);
 490
 491             /* Calculate displacement vector */
 492             dx00             = _mm_sub_pd(ix0,jx0);
 493             dy00             = _mm_sub_pd(iy0,jy0);
 494             dz00             = _mm_sub_pd(iz0,jz0);
 495
 496             /* Calculate squared distance and things based on it */
 497             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 498
 499             rinv00           = sse2_invsqrt_d(rsq00);
 500
 501             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 502
 503             /* Load parameters for j particles */
 504             jq0              = _mm_load_sd(charge+jnrA+0);
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             r00              = _mm_mul_pd(rsq00,rinv00);
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq00             = _mm_mul_pd(iq0,jq0);
 514
 515             /* EWALD ELECTROSTATICS */
 516
 517             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 518             ewrt             = _mm_mul_pd(r00,ewtabscale);
 519             ewitab           = _mm_cvttpd_epi32(ewrt);
 520             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 521             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 522             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 523             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 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,dx00);
 531             ty               = _mm_mul_pd(fscal,dy00);
 532             tz               = _mm_mul_pd(fscal,dz00);
 533
 534             /* Update vectorial force */
 535             fix0             = _mm_add_pd(fix0,tx);
 536             fiy0             = _mm_add_pd(fiy0,ty);
 537             fiz0             = _mm_add_pd(fiz0,tz);
 538
 539             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 540
 541             /* Inner loop uses 36 flops */
 542         }
 543
 544         /* End of innermost loop */
 545
 546         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 547                                               f+i_coord_offset,fshift+i_shift_offset);
 548
 549         /* Increment number of inner iterations */
 550         inneriter                  += j_index_end - j_index_start;
 551
 552         /* Outer loop uses 7 flops */
 553     }
 554
 555     /* Increment number of outer iterations */
 556     outeriter        += nri;
 557
 558     /* Update outer/inner flops */
 559
 560     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*36);
 561 }