src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwNone_GeomP1P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_sse4_1_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     __m128i          ewitab;
  74     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  75     real             *ewtab;
  76     __m128d          dummy_mask,cutoff_mask;
  77     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  78     __m128d          one     = _mm_set1_pd(1.0);
  79     __m128d          two     = _mm_set1_pd(2.0);
  80     x                = xx[0];
  81     f                = ff[0];
  82
  83     nri              = nlist->nri;
  84     iinr             = nlist->iinr;
  85     jindex           = nlist->jindex;
  86     jjnr             = nlist->jjnr;
  87     shiftidx         = nlist->shift;
  88     gid              = nlist->gid;
  89     shiftvec         = fr->shift_vec[0];
  90     fshift           = fr->fshift[0];
  91     facel            = _mm_set1_pd(fr->epsfac);
  92     charge           = mdatoms->chargeA;
  93
  94     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
  95     ewtab            = fr->ic->tabq_coul_FDV0;
  96     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
  97     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
  98
  99     /* Avoid stupid compiler warnings */
 100     jnrA = jnrB = 0;
 101     j_coord_offsetA = 0;
 102     j_coord_offsetB = 0;
 103
 104     outeriter        = 0;
 105     inneriter        = 0;
 106
 107     /* Start outer loop over neighborlists */
 108     for(iidx=0; iidx<nri; iidx++)
 109     {
 110         /* Load shift vector for this list */
 111         i_shift_offset   = DIM*shiftidx[iidx];
 112
 113         /* Load limits for loop over neighbors */
 114         j_index_start    = jindex[iidx];
 115         j_index_end      = jindex[iidx+1];
 116
 117         /* Get outer coordinate index */
 118         inr              = iinr[iidx];
 119         i_coord_offset   = DIM*inr;
 120
 121         /* Load i particle coords and add shift vector */
 122         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 123
 124         fix0             = _mm_setzero_pd();
 125         fiy0             = _mm_setzero_pd();
 126         fiz0             = _mm_setzero_pd();
 127
 128         /* Load parameters for i particles */
 129         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 130
 131         /* Reset potential sums */
 132         velecsum         = _mm_setzero_pd();
 133
 134         /* Start inner kernel loop */
 135         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 136         {
 137
 138             /* Get j neighbor index, and coordinate index */
 139             jnrA             = jjnr[jidx];
 140             jnrB             = jjnr[jidx+1];
 141             j_coord_offsetA  = DIM*jnrA;
 142             j_coord_offsetB  = DIM*jnrB;
 143
 144             /* load j atom coordinates */
 145             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 146                                               &jx0,&jy0,&jz0);
 147
 148             /* Calculate displacement vector */
 149             dx00             = _mm_sub_pd(ix0,jx0);
 150             dy00             = _mm_sub_pd(iy0,jy0);
 151             dz00             = _mm_sub_pd(iz0,jz0);
 152
 153             /* Calculate squared distance and things based on it */
 154             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 155
 156             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 157
 158             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 159
 160             /* Load parameters for j particles */
 161             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 162
 163             /**************************
 164              * CALCULATE INTERACTIONS *
 165              **************************/
 166
 167             r00              = _mm_mul_pd(rsq00,rinv00);
 168
 169             /* Compute parameters for interactions between i and j atoms */
 170             qq00             = _mm_mul_pd(iq0,jq0);
 171
 172             /* EWALD ELECTROSTATICS */
 173
 174             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 175             ewrt             = _mm_mul_pd(r00,ewtabscale);
 176             ewitab           = _mm_cvttpd_epi32(ewrt);
 177             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 178             ewitab           = _mm_slli_epi32(ewitab,2);
 179             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 180             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 181             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 182             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 183             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 184             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 185             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 186             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 187             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 188             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 189
 190             /* Update potential sum for this i atom from the interaction with this j atom. */
 191             velecsum         = _mm_add_pd(velecsum,velec);
 192
 193             fscal            = felec;
 194
 195             /* Calculate temporary vectorial force */
 196             tx               = _mm_mul_pd(fscal,dx00);
 197             ty               = _mm_mul_pd(fscal,dy00);
 198             tz               = _mm_mul_pd(fscal,dz00);
 199
 200             /* Update vectorial force */
 201             fix0             = _mm_add_pd(fix0,tx);
 202             fiy0             = _mm_add_pd(fiy0,ty);
 203             fiz0             = _mm_add_pd(fiz0,tz);
 204
 205             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 206
 207             /* Inner loop uses 41 flops */
 208         }
 209
 210         if(jidx<j_index_end)
 211         {
 212
 213             jnrA             = jjnr[jidx];
 214             j_coord_offsetA  = DIM*jnrA;
 215
 216             /* load j atom coordinates */
 217             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 218                                               &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm_sub_pd(ix0,jx0);
 222             dy00             = _mm_sub_pd(iy0,jy0);
 223             dz00             = _mm_sub_pd(iz0,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 227
 228             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 229
 230             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 231
 232             /* Load parameters for j particles */
 233             jq0              = _mm_load_sd(charge+jnrA+0);
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_pd(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm_mul_pd(iq0,jq0);
 243
 244             /* EWALD ELECTROSTATICS */
 245
 246             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 247             ewrt             = _mm_mul_pd(r00,ewtabscale);
 248             ewitab           = _mm_cvttpd_epi32(ewrt);
 249             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 250             ewitab           = _mm_slli_epi32(ewitab,2);
 251             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 252             ewtabD           = _mm_setzero_pd();
 253             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 254             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 255             ewtabFn          = _mm_setzero_pd();
 256             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 257             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 258             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 259             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 260             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 264             velecsum         = _mm_add_pd(velecsum,velec);
 265
 266             fscal            = felec;
 267
 268             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm_mul_pd(fscal,dx00);
 272             ty               = _mm_mul_pd(fscal,dy00);
 273             tz               = _mm_mul_pd(fscal,dz00);
 274
 275             /* Update vectorial force */
 276             fix0             = _mm_add_pd(fix0,tx);
 277             fiy0             = _mm_add_pd(fiy0,ty);
 278             fiz0             = _mm_add_pd(fiz0,tz);
 279
 280             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 281
 282             /* Inner loop uses 41 flops */
 283         }
 284
 285         /* End of innermost loop */
 286
 287         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 288                                               f+i_coord_offset,fshift+i_shift_offset);
 289
 290         ggid                        = gid[iidx];
 291         /* Update potential energies */
 292         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 293
 294         /* Increment number of inner iterations */
 295         inneriter                  += j_index_end - j_index_start;
 296
 297         /* Outer loop uses 8 flops */
 298     }
 299
 300     /* Increment number of outer iterations */
 301     outeriter        += nri;
 302
 303     /* Update outer/inner flops */
 304
 305     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*41);
 306 }
 307 /*
 308  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_double
 309  * Electrostatics interaction: Ewald
 310  * VdW interaction:            None
 311  * Geometry:                   Particle-Particle
 312  * Calculate force/pot:        Force
 313  */
 314 void
 315 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_double
 316                     (t_nblist * gmx_restrict                nlist,
 317                      rvec * gmx_restrict                    xx,
 318                      rvec * gmx_restrict                    ff,
 319                      t_forcerec * gmx_restrict              fr,
 320                      t_mdatoms * gmx_restrict               mdatoms,
 321                      nb_kernel_data_t * gmx_restrict        kernel_data,
 322                      t_nrnb * gmx_restrict                  nrnb)
 323 {
 324     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 325      * just 0 for non-waters.
 326      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 327      * jnr indices corresponding to data put in the four positions in the SIMD register.
 328      */
 329     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 330     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 331     int              jnrA,jnrB;
 332     int              j_coord_offsetA,j_coord_offsetB;
 333     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 334     real             rcutoff_scalar;
 335     real             *shiftvec,*fshift,*x,*f;
 336     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 337     int              vdwioffset0;
 338     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 339     int              vdwjidx0A,vdwjidx0B;
 340     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 341     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 342     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 343     real             *charge;
 344     __m128i          ewitab;
 345     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 346     real             *ewtab;
 347     __m128d          dummy_mask,cutoff_mask;
 348     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 349     __m128d          one     = _mm_set1_pd(1.0);
 350     __m128d          two     = _mm_set1_pd(2.0);
 351     x                = xx[0];
 352     f                = ff[0];
 353
 354     nri              = nlist->nri;
 355     iinr             = nlist->iinr;
 356     jindex           = nlist->jindex;
 357     jjnr             = nlist->jjnr;
 358     shiftidx         = nlist->shift;
 359     gid              = nlist->gid;
 360     shiftvec         = fr->shift_vec[0];
 361     fshift           = fr->fshift[0];
 362     facel            = _mm_set1_pd(fr->epsfac);
 363     charge           = mdatoms->chargeA;
 364
 365     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 366     ewtab            = fr->ic->tabq_coul_F;
 367     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 368     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 369
 370     /* Avoid stupid compiler warnings */
 371     jnrA = jnrB = 0;
 372     j_coord_offsetA = 0;
 373     j_coord_offsetB = 0;
 374
 375     outeriter        = 0;
 376     inneriter        = 0;
 377
 378     /* Start outer loop over neighborlists */
 379     for(iidx=0; iidx<nri; iidx++)
 380     {
 381         /* Load shift vector for this list */
 382         i_shift_offset   = DIM*shiftidx[iidx];
 383
 384         /* Load limits for loop over neighbors */
 385         j_index_start    = jindex[iidx];
 386         j_index_end      = jindex[iidx+1];
 387
 388         /* Get outer coordinate index */
 389         inr              = iinr[iidx];
 390         i_coord_offset   = DIM*inr;
 391
 392         /* Load i particle coords and add shift vector */
 393         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 394
 395         fix0             = _mm_setzero_pd();
 396         fiy0             = _mm_setzero_pd();
 397         fiz0             = _mm_setzero_pd();
 398
 399         /* Load parameters for i particles */
 400         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 401
 402         /* Start inner kernel loop */
 403         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrA             = jjnr[jidx];
 408             jnrB             = jjnr[jidx+1];
 409             j_coord_offsetA  = DIM*jnrA;
 410             j_coord_offsetB  = DIM*jnrB;
 411
 412             /* load j atom coordinates */
 413             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 414                                               &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _mm_sub_pd(ix0,jx0);
 418             dy00             = _mm_sub_pd(iy0,jy0);
 419             dz00             = _mm_sub_pd(iz0,jz0);
 420
 421             /* Calculate squared distance and things based on it */
 422             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 423
 424             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 425
 426             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 427
 428             /* Load parameters for j particles */
 429             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 430
 431             /**************************
 432              * CALCULATE INTERACTIONS *
 433              **************************/
 434
 435             r00              = _mm_mul_pd(rsq00,rinv00);
 436
 437             /* Compute parameters for interactions between i and j atoms */
 438             qq00             = _mm_mul_pd(iq0,jq0);
 439
 440             /* EWALD ELECTROSTATICS */
 441
 442             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 443             ewrt             = _mm_mul_pd(r00,ewtabscale);
 444             ewitab           = _mm_cvttpd_epi32(ewrt);
 445             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 446             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 447                                          &ewtabF,&ewtabFn);
 448             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 449             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 450
 451             fscal            = felec;
 452
 453             /* Calculate temporary vectorial force */
 454             tx               = _mm_mul_pd(fscal,dx00);
 455             ty               = _mm_mul_pd(fscal,dy00);
 456             tz               = _mm_mul_pd(fscal,dz00);
 457
 458             /* Update vectorial force */
 459             fix0             = _mm_add_pd(fix0,tx);
 460             fiy0             = _mm_add_pd(fiy0,ty);
 461             fiz0             = _mm_add_pd(fiz0,tz);
 462
 463             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 464
 465             /* Inner loop uses 36 flops */
 466         }
 467
 468         if(jidx<j_index_end)
 469         {
 470
 471             jnrA             = jjnr[jidx];
 472             j_coord_offsetA  = DIM*jnrA;
 473
 474             /* load j atom coordinates */
 475             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 476                                               &jx0,&jy0,&jz0);
 477
 478             /* Calculate displacement vector */
 479             dx00             = _mm_sub_pd(ix0,jx0);
 480             dy00             = _mm_sub_pd(iy0,jy0);
 481             dz00             = _mm_sub_pd(iz0,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 485
 486             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 487
 488             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 489
 490             /* Load parameters for j particles */
 491             jq0              = _mm_load_sd(charge+jnrA+0);
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             r00              = _mm_mul_pd(rsq00,rinv00);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq00             = _mm_mul_pd(iq0,jq0);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 505             ewrt             = _mm_mul_pd(r00,ewtabscale);
 506             ewitab           = _mm_cvttpd_epi32(ewrt);
 507             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 508             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 509             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 510             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 511
 512             fscal            = felec;
 513
 514             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm_mul_pd(fscal,dx00);
 518             ty               = _mm_mul_pd(fscal,dy00);
 519             tz               = _mm_mul_pd(fscal,dz00);
 520
 521             /* Update vectorial force */
 522             fix0             = _mm_add_pd(fix0,tx);
 523             fiy0             = _mm_add_pd(fiy0,ty);
 524             fiz0             = _mm_add_pd(fiz0,tz);
 525
 526             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 527
 528             /* Inner loop uses 36 flops */
 529         }
 530
 531         /* End of innermost loop */
 532
 533         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 534                                               f+i_coord_offset,fshift+i_shift_offset);
 535
 536         /* Increment number of inner iterations */
 537         inneriter                  += j_index_end - j_index_start;
 538
 539         /* Outer loop uses 7 flops */
 540     }
 541
 542     /* Increment number of outer iterations */
 543     outeriter        += nri;
 544
 545     /* Update outer/inner flops */
 546
 547     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*36);
 548 }