src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwNone_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          ewitab;
  90     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  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     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     /* Start outer loop over neighborlists */
 124     for(iidx=0; iidx<nri; iidx++)
 125     {
 126         /* Load shift vector for this list */
 127         i_shift_offset   = DIM*shiftidx[iidx];
 128
 129         /* Load limits for loop over neighbors */
 130         j_index_start    = jindex[iidx];
 131         j_index_end      = jindex[iidx+1];
 132
 133         /* Get outer coordinate index */
 134         inr              = iinr[iidx];
 135         i_coord_offset   = DIM*inr;
 136
 137         /* Load i particle coords and add shift vector */
 138         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 139
 140         fix0             = _mm_setzero_pd();
 141         fiy0             = _mm_setzero_pd();
 142         fiz0             = _mm_setzero_pd();
 143
 144         /* Load parameters for i particles */
 145         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 146
 147         /* Reset potential sums */
 148         velecsum         = _mm_setzero_pd();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             j_coord_offsetA  = DIM*jnrA;
 158             j_coord_offsetB  = DIM*jnrB;
 159
 160             /* load j atom coordinates */
 161             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 162                                               &jx0,&jy0,&jz0);
 163
 164             /* Calculate displacement vector */
 165             dx00             = _mm_sub_pd(ix0,jx0);
 166             dy00             = _mm_sub_pd(iy0,jy0);
 167             dz00             = _mm_sub_pd(iz0,jz0);
 168
 169             /* Calculate squared distance and things based on it */
 170             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 171
 172             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 173
 174             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 175
 176             /* Load parameters for j particles */
 177             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             r00              = _mm_mul_pd(rsq00,rinv00);
 184
 185             /* Compute parameters for interactions between i and j atoms */
 186             qq00             = _mm_mul_pd(iq0,jq0);
 187
 188             /* EWALD ELECTROSTATICS */
 189
 190             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 191             ewrt             = _mm_mul_pd(r00,ewtabscale);
 192             ewitab           = _mm_cvttpd_epi32(ewrt);
 193 #ifdef __XOP__
 194             eweps            = _mm_frcz_pd(ewrt);
 195 #else
 196             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 197 #endif
 198             twoeweps         = _mm_add_pd(eweps,eweps);
 199             ewitab           = _mm_slli_epi32(ewitab,2);
 200             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 201             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 202             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 203             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 204             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 205             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 206             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 207             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 208             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 209             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             velecsum         = _mm_add_pd(velecsum,velec);
 213
 214             fscal            = felec;
 215
 216             /* Update vectorial force */
 217             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 218             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 219             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 220
 221             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 222                                                    _mm_mul_pd(dx00,fscal),
 223                                                    _mm_mul_pd(dy00,fscal),
 224                                                    _mm_mul_pd(dz00,fscal));
 225
 226             /* Inner loop uses 44 flops */
 227         }
 228
 229         if(jidx<j_index_end)
 230         {
 231
 232             jnrA             = jjnr[jidx];
 233             j_coord_offsetA  = DIM*jnrA;
 234
 235             /* load j atom coordinates */
 236             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 237                                               &jx0,&jy0,&jz0);
 238
 239             /* Calculate displacement vector */
 240             dx00             = _mm_sub_pd(ix0,jx0);
 241             dy00             = _mm_sub_pd(iy0,jy0);
 242             dz00             = _mm_sub_pd(iz0,jz0);
 243
 244             /* Calculate squared distance and things based on it */
 245             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 246
 247             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 248
 249             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 250
 251             /* Load parameters for j particles */
 252             jq0              = _mm_load_sd(charge+jnrA+0);
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             r00              = _mm_mul_pd(rsq00,rinv00);
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq00             = _mm_mul_pd(iq0,jq0);
 262
 263             /* EWALD ELECTROSTATICS */
 264
 265             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 266             ewrt             = _mm_mul_pd(r00,ewtabscale);
 267             ewitab           = _mm_cvttpd_epi32(ewrt);
 268 #ifdef __XOP__
 269             eweps            = _mm_frcz_pd(ewrt);
 270 #else
 271             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 272 #endif
 273             twoeweps         = _mm_add_pd(eweps,eweps);
 274             ewitab           = _mm_slli_epi32(ewitab,2);
 275             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 276             ewtabD           = _mm_setzero_pd();
 277             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 278             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 279             ewtabFn          = _mm_setzero_pd();
 280             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 281             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 282             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 283             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 284             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 288             velecsum         = _mm_add_pd(velecsum,velec);
 289
 290             fscal            = felec;
 291
 292             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 293
 294             /* Update vectorial force */
 295             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 296             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 297             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 298
 299             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 300                                                    _mm_mul_pd(dx00,fscal),
 301                                                    _mm_mul_pd(dy00,fscal),
 302                                                    _mm_mul_pd(dz00,fscal));
 303
 304             /* Inner loop uses 44 flops */
 305         }
 306
 307         /* End of innermost loop */
 308
 309         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 310                                               f+i_coord_offset,fshift+i_shift_offset);
 311
 312         ggid                        = gid[iidx];
 313         /* Update potential energies */
 314         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 315
 316         /* Increment number of inner iterations */
 317         inneriter                  += j_index_end - j_index_start;
 318
 319         /* Outer loop uses 8 flops */
 320     }
 321
 322     /* Increment number of outer iterations */
 323     outeriter        += nri;
 324
 325     /* Update outer/inner flops */
 326
 327     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
 328 }
 329 /*
 330  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_double
 331  * Electrostatics interaction: Ewald
 332  * VdW interaction:            None
 333  * Geometry:                   Particle-Particle
 334  * Calculate force/pot:        Force
 335  */
 336 void
 337 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_double
 338                     (t_nblist                    * gmx_restrict       nlist,
 339                      rvec                        * gmx_restrict          xx,
 340                      rvec                        * gmx_restrict          ff,
 341                      t_forcerec                  * gmx_restrict          fr,
 342                      t_mdatoms                   * gmx_restrict     mdatoms,
 343                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 344                      t_nrnb                      * gmx_restrict        nrnb)
 345 {
 346     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 347      * just 0 for non-waters.
 348      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 349      * jnr indices corresponding to data put in the four positions in the SIMD register.
 350      */
 351     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 352     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 353     int              jnrA,jnrB;
 354     int              j_coord_offsetA,j_coord_offsetB;
 355     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 356     real             rcutoff_scalar;
 357     real             *shiftvec,*fshift,*x,*f;
 358     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 359     int              vdwioffset0;
 360     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 361     int              vdwjidx0A,vdwjidx0B;
 362     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 363     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 364     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 365     real             *charge;
 366     __m128i          ewitab;
 367     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 368     real             *ewtab;
 369     __m128d          dummy_mask,cutoff_mask;
 370     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 371     __m128d          one     = _mm_set1_pd(1.0);
 372     __m128d          two     = _mm_set1_pd(2.0);
 373     x                = xx[0];
 374     f                = ff[0];
 375
 376     nri              = nlist->nri;
 377     iinr             = nlist->iinr;
 378     jindex           = nlist->jindex;
 379     jjnr             = nlist->jjnr;
 380     shiftidx         = nlist->shift;
 381     gid              = nlist->gid;
 382     shiftvec         = fr->shift_vec[0];
 383     fshift           = fr->fshift[0];
 384     facel            = _mm_set1_pd(fr->epsfac);
 385     charge           = mdatoms->chargeA;
 386
 387     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 388     ewtab            = fr->ic->tabq_coul_F;
 389     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 390     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 391
 392     /* Avoid stupid compiler warnings */
 393     jnrA = jnrB = 0;
 394     j_coord_offsetA = 0;
 395     j_coord_offsetB = 0;
 396
 397     outeriter        = 0;
 398     inneriter        = 0;
 399
 400     /* Start outer loop over neighborlists */
 401     for(iidx=0; iidx<nri; iidx++)
 402     {
 403         /* Load shift vector for this list */
 404         i_shift_offset   = DIM*shiftidx[iidx];
 405
 406         /* Load limits for loop over neighbors */
 407         j_index_start    = jindex[iidx];
 408         j_index_end      = jindex[iidx+1];
 409
 410         /* Get outer coordinate index */
 411         inr              = iinr[iidx];
 412         i_coord_offset   = DIM*inr;
 413
 414         /* Load i particle coords and add shift vector */
 415         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 416
 417         fix0             = _mm_setzero_pd();
 418         fiy0             = _mm_setzero_pd();
 419         fiz0             = _mm_setzero_pd();
 420
 421         /* Load parameters for i particles */
 422         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 423
 424         /* Start inner kernel loop */
 425         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 426         {
 427
 428             /* Get j neighbor index, and coordinate index */
 429             jnrA             = jjnr[jidx];
 430             jnrB             = jjnr[jidx+1];
 431             j_coord_offsetA  = DIM*jnrA;
 432             j_coord_offsetB  = DIM*jnrB;
 433
 434             /* load j atom coordinates */
 435             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 436                                               &jx0,&jy0,&jz0);
 437
 438             /* Calculate displacement vector */
 439             dx00             = _mm_sub_pd(ix0,jx0);
 440             dy00             = _mm_sub_pd(iy0,jy0);
 441             dz00             = _mm_sub_pd(iz0,jz0);
 442
 443             /* Calculate squared distance and things based on it */
 444             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 445
 446             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 447
 448             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 449
 450             /* Load parameters for j particles */
 451             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             r00              = _mm_mul_pd(rsq00,rinv00);
 458
 459             /* Compute parameters for interactions between i and j atoms */
 460             qq00             = _mm_mul_pd(iq0,jq0);
 461
 462             /* EWALD ELECTROSTATICS */
 463
 464             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 465             ewrt             = _mm_mul_pd(r00,ewtabscale);
 466             ewitab           = _mm_cvttpd_epi32(ewrt);
 467 #ifdef __XOP__
 468             eweps            = _mm_frcz_pd(ewrt);
 469 #else
 470             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 471 #endif
 472             twoeweps         = _mm_add_pd(eweps,eweps);
 473             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 474                                          &ewtabF,&ewtabFn);
 475             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 476             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 477
 478             fscal            = felec;
 479
 480             /* Update vectorial force */
 481             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 482             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 483             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 484
 485             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 486                                                    _mm_mul_pd(dx00,fscal),
 487                                                    _mm_mul_pd(dy00,fscal),
 488                                                    _mm_mul_pd(dz00,fscal));
 489
 490             /* Inner loop uses 39 flops */
 491         }
 492
 493         if(jidx<j_index_end)
 494         {
 495
 496             jnrA             = jjnr[jidx];
 497             j_coord_offsetA  = DIM*jnrA;
 498
 499             /* load j atom coordinates */
 500             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 501                                               &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _mm_sub_pd(ix0,jx0);
 505             dy00             = _mm_sub_pd(iy0,jy0);
 506             dz00             = _mm_sub_pd(iz0,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 510
 511             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 512
 513             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 514
 515             /* Load parameters for j particles */
 516             jq0              = _mm_load_sd(charge+jnrA+0);
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             r00              = _mm_mul_pd(rsq00,rinv00);
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             qq00             = _mm_mul_pd(iq0,jq0);
 526
 527             /* EWALD ELECTROSTATICS */
 528
 529             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 530             ewrt             = _mm_mul_pd(r00,ewtabscale);
 531             ewitab           = _mm_cvttpd_epi32(ewrt);
 532 #ifdef __XOP__
 533             eweps            = _mm_frcz_pd(ewrt);
 534 #else
 535             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 536 #endif
 537             twoeweps         = _mm_add_pd(eweps,eweps);
 538             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 539             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 540             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 541
 542             fscal            = felec;
 543
 544             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 545
 546             /* Update vectorial force */
 547             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 548             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 549             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 550
 551             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 552                                                    _mm_mul_pd(dx00,fscal),
 553                                                    _mm_mul_pd(dy00,fscal),
 554                                                    _mm_mul_pd(dz00,fscal));
 555
 556             /* Inner loop uses 39 flops */
 557         }
 558
 559         /* End of innermost loop */
 560
 561         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 562                                               f+i_coord_offset,fshift+i_shift_offset);
 563
 564         /* Increment number of inner iterations */
 565         inneriter                  += j_index_end - j_index_start;
 566
 567         /* Outer loop uses 7 flops */
 568     }
 569
 570     /* Increment number of outer iterations */
 571     outeriter        += nri;
 572
 573     /* Update outer/inner flops */
 574
 575     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 576 }