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) 2012,2013,2014, 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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     __m128i          ewitab;
  88     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  89     real             *ewtab;
  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->epsfac);
 106     charge           = mdatoms->chargeA;
 107
 108     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 109     ewtab            = fr->ic->tabq_coul_FDV0;
 110     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 111     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 137
 138         fix0             = _mm_setzero_pd();
 139         fiy0             = _mm_setzero_pd();
 140         fiz0             = _mm_setzero_pd();
 141
 142         /* Load parameters for i particles */
 143         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 144
 145         /* Reset potential sums */
 146         velecsum         = _mm_setzero_pd();
 147
 148         /* Start inner kernel loop */
 149         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 150         {
 151
 152             /* Get j neighbor index, and coordinate index */
 153             jnrA             = jjnr[jidx];
 154             jnrB             = jjnr[jidx+1];
 155             j_coord_offsetA  = DIM*jnrA;
 156             j_coord_offsetB  = DIM*jnrB;
 157
 158             /* load j atom coordinates */
 159             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 160                                               &jx0,&jy0,&jz0);
 161
 162             /* Calculate displacement vector */
 163             dx00             = _mm_sub_pd(ix0,jx0);
 164             dy00             = _mm_sub_pd(iy0,jy0);
 165             dz00             = _mm_sub_pd(iz0,jz0);
 166
 167             /* Calculate squared distance and things based on it */
 168             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 169
 170             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 171
 172             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 173
 174             /* Load parameters for j particles */
 175             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 176
 177             /**************************
 178              * CALCULATE INTERACTIONS *
 179              **************************/
 180
 181             r00              = _mm_mul_pd(rsq00,rinv00);
 182
 183             /* Compute parameters for interactions between i and j atoms */
 184             qq00             = _mm_mul_pd(iq0,jq0);
 185
 186             /* EWALD ELECTROSTATICS */
 187
 188             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 189             ewrt             = _mm_mul_pd(r00,ewtabscale);
 190             ewitab           = _mm_cvttpd_epi32(ewrt);
 191 #ifdef __XOP__
 192             eweps            = _mm_frcz_pd(ewrt);
 193 #else
 194             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 195 #endif
 196             twoeweps         = _mm_add_pd(eweps,eweps);
 197             ewitab           = _mm_slli_epi32(ewitab,2);
 198             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 199             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 200             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 201             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 202             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 203             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 204             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 205             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 206             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 207             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 208
 209             /* Update potential sum for this i atom from the interaction with this j atom. */
 210             velecsum         = _mm_add_pd(velecsum,velec);
 211
 212             fscal            = felec;
 213
 214             /* Update vectorial force */
 215             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 216             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 217             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 218
 219             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 220                                                    _mm_mul_pd(dx00,fscal),
 221                                                    _mm_mul_pd(dy00,fscal),
 222                                                    _mm_mul_pd(dz00,fscal));
 223
 224             /* Inner loop uses 44 flops */
 225         }
 226
 227         if(jidx<j_index_end)
 228         {
 229
 230             jnrA             = jjnr[jidx];
 231             j_coord_offsetA  = DIM*jnrA;
 232
 233             /* load j atom coordinates */
 234             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 235                                               &jx0,&jy0,&jz0);
 236
 237             /* Calculate displacement vector */
 238             dx00             = _mm_sub_pd(ix0,jx0);
 239             dy00             = _mm_sub_pd(iy0,jy0);
 240             dz00             = _mm_sub_pd(iz0,jz0);
 241
 242             /* Calculate squared distance and things based on it */
 243             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 244
 245             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 246
 247             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 248
 249             /* Load parameters for j particles */
 250             jq0              = _mm_load_sd(charge+jnrA+0);
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             r00              = _mm_mul_pd(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq00             = _mm_mul_pd(iq0,jq0);
 260
 261             /* EWALD ELECTROSTATICS */
 262
 263             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 264             ewrt             = _mm_mul_pd(r00,ewtabscale);
 265             ewitab           = _mm_cvttpd_epi32(ewrt);
 266 #ifdef __XOP__
 267             eweps            = _mm_frcz_pd(ewrt);
 268 #else
 269             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 270 #endif
 271             twoeweps         = _mm_add_pd(eweps,eweps);
 272             ewitab           = _mm_slli_epi32(ewitab,2);
 273             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 274             ewtabD           = _mm_setzero_pd();
 275             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 276             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 277             ewtabFn          = _mm_setzero_pd();
 278             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 279             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 280             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 281             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 282             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 286             velecsum         = _mm_add_pd(velecsum,velec);
 287
 288             fscal            = felec;
 289
 290             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 294             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 295             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 296
 297             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 298                                                    _mm_mul_pd(dx00,fscal),
 299                                                    _mm_mul_pd(dy00,fscal),
 300                                                    _mm_mul_pd(dz00,fscal));
 301
 302             /* Inner loop uses 44 flops */
 303         }
 304
 305         /* End of innermost loop */
 306
 307         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 308                                               f+i_coord_offset,fshift+i_shift_offset);
 309
 310         ggid                        = gid[iidx];
 311         /* Update potential energies */
 312         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 313
 314         /* Increment number of inner iterations */
 315         inneriter                  += j_index_end - j_index_start;
 316
 317         /* Outer loop uses 8 flops */
 318     }
 319
 320     /* Increment number of outer iterations */
 321     outeriter        += nri;
 322
 323     /* Update outer/inner flops */
 324
 325     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
 326 }
 327 /*
 328  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_double
 329  * Electrostatics interaction: Ewald
 330  * VdW interaction:            None
 331  * Geometry:                   Particle-Particle
 332  * Calculate force/pot:        Force
 333  */
 334 void
 335 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_double
 336                     (t_nblist                    * gmx_restrict       nlist,
 337                      rvec                        * gmx_restrict          xx,
 338                      rvec                        * gmx_restrict          ff,
 339                      t_forcerec                  * gmx_restrict          fr,
 340                      t_mdatoms                   * gmx_restrict     mdatoms,
 341                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 342                      t_nrnb                      * gmx_restrict        nrnb)
 343 {
 344     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 345      * just 0 for non-waters.
 346      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 347      * jnr indices corresponding to data put in the four positions in the SIMD register.
 348      */
 349     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 350     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 351     int              jnrA,jnrB;
 352     int              j_coord_offsetA,j_coord_offsetB;
 353     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 354     real             rcutoff_scalar;
 355     real             *shiftvec,*fshift,*x,*f;
 356     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 357     int              vdwioffset0;
 358     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 359     int              vdwjidx0A,vdwjidx0B;
 360     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 361     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 362     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 363     real             *charge;
 364     __m128i          ewitab;
 365     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 366     real             *ewtab;
 367     __m128d          dummy_mask,cutoff_mask;
 368     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 369     __m128d          one     = _mm_set1_pd(1.0);
 370     __m128d          two     = _mm_set1_pd(2.0);
 371     x                = xx[0];
 372     f                = ff[0];
 373
 374     nri              = nlist->nri;
 375     iinr             = nlist->iinr;
 376     jindex           = nlist->jindex;
 377     jjnr             = nlist->jjnr;
 378     shiftidx         = nlist->shift;
 379     gid              = nlist->gid;
 380     shiftvec         = fr->shift_vec[0];
 381     fshift           = fr->fshift[0];
 382     facel            = _mm_set1_pd(fr->epsfac);
 383     charge           = mdatoms->chargeA;
 384
 385     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 386     ewtab            = fr->ic->tabq_coul_F;
 387     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 388     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 389
 390     /* Avoid stupid compiler warnings */
 391     jnrA = jnrB = 0;
 392     j_coord_offsetA = 0;
 393     j_coord_offsetB = 0;
 394
 395     outeriter        = 0;
 396     inneriter        = 0;
 397
 398     /* Start outer loop over neighborlists */
 399     for(iidx=0; iidx<nri; iidx++)
 400     {
 401         /* Load shift vector for this list */
 402         i_shift_offset   = DIM*shiftidx[iidx];
 403
 404         /* Load limits for loop over neighbors */
 405         j_index_start    = jindex[iidx];
 406         j_index_end      = jindex[iidx+1];
 407
 408         /* Get outer coordinate index */
 409         inr              = iinr[iidx];
 410         i_coord_offset   = DIM*inr;
 411
 412         /* Load i particle coords and add shift vector */
 413         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 414
 415         fix0             = _mm_setzero_pd();
 416         fiy0             = _mm_setzero_pd();
 417         fiz0             = _mm_setzero_pd();
 418
 419         /* Load parameters for i particles */
 420         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 421
 422         /* Start inner kernel loop */
 423         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 424         {
 425
 426             /* Get j neighbor index, and coordinate index */
 427             jnrA             = jjnr[jidx];
 428             jnrB             = jjnr[jidx+1];
 429             j_coord_offsetA  = DIM*jnrA;
 430             j_coord_offsetB  = DIM*jnrB;
 431
 432             /* load j atom coordinates */
 433             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 434                                               &jx0,&jy0,&jz0);
 435
 436             /* Calculate displacement vector */
 437             dx00             = _mm_sub_pd(ix0,jx0);
 438             dy00             = _mm_sub_pd(iy0,jy0);
 439             dz00             = _mm_sub_pd(iz0,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 443
 444             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 445
 446             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 447
 448             /* Load parameters for j particles */
 449             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             r00              = _mm_mul_pd(rsq00,rinv00);
 456
 457             /* Compute parameters for interactions between i and j atoms */
 458             qq00             = _mm_mul_pd(iq0,jq0);
 459
 460             /* EWALD ELECTROSTATICS */
 461
 462             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 463             ewrt             = _mm_mul_pd(r00,ewtabscale);
 464             ewitab           = _mm_cvttpd_epi32(ewrt);
 465 #ifdef __XOP__
 466             eweps            = _mm_frcz_pd(ewrt);
 467 #else
 468             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 469 #endif
 470             twoeweps         = _mm_add_pd(eweps,eweps);
 471             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 472                                          &ewtabF,&ewtabFn);
 473             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 474             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 475
 476             fscal            = felec;
 477
 478             /* Update vectorial force */
 479             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 480             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 481             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 482
 483             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 484                                                    _mm_mul_pd(dx00,fscal),
 485                                                    _mm_mul_pd(dy00,fscal),
 486                                                    _mm_mul_pd(dz00,fscal));
 487
 488             /* Inner loop uses 39 flops */
 489         }
 490
 491         if(jidx<j_index_end)
 492         {
 493
 494             jnrA             = jjnr[jidx];
 495             j_coord_offsetA  = DIM*jnrA;
 496
 497             /* load j atom coordinates */
 498             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 499                                               &jx0,&jy0,&jz0);
 500
 501             /* Calculate displacement vector */
 502             dx00             = _mm_sub_pd(ix0,jx0);
 503             dy00             = _mm_sub_pd(iy0,jy0);
 504             dz00             = _mm_sub_pd(iz0,jz0);
 505
 506             /* Calculate squared distance and things based on it */
 507             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 508
 509             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 510
 511             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 512
 513             /* Load parameters for j particles */
 514             jq0              = _mm_load_sd(charge+jnrA+0);
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             r00              = _mm_mul_pd(rsq00,rinv00);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             qq00             = _mm_mul_pd(iq0,jq0);
 524
 525             /* EWALD ELECTROSTATICS */
 526
 527             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 528             ewrt             = _mm_mul_pd(r00,ewtabscale);
 529             ewitab           = _mm_cvttpd_epi32(ewrt);
 530 #ifdef __XOP__
 531             eweps            = _mm_frcz_pd(ewrt);
 532 #else
 533             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 534 #endif
 535             twoeweps         = _mm_add_pd(eweps,eweps);
 536             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 537             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 538             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 539
 540             fscal            = felec;
 541
 542             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 543
 544             /* Update vectorial force */
 545             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 546             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 547             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 548
 549             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 550                                                    _mm_mul_pd(dx00,fscal),
 551                                                    _mm_mul_pd(dy00,fscal),
 552                                                    _mm_mul_pd(dz00,fscal));
 553
 554             /* Inner loop uses 39 flops */
 555         }
 556
 557         /* End of innermost loop */
 558
 559         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 560                                               f+i_coord_offset,fshift+i_shift_offset);
 561
 562         /* Increment number of inner iterations */
 563         inneriter                  += j_index_end - j_index_start;
 564
 565         /* Outer loop uses 7 flops */
 566     }
 567
 568     /* Increment number of outer iterations */
 569     outeriter        += nri;
 570
 571     /* Update outer/inner flops */
 572
 573     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 574 }