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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_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_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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,twoeweps,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           = avx128fma_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 #ifdef __XOP__
 191             eweps            = _mm_frcz_pd(ewrt);
 192 #else
 193             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 194 #endif
 195             twoeweps         = _mm_add_pd(eweps,eweps);
 196             ewitab           = _mm_slli_epi32(ewitab,2);
 197             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 198             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 199             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 200             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 201             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 202             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 203             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 204             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 205             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 206             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 207
 208             /* Update potential sum for this i atom from the interaction with this j atom. */
 209             velecsum         = _mm_add_pd(velecsum,velec);
 210
 211             fscal            = felec;
 212
 213             /* Update vectorial force */
 214             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 215             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 216             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 217
 218             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 219                                                    _mm_mul_pd(dx00,fscal),
 220                                                    _mm_mul_pd(dy00,fscal),
 221                                                    _mm_mul_pd(dz00,fscal));
 222
 223             /* Inner loop uses 44 flops */
 224         }
 225
 226         if(jidx<j_index_end)
 227         {
 228
 229             jnrA             = jjnr[jidx];
 230             j_coord_offsetA  = DIM*jnrA;
 231
 232             /* load j atom coordinates */
 233             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 234                                               &jx0,&jy0,&jz0);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm_sub_pd(ix0,jx0);
 238             dy00             = _mm_sub_pd(iy0,jy0);
 239             dz00             = _mm_sub_pd(iz0,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 243
 244             rinv00           = avx128fma_invsqrt_d(rsq00);
 245
 246             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 247
 248             /* Load parameters for j particles */
 249             jq0              = _mm_load_sd(charge+jnrA+0);
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             r00              = _mm_mul_pd(rsq00,rinv00);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq00             = _mm_mul_pd(iq0,jq0);
 259
 260             /* EWALD ELECTROSTATICS */
 261
 262             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 263             ewrt             = _mm_mul_pd(r00,ewtabscale);
 264             ewitab           = _mm_cvttpd_epi32(ewrt);
 265 #ifdef __XOP__
 266             eweps            = _mm_frcz_pd(ewrt);
 267 #else
 268             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 269 #endif
 270             twoeweps         = _mm_add_pd(eweps,eweps);
 271             ewitab           = _mm_slli_epi32(ewitab,2);
 272             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 273             ewtabD           = _mm_setzero_pd();
 274             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 275             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 276             ewtabFn          = _mm_setzero_pd();
 277             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 278             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 279             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 280             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 281             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 285             velecsum         = _mm_add_pd(velecsum,velec);
 286
 287             fscal            = felec;
 288
 289             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 293             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 294             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 295
 296             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 297                                                    _mm_mul_pd(dx00,fscal),
 298                                                    _mm_mul_pd(dy00,fscal),
 299                                                    _mm_mul_pd(dz00,fscal));
 300
 301             /* Inner loop uses 44 flops */
 302         }
 303
 304         /* End of innermost loop */
 305
 306         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 307                                               f+i_coord_offset,fshift+i_shift_offset);
 308
 309         ggid                        = gid[iidx];
 310         /* Update potential energies */
 311         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 312
 313         /* Increment number of inner iterations */
 314         inneriter                  += j_index_end - j_index_start;
 315
 316         /* Outer loop uses 8 flops */
 317     }
 318
 319     /* Increment number of outer iterations */
 320     outeriter        += nri;
 321
 322     /* Update outer/inner flops */
 323
 324     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
 325 }
 326 /*
 327  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_double
 328  * Electrostatics interaction: Ewald
 329  * VdW interaction:            None
 330  * Geometry:                   Particle-Particle
 331  * Calculate force/pot:        Force
 332  */
 333 void
 334 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_double
 335                     (t_nblist                    * gmx_restrict       nlist,
 336                      rvec                        * gmx_restrict          xx,
 337                      rvec                        * gmx_restrict          ff,
 338                      struct t_forcerec           * gmx_restrict          fr,
 339                      t_mdatoms                   * gmx_restrict     mdatoms,
 340                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 341                      t_nrnb                      * gmx_restrict        nrnb)
 342 {
 343     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 344      * just 0 for non-waters.
 345      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 346      * jnr indices corresponding to data put in the four positions in the SIMD register.
 347      */
 348     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 349     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 350     int              jnrA,jnrB;
 351     int              j_coord_offsetA,j_coord_offsetB;
 352     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 353     real             rcutoff_scalar;
 354     real             *shiftvec,*fshift,*x,*f;
 355     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 356     int              vdwioffset0;
 357     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 358     int              vdwjidx0A,vdwjidx0B;
 359     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 360     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 361     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 362     real             *charge;
 363     __m128i          ewitab;
 364     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 365     real             *ewtab;
 366     __m128d          dummy_mask,cutoff_mask;
 367     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 368     __m128d          one     = _mm_set1_pd(1.0);
 369     __m128d          two     = _mm_set1_pd(2.0);
 370     x                = xx[0];
 371     f                = ff[0];
 372
 373     nri              = nlist->nri;
 374     iinr             = nlist->iinr;
 375     jindex           = nlist->jindex;
 376     jjnr             = nlist->jjnr;
 377     shiftidx         = nlist->shift;
 378     gid              = nlist->gid;
 379     shiftvec         = fr->shift_vec[0];
 380     fshift           = fr->fshift[0];
 381     facel            = _mm_set1_pd(fr->ic->epsfac);
 382     charge           = mdatoms->chargeA;
 383
 384     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 385     ewtab            = fr->ic->tabq_coul_F;
 386     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 387     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 388
 389     /* Avoid stupid compiler warnings */
 390     jnrA = jnrB = 0;
 391     j_coord_offsetA = 0;
 392     j_coord_offsetB = 0;
 393
 394     outeriter        = 0;
 395     inneriter        = 0;
 396
 397     /* Start outer loop over neighborlists */
 398     for(iidx=0; iidx<nri; iidx++)
 399     {
 400         /* Load shift vector for this list */
 401         i_shift_offset   = DIM*shiftidx[iidx];
 402
 403         /* Load limits for loop over neighbors */
 404         j_index_start    = jindex[iidx];
 405         j_index_end      = jindex[iidx+1];
 406
 407         /* Get outer coordinate index */
 408         inr              = iinr[iidx];
 409         i_coord_offset   = DIM*inr;
 410
 411         /* Load i particle coords and add shift vector */
 412         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 413
 414         fix0             = _mm_setzero_pd();
 415         fiy0             = _mm_setzero_pd();
 416         fiz0             = _mm_setzero_pd();
 417
 418         /* Load parameters for i particles */
 419         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 420
 421         /* Start inner kernel loop */
 422         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrA             = jjnr[jidx];
 427             jnrB             = jjnr[jidx+1];
 428             j_coord_offsetA  = DIM*jnrA;
 429             j_coord_offsetB  = DIM*jnrB;
 430
 431             /* load j atom coordinates */
 432             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 433                                               &jx0,&jy0,&jz0);
 434
 435             /* Calculate displacement vector */
 436             dx00             = _mm_sub_pd(ix0,jx0);
 437             dy00             = _mm_sub_pd(iy0,jy0);
 438             dz00             = _mm_sub_pd(iz0,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 442
 443             rinv00           = avx128fma_invsqrt_d(rsq00);
 444
 445             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 446
 447             /* Load parameters for j particles */
 448             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             r00              = _mm_mul_pd(rsq00,rinv00);
 455
 456             /* Compute parameters for interactions between i and j atoms */
 457             qq00             = _mm_mul_pd(iq0,jq0);
 458
 459             /* EWALD ELECTROSTATICS */
 460
 461             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 462             ewrt             = _mm_mul_pd(r00,ewtabscale);
 463             ewitab           = _mm_cvttpd_epi32(ewrt);
 464 #ifdef __XOP__
 465             eweps            = _mm_frcz_pd(ewrt);
 466 #else
 467             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 468 #endif
 469             twoeweps         = _mm_add_pd(eweps,eweps);
 470             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 471                                          &ewtabF,&ewtabFn);
 472             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 473             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 474
 475             fscal            = felec;
 476
 477             /* Update vectorial force */
 478             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 479             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 480             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 481
 482             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 483                                                    _mm_mul_pd(dx00,fscal),
 484                                                    _mm_mul_pd(dy00,fscal),
 485                                                    _mm_mul_pd(dz00,fscal));
 486
 487             /* Inner loop uses 39 flops */
 488         }
 489
 490         if(jidx<j_index_end)
 491         {
 492
 493             jnrA             = jjnr[jidx];
 494             j_coord_offsetA  = DIM*jnrA;
 495
 496             /* load j atom coordinates */
 497             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 498                                               &jx0,&jy0,&jz0);
 499
 500             /* Calculate displacement vector */
 501             dx00             = _mm_sub_pd(ix0,jx0);
 502             dy00             = _mm_sub_pd(iy0,jy0);
 503             dz00             = _mm_sub_pd(iz0,jz0);
 504
 505             /* Calculate squared distance and things based on it */
 506             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 507
 508             rinv00           = avx128fma_invsqrt_d(rsq00);
 509
 510             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 511
 512             /* Load parameters for j particles */
 513             jq0              = _mm_load_sd(charge+jnrA+0);
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r00              = _mm_mul_pd(rsq00,rinv00);
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq00             = _mm_mul_pd(iq0,jq0);
 523
 524             /* EWALD ELECTROSTATICS */
 525
 526             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 527             ewrt             = _mm_mul_pd(r00,ewtabscale);
 528             ewitab           = _mm_cvttpd_epi32(ewrt);
 529 #ifdef __XOP__
 530             eweps            = _mm_frcz_pd(ewrt);
 531 #else
 532             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 533 #endif
 534             twoeweps         = _mm_add_pd(eweps,eweps);
 535             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 536             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 537             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 538
 539             fscal            = felec;
 540
 541             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 542
 543             /* Update vectorial force */
 544             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 545             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 546             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 547
 548             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 549                                                    _mm_mul_pd(dx00,fscal),
 550                                                    _mm_mul_pd(dy00,fscal),
 551                                                    _mm_mul_pd(dz00,fscal));
 552
 553             /* Inner loop uses 39 flops */
 554         }
 555
 556         /* End of innermost loop */
 557
 558         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 559                                               f+i_coord_offset,fshift+i_shift_offset);
 560
 561         /* Increment number of inner iterations */
 562         inneriter                  += j_index_end - j_index_start;
 563
 564         /* Outer loop uses 7 flops */
 565     }
 566
 567     /* Increment number of outer iterations */
 568     outeriter        += nri;
 569
 570     /* Update outer/inner flops */
 571
 572     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 573 }