src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_ElecEwSh_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_ElecEwSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 114     rcutoff_scalar   = fr->rcoulomb;
 115     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 116     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_pd();
 144         fiy0             = _mm_setzero_pd();
 145         fiz0             = _mm_setzero_pd();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_pd();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_pd(ix0,jx0);
 169             dy00             = _mm_sub_pd(iy0,jy0);
 170             dz00             = _mm_sub_pd(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 174
 175             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 176
 177             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 178
 179             /* Load parameters for j particles */
 180             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 181
 182             /**************************
 183              * CALCULATE INTERACTIONS *
 184              **************************/
 185
 186             if (gmx_mm_any_lt(rsq00,rcutoff2))
 187             {
 188
 189             r00              = _mm_mul_pd(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_pd(iq0,jq0);
 193
 194             /* EWALD ELECTROSTATICS */
 195
 196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 197             ewrt             = _mm_mul_pd(r00,ewtabscale);
 198             ewitab           = _mm_cvttpd_epi32(ewrt);
 199 #ifdef __XOP__
 200             eweps            = _mm_frcz_pd(ewrt);
 201 #else
 202             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 203 #endif
 204             twoeweps         = _mm_add_pd(eweps,eweps);
 205             ewitab           = _mm_slli_epi32(ewitab,2);
 206             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 207             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 208             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 209             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 210             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 211             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 212             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 213             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 214             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 215             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 216
 217             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 218
 219             /* Update potential sum for this i atom from the interaction with this j atom. */
 220             velec            = _mm_and_pd(velec,cutoff_mask);
 221             velecsum         = _mm_add_pd(velecsum,velec);
 222
 223             fscal            = felec;
 224
 225             fscal            = _mm_and_pd(fscal,cutoff_mask);
 226
 227             /* Update vectorial force */
 228             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 229             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 230             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 231
 232             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 233                                                    _mm_mul_pd(dx00,fscal),
 234                                                    _mm_mul_pd(dy00,fscal),
 235                                                    _mm_mul_pd(dz00,fscal));
 236
 237             }
 238
 239             /* Inner loop uses 49 flops */
 240         }
 241
 242         if(jidx<j_index_end)
 243         {
 244
 245             jnrA             = jjnr[jidx];
 246             j_coord_offsetA  = DIM*jnrA;
 247
 248             /* load j atom coordinates */
 249             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 250                                               &jx0,&jy0,&jz0);
 251
 252             /* Calculate displacement vector */
 253             dx00             = _mm_sub_pd(ix0,jx0);
 254             dy00             = _mm_sub_pd(iy0,jy0);
 255             dz00             = _mm_sub_pd(iz0,jz0);
 256
 257             /* Calculate squared distance and things based on it */
 258             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 259
 260             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 261
 262             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 263
 264             /* Load parameters for j particles */
 265             jq0              = _mm_load_sd(charge+jnrA+0);
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             if (gmx_mm_any_lt(rsq00,rcutoff2))
 272             {
 273
 274             r00              = _mm_mul_pd(rsq00,rinv00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq00             = _mm_mul_pd(iq0,jq0);
 278
 279             /* EWALD ELECTROSTATICS */
 280
 281             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 282             ewrt             = _mm_mul_pd(r00,ewtabscale);
 283             ewitab           = _mm_cvttpd_epi32(ewrt);
 284 #ifdef __XOP__
 285             eweps            = _mm_frcz_pd(ewrt);
 286 #else
 287             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 288 #endif
 289             twoeweps         = _mm_add_pd(eweps,eweps);
 290             ewitab           = _mm_slli_epi32(ewitab,2);
 291             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 292             ewtabD           = _mm_setzero_pd();
 293             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 294             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 295             ewtabFn          = _mm_setzero_pd();
 296             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 297             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 298             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 299             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 300             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 301
 302             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_and_pd(velec,cutoff_mask);
 306             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 307             velecsum         = _mm_add_pd(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             fscal            = _mm_and_pd(fscal,cutoff_mask);
 312
 313             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 314
 315             /* Update vectorial force */
 316             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 317             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 318             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 319
 320             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 321                                                    _mm_mul_pd(dx00,fscal),
 322                                                    _mm_mul_pd(dy00,fscal),
 323                                                    _mm_mul_pd(dz00,fscal));
 324
 325             }
 326
 327             /* Inner loop uses 49 flops */
 328         }
 329
 330         /* End of innermost loop */
 331
 332         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 333                                               f+i_coord_offset,fshift+i_shift_offset);
 334
 335         ggid                        = gid[iidx];
 336         /* Update potential energies */
 337         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 338
 339         /* Increment number of inner iterations */
 340         inneriter                  += j_index_end - j_index_start;
 341
 342         /* Outer loop uses 8 flops */
 343     }
 344
 345     /* Increment number of outer iterations */
 346     outeriter        += nri;
 347
 348     /* Update outer/inner flops */
 349
 350     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*49);
 351 }
 352 /*
 353  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_double
 354  * Electrostatics interaction: Ewald
 355  * VdW interaction:            None
 356  * Geometry:                   Particle-Particle
 357  * Calculate force/pot:        Force
 358  */
 359 void
 360 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_double
 361                     (t_nblist                    * gmx_restrict       nlist,
 362                      rvec                        * gmx_restrict          xx,
 363                      rvec                        * gmx_restrict          ff,
 364                      t_forcerec                  * gmx_restrict          fr,
 365                      t_mdatoms                   * gmx_restrict     mdatoms,
 366                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 367                      t_nrnb                      * gmx_restrict        nrnb)
 368 {
 369     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 370      * just 0 for non-waters.
 371      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 372      * jnr indices corresponding to data put in the four positions in the SIMD register.
 373      */
 374     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 375     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 376     int              jnrA,jnrB;
 377     int              j_coord_offsetA,j_coord_offsetB;
 378     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 379     real             rcutoff_scalar;
 380     real             *shiftvec,*fshift,*x,*f;
 381     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 382     int              vdwioffset0;
 383     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 384     int              vdwjidx0A,vdwjidx0B;
 385     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 386     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 387     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 388     real             *charge;
 389     __m128i          ewitab;
 390     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 391     real             *ewtab;
 392     __m128d          dummy_mask,cutoff_mask;
 393     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 394     __m128d          one     = _mm_set1_pd(1.0);
 395     __m128d          two     = _mm_set1_pd(2.0);
 396     x                = xx[0];
 397     f                = ff[0];
 398
 399     nri              = nlist->nri;
 400     iinr             = nlist->iinr;
 401     jindex           = nlist->jindex;
 402     jjnr             = nlist->jjnr;
 403     shiftidx         = nlist->shift;
 404     gid              = nlist->gid;
 405     shiftvec         = fr->shift_vec[0];
 406     fshift           = fr->fshift[0];
 407     facel            = _mm_set1_pd(fr->epsfac);
 408     charge           = mdatoms->chargeA;
 409
 410     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 411     ewtab            = fr->ic->tabq_coul_F;
 412     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 413     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 414
 415     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 416     rcutoff_scalar   = fr->rcoulomb;
 417     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 418     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 419
 420     /* Avoid stupid compiler warnings */
 421     jnrA = jnrB = 0;
 422     j_coord_offsetA = 0;
 423     j_coord_offsetB = 0;
 424
 425     outeriter        = 0;
 426     inneriter        = 0;
 427
 428     /* Start outer loop over neighborlists */
 429     for(iidx=0; iidx<nri; iidx++)
 430     {
 431         /* Load shift vector for this list */
 432         i_shift_offset   = DIM*shiftidx[iidx];
 433
 434         /* Load limits for loop over neighbors */
 435         j_index_start    = jindex[iidx];
 436         j_index_end      = jindex[iidx+1];
 437
 438         /* Get outer coordinate index */
 439         inr              = iinr[iidx];
 440         i_coord_offset   = DIM*inr;
 441
 442         /* Load i particle coords and add shift vector */
 443         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 444
 445         fix0             = _mm_setzero_pd();
 446         fiy0             = _mm_setzero_pd();
 447         fiz0             = _mm_setzero_pd();
 448
 449         /* Load parameters for i particles */
 450         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 451
 452         /* Start inner kernel loop */
 453         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 454         {
 455
 456             /* Get j neighbor index, and coordinate index */
 457             jnrA             = jjnr[jidx];
 458             jnrB             = jjnr[jidx+1];
 459             j_coord_offsetA  = DIM*jnrA;
 460             j_coord_offsetB  = DIM*jnrB;
 461
 462             /* load j atom coordinates */
 463             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 464                                               &jx0,&jy0,&jz0);
 465
 466             /* Calculate displacement vector */
 467             dx00             = _mm_sub_pd(ix0,jx0);
 468             dy00             = _mm_sub_pd(iy0,jy0);
 469             dz00             = _mm_sub_pd(iz0,jz0);
 470
 471             /* Calculate squared distance and things based on it */
 472             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 473
 474             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 475
 476             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 477
 478             /* Load parameters for j particles */
 479             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             if (gmx_mm_any_lt(rsq00,rcutoff2))
 486             {
 487
 488             r00              = _mm_mul_pd(rsq00,rinv00);
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             qq00             = _mm_mul_pd(iq0,jq0);
 492
 493             /* EWALD ELECTROSTATICS */
 494
 495             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 496             ewrt             = _mm_mul_pd(r00,ewtabscale);
 497             ewitab           = _mm_cvttpd_epi32(ewrt);
 498 #ifdef __XOP__
 499             eweps            = _mm_frcz_pd(ewrt);
 500 #else
 501             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 502 #endif
 503             twoeweps         = _mm_add_pd(eweps,eweps);
 504             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 505                                          &ewtabF,&ewtabFn);
 506             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 507             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 508
 509             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 510
 511             fscal            = felec;
 512
 513             fscal            = _mm_and_pd(fscal,cutoff_mask);
 514
 515             /* Update vectorial force */
 516             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 517             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 518             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 519
 520             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 521                                                    _mm_mul_pd(dx00,fscal),
 522                                                    _mm_mul_pd(dy00,fscal),
 523                                                    _mm_mul_pd(dz00,fscal));
 524
 525             }
 526
 527             /* Inner loop uses 42 flops */
 528         }
 529
 530         if(jidx<j_index_end)
 531         {
 532
 533             jnrA             = jjnr[jidx];
 534             j_coord_offsetA  = DIM*jnrA;
 535
 536             /* load j atom coordinates */
 537             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 538                                               &jx0,&jy0,&jz0);
 539
 540             /* Calculate displacement vector */
 541             dx00             = _mm_sub_pd(ix0,jx0);
 542             dy00             = _mm_sub_pd(iy0,jy0);
 543             dz00             = _mm_sub_pd(iz0,jz0);
 544
 545             /* Calculate squared distance and things based on it */
 546             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 547
 548             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 549
 550             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 551
 552             /* Load parameters for j particles */
 553             jq0              = _mm_load_sd(charge+jnrA+0);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm_any_lt(rsq00,rcutoff2))
 560             {
 561
 562             r00              = _mm_mul_pd(rsq00,rinv00);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq00             = _mm_mul_pd(iq0,jq0);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 570             ewrt             = _mm_mul_pd(r00,ewtabscale);
 571             ewitab           = _mm_cvttpd_epi32(ewrt);
 572 #ifdef __XOP__
 573             eweps            = _mm_frcz_pd(ewrt);
 574 #else
 575             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 576 #endif
 577             twoeweps         = _mm_add_pd(eweps,eweps);
 578             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 579             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 580             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 581
 582             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 583
 584             fscal            = felec;
 585
 586             fscal            = _mm_and_pd(fscal,cutoff_mask);
 587
 588             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 589
 590             /* Update vectorial force */
 591             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 592             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 593             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 594
 595             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 596                                                    _mm_mul_pd(dx00,fscal),
 597                                                    _mm_mul_pd(dy00,fscal),
 598                                                    _mm_mul_pd(dz00,fscal));
 599
 600             }
 601
 602             /* Inner loop uses 42 flops */
 603         }
 604
 605         /* End of innermost loop */
 606
 607         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 608                                               f+i_coord_offset,fshift+i_shift_offset);
 609
 610         /* Increment number of inner iterations */
 611         inneriter                  += j_index_end - j_index_start;
 612
 613         /* Outer loop uses 7 flops */
 614     }
 615
 616     /* Increment number of outer iterations */
 617     outeriter        += nri;
 618
 619     /* Update outer/inner flops */
 620
 621     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*42);
 622 }