src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_sse4_1_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 sse4_1_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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse4_1_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,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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 116     rcutoff_scalar   = fr->rcoulomb;
 117     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 118     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_pd();
 146         fiy0             = _mm_setzero_pd();
 147         fiz0             = _mm_setzero_pd();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_pd();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164
 165             /* load j atom coordinates */
 166             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx00             = _mm_sub_pd(ix0,jx0);
 171             dy00             = _mm_sub_pd(iy0,jy0);
 172             dz00             = _mm_sub_pd(iz0,jz0);
 173
 174             /* Calculate squared distance and things based on it */
 175             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 176
 177             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 178
 179             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 183
 184             /**************************
 185              * CALCULATE INTERACTIONS *
 186              **************************/
 187
 188             if (gmx_mm_any_lt(rsq00,rcutoff2))
 189             {
 190
 191             r00              = _mm_mul_pd(rsq00,rinv00);
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_pd(iq0,jq0);
 195
 196             /* EWALD ELECTROSTATICS */
 197
 198             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 199             ewrt             = _mm_mul_pd(r00,ewtabscale);
 200             ewitab           = _mm_cvttpd_epi32(ewrt);
 201             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 202             ewitab           = _mm_slli_epi32(ewitab,2);
 203             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 204             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 205             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 206             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 207             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 208             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 209             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 210             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 211             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 212             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 213
 214             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 215
 216             /* Update potential sum for this i atom from the interaction with this j atom. */
 217             velec            = _mm_and_pd(velec,cutoff_mask);
 218             velecsum         = _mm_add_pd(velecsum,velec);
 219
 220             fscal            = felec;
 221
 222             fscal            = _mm_and_pd(fscal,cutoff_mask);
 223
 224             /* Calculate temporary vectorial force */
 225             tx               = _mm_mul_pd(fscal,dx00);
 226             ty               = _mm_mul_pd(fscal,dy00);
 227             tz               = _mm_mul_pd(fscal,dz00);
 228
 229             /* Update vectorial force */
 230             fix0             = _mm_add_pd(fix0,tx);
 231             fiy0             = _mm_add_pd(fiy0,ty);
 232             fiz0             = _mm_add_pd(fiz0,tz);
 233
 234             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 235
 236             }
 237
 238             /* Inner loop uses 46 flops */
 239         }
 240
 241         if(jidx<j_index_end)
 242         {
 243
 244             jnrA             = jjnr[jidx];
 245             j_coord_offsetA  = DIM*jnrA;
 246
 247             /* load j atom coordinates */
 248             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 249                                               &jx0,&jy0,&jz0);
 250
 251             /* Calculate displacement vector */
 252             dx00             = _mm_sub_pd(ix0,jx0);
 253             dy00             = _mm_sub_pd(iy0,jy0);
 254             dz00             = _mm_sub_pd(iz0,jz0);
 255
 256             /* Calculate squared distance and things based on it */
 257             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 258
 259             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 260
 261             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 262
 263             /* Load parameters for j particles */
 264             jq0              = _mm_load_sd(charge+jnrA+0);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             if (gmx_mm_any_lt(rsq00,rcutoff2))
 271             {
 272
 273             r00              = _mm_mul_pd(rsq00,rinv00);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             qq00             = _mm_mul_pd(iq0,jq0);
 277
 278             /* EWALD ELECTROSTATICS */
 279
 280             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 281             ewrt             = _mm_mul_pd(r00,ewtabscale);
 282             ewitab           = _mm_cvttpd_epi32(ewrt);
 283             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 284             ewitab           = _mm_slli_epi32(ewitab,2);
 285             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 286             ewtabD           = _mm_setzero_pd();
 287             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 288             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 289             ewtabFn          = _mm_setzero_pd();
 290             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 291             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 292             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 293             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 294             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 295
 296             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_and_pd(velec,cutoff_mask);
 300             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 301             velecsum         = _mm_add_pd(velecsum,velec);
 302
 303             fscal            = felec;
 304
 305             fscal            = _mm_and_pd(fscal,cutoff_mask);
 306
 307             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm_mul_pd(fscal,dx00);
 311             ty               = _mm_mul_pd(fscal,dy00);
 312             tz               = _mm_mul_pd(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_add_pd(fix0,tx);
 316             fiy0             = _mm_add_pd(fiy0,ty);
 317             fiz0             = _mm_add_pd(fiz0,tz);
 318
 319             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 320
 321             }
 322
 323             /* Inner loop uses 46 flops */
 324         }
 325
 326         /* End of innermost loop */
 327
 328         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 329                                               f+i_coord_offset,fshift+i_shift_offset);
 330
 331         ggid                        = gid[iidx];
 332         /* Update potential energies */
 333         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 334
 335         /* Increment number of inner iterations */
 336         inneriter                  += j_index_end - j_index_start;
 337
 338         /* Outer loop uses 8 flops */
 339     }
 340
 341     /* Increment number of outer iterations */
 342     outeriter        += nri;
 343
 344     /* Update outer/inner flops */
 345
 346     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*46);
 347 }
 348 /*
 349  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_double
 350  * Electrostatics interaction: Ewald
 351  * VdW interaction:            None
 352  * Geometry:                   Particle-Particle
 353  * Calculate force/pot:        Force
 354  */
 355 void
 356 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_double
 357                     (t_nblist                    * gmx_restrict       nlist,
 358                      rvec                        * gmx_restrict          xx,
 359                      rvec                        * gmx_restrict          ff,
 360                      t_forcerec                  * gmx_restrict          fr,
 361                      t_mdatoms                   * gmx_restrict     mdatoms,
 362                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 363                      t_nrnb                      * gmx_restrict        nrnb)
 364 {
 365     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 366      * just 0 for non-waters.
 367      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 368      * jnr indices corresponding to data put in the four positions in the SIMD register.
 369      */
 370     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 371     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 372     int              jnrA,jnrB;
 373     int              j_coord_offsetA,j_coord_offsetB;
 374     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 375     real             rcutoff_scalar;
 376     real             *shiftvec,*fshift,*x,*f;
 377     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 378     int              vdwioffset0;
 379     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 380     int              vdwjidx0A,vdwjidx0B;
 381     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 382     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 383     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 384     real             *charge;
 385     __m128i          ewitab;
 386     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 387     real             *ewtab;
 388     __m128d          dummy_mask,cutoff_mask;
 389     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 390     __m128d          one     = _mm_set1_pd(1.0);
 391     __m128d          two     = _mm_set1_pd(2.0);
 392     x                = xx[0];
 393     f                = ff[0];
 394
 395     nri              = nlist->nri;
 396     iinr             = nlist->iinr;
 397     jindex           = nlist->jindex;
 398     jjnr             = nlist->jjnr;
 399     shiftidx         = nlist->shift;
 400     gid              = nlist->gid;
 401     shiftvec         = fr->shift_vec[0];
 402     fshift           = fr->fshift[0];
 403     facel            = _mm_set1_pd(fr->epsfac);
 404     charge           = mdatoms->chargeA;
 405
 406     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 407     ewtab            = fr->ic->tabq_coul_F;
 408     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 409     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 410
 411     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 412     rcutoff_scalar   = fr->rcoulomb;
 413     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 414     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 415
 416     /* Avoid stupid compiler warnings */
 417     jnrA = jnrB = 0;
 418     j_coord_offsetA = 0;
 419     j_coord_offsetB = 0;
 420
 421     outeriter        = 0;
 422     inneriter        = 0;
 423
 424     /* Start outer loop over neighborlists */
 425     for(iidx=0; iidx<nri; iidx++)
 426     {
 427         /* Load shift vector for this list */
 428         i_shift_offset   = DIM*shiftidx[iidx];
 429
 430         /* Load limits for loop over neighbors */
 431         j_index_start    = jindex[iidx];
 432         j_index_end      = jindex[iidx+1];
 433
 434         /* Get outer coordinate index */
 435         inr              = iinr[iidx];
 436         i_coord_offset   = DIM*inr;
 437
 438         /* Load i particle coords and add shift vector */
 439         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 440
 441         fix0             = _mm_setzero_pd();
 442         fiy0             = _mm_setzero_pd();
 443         fiz0             = _mm_setzero_pd();
 444
 445         /* Load parameters for i particles */
 446         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 447
 448         /* Start inner kernel loop */
 449         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 450         {
 451
 452             /* Get j neighbor index, and coordinate index */
 453             jnrA             = jjnr[jidx];
 454             jnrB             = jjnr[jidx+1];
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457
 458             /* load j atom coordinates */
 459             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 460                                               &jx0,&jy0,&jz0);
 461
 462             /* Calculate displacement vector */
 463             dx00             = _mm_sub_pd(ix0,jx0);
 464             dy00             = _mm_sub_pd(iy0,jy0);
 465             dz00             = _mm_sub_pd(iz0,jz0);
 466
 467             /* Calculate squared distance and things based on it */
 468             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 469
 470             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 471
 472             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 473
 474             /* Load parameters for j particles */
 475             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm_any_lt(rsq00,rcutoff2))
 482             {
 483
 484             r00              = _mm_mul_pd(rsq00,rinv00);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq00             = _mm_mul_pd(iq0,jq0);
 488
 489             /* EWALD ELECTROSTATICS */
 490
 491             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 492             ewrt             = _mm_mul_pd(r00,ewtabscale);
 493             ewitab           = _mm_cvttpd_epi32(ewrt);
 494             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 495             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 496                                          &ewtabF,&ewtabFn);
 497             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 498             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 499
 500             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 501
 502             fscal            = felec;
 503
 504             fscal            = _mm_and_pd(fscal,cutoff_mask);
 505
 506             /* Calculate temporary vectorial force */
 507             tx               = _mm_mul_pd(fscal,dx00);
 508             ty               = _mm_mul_pd(fscal,dy00);
 509             tz               = _mm_mul_pd(fscal,dz00);
 510
 511             /* Update vectorial force */
 512             fix0             = _mm_add_pd(fix0,tx);
 513             fiy0             = _mm_add_pd(fiy0,ty);
 514             fiz0             = _mm_add_pd(fiz0,tz);
 515
 516             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 517
 518             }
 519
 520             /* Inner loop uses 39 flops */
 521         }
 522
 523         if(jidx<j_index_end)
 524         {
 525
 526             jnrA             = jjnr[jidx];
 527             j_coord_offsetA  = DIM*jnrA;
 528
 529             /* load j atom coordinates */
 530             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 531                                               &jx0,&jy0,&jz0);
 532
 533             /* Calculate displacement vector */
 534             dx00             = _mm_sub_pd(ix0,jx0);
 535             dy00             = _mm_sub_pd(iy0,jy0);
 536             dz00             = _mm_sub_pd(iz0,jz0);
 537
 538             /* Calculate squared distance and things based on it */
 539             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 540
 541             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 542
 543             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 544
 545             /* Load parameters for j particles */
 546             jq0              = _mm_load_sd(charge+jnrA+0);
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             if (gmx_mm_any_lt(rsq00,rcutoff2))
 553             {
 554
 555             r00              = _mm_mul_pd(rsq00,rinv00);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq00             = _mm_mul_pd(iq0,jq0);
 559
 560             /* EWALD ELECTROSTATICS */
 561
 562             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 563             ewrt             = _mm_mul_pd(r00,ewtabscale);
 564             ewitab           = _mm_cvttpd_epi32(ewrt);
 565             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 566             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 567             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 568             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 569
 570             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm_and_pd(fscal,cutoff_mask);
 575
 576             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 577
 578             /* Calculate temporary vectorial force */
 579             tx               = _mm_mul_pd(fscal,dx00);
 580             ty               = _mm_mul_pd(fscal,dy00);
 581             tz               = _mm_mul_pd(fscal,dz00);
 582
 583             /* Update vectorial force */
 584             fix0             = _mm_add_pd(fix0,tx);
 585             fiy0             = _mm_add_pd(fiy0,ty);
 586             fiz0             = _mm_add_pd(fiz0,tz);
 587
 588             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 589
 590             }
 591
 592             /* Inner loop uses 39 flops */
 593         }
 594
 595         /* End of innermost loop */
 596
 597         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 598                                               f+i_coord_offset,fshift+i_shift_offset);
 599
 600         /* Increment number of inner iterations */
 601         inneriter                  += j_index_end - j_index_start;
 602
 603         /* Outer loop uses 7 flops */
 604     }
 605
 606     /* Increment number of outer iterations */
 607     outeriter        += nri;
 608
 609     /* Update outer/inner flops */
 610
 611     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 612 }