src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse2_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_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse2_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,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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 113     rcutoff_scalar   = fr->ic->rcoulomb;
 114     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 115     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm_setzero_pd();
 143         fiy0             = _mm_setzero_pd();
 144         fiz0             = _mm_setzero_pd();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm_setzero_pd();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161
 162             /* load j atom coordinates */
 163             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 164                                               &jx0,&jy0,&jz0);
 165
 166             /* Calculate displacement vector */
 167             dx00             = _mm_sub_pd(ix0,jx0);
 168             dy00             = _mm_sub_pd(iy0,jy0);
 169             dz00             = _mm_sub_pd(iz0,jz0);
 170
 171             /* Calculate squared distance and things based on it */
 172             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 173
 174             rinv00           = sse2_invsqrt_d(rsq00);
 175
 176             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 177
 178             /* Load parameters for j particles */
 179             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 180
 181             /**************************
 182              * CALCULATE INTERACTIONS *
 183              **************************/
 184
 185             if (gmx_mm_any_lt(rsq00,rcutoff2))
 186             {
 187
 188             r00              = _mm_mul_pd(rsq00,rinv00);
 189
 190             /* Compute parameters for interactions between i and j atoms */
 191             qq00             = _mm_mul_pd(iq0,jq0);
 192
 193             /* EWALD ELECTROSTATICS */
 194
 195             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 196             ewrt             = _mm_mul_pd(r00,ewtabscale);
 197             ewitab           = _mm_cvttpd_epi32(ewrt);
 198             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 199             ewitab           = _mm_slli_epi32(ewitab,2);
 200             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 201             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 202             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 203             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 204             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 205             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 206             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 207             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 208             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 209             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 210
 211             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             velec            = _mm_and_pd(velec,cutoff_mask);
 215             velecsum         = _mm_add_pd(velecsum,velec);
 216
 217             fscal            = felec;
 218
 219             fscal            = _mm_and_pd(fscal,cutoff_mask);
 220
 221             /* Calculate temporary vectorial force */
 222             tx               = _mm_mul_pd(fscal,dx00);
 223             ty               = _mm_mul_pd(fscal,dy00);
 224             tz               = _mm_mul_pd(fscal,dz00);
 225
 226             /* Update vectorial force */
 227             fix0             = _mm_add_pd(fix0,tx);
 228             fiy0             = _mm_add_pd(fiy0,ty);
 229             fiz0             = _mm_add_pd(fiz0,tz);
 230
 231             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 232
 233             }
 234
 235             /* Inner loop uses 46 flops */
 236         }
 237
 238         if(jidx<j_index_end)
 239         {
 240
 241             jnrA             = jjnr[jidx];
 242             j_coord_offsetA  = DIM*jnrA;
 243
 244             /* load j atom coordinates */
 245             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 246                                               &jx0,&jy0,&jz0);
 247
 248             /* Calculate displacement vector */
 249             dx00             = _mm_sub_pd(ix0,jx0);
 250             dy00             = _mm_sub_pd(iy0,jy0);
 251             dz00             = _mm_sub_pd(iz0,jz0);
 252
 253             /* Calculate squared distance and things based on it */
 254             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 255
 256             rinv00           = sse2_invsqrt_d(rsq00);
 257
 258             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 259
 260             /* Load parameters for j particles */
 261             jq0              = _mm_load_sd(charge+jnrA+0);
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             if (gmx_mm_any_lt(rsq00,rcutoff2))
 268             {
 269
 270             r00              = _mm_mul_pd(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm_mul_pd(iq0,jq0);
 274
 275             /* EWALD ELECTROSTATICS */
 276
 277             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 278             ewrt             = _mm_mul_pd(r00,ewtabscale);
 279             ewitab           = _mm_cvttpd_epi32(ewrt);
 280             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 281             ewitab           = _mm_slli_epi32(ewitab,2);
 282             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 283             ewtabD           = _mm_setzero_pd();
 284             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 285             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 286             ewtabFn          = _mm_setzero_pd();
 287             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 288             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 289             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 290             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 291             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 292
 293             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velec            = _mm_and_pd(velec,cutoff_mask);
 297             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 298             velecsum         = _mm_add_pd(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             fscal            = _mm_and_pd(fscal,cutoff_mask);
 303
 304             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_pd(fscal,dx00);
 308             ty               = _mm_mul_pd(fscal,dy00);
 309             tz               = _mm_mul_pd(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_add_pd(fix0,tx);
 313             fiy0             = _mm_add_pd(fiy0,ty);
 314             fiz0             = _mm_add_pd(fiz0,tz);
 315
 316             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 317
 318             }
 319
 320             /* Inner loop uses 46 flops */
 321         }
 322
 323         /* End of innermost loop */
 324
 325         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 326                                               f+i_coord_offset,fshift+i_shift_offset);
 327
 328         ggid                        = gid[iidx];
 329         /* Update potential energies */
 330         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 331
 332         /* Increment number of inner iterations */
 333         inneriter                  += j_index_end - j_index_start;
 334
 335         /* Outer loop uses 8 flops */
 336     }
 337
 338     /* Increment number of outer iterations */
 339     outeriter        += nri;
 340
 341     /* Update outer/inner flops */
 342
 343     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*46);
 344 }
 345 /*
 346  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_double
 347  * Electrostatics interaction: Ewald
 348  * VdW interaction:            None
 349  * Geometry:                   Particle-Particle
 350  * Calculate force/pot:        Force
 351  */
 352 void
 353 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_double
 354                     (t_nblist                    * gmx_restrict       nlist,
 355                      rvec                        * gmx_restrict          xx,
 356                      rvec                        * gmx_restrict          ff,
 357                      struct t_forcerec           * gmx_restrict          fr,
 358                      t_mdatoms                   * gmx_restrict     mdatoms,
 359                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 360                      t_nrnb                      * gmx_restrict        nrnb)
 361 {
 362     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 363      * just 0 for non-waters.
 364      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 365      * jnr indices corresponding to data put in the four positions in the SIMD register.
 366      */
 367     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 368     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 369     int              jnrA,jnrB;
 370     int              j_coord_offsetA,j_coord_offsetB;
 371     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 372     real             rcutoff_scalar;
 373     real             *shiftvec,*fshift,*x,*f;
 374     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 375     int              vdwioffset0;
 376     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 377     int              vdwjidx0A,vdwjidx0B;
 378     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 379     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 380     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 381     real             *charge;
 382     __m128i          ewitab;
 383     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 384     real             *ewtab;
 385     __m128d          dummy_mask,cutoff_mask;
 386     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 387     __m128d          one     = _mm_set1_pd(1.0);
 388     __m128d          two     = _mm_set1_pd(2.0);
 389     x                = xx[0];
 390     f                = ff[0];
 391
 392     nri              = nlist->nri;
 393     iinr             = nlist->iinr;
 394     jindex           = nlist->jindex;
 395     jjnr             = nlist->jjnr;
 396     shiftidx         = nlist->shift;
 397     gid              = nlist->gid;
 398     shiftvec         = fr->shift_vec[0];
 399     fshift           = fr->fshift[0];
 400     facel            = _mm_set1_pd(fr->ic->epsfac);
 401     charge           = mdatoms->chargeA;
 402
 403     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 404     ewtab            = fr->ic->tabq_coul_F;
 405     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 406     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 407
 408     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 409     rcutoff_scalar   = fr->ic->rcoulomb;
 410     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 411     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 412
 413     /* Avoid stupid compiler warnings */
 414     jnrA = jnrB = 0;
 415     j_coord_offsetA = 0;
 416     j_coord_offsetB = 0;
 417
 418     outeriter        = 0;
 419     inneriter        = 0;
 420
 421     /* Start outer loop over neighborlists */
 422     for(iidx=0; iidx<nri; iidx++)
 423     {
 424         /* Load shift vector for this list */
 425         i_shift_offset   = DIM*shiftidx[iidx];
 426
 427         /* Load limits for loop over neighbors */
 428         j_index_start    = jindex[iidx];
 429         j_index_end      = jindex[iidx+1];
 430
 431         /* Get outer coordinate index */
 432         inr              = iinr[iidx];
 433         i_coord_offset   = DIM*inr;
 434
 435         /* Load i particle coords and add shift vector */
 436         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 437
 438         fix0             = _mm_setzero_pd();
 439         fiy0             = _mm_setzero_pd();
 440         fiz0             = _mm_setzero_pd();
 441
 442         /* Load parameters for i particles */
 443         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 444
 445         /* Start inner kernel loop */
 446         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 447         {
 448
 449             /* Get j neighbor index, and coordinate index */
 450             jnrA             = jjnr[jidx];
 451             jnrB             = jjnr[jidx+1];
 452             j_coord_offsetA  = DIM*jnrA;
 453             j_coord_offsetB  = DIM*jnrB;
 454
 455             /* load j atom coordinates */
 456             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 457                                               &jx0,&jy0,&jz0);
 458
 459             /* Calculate displacement vector */
 460             dx00             = _mm_sub_pd(ix0,jx0);
 461             dy00             = _mm_sub_pd(iy0,jy0);
 462             dz00             = _mm_sub_pd(iz0,jz0);
 463
 464             /* Calculate squared distance and things based on it */
 465             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 466
 467             rinv00           = sse2_invsqrt_d(rsq00);
 468
 469             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 470
 471             /* Load parameters for j particles */
 472             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             if (gmx_mm_any_lt(rsq00,rcutoff2))
 479             {
 480
 481             r00              = _mm_mul_pd(rsq00,rinv00);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq00             = _mm_mul_pd(iq0,jq0);
 485
 486             /* EWALD ELECTROSTATICS */
 487
 488             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 489             ewrt             = _mm_mul_pd(r00,ewtabscale);
 490             ewitab           = _mm_cvttpd_epi32(ewrt);
 491             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 492             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 493                                          &ewtabF,&ewtabFn);
 494             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 495             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 496
 497             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 498
 499             fscal            = felec;
 500
 501             fscal            = _mm_and_pd(fscal,cutoff_mask);
 502
 503             /* Calculate temporary vectorial force */
 504             tx               = _mm_mul_pd(fscal,dx00);
 505             ty               = _mm_mul_pd(fscal,dy00);
 506             tz               = _mm_mul_pd(fscal,dz00);
 507
 508             /* Update vectorial force */
 509             fix0             = _mm_add_pd(fix0,tx);
 510             fiy0             = _mm_add_pd(fiy0,ty);
 511             fiz0             = _mm_add_pd(fiz0,tz);
 512
 513             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 514
 515             }
 516
 517             /* Inner loop uses 39 flops */
 518         }
 519
 520         if(jidx<j_index_end)
 521         {
 522
 523             jnrA             = jjnr[jidx];
 524             j_coord_offsetA  = DIM*jnrA;
 525
 526             /* load j atom coordinates */
 527             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 528                                               &jx0,&jy0,&jz0);
 529
 530             /* Calculate displacement vector */
 531             dx00             = _mm_sub_pd(ix0,jx0);
 532             dy00             = _mm_sub_pd(iy0,jy0);
 533             dz00             = _mm_sub_pd(iz0,jz0);
 534
 535             /* Calculate squared distance and things based on it */
 536             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 537
 538             rinv00           = sse2_invsqrt_d(rsq00);
 539
 540             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 541
 542             /* Load parameters for j particles */
 543             jq0              = _mm_load_sd(charge+jnrA+0);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq00,rcutoff2))
 550             {
 551
 552             r00              = _mm_mul_pd(rsq00,rinv00);
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq00             = _mm_mul_pd(iq0,jq0);
 556
 557             /* EWALD ELECTROSTATICS */
 558
 559             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 560             ewrt             = _mm_mul_pd(r00,ewtabscale);
 561             ewitab           = _mm_cvttpd_epi32(ewrt);
 562             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 563             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 564             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 565             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 566
 567             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm_and_pd(fscal,cutoff_mask);
 572
 573             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_pd(fscal,dx00);
 577             ty               = _mm_mul_pd(fscal,dy00);
 578             tz               = _mm_mul_pd(fscal,dz00);
 579
 580             /* Update vectorial force */
 581             fix0             = _mm_add_pd(fix0,tx);
 582             fiy0             = _mm_add_pd(fiy0,ty);
 583             fiz0             = _mm_add_pd(fiz0,tz);
 584
 585             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 586
 587             }
 588
 589             /* Inner loop uses 39 flops */
 590         }
 591
 592         /* End of innermost loop */
 593
 594         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 595                                               f+i_coord_offset,fshift+i_shift_offset);
 596
 597         /* Increment number of inner iterations */
 598         inneriter                  += j_index_end - j_index_start;
 599
 600         /* Outer loop uses 7 flops */
 601     }
 602
 603     /* Increment number of outer iterations */
 604     outeriter        += nri;
 605
 606     /* Update outer/inner flops */
 607
 608     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 609 }