src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwNone_GeomP1P1_avx_256_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_256_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_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_256_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,C,D refer to j loop unrolling done with AVX, e.g. for the four 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,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m256d          dummy_mask,cutoff_mask;
  94     __m128           tmpmask0,tmpmask1;
  95     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  96     __m256d          one     = _mm256_set1_pd(1.0);
  97     __m256d          two     = _mm256_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm256_set1_pd(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111
 112     /* Avoid stupid compiler warnings */
 113     jnrA = jnrB = jnrC = jnrD = 0;
 114     j_coord_offsetA = 0;
 115     j_coord_offsetB = 0;
 116     j_coord_offsetC = 0;
 117     j_coord_offsetD = 0;
 118
 119     outeriter        = 0;
 120     inneriter        = 0;
 121
 122     for(iidx=0;iidx<4*DIM;iidx++)
 123     {
 124         scratch[iidx] = 0.0;
 125     }
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _mm256_setzero_pd();
 145         fiy0             = _mm256_setzero_pd();
 146         fiz0             = _mm256_setzero_pd();
 147
 148         /* Load parameters for i particles */
 149         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 150
 151         /* Reset potential sums */
 152         velecsum         = _mm256_setzero_pd();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             jnrC             = jjnr[jidx+2];
 162             jnrD             = jjnr[jidx+3];
 163             j_coord_offsetA  = DIM*jnrA;
 164             j_coord_offsetB  = DIM*jnrB;
 165             j_coord_offsetC  = DIM*jnrC;
 166             j_coord_offsetD  = DIM*jnrD;
 167
 168             /* load j atom coordinates */
 169             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 171                                                  &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm256_sub_pd(ix0,jx0);
 175             dy00             = _mm256_sub_pd(iy0,jy0);
 176             dz00             = _mm256_sub_pd(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 182
 183             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 187                                                                  charge+jnrC+0,charge+jnrD+0);
 188
 189             /**************************
 190              * CALCULATE INTERACTIONS *
 191              **************************/
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm256_mul_pd(iq0,jq0);
 195
 196             /* COULOMB ELECTROSTATICS */
 197             velec            = _mm256_mul_pd(qq00,rinv00);
 198             felec            = _mm256_mul_pd(velec,rinvsq00);
 199
 200             /* Update potential sum for this i atom from the interaction with this j atom. */
 201             velecsum         = _mm256_add_pd(velecsum,velec);
 202
 203             fscal            = felec;
 204
 205             /* Calculate temporary vectorial force */
 206             tx               = _mm256_mul_pd(fscal,dx00);
 207             ty               = _mm256_mul_pd(fscal,dy00);
 208             tz               = _mm256_mul_pd(fscal,dz00);
 209
 210             /* Update vectorial force */
 211             fix0             = _mm256_add_pd(fix0,tx);
 212             fiy0             = _mm256_add_pd(fiy0,ty);
 213             fiz0             = _mm256_add_pd(fiz0,tz);
 214
 215             fjptrA             = f+j_coord_offsetA;
 216             fjptrB             = f+j_coord_offsetB;
 217             fjptrC             = f+j_coord_offsetC;
 218             fjptrD             = f+j_coord_offsetD;
 219             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 220
 221             /* Inner loop uses 27 flops */
 222         }
 223
 224         if(jidx<j_index_end)
 225         {
 226
 227             /* Get j neighbor index, and coordinate index */
 228             jnrlistA         = jjnr[jidx];
 229             jnrlistB         = jjnr[jidx+1];
 230             jnrlistC         = jjnr[jidx+2];
 231             jnrlistD         = jjnr[jidx+3];
 232             /* Sign of each element will be negative for non-real atoms.
 233              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 234              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 235              */
 236             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 237
 238             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 239             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 240             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 241
 242             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 243             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 244             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 245             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 246             j_coord_offsetA  = DIM*jnrA;
 247             j_coord_offsetB  = DIM*jnrB;
 248             j_coord_offsetC  = DIM*jnrC;
 249             j_coord_offsetD  = DIM*jnrD;
 250
 251             /* load j atom coordinates */
 252             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 253                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 254                                                  &jx0,&jy0,&jz0);
 255
 256             /* Calculate displacement vector */
 257             dx00             = _mm256_sub_pd(ix0,jx0);
 258             dy00             = _mm256_sub_pd(iy0,jy0);
 259             dz00             = _mm256_sub_pd(iz0,jz0);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 263
 264             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 265
 266             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 267
 268             /* Load parameters for j particles */
 269             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 270                                                                  charge+jnrC+0,charge+jnrD+0);
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq00             = _mm256_mul_pd(iq0,jq0);
 278
 279             /* COULOMB ELECTROSTATICS */
 280             velec            = _mm256_mul_pd(qq00,rinv00);
 281             felec            = _mm256_mul_pd(velec,rinvsq00);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velec            = _mm256_andnot_pd(dummy_mask,velec);
 285             velecsum         = _mm256_add_pd(velecsum,velec);
 286
 287             fscal            = felec;
 288
 289             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm256_mul_pd(fscal,dx00);
 293             ty               = _mm256_mul_pd(fscal,dy00);
 294             tz               = _mm256_mul_pd(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm256_add_pd(fix0,tx);
 298             fiy0             = _mm256_add_pd(fiy0,ty);
 299             fiz0             = _mm256_add_pd(fiz0,tz);
 300
 301             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 302             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 303             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 304             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 305             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 306
 307             /* Inner loop uses 27 flops */
 308         }
 309
 310         /* End of innermost loop */
 311
 312         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 313                                                  f+i_coord_offset,fshift+i_shift_offset);
 314
 315         ggid                        = gid[iidx];
 316         /* Update potential energies */
 317         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 318
 319         /* Increment number of inner iterations */
 320         inneriter                  += j_index_end - j_index_start;
 321
 322         /* Outer loop uses 8 flops */
 323     }
 324
 325     /* Increment number of outer iterations */
 326     outeriter        += nri;
 327
 328     /* Update outer/inner flops */
 329
 330     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*27);
 331 }
 332 /*
 333  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_double
 334  * Electrostatics interaction: Coulomb
 335  * VdW interaction:            None
 336  * Geometry:                   Particle-Particle
 337  * Calculate force/pot:        Force
 338  */
 339 void
 340 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_256_double
 341                     (t_nblist                    * gmx_restrict       nlist,
 342                      rvec                        * gmx_restrict          xx,
 343                      rvec                        * gmx_restrict          ff,
 344                      t_forcerec                  * gmx_restrict          fr,
 345                      t_mdatoms                   * gmx_restrict     mdatoms,
 346                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 347                      t_nrnb                      * gmx_restrict        nrnb)
 348 {
 349     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 350      * just 0 for non-waters.
 351      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 352      * jnr indices corresponding to data put in the four positions in the SIMD register.
 353      */
 354     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 355     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 356     int              jnrA,jnrB,jnrC,jnrD;
 357     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 358     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 359     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 360     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 361     real             rcutoff_scalar;
 362     real             *shiftvec,*fshift,*x,*f;
 363     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 364     real             scratch[4*DIM];
 365     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 366     real *           vdwioffsetptr0;
 367     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 368     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 369     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 370     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 371     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 372     real             *charge;
 373     __m256d          dummy_mask,cutoff_mask;
 374     __m128           tmpmask0,tmpmask1;
 375     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 376     __m256d          one     = _mm256_set1_pd(1.0);
 377     __m256d          two     = _mm256_set1_pd(2.0);
 378     x                = xx[0];
 379     f                = ff[0];
 380
 381     nri              = nlist->nri;
 382     iinr             = nlist->iinr;
 383     jindex           = nlist->jindex;
 384     jjnr             = nlist->jjnr;
 385     shiftidx         = nlist->shift;
 386     gid              = nlist->gid;
 387     shiftvec         = fr->shift_vec[0];
 388     fshift           = fr->fshift[0];
 389     facel            = _mm256_set1_pd(fr->epsfac);
 390     charge           = mdatoms->chargeA;
 391
 392     /* Avoid stupid compiler warnings */
 393     jnrA = jnrB = jnrC = jnrD = 0;
 394     j_coord_offsetA = 0;
 395     j_coord_offsetB = 0;
 396     j_coord_offsetC = 0;
 397     j_coord_offsetD = 0;
 398
 399     outeriter        = 0;
 400     inneriter        = 0;
 401
 402     for(iidx=0;iidx<4*DIM;iidx++)
 403     {
 404         scratch[iidx] = 0.0;
 405     }
 406
 407     /* Start outer loop over neighborlists */
 408     for(iidx=0; iidx<nri; iidx++)
 409     {
 410         /* Load shift vector for this list */
 411         i_shift_offset   = DIM*shiftidx[iidx];
 412
 413         /* Load limits for loop over neighbors */
 414         j_index_start    = jindex[iidx];
 415         j_index_end      = jindex[iidx+1];
 416
 417         /* Get outer coordinate index */
 418         inr              = iinr[iidx];
 419         i_coord_offset   = DIM*inr;
 420
 421         /* Load i particle coords and add shift vector */
 422         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 423
 424         fix0             = _mm256_setzero_pd();
 425         fiy0             = _mm256_setzero_pd();
 426         fiz0             = _mm256_setzero_pd();
 427
 428         /* Load parameters for i particles */
 429         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 430
 431         /* Start inner kernel loop */
 432         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 433         {
 434
 435             /* Get j neighbor index, and coordinate index */
 436             jnrA             = jjnr[jidx];
 437             jnrB             = jjnr[jidx+1];
 438             jnrC             = jjnr[jidx+2];
 439             jnrD             = jjnr[jidx+3];
 440             j_coord_offsetA  = DIM*jnrA;
 441             j_coord_offsetB  = DIM*jnrB;
 442             j_coord_offsetC  = DIM*jnrC;
 443             j_coord_offsetD  = DIM*jnrD;
 444
 445             /* load j atom coordinates */
 446             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 447                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 448                                                  &jx0,&jy0,&jz0);
 449
 450             /* Calculate displacement vector */
 451             dx00             = _mm256_sub_pd(ix0,jx0);
 452             dy00             = _mm256_sub_pd(iy0,jy0);
 453             dz00             = _mm256_sub_pd(iz0,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 457
 458             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 459
 460             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 461
 462             /* Load parameters for j particles */
 463             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 464                                                                  charge+jnrC+0,charge+jnrD+0);
 465
 466             /**************************
 467              * CALCULATE INTERACTIONS *
 468              **************************/
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq00             = _mm256_mul_pd(iq0,jq0);
 472
 473             /* COULOMB ELECTROSTATICS */
 474             velec            = _mm256_mul_pd(qq00,rinv00);
 475             felec            = _mm256_mul_pd(velec,rinvsq00);
 476
 477             fscal            = felec;
 478
 479             /* Calculate temporary vectorial force */
 480             tx               = _mm256_mul_pd(fscal,dx00);
 481             ty               = _mm256_mul_pd(fscal,dy00);
 482             tz               = _mm256_mul_pd(fscal,dz00);
 483
 484             /* Update vectorial force */
 485             fix0             = _mm256_add_pd(fix0,tx);
 486             fiy0             = _mm256_add_pd(fiy0,ty);
 487             fiz0             = _mm256_add_pd(fiz0,tz);
 488
 489             fjptrA             = f+j_coord_offsetA;
 490             fjptrB             = f+j_coord_offsetB;
 491             fjptrC             = f+j_coord_offsetC;
 492             fjptrD             = f+j_coord_offsetD;
 493             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 494
 495             /* Inner loop uses 26 flops */
 496         }
 497
 498         if(jidx<j_index_end)
 499         {
 500
 501             /* Get j neighbor index, and coordinate index */
 502             jnrlistA         = jjnr[jidx];
 503             jnrlistB         = jjnr[jidx+1];
 504             jnrlistC         = jjnr[jidx+2];
 505             jnrlistD         = jjnr[jidx+3];
 506             /* Sign of each element will be negative for non-real atoms.
 507              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 508              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 509              */
 510             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 511
 512             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 513             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 514             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 515
 516             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 517             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 518             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 519             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 520             j_coord_offsetA  = DIM*jnrA;
 521             j_coord_offsetB  = DIM*jnrB;
 522             j_coord_offsetC  = DIM*jnrC;
 523             j_coord_offsetD  = DIM*jnrD;
 524
 525             /* load j atom coordinates */
 526             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 527                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 528                                                  &jx0,&jy0,&jz0);
 529
 530             /* Calculate displacement vector */
 531             dx00             = _mm256_sub_pd(ix0,jx0);
 532             dy00             = _mm256_sub_pd(iy0,jy0);
 533             dz00             = _mm256_sub_pd(iz0,jz0);
 534
 535             /* Calculate squared distance and things based on it */
 536             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 537
 538             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 539
 540             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 541
 542             /* Load parameters for j particles */
 543             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 544                                                                  charge+jnrC+0,charge+jnrD+0);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq00             = _mm256_mul_pd(iq0,jq0);
 552
 553             /* COULOMB ELECTROSTATICS */
 554             velec            = _mm256_mul_pd(qq00,rinv00);
 555             felec            = _mm256_mul_pd(velec,rinvsq00);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm256_mul_pd(fscal,dx00);
 563             ty               = _mm256_mul_pd(fscal,dy00);
 564             tz               = _mm256_mul_pd(fscal,dz00);
 565
 566             /* Update vectorial force */
 567             fix0             = _mm256_add_pd(fix0,tx);
 568             fiy0             = _mm256_add_pd(fiy0,ty);
 569             fiz0             = _mm256_add_pd(fiz0,tz);
 570
 571             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 572             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 573             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 574             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 575             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 576
 577             /* Inner loop uses 26 flops */
 578         }
 579
 580         /* End of innermost loop */
 581
 582         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 583                                                  f+i_coord_offset,fshift+i_shift_offset);
 584
 585         /* Increment number of inner iterations */
 586         inneriter                  += j_index_end - j_index_start;
 587
 588         /* Outer loop uses 7 flops */
 589     }
 590
 591     /* Increment number of outer iterations */
 592     outeriter        += nri;
 593
 594     /* Update outer/inner flops */
 595
 596     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*26);
 597 }