src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_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_ElecRF_VdwNone_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_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     krf              = _mm256_set1_pd(fr->ic->k_rf);
 112     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 113     crf              = _mm256_set1_pd(fr->ic->c_rf);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = jnrC = jnrD = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119     j_coord_offsetC = 0;
 120     j_coord_offsetD = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm256_setzero_pd();
 148         fiy0             = _mm256_setzero_pd();
 149         fiz0             = _mm256_setzero_pd();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm256_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168             j_coord_offsetC  = DIM*jnrC;
 169             j_coord_offsetD  = DIM*jnrD;
 170
 171             /* load j atom coordinates */
 172             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 174                                                  &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm256_sub_pd(ix0,jx0);
 178             dy00             = _mm256_sub_pd(iy0,jy0);
 179             dz00             = _mm256_sub_pd(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 183
 184             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 185
 186             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 190                                                                  charge+jnrC+0,charge+jnrD+0);
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             qq00             = _mm256_mul_pd(iq0,jq0);
 198
 199             /* REACTION-FIELD ELECTROSTATICS */
 200             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 201             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 202
 203             /* Update potential sum for this i atom from the interaction with this j atom. */
 204             velecsum         = _mm256_add_pd(velecsum,velec);
 205
 206             fscal            = felec;
 207
 208             /* Calculate temporary vectorial force */
 209             tx               = _mm256_mul_pd(fscal,dx00);
 210             ty               = _mm256_mul_pd(fscal,dy00);
 211             tz               = _mm256_mul_pd(fscal,dz00);
 212
 213             /* Update vectorial force */
 214             fix0             = _mm256_add_pd(fix0,tx);
 215             fiy0             = _mm256_add_pd(fiy0,ty);
 216             fiz0             = _mm256_add_pd(fiz0,tz);
 217
 218             fjptrA             = f+j_coord_offsetA;
 219             fjptrB             = f+j_coord_offsetB;
 220             fjptrC             = f+j_coord_offsetC;
 221             fjptrD             = f+j_coord_offsetD;
 222             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 223
 224             /* Inner loop uses 32 flops */
 225         }
 226
 227         if(jidx<j_index_end)
 228         {
 229
 230             /* Get j neighbor index, and coordinate index */
 231             jnrlistA         = jjnr[jidx];
 232             jnrlistB         = jjnr[jidx+1];
 233             jnrlistC         = jjnr[jidx+2];
 234             jnrlistD         = jjnr[jidx+3];
 235             /* Sign of each element will be negative for non-real atoms.
 236              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 237              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 238              */
 239             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 240
 241             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 242             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 243             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 244
 245             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 246             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 247             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 248             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 249             j_coord_offsetA  = DIM*jnrA;
 250             j_coord_offsetB  = DIM*jnrB;
 251             j_coord_offsetC  = DIM*jnrC;
 252             j_coord_offsetD  = DIM*jnrD;
 253
 254             /* load j atom coordinates */
 255             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 256                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 257                                                  &jx0,&jy0,&jz0);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm256_sub_pd(ix0,jx0);
 261             dy00             = _mm256_sub_pd(iy0,jy0);
 262             dz00             = _mm256_sub_pd(iz0,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 266
 267             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 268
 269             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 270
 271             /* Load parameters for j particles */
 272             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 273                                                                  charge+jnrC+0,charge+jnrD+0);
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq00             = _mm256_mul_pd(iq0,jq0);
 281
 282             /* REACTION-FIELD ELECTROSTATICS */
 283             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 284             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velec            = _mm256_andnot_pd(dummy_mask,velec);
 288             velecsum         = _mm256_add_pd(velecsum,velec);
 289
 290             fscal            = felec;
 291
 292             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm256_mul_pd(fscal,dx00);
 296             ty               = _mm256_mul_pd(fscal,dy00);
 297             tz               = _mm256_mul_pd(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm256_add_pd(fix0,tx);
 301             fiy0             = _mm256_add_pd(fiy0,ty);
 302             fiz0             = _mm256_add_pd(fiz0,tz);
 303
 304             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 305             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 306             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 307             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 308             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 309
 310             /* Inner loop uses 32 flops */
 311         }
 312
 313         /* End of innermost loop */
 314
 315         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 316                                                  f+i_coord_offset,fshift+i_shift_offset);
 317
 318         ggid                        = gid[iidx];
 319         /* Update potential energies */
 320         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 321
 322         /* Increment number of inner iterations */
 323         inneriter                  += j_index_end - j_index_start;
 324
 325         /* Outer loop uses 8 flops */
 326     }
 327
 328     /* Increment number of outer iterations */
 329     outeriter        += nri;
 330
 331     /* Update outer/inner flops */
 332
 333     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 334 }
 335 /*
 336  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_double
 337  * Electrostatics interaction: ReactionField
 338  * VdW interaction:            None
 339  * Geometry:                   Particle-Particle
 340  * Calculate force/pot:        Force
 341  */
 342 void
 343 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_double
 344                     (t_nblist                    * gmx_restrict       nlist,
 345                      rvec                        * gmx_restrict          xx,
 346                      rvec                        * gmx_restrict          ff,
 347                      t_forcerec                  * gmx_restrict          fr,
 348                      t_mdatoms                   * gmx_restrict     mdatoms,
 349                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 350                      t_nrnb                      * gmx_restrict        nrnb)
 351 {
 352     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 353      * just 0 for non-waters.
 354      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 355      * jnr indices corresponding to data put in the four positions in the SIMD register.
 356      */
 357     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 358     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 359     int              jnrA,jnrB,jnrC,jnrD;
 360     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 361     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 362     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 363     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 364     real             rcutoff_scalar;
 365     real             *shiftvec,*fshift,*x,*f;
 366     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 367     real             scratch[4*DIM];
 368     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 369     real *           vdwioffsetptr0;
 370     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 371     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 372     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 373     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 374     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 375     real             *charge;
 376     __m256d          dummy_mask,cutoff_mask;
 377     __m128           tmpmask0,tmpmask1;
 378     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 379     __m256d          one     = _mm256_set1_pd(1.0);
 380     __m256d          two     = _mm256_set1_pd(2.0);
 381     x                = xx[0];
 382     f                = ff[0];
 383
 384     nri              = nlist->nri;
 385     iinr             = nlist->iinr;
 386     jindex           = nlist->jindex;
 387     jjnr             = nlist->jjnr;
 388     shiftidx         = nlist->shift;
 389     gid              = nlist->gid;
 390     shiftvec         = fr->shift_vec[0];
 391     fshift           = fr->fshift[0];
 392     facel            = _mm256_set1_pd(fr->epsfac);
 393     charge           = mdatoms->chargeA;
 394     krf              = _mm256_set1_pd(fr->ic->k_rf);
 395     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 396     crf              = _mm256_set1_pd(fr->ic->c_rf);
 397
 398     /* Avoid stupid compiler warnings */
 399     jnrA = jnrB = jnrC = jnrD = 0;
 400     j_coord_offsetA = 0;
 401     j_coord_offsetB = 0;
 402     j_coord_offsetC = 0;
 403     j_coord_offsetD = 0;
 404
 405     outeriter        = 0;
 406     inneriter        = 0;
 407
 408     for(iidx=0;iidx<4*DIM;iidx++)
 409     {
 410         scratch[iidx] = 0.0;
 411     }
 412
 413     /* Start outer loop over neighborlists */
 414     for(iidx=0; iidx<nri; iidx++)
 415     {
 416         /* Load shift vector for this list */
 417         i_shift_offset   = DIM*shiftidx[iidx];
 418
 419         /* Load limits for loop over neighbors */
 420         j_index_start    = jindex[iidx];
 421         j_index_end      = jindex[iidx+1];
 422
 423         /* Get outer coordinate index */
 424         inr              = iinr[iidx];
 425         i_coord_offset   = DIM*inr;
 426
 427         /* Load i particle coords and add shift vector */
 428         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 429
 430         fix0             = _mm256_setzero_pd();
 431         fiy0             = _mm256_setzero_pd();
 432         fiz0             = _mm256_setzero_pd();
 433
 434         /* Load parameters for i particles */
 435         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 436
 437         /* Start inner kernel loop */
 438         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 439         {
 440
 441             /* Get j neighbor index, and coordinate index */
 442             jnrA             = jjnr[jidx];
 443             jnrB             = jjnr[jidx+1];
 444             jnrC             = jjnr[jidx+2];
 445             jnrD             = jjnr[jidx+3];
 446             j_coord_offsetA  = DIM*jnrA;
 447             j_coord_offsetB  = DIM*jnrB;
 448             j_coord_offsetC  = DIM*jnrC;
 449             j_coord_offsetD  = DIM*jnrD;
 450
 451             /* load j atom coordinates */
 452             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 453                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 454                                                  &jx0,&jy0,&jz0);
 455
 456             /* Calculate displacement vector */
 457             dx00             = _mm256_sub_pd(ix0,jx0);
 458             dy00             = _mm256_sub_pd(iy0,jy0);
 459             dz00             = _mm256_sub_pd(iz0,jz0);
 460
 461             /* Calculate squared distance and things based on it */
 462             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 463
 464             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 465
 466             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 467
 468             /* Load parameters for j particles */
 469             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 470                                                                  charge+jnrC+0,charge+jnrD+0);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             /* Compute parameters for interactions between i and j atoms */
 477             qq00             = _mm256_mul_pd(iq0,jq0);
 478
 479             /* REACTION-FIELD ELECTROSTATICS */
 480             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 481
 482             fscal            = felec;
 483
 484             /* Calculate temporary vectorial force */
 485             tx               = _mm256_mul_pd(fscal,dx00);
 486             ty               = _mm256_mul_pd(fscal,dy00);
 487             tz               = _mm256_mul_pd(fscal,dz00);
 488
 489             /* Update vectorial force */
 490             fix0             = _mm256_add_pd(fix0,tx);
 491             fiy0             = _mm256_add_pd(fiy0,ty);
 492             fiz0             = _mm256_add_pd(fiz0,tz);
 493
 494             fjptrA             = f+j_coord_offsetA;
 495             fjptrB             = f+j_coord_offsetB;
 496             fjptrC             = f+j_coord_offsetC;
 497             fjptrD             = f+j_coord_offsetD;
 498             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 499
 500             /* Inner loop uses 27 flops */
 501         }
 502
 503         if(jidx<j_index_end)
 504         {
 505
 506             /* Get j neighbor index, and coordinate index */
 507             jnrlistA         = jjnr[jidx];
 508             jnrlistB         = jjnr[jidx+1];
 509             jnrlistC         = jjnr[jidx+2];
 510             jnrlistD         = jjnr[jidx+3];
 511             /* Sign of each element will be negative for non-real atoms.
 512              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 513              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 514              */
 515             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 516
 517             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 518             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 519             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 520
 521             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 522             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 523             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 524             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 525             j_coord_offsetA  = DIM*jnrA;
 526             j_coord_offsetB  = DIM*jnrB;
 527             j_coord_offsetC  = DIM*jnrC;
 528             j_coord_offsetD  = DIM*jnrD;
 529
 530             /* load j atom coordinates */
 531             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 533                                                  &jx0,&jy0,&jz0);
 534
 535             /* Calculate displacement vector */
 536             dx00             = _mm256_sub_pd(ix0,jx0);
 537             dy00             = _mm256_sub_pd(iy0,jy0);
 538             dz00             = _mm256_sub_pd(iz0,jz0);
 539
 540             /* Calculate squared distance and things based on it */
 541             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 542
 543             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 544
 545             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 546
 547             /* Load parameters for j particles */
 548             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 549                                                                  charge+jnrC+0,charge+jnrD+0);
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq00             = _mm256_mul_pd(iq0,jq0);
 557
 558             /* REACTION-FIELD ELECTROSTATICS */
 559             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm256_mul_pd(fscal,dx00);
 567             ty               = _mm256_mul_pd(fscal,dy00);
 568             tz               = _mm256_mul_pd(fscal,dz00);
 569
 570             /* Update vectorial force */
 571             fix0             = _mm256_add_pd(fix0,tx);
 572             fiy0             = _mm256_add_pd(fiy0,ty);
 573             fiz0             = _mm256_add_pd(fiz0,tz);
 574
 575             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 576             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 577             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 578             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 579             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 580
 581             /* Inner loop uses 27 flops */
 582         }
 583
 584         /* End of innermost loop */
 585
 586         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 587                                                  f+i_coord_offset,fshift+i_shift_offset);
 588
 589         /* Increment number of inner iterations */
 590         inneriter                  += j_index_end - j_index_start;
 591
 592         /* Outer loop uses 7 flops */
 593     }
 594
 595     /* Increment number of outer iterations */
 596     outeriter        += nri;
 597
 598     /* Update outer/inner flops */
 599
 600     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 601 }