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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  90     real             *charge;
  91     __m256d          dummy_mask,cutoff_mask;
  92     __m128           tmpmask0,tmpmask1;
  93     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  94     __m256d          one     = _mm256_set1_pd(1.0);
  95     __m256d          two     = _mm256_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            = _mm256_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm256_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm256_set1_pd(fr->ic->c_rf);
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     for(iidx=0;iidx<4*DIM;iidx++)
 124     {
 125         scratch[iidx] = 0.0;
 126     }
 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_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm256_setzero_pd();
 146         fiy0             = _mm256_setzero_pd();
 147         fiz0             = _mm256_setzero_pd();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm256_setzero_pd();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             jnrC             = jjnr[jidx+2];
 163             jnrD             = jjnr[jidx+3];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 172                                                  &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm256_sub_pd(ix0,jx0);
 176             dy00             = _mm256_sub_pd(iy0,jy0);
 177             dz00             = _mm256_sub_pd(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 183
 184             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 188                                                                  charge+jnrC+0,charge+jnrD+0);
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _mm256_mul_pd(iq0,jq0);
 196
 197             /* REACTION-FIELD ELECTROSTATICS */
 198             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 199             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 200
 201             /* Update potential sum for this i atom from the interaction with this j atom. */
 202             velecsum         = _mm256_add_pd(velecsum,velec);
 203
 204             fscal            = felec;
 205
 206             /* Calculate temporary vectorial force */
 207             tx               = _mm256_mul_pd(fscal,dx00);
 208             ty               = _mm256_mul_pd(fscal,dy00);
 209             tz               = _mm256_mul_pd(fscal,dz00);
 210
 211             /* Update vectorial force */
 212             fix0             = _mm256_add_pd(fix0,tx);
 213             fiy0             = _mm256_add_pd(fiy0,ty);
 214             fiz0             = _mm256_add_pd(fiz0,tz);
 215
 216             fjptrA             = f+j_coord_offsetA;
 217             fjptrB             = f+j_coord_offsetB;
 218             fjptrC             = f+j_coord_offsetC;
 219             fjptrD             = f+j_coord_offsetD;
 220             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 221
 222             /* Inner loop uses 32 flops */
 223         }
 224
 225         if(jidx<j_index_end)
 226         {
 227
 228             /* Get j neighbor index, and coordinate index */
 229             jnrlistA         = jjnr[jidx];
 230             jnrlistB         = jjnr[jidx+1];
 231             jnrlistC         = jjnr[jidx+2];
 232             jnrlistD         = jjnr[jidx+3];
 233             /* Sign of each element will be negative for non-real atoms.
 234              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 235              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 236              */
 237             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 238
 239             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 240             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 241             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 242
 243             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 244             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 245             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 246             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 247             j_coord_offsetA  = DIM*jnrA;
 248             j_coord_offsetB  = DIM*jnrB;
 249             j_coord_offsetC  = DIM*jnrC;
 250             j_coord_offsetD  = DIM*jnrD;
 251
 252             /* load j atom coordinates */
 253             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 254                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 255                                                  &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm256_sub_pd(ix0,jx0);
 259             dy00             = _mm256_sub_pd(iy0,jy0);
 260             dz00             = _mm256_sub_pd(iz0,jz0);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 264
 265             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 266
 267             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 268
 269             /* Load parameters for j particles */
 270             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 271                                                                  charge+jnrC+0,charge+jnrD+0);
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq00             = _mm256_mul_pd(iq0,jq0);
 279
 280             /* REACTION-FIELD ELECTROSTATICS */
 281             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 282             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velec            = _mm256_andnot_pd(dummy_mask,velec);
 286             velecsum         = _mm256_add_pd(velecsum,velec);
 287
 288             fscal            = felec;
 289
 290             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm256_mul_pd(fscal,dx00);
 294             ty               = _mm256_mul_pd(fscal,dy00);
 295             tz               = _mm256_mul_pd(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm256_add_pd(fix0,tx);
 299             fiy0             = _mm256_add_pd(fiy0,ty);
 300             fiz0             = _mm256_add_pd(fiz0,tz);
 301
 302             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 303             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 304             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 305             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 306             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 307
 308             /* Inner loop uses 32 flops */
 309         }
 310
 311         /* End of innermost loop */
 312
 313         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 314                                                  f+i_coord_offset,fshift+i_shift_offset);
 315
 316         ggid                        = gid[iidx];
 317         /* Update potential energies */
 318         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 319
 320         /* Increment number of inner iterations */
 321         inneriter                  += j_index_end - j_index_start;
 322
 323         /* Outer loop uses 8 flops */
 324     }
 325
 326     /* Increment number of outer iterations */
 327     outeriter        += nri;
 328
 329     /* Update outer/inner flops */
 330
 331     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 332 }
 333 /*
 334  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_double
 335  * Electrostatics interaction: ReactionField
 336  * VdW interaction:            None
 337  * Geometry:                   Particle-Particle
 338  * Calculate force/pot:        Force
 339  */
 340 void
 341 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_double
 342                     (t_nblist                    * gmx_restrict       nlist,
 343                      rvec                        * gmx_restrict          xx,
 344                      rvec                        * gmx_restrict          ff,
 345                      t_forcerec                  * gmx_restrict          fr,
 346                      t_mdatoms                   * gmx_restrict     mdatoms,
 347                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 348                      t_nrnb                      * gmx_restrict        nrnb)
 349 {
 350     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 351      * just 0 for non-waters.
 352      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 353      * jnr indices corresponding to data put in the four positions in the SIMD register.
 354      */
 355     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 356     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 357     int              jnrA,jnrB,jnrC,jnrD;
 358     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 359     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 360     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 361     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 362     real             rcutoff_scalar;
 363     real             *shiftvec,*fshift,*x,*f;
 364     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 365     real             scratch[4*DIM];
 366     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 367     real *           vdwioffsetptr0;
 368     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 369     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 370     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 371     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 372     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 373     real             *charge;
 374     __m256d          dummy_mask,cutoff_mask;
 375     __m128           tmpmask0,tmpmask1;
 376     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 377     __m256d          one     = _mm256_set1_pd(1.0);
 378     __m256d          two     = _mm256_set1_pd(2.0);
 379     x                = xx[0];
 380     f                = ff[0];
 381
 382     nri              = nlist->nri;
 383     iinr             = nlist->iinr;
 384     jindex           = nlist->jindex;
 385     jjnr             = nlist->jjnr;
 386     shiftidx         = nlist->shift;
 387     gid              = nlist->gid;
 388     shiftvec         = fr->shift_vec[0];
 389     fshift           = fr->fshift[0];
 390     facel            = _mm256_set1_pd(fr->epsfac);
 391     charge           = mdatoms->chargeA;
 392     krf              = _mm256_set1_pd(fr->ic->k_rf);
 393     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 394     crf              = _mm256_set1_pd(fr->ic->c_rf);
 395
 396     /* Avoid stupid compiler warnings */
 397     jnrA = jnrB = jnrC = jnrD = 0;
 398     j_coord_offsetA = 0;
 399     j_coord_offsetB = 0;
 400     j_coord_offsetC = 0;
 401     j_coord_offsetD = 0;
 402
 403     outeriter        = 0;
 404     inneriter        = 0;
 405
 406     for(iidx=0;iidx<4*DIM;iidx++)
 407     {
 408         scratch[iidx] = 0.0;
 409     }
 410
 411     /* Start outer loop over neighborlists */
 412     for(iidx=0; iidx<nri; iidx++)
 413     {
 414         /* Load shift vector for this list */
 415         i_shift_offset   = DIM*shiftidx[iidx];
 416
 417         /* Load limits for loop over neighbors */
 418         j_index_start    = jindex[iidx];
 419         j_index_end      = jindex[iidx+1];
 420
 421         /* Get outer coordinate index */
 422         inr              = iinr[iidx];
 423         i_coord_offset   = DIM*inr;
 424
 425         /* Load i particle coords and add shift vector */
 426         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 427
 428         fix0             = _mm256_setzero_pd();
 429         fiy0             = _mm256_setzero_pd();
 430         fiz0             = _mm256_setzero_pd();
 431
 432         /* Load parameters for i particles */
 433         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 434
 435         /* Start inner kernel loop */
 436         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 437         {
 438
 439             /* Get j neighbor index, and coordinate index */
 440             jnrA             = jjnr[jidx];
 441             jnrB             = jjnr[jidx+1];
 442             jnrC             = jjnr[jidx+2];
 443             jnrD             = jjnr[jidx+3];
 444             j_coord_offsetA  = DIM*jnrA;
 445             j_coord_offsetB  = DIM*jnrB;
 446             j_coord_offsetC  = DIM*jnrC;
 447             j_coord_offsetD  = DIM*jnrD;
 448
 449             /* load j atom coordinates */
 450             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 451                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 452                                                  &jx0,&jy0,&jz0);
 453
 454             /* Calculate displacement vector */
 455             dx00             = _mm256_sub_pd(ix0,jx0);
 456             dy00             = _mm256_sub_pd(iy0,jy0);
 457             dz00             = _mm256_sub_pd(iz0,jz0);
 458
 459             /* Calculate squared distance and things based on it */
 460             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 461
 462             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 463
 464             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 465
 466             /* Load parameters for j particles */
 467             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 468                                                                  charge+jnrC+0,charge+jnrD+0);
 469
 470             /**************************
 471              * CALCULATE INTERACTIONS *
 472              **************************/
 473
 474             /* Compute parameters for interactions between i and j atoms */
 475             qq00             = _mm256_mul_pd(iq0,jq0);
 476
 477             /* REACTION-FIELD ELECTROSTATICS */
 478             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 479
 480             fscal            = felec;
 481
 482             /* Calculate temporary vectorial force */
 483             tx               = _mm256_mul_pd(fscal,dx00);
 484             ty               = _mm256_mul_pd(fscal,dy00);
 485             tz               = _mm256_mul_pd(fscal,dz00);
 486
 487             /* Update vectorial force */
 488             fix0             = _mm256_add_pd(fix0,tx);
 489             fiy0             = _mm256_add_pd(fiy0,ty);
 490             fiz0             = _mm256_add_pd(fiz0,tz);
 491
 492             fjptrA             = f+j_coord_offsetA;
 493             fjptrB             = f+j_coord_offsetB;
 494             fjptrC             = f+j_coord_offsetC;
 495             fjptrD             = f+j_coord_offsetD;
 496             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 497
 498             /* Inner loop uses 27 flops */
 499         }
 500
 501         if(jidx<j_index_end)
 502         {
 503
 504             /* Get j neighbor index, and coordinate index */
 505             jnrlistA         = jjnr[jidx];
 506             jnrlistB         = jjnr[jidx+1];
 507             jnrlistC         = jjnr[jidx+2];
 508             jnrlistD         = jjnr[jidx+3];
 509             /* Sign of each element will be negative for non-real atoms.
 510              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 511              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 512              */
 513             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 514
 515             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 516             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 517             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 518
 519             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 520             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 521             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 522             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 523             j_coord_offsetA  = DIM*jnrA;
 524             j_coord_offsetB  = DIM*jnrB;
 525             j_coord_offsetC  = DIM*jnrC;
 526             j_coord_offsetD  = DIM*jnrD;
 527
 528             /* load j atom coordinates */
 529             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 530                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 531                                                  &jx0,&jy0,&jz0);
 532
 533             /* Calculate displacement vector */
 534             dx00             = _mm256_sub_pd(ix0,jx0);
 535             dy00             = _mm256_sub_pd(iy0,jy0);
 536             dz00             = _mm256_sub_pd(iz0,jz0);
 537
 538             /* Calculate squared distance and things based on it */
 539             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 540
 541             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 542
 543             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 544
 545             /* Load parameters for j particles */
 546             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 547                                                                  charge+jnrC+0,charge+jnrD+0);
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq00             = _mm256_mul_pd(iq0,jq0);
 555
 556             /* REACTION-FIELD ELECTROSTATICS */
 557             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 558
 559             fscal            = felec;
 560
 561             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm256_mul_pd(fscal,dx00);
 565             ty               = _mm256_mul_pd(fscal,dy00);
 566             tz               = _mm256_mul_pd(fscal,dz00);
 567
 568             /* Update vectorial force */
 569             fix0             = _mm256_add_pd(fix0,tx);
 570             fiy0             = _mm256_add_pd(fiy0,ty);
 571             fiz0             = _mm256_add_pd(fiz0,tz);
 572
 573             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 574             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 575             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 576             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 577             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 578
 579             /* Inner loop uses 27 flops */
 580         }
 581
 582         /* End of innermost loop */
 583
 584         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 585                                                  f+i_coord_offset,fshift+i_shift_offset);
 586
 587         /* Increment number of inner iterations */
 588         inneriter                  += j_index_end - j_index_start;
 589
 590         /* Outer loop uses 7 flops */
 591     }
 592
 593     /* Increment number of outer iterations */
 594     outeriter        += nri;
 595
 596     /* Update outer/inner flops */
 597
 598     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 599 }