src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwNone_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 116     rcutoff_scalar   = fr->rcoulomb;
 117     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 118     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = jnrC = jnrD = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124     j_coord_offsetC = 0;
 125     j_coord_offsetD = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     for(iidx=0;iidx<4*DIM;iidx++)
 131     {
 132         scratch[iidx] = 0.0;
 133     }
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm256_setzero_pd();
 153         fiy0             = _mm256_setzero_pd();
 154         fiz0             = _mm256_setzero_pd();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm256_setzero_pd();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173             j_coord_offsetC  = DIM*jnrC;
 174             j_coord_offsetD  = DIM*jnrD;
 175
 176             /* load j atom coordinates */
 177             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 179                                                  &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm256_sub_pd(ix0,jx0);
 183             dy00             = _mm256_sub_pd(iy0,jy0);
 184             dz00             = _mm256_sub_pd(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 188
 189             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 190
 191             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 195                                                                  charge+jnrC+0,charge+jnrD+0);
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 202             {
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm256_mul_pd(iq0,jq0);
 206
 207             /* REACTION-FIELD ELECTROSTATICS */
 208             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 209             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 210
 211             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             velec            = _mm256_and_pd(velec,cutoff_mask);
 215             velecsum         = _mm256_add_pd(velecsum,velec);
 216
 217             fscal            = felec;
 218
 219             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 220
 221             /* Calculate temporary vectorial force */
 222             tx               = _mm256_mul_pd(fscal,dx00);
 223             ty               = _mm256_mul_pd(fscal,dy00);
 224             tz               = _mm256_mul_pd(fscal,dz00);
 225
 226             /* Update vectorial force */
 227             fix0             = _mm256_add_pd(fix0,tx);
 228             fiy0             = _mm256_add_pd(fiy0,ty);
 229             fiz0             = _mm256_add_pd(fiz0,tz);
 230
 231             fjptrA             = f+j_coord_offsetA;
 232             fjptrB             = f+j_coord_offsetB;
 233             fjptrC             = f+j_coord_offsetC;
 234             fjptrD             = f+j_coord_offsetD;
 235             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 236
 237             }
 238
 239             /* Inner loop uses 36 flops */
 240         }
 241
 242         if(jidx<j_index_end)
 243         {
 244
 245             /* Get j neighbor index, and coordinate index */
 246             jnrlistA         = jjnr[jidx];
 247             jnrlistB         = jjnr[jidx+1];
 248             jnrlistC         = jjnr[jidx+2];
 249             jnrlistD         = jjnr[jidx+3];
 250             /* Sign of each element will be negative for non-real atoms.
 251              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 252              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 253              */
 254             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 255
 256             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 257             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 258             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 259
 260             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 261             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 262             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 263             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 264             j_coord_offsetA  = DIM*jnrA;
 265             j_coord_offsetB  = DIM*jnrB;
 266             j_coord_offsetC  = DIM*jnrC;
 267             j_coord_offsetD  = DIM*jnrD;
 268
 269             /* load j atom coordinates */
 270             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 271                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 272                                                  &jx0,&jy0,&jz0);
 273
 274             /* Calculate displacement vector */
 275             dx00             = _mm256_sub_pd(ix0,jx0);
 276             dy00             = _mm256_sub_pd(iy0,jy0);
 277             dz00             = _mm256_sub_pd(iz0,jz0);
 278
 279             /* Calculate squared distance and things based on it */
 280             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 281
 282             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 283
 284             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 285
 286             /* Load parameters for j particles */
 287             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 288                                                                  charge+jnrC+0,charge+jnrD+0);
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 295             {
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq00             = _mm256_mul_pd(iq0,jq0);
 299
 300             /* REACTION-FIELD ELECTROSTATICS */
 301             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 302             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 303
 304             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velec            = _mm256_and_pd(velec,cutoff_mask);
 308             velec            = _mm256_andnot_pd(dummy_mask,velec);
 309             velecsum         = _mm256_add_pd(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 314
 315             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm256_mul_pd(fscal,dx00);
 319             ty               = _mm256_mul_pd(fscal,dy00);
 320             tz               = _mm256_mul_pd(fscal,dz00);
 321
 322             /* Update vectorial force */
 323             fix0             = _mm256_add_pd(fix0,tx);
 324             fiy0             = _mm256_add_pd(fiy0,ty);
 325             fiz0             = _mm256_add_pd(fiz0,tz);
 326
 327             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 328             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 329             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 330             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 331             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 332
 333             }
 334
 335             /* Inner loop uses 36 flops */
 336         }
 337
 338         /* End of innermost loop */
 339
 340         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 341                                                  f+i_coord_offset,fshift+i_shift_offset);
 342
 343         ggid                        = gid[iidx];
 344         /* Update potential energies */
 345         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 346
 347         /* Increment number of inner iterations */
 348         inneriter                  += j_index_end - j_index_start;
 349
 350         /* Outer loop uses 8 flops */
 351     }
 352
 353     /* Increment number of outer iterations */
 354     outeriter        += nri;
 355
 356     /* Update outer/inner flops */
 357
 358     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 359 }
 360 /*
 361  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_double
 362  * Electrostatics interaction: ReactionField
 363  * VdW interaction:            None
 364  * Geometry:                   Particle-Particle
 365  * Calculate force/pot:        Force
 366  */
 367 void
 368 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_double
 369                     (t_nblist                    * gmx_restrict       nlist,
 370                      rvec                        * gmx_restrict          xx,
 371                      rvec                        * gmx_restrict          ff,
 372                      t_forcerec                  * gmx_restrict          fr,
 373                      t_mdatoms                   * gmx_restrict     mdatoms,
 374                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 375                      t_nrnb                      * gmx_restrict        nrnb)
 376 {
 377     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 378      * just 0 for non-waters.
 379      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 380      * jnr indices corresponding to data put in the four positions in the SIMD register.
 381      */
 382     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 383     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 384     int              jnrA,jnrB,jnrC,jnrD;
 385     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 386     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 387     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 388     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 389     real             rcutoff_scalar;
 390     real             *shiftvec,*fshift,*x,*f;
 391     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 392     real             scratch[4*DIM];
 393     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 394     real *           vdwioffsetptr0;
 395     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 396     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 397     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 398     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 399     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 400     real             *charge;
 401     __m256d          dummy_mask,cutoff_mask;
 402     __m128           tmpmask0,tmpmask1;
 403     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 404     __m256d          one     = _mm256_set1_pd(1.0);
 405     __m256d          two     = _mm256_set1_pd(2.0);
 406     x                = xx[0];
 407     f                = ff[0];
 408
 409     nri              = nlist->nri;
 410     iinr             = nlist->iinr;
 411     jindex           = nlist->jindex;
 412     jjnr             = nlist->jjnr;
 413     shiftidx         = nlist->shift;
 414     gid              = nlist->gid;
 415     shiftvec         = fr->shift_vec[0];
 416     fshift           = fr->fshift[0];
 417     facel            = _mm256_set1_pd(fr->epsfac);
 418     charge           = mdatoms->chargeA;
 419     krf              = _mm256_set1_pd(fr->ic->k_rf);
 420     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 421     crf              = _mm256_set1_pd(fr->ic->c_rf);
 422
 423     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 424     rcutoff_scalar   = fr->rcoulomb;
 425     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 426     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 427
 428     /* Avoid stupid compiler warnings */
 429     jnrA = jnrB = jnrC = jnrD = 0;
 430     j_coord_offsetA = 0;
 431     j_coord_offsetB = 0;
 432     j_coord_offsetC = 0;
 433     j_coord_offsetD = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     for(iidx=0;iidx<4*DIM;iidx++)
 439     {
 440         scratch[iidx] = 0.0;
 441     }
 442
 443     /* Start outer loop over neighborlists */
 444     for(iidx=0; iidx<nri; iidx++)
 445     {
 446         /* Load shift vector for this list */
 447         i_shift_offset   = DIM*shiftidx[iidx];
 448
 449         /* Load limits for loop over neighbors */
 450         j_index_start    = jindex[iidx];
 451         j_index_end      = jindex[iidx+1];
 452
 453         /* Get outer coordinate index */
 454         inr              = iinr[iidx];
 455         i_coord_offset   = DIM*inr;
 456
 457         /* Load i particle coords and add shift vector */
 458         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 459
 460         fix0             = _mm256_setzero_pd();
 461         fiy0             = _mm256_setzero_pd();
 462         fiz0             = _mm256_setzero_pd();
 463
 464         /* Load parameters for i particles */
 465         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 466
 467         /* Start inner kernel loop */
 468         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 469         {
 470
 471             /* Get j neighbor index, and coordinate index */
 472             jnrA             = jjnr[jidx];
 473             jnrB             = jjnr[jidx+1];
 474             jnrC             = jjnr[jidx+2];
 475             jnrD             = jjnr[jidx+3];
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478             j_coord_offsetC  = DIM*jnrC;
 479             j_coord_offsetD  = DIM*jnrD;
 480
 481             /* load j atom coordinates */
 482             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 484                                                  &jx0,&jy0,&jz0);
 485
 486             /* Calculate displacement vector */
 487             dx00             = _mm256_sub_pd(ix0,jx0);
 488             dy00             = _mm256_sub_pd(iy0,jy0);
 489             dz00             = _mm256_sub_pd(iz0,jz0);
 490
 491             /* Calculate squared distance and things based on it */
 492             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 493
 494             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 495
 496             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 497
 498             /* Load parameters for j particles */
 499             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 500                                                                  charge+jnrC+0,charge+jnrD+0);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 507             {
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq00             = _mm256_mul_pd(iq0,jq0);
 511
 512             /* REACTION-FIELD ELECTROSTATICS */
 513             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 514
 515             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 516
 517             fscal            = felec;
 518
 519             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_pd(fscal,dx00);
 523             ty               = _mm256_mul_pd(fscal,dy00);
 524             tz               = _mm256_mul_pd(fscal,dz00);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm256_add_pd(fix0,tx);
 528             fiy0             = _mm256_add_pd(fiy0,ty);
 529             fiz0             = _mm256_add_pd(fiz0,tz);
 530
 531             fjptrA             = f+j_coord_offsetA;
 532             fjptrB             = f+j_coord_offsetB;
 533             fjptrC             = f+j_coord_offsetC;
 534             fjptrD             = f+j_coord_offsetD;
 535             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 536
 537             }
 538
 539             /* Inner loop uses 30 flops */
 540         }
 541
 542         if(jidx<j_index_end)
 543         {
 544
 545             /* Get j neighbor index, and coordinate index */
 546             jnrlistA         = jjnr[jidx];
 547             jnrlistB         = jjnr[jidx+1];
 548             jnrlistC         = jjnr[jidx+2];
 549             jnrlistD         = jjnr[jidx+3];
 550             /* Sign of each element will be negative for non-real atoms.
 551              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 552              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 553              */
 554             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 555
 556             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 557             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 558             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 559
 560             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 561             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 562             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 563             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 564             j_coord_offsetA  = DIM*jnrA;
 565             j_coord_offsetB  = DIM*jnrB;
 566             j_coord_offsetC  = DIM*jnrC;
 567             j_coord_offsetD  = DIM*jnrD;
 568
 569             /* load j atom coordinates */
 570             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 571                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 572                                                  &jx0,&jy0,&jz0);
 573
 574             /* Calculate displacement vector */
 575             dx00             = _mm256_sub_pd(ix0,jx0);
 576             dy00             = _mm256_sub_pd(iy0,jy0);
 577             dz00             = _mm256_sub_pd(iz0,jz0);
 578
 579             /* Calculate squared distance and things based on it */
 580             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 581
 582             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 583
 584             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 585
 586             /* Load parameters for j particles */
 587             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 588                                                                  charge+jnrC+0,charge+jnrD+0);
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 595             {
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq00             = _mm256_mul_pd(iq0,jq0);
 599
 600             /* REACTION-FIELD ELECTROSTATICS */
 601             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 602
 603             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 608
 609             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm256_mul_pd(fscal,dx00);
 613             ty               = _mm256_mul_pd(fscal,dy00);
 614             tz               = _mm256_mul_pd(fscal,dz00);
 615
 616             /* Update vectorial force */
 617             fix0             = _mm256_add_pd(fix0,tx);
 618             fiy0             = _mm256_add_pd(fiy0,ty);
 619             fiz0             = _mm256_add_pd(fiz0,tz);
 620
 621             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 622             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 623             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 624             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 625             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 626
 627             }
 628
 629             /* Inner loop uses 30 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 635                                                  f+i_coord_offset,fshift+i_shift_offset);
 636
 637         /* Increment number of inner iterations */
 638         inneriter                  += j_index_end - j_index_start;
 639
 640         /* Outer loop uses 7 flops */
 641     }
 642
 643     /* Increment number of outer iterations */
 644     outeriter        += nri;
 645
 646     /* Update outer/inner flops */
 647
 648     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 649 }