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) 2012,2013,2014, 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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 114     rcutoff_scalar   = fr->rcoulomb;
 115     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 116     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = jnrC = jnrD = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122     j_coord_offsetC = 0;
 123     j_coord_offsetD = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     for(iidx=0;iidx<4*DIM;iidx++)
 129     {
 130         scratch[iidx] = 0.0;
 131     }
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 149
 150         fix0             = _mm256_setzero_pd();
 151         fiy0             = _mm256_setzero_pd();
 152         fiz0             = _mm256_setzero_pd();
 153
 154         /* Load parameters for i particles */
 155         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm256_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             jnrC             = jjnr[jidx+2];
 168             jnrD             = jjnr[jidx+3];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171             j_coord_offsetC  = DIM*jnrC;
 172             j_coord_offsetD  = DIM*jnrD;
 173
 174             /* load j atom coordinates */
 175             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 177                                                  &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx00             = _mm256_sub_pd(ix0,jx0);
 181             dy00             = _mm256_sub_pd(iy0,jy0);
 182             dz00             = _mm256_sub_pd(iz0,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 186
 187             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 188
 189             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 190
 191             /* Load parameters for j particles */
 192             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 193                                                                  charge+jnrC+0,charge+jnrD+0);
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 200             {
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             qq00             = _mm256_mul_pd(iq0,jq0);
 204
 205             /* REACTION-FIELD ELECTROSTATICS */
 206             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 207             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 208
 209             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             velec            = _mm256_and_pd(velec,cutoff_mask);
 213             velecsum         = _mm256_add_pd(velecsum,velec);
 214
 215             fscal            = felec;
 216
 217             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 218
 219             /* Calculate temporary vectorial force */
 220             tx               = _mm256_mul_pd(fscal,dx00);
 221             ty               = _mm256_mul_pd(fscal,dy00);
 222             tz               = _mm256_mul_pd(fscal,dz00);
 223
 224             /* Update vectorial force */
 225             fix0             = _mm256_add_pd(fix0,tx);
 226             fiy0             = _mm256_add_pd(fiy0,ty);
 227             fiz0             = _mm256_add_pd(fiz0,tz);
 228
 229             fjptrA             = f+j_coord_offsetA;
 230             fjptrB             = f+j_coord_offsetB;
 231             fjptrC             = f+j_coord_offsetC;
 232             fjptrD             = f+j_coord_offsetD;
 233             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 234
 235             }
 236
 237             /* Inner loop uses 36 flops */
 238         }
 239
 240         if(jidx<j_index_end)
 241         {
 242
 243             /* Get j neighbor index, and coordinate index */
 244             jnrlistA         = jjnr[jidx];
 245             jnrlistB         = jjnr[jidx+1];
 246             jnrlistC         = jjnr[jidx+2];
 247             jnrlistD         = jjnr[jidx+3];
 248             /* Sign of each element will be negative for non-real atoms.
 249              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 250              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 251              */
 252             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 253
 254             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 255             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 256             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 257
 258             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 259             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 260             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 261             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 262             j_coord_offsetA  = DIM*jnrA;
 263             j_coord_offsetB  = DIM*jnrB;
 264             j_coord_offsetC  = DIM*jnrC;
 265             j_coord_offsetD  = DIM*jnrD;
 266
 267             /* load j atom coordinates */
 268             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 269                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 270                                                  &jx0,&jy0,&jz0);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _mm256_sub_pd(ix0,jx0);
 274             dy00             = _mm256_sub_pd(iy0,jy0);
 275             dz00             = _mm256_sub_pd(iz0,jz0);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 279
 280             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 281
 282             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 283
 284             /* Load parameters for j particles */
 285             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 286                                                                  charge+jnrC+0,charge+jnrD+0);
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 293             {
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             qq00             = _mm256_mul_pd(iq0,jq0);
 297
 298             /* REACTION-FIELD ELECTROSTATICS */
 299             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 300             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 301
 302             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm256_and_pd(velec,cutoff_mask);
 306             velec            = _mm256_andnot_pd(dummy_mask,velec);
 307             velecsum         = _mm256_add_pd(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 312
 313             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm256_mul_pd(fscal,dx00);
 317             ty               = _mm256_mul_pd(fscal,dy00);
 318             tz               = _mm256_mul_pd(fscal,dz00);
 319
 320             /* Update vectorial force */
 321             fix0             = _mm256_add_pd(fix0,tx);
 322             fiy0             = _mm256_add_pd(fiy0,ty);
 323             fiz0             = _mm256_add_pd(fiz0,tz);
 324
 325             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 326             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 327             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 328             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 329             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 330
 331             }
 332
 333             /* Inner loop uses 36 flops */
 334         }
 335
 336         /* End of innermost loop */
 337
 338         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 339                                                  f+i_coord_offset,fshift+i_shift_offset);
 340
 341         ggid                        = gid[iidx];
 342         /* Update potential energies */
 343         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 344
 345         /* Increment number of inner iterations */
 346         inneriter                  += j_index_end - j_index_start;
 347
 348         /* Outer loop uses 8 flops */
 349     }
 350
 351     /* Increment number of outer iterations */
 352     outeriter        += nri;
 353
 354     /* Update outer/inner flops */
 355
 356     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 357 }
 358 /*
 359  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_double
 360  * Electrostatics interaction: ReactionField
 361  * VdW interaction:            None
 362  * Geometry:                   Particle-Particle
 363  * Calculate force/pot:        Force
 364  */
 365 void
 366 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_double
 367                     (t_nblist                    * gmx_restrict       nlist,
 368                      rvec                        * gmx_restrict          xx,
 369                      rvec                        * gmx_restrict          ff,
 370                      t_forcerec                  * gmx_restrict          fr,
 371                      t_mdatoms                   * gmx_restrict     mdatoms,
 372                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 373                      t_nrnb                      * gmx_restrict        nrnb)
 374 {
 375     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 376      * just 0 for non-waters.
 377      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 378      * jnr indices corresponding to data put in the four positions in the SIMD register.
 379      */
 380     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 381     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 382     int              jnrA,jnrB,jnrC,jnrD;
 383     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 384     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 385     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 386     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 387     real             rcutoff_scalar;
 388     real             *shiftvec,*fshift,*x,*f;
 389     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 390     real             scratch[4*DIM];
 391     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 392     real *           vdwioffsetptr0;
 393     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 394     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 395     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 396     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 397     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 398     real             *charge;
 399     __m256d          dummy_mask,cutoff_mask;
 400     __m128           tmpmask0,tmpmask1;
 401     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 402     __m256d          one     = _mm256_set1_pd(1.0);
 403     __m256d          two     = _mm256_set1_pd(2.0);
 404     x                = xx[0];
 405     f                = ff[0];
 406
 407     nri              = nlist->nri;
 408     iinr             = nlist->iinr;
 409     jindex           = nlist->jindex;
 410     jjnr             = nlist->jjnr;
 411     shiftidx         = nlist->shift;
 412     gid              = nlist->gid;
 413     shiftvec         = fr->shift_vec[0];
 414     fshift           = fr->fshift[0];
 415     facel            = _mm256_set1_pd(fr->epsfac);
 416     charge           = mdatoms->chargeA;
 417     krf              = _mm256_set1_pd(fr->ic->k_rf);
 418     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 419     crf              = _mm256_set1_pd(fr->ic->c_rf);
 420
 421     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 422     rcutoff_scalar   = fr->rcoulomb;
 423     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 424     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 425
 426     /* Avoid stupid compiler warnings */
 427     jnrA = jnrB = jnrC = jnrD = 0;
 428     j_coord_offsetA = 0;
 429     j_coord_offsetB = 0;
 430     j_coord_offsetC = 0;
 431     j_coord_offsetD = 0;
 432
 433     outeriter        = 0;
 434     inneriter        = 0;
 435
 436     for(iidx=0;iidx<4*DIM;iidx++)
 437     {
 438         scratch[iidx] = 0.0;
 439     }
 440
 441     /* Start outer loop over neighborlists */
 442     for(iidx=0; iidx<nri; iidx++)
 443     {
 444         /* Load shift vector for this list */
 445         i_shift_offset   = DIM*shiftidx[iidx];
 446
 447         /* Load limits for loop over neighbors */
 448         j_index_start    = jindex[iidx];
 449         j_index_end      = jindex[iidx+1];
 450
 451         /* Get outer coordinate index */
 452         inr              = iinr[iidx];
 453         i_coord_offset   = DIM*inr;
 454
 455         /* Load i particle coords and add shift vector */
 456         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 457
 458         fix0             = _mm256_setzero_pd();
 459         fiy0             = _mm256_setzero_pd();
 460         fiz0             = _mm256_setzero_pd();
 461
 462         /* Load parameters for i particles */
 463         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 464
 465         /* Start inner kernel loop */
 466         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrA             = jjnr[jidx];
 471             jnrB             = jjnr[jidx+1];
 472             jnrC             = jjnr[jidx+2];
 473             jnrD             = jjnr[jidx+3];
 474             j_coord_offsetA  = DIM*jnrA;
 475             j_coord_offsetB  = DIM*jnrB;
 476             j_coord_offsetC  = DIM*jnrC;
 477             j_coord_offsetD  = DIM*jnrD;
 478
 479             /* load j atom coordinates */
 480             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 481                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 482                                                  &jx0,&jy0,&jz0);
 483
 484             /* Calculate displacement vector */
 485             dx00             = _mm256_sub_pd(ix0,jx0);
 486             dy00             = _mm256_sub_pd(iy0,jy0);
 487             dz00             = _mm256_sub_pd(iz0,jz0);
 488
 489             /* Calculate squared distance and things based on it */
 490             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 491
 492             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 493
 494             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 495
 496             /* Load parameters for j particles */
 497             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 498                                                                  charge+jnrC+0,charge+jnrD+0);
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 505             {
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             qq00             = _mm256_mul_pd(iq0,jq0);
 509
 510             /* REACTION-FIELD ELECTROSTATICS */
 511             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 512
 513             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 518
 519             /* Calculate temporary vectorial force */
 520             tx               = _mm256_mul_pd(fscal,dx00);
 521             ty               = _mm256_mul_pd(fscal,dy00);
 522             tz               = _mm256_mul_pd(fscal,dz00);
 523
 524             /* Update vectorial force */
 525             fix0             = _mm256_add_pd(fix0,tx);
 526             fiy0             = _mm256_add_pd(fiy0,ty);
 527             fiz0             = _mm256_add_pd(fiz0,tz);
 528
 529             fjptrA             = f+j_coord_offsetA;
 530             fjptrB             = f+j_coord_offsetB;
 531             fjptrC             = f+j_coord_offsetC;
 532             fjptrD             = f+j_coord_offsetD;
 533             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 534
 535             }
 536
 537             /* Inner loop uses 30 flops */
 538         }
 539
 540         if(jidx<j_index_end)
 541         {
 542
 543             /* Get j neighbor index, and coordinate index */
 544             jnrlistA         = jjnr[jidx];
 545             jnrlistB         = jjnr[jidx+1];
 546             jnrlistC         = jjnr[jidx+2];
 547             jnrlistD         = jjnr[jidx+3];
 548             /* Sign of each element will be negative for non-real atoms.
 549              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 550              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 551              */
 552             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 553
 554             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 555             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 556             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 557
 558             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 559             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 560             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 561             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 562             j_coord_offsetA  = DIM*jnrA;
 563             j_coord_offsetB  = DIM*jnrB;
 564             j_coord_offsetC  = DIM*jnrC;
 565             j_coord_offsetD  = DIM*jnrD;
 566
 567             /* load j atom coordinates */
 568             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 569                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 570                                                  &jx0,&jy0,&jz0);
 571
 572             /* Calculate displacement vector */
 573             dx00             = _mm256_sub_pd(ix0,jx0);
 574             dy00             = _mm256_sub_pd(iy0,jy0);
 575             dz00             = _mm256_sub_pd(iz0,jz0);
 576
 577             /* Calculate squared distance and things based on it */
 578             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 579
 580             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 581
 582             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 583
 584             /* Load parameters for j particles */
 585             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 586                                                                  charge+jnrC+0,charge+jnrD+0);
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 593             {
 594
 595             /* Compute parameters for interactions between i and j atoms */
 596             qq00             = _mm256_mul_pd(iq0,jq0);
 597
 598             /* REACTION-FIELD ELECTROSTATICS */
 599             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 600
 601             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 606
 607             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm256_mul_pd(fscal,dx00);
 611             ty               = _mm256_mul_pd(fscal,dy00);
 612             tz               = _mm256_mul_pd(fscal,dz00);
 613
 614             /* Update vectorial force */
 615             fix0             = _mm256_add_pd(fix0,tx);
 616             fiy0             = _mm256_add_pd(fiy0,ty);
 617             fiz0             = _mm256_add_pd(fiz0,tz);
 618
 619             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 620             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 621             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 622             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 623             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 624
 625             }
 626
 627             /* Inner loop uses 30 flops */
 628         }
 629
 630         /* End of innermost loop */
 631
 632         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 633                                                  f+i_coord_offset,fshift+i_shift_offset);
 634
 635         /* Increment number of inner iterations */
 636         inneriter                  += j_index_end - j_index_start;
 637
 638         /* Outer loop uses 7 flops */
 639     }
 640
 641     /* Increment number of outer iterations */
 642     outeriter        += nri;
 643
 644     /* Update outer/inner flops */
 645
 646     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 647 }