src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwNone_GeomP1P1_avx_256_single.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_single 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_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_256_single
  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_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m256           dummy_mask,cutoff_mask;
  94     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  95     __m256           one     = _mm256_set1_ps(1.0);
  96     __m256           two     = _mm256_set1_ps(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm256_set1_ps(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     krf              = _mm256_set1_ps(fr->ic->k_rf);
 111     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 112     crf              = _mm256_set1_ps(fr->ic->c_rf);
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120     j_coord_offsetE = 0;
 121     j_coord_offsetF = 0;
 122     j_coord_offsetG = 0;
 123     j_coord_offsetH = 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_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 149
 150         fix0             = _mm256_setzero_ps();
 151         fiy0             = _mm256_setzero_ps();
 152         fiz0             = _mm256_setzero_ps();
 153
 154         /* Load parameters for i particles */
 155         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm256_setzero_ps();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 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             jnrE             = jjnr[jidx+4];
 170             jnrF             = jjnr[jidx+5];
 171             jnrG             = jjnr[jidx+6];
 172             jnrH             = jjnr[jidx+7];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175             j_coord_offsetC  = DIM*jnrC;
 176             j_coord_offsetD  = DIM*jnrD;
 177             j_coord_offsetE  = DIM*jnrE;
 178             j_coord_offsetF  = DIM*jnrF;
 179             j_coord_offsetG  = DIM*jnrG;
 180             j_coord_offsetH  = DIM*jnrH;
 181
 182             /* load j atom coordinates */
 183             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 184                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 185                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 186                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 187                                                  &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm256_sub_ps(ix0,jx0);
 191             dy00             = _mm256_sub_ps(iy0,jy0);
 192             dz00             = _mm256_sub_ps(iz0,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 196
 197             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 198
 199             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 203                                                                  charge+jnrC+0,charge+jnrD+0,
 204                                                                  charge+jnrE+0,charge+jnrF+0,
 205                                                                  charge+jnrG+0,charge+jnrH+0);
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             /* Compute parameters for interactions between i and j atoms */
 212             qq00             = _mm256_mul_ps(iq0,jq0);
 213
 214             /* REACTION-FIELD ELECTROSTATICS */
 215             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 216             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 217
 218             /* Update potential sum for this i atom from the interaction with this j atom. */
 219             velecsum         = _mm256_add_ps(velecsum,velec);
 220
 221             fscal            = felec;
 222
 223             /* Calculate temporary vectorial force */
 224             tx               = _mm256_mul_ps(fscal,dx00);
 225             ty               = _mm256_mul_ps(fscal,dy00);
 226             tz               = _mm256_mul_ps(fscal,dz00);
 227
 228             /* Update vectorial force */
 229             fix0             = _mm256_add_ps(fix0,tx);
 230             fiy0             = _mm256_add_ps(fiy0,ty);
 231             fiz0             = _mm256_add_ps(fiz0,tz);
 232
 233             fjptrA             = f+j_coord_offsetA;
 234             fjptrB             = f+j_coord_offsetB;
 235             fjptrC             = f+j_coord_offsetC;
 236             fjptrD             = f+j_coord_offsetD;
 237             fjptrE             = f+j_coord_offsetE;
 238             fjptrF             = f+j_coord_offsetF;
 239             fjptrG             = f+j_coord_offsetG;
 240             fjptrH             = f+j_coord_offsetH;
 241             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 242
 243             /* Inner loop uses 32 flops */
 244         }
 245
 246         if(jidx<j_index_end)
 247         {
 248
 249             /* Get j neighbor index, and coordinate index */
 250             jnrlistA         = jjnr[jidx];
 251             jnrlistB         = jjnr[jidx+1];
 252             jnrlistC         = jjnr[jidx+2];
 253             jnrlistD         = jjnr[jidx+3];
 254             jnrlistE         = jjnr[jidx+4];
 255             jnrlistF         = jjnr[jidx+5];
 256             jnrlistG         = jjnr[jidx+6];
 257             jnrlistH         = jjnr[jidx+7];
 258             /* Sign of each element will be negative for non-real atoms.
 259              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 260              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 261              */
 262             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 263                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 264
 265             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 266             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 267             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 268             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 269             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 270             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 271             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 272             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 273             j_coord_offsetA  = DIM*jnrA;
 274             j_coord_offsetB  = DIM*jnrB;
 275             j_coord_offsetC  = DIM*jnrC;
 276             j_coord_offsetD  = DIM*jnrD;
 277             j_coord_offsetE  = DIM*jnrE;
 278             j_coord_offsetF  = DIM*jnrF;
 279             j_coord_offsetG  = DIM*jnrG;
 280             j_coord_offsetH  = DIM*jnrH;
 281
 282             /* load j atom coordinates */
 283             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 284                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 285                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 286                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 287                                                  &jx0,&jy0,&jz0);
 288
 289             /* Calculate displacement vector */
 290             dx00             = _mm256_sub_ps(ix0,jx0);
 291             dy00             = _mm256_sub_ps(iy0,jy0);
 292             dz00             = _mm256_sub_ps(iz0,jz0);
 293
 294             /* Calculate squared distance and things based on it */
 295             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 296
 297             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 298
 299             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 300
 301             /* Load parameters for j particles */
 302             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 303                                                                  charge+jnrC+0,charge+jnrD+0,
 304                                                                  charge+jnrE+0,charge+jnrF+0,
 305                                                                  charge+jnrG+0,charge+jnrH+0);
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq00             = _mm256_mul_ps(iq0,jq0);
 313
 314             /* REACTION-FIELD ELECTROSTATICS */
 315             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 316             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm256_andnot_ps(dummy_mask,velec);
 320             velecsum         = _mm256_add_ps(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm256_mul_ps(fscal,dx00);
 328             ty               = _mm256_mul_ps(fscal,dy00);
 329             tz               = _mm256_mul_ps(fscal,dz00);
 330
 331             /* Update vectorial force */
 332             fix0             = _mm256_add_ps(fix0,tx);
 333             fiy0             = _mm256_add_ps(fiy0,ty);
 334             fiz0             = _mm256_add_ps(fiz0,tz);
 335
 336             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 337             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 338             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 339             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 340             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 341             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 342             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 343             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 344             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 345
 346             /* Inner loop uses 32 flops */
 347         }
 348
 349         /* End of innermost loop */
 350
 351         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 352                                                  f+i_coord_offset,fshift+i_shift_offset);
 353
 354         ggid                        = gid[iidx];
 355         /* Update potential energies */
 356         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 357
 358         /* Increment number of inner iterations */
 359         inneriter                  += j_index_end - j_index_start;
 360
 361         /* Outer loop uses 8 flops */
 362     }
 363
 364     /* Increment number of outer iterations */
 365     outeriter        += nri;
 366
 367     /* Update outer/inner flops */
 368
 369     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 370 }
 371 /*
 372  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_single
 373  * Electrostatics interaction: ReactionField
 374  * VdW interaction:            None
 375  * Geometry:                   Particle-Particle
 376  * Calculate force/pot:        Force
 377  */
 378 void
 379 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_256_single
 380                     (t_nblist                    * gmx_restrict       nlist,
 381                      rvec                        * gmx_restrict          xx,
 382                      rvec                        * gmx_restrict          ff,
 383                      t_forcerec                  * gmx_restrict          fr,
 384                      t_mdatoms                   * gmx_restrict     mdatoms,
 385                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 386                      t_nrnb                      * gmx_restrict        nrnb)
 387 {
 388     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 389      * just 0 for non-waters.
 390      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 391      * jnr indices corresponding to data put in the four positions in the SIMD register.
 392      */
 393     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 394     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 395     int              jnrA,jnrB,jnrC,jnrD;
 396     int              jnrE,jnrF,jnrG,jnrH;
 397     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 398     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 399     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 400     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 401     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 402     real             rcutoff_scalar;
 403     real             *shiftvec,*fshift,*x,*f;
 404     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 405     real             scratch[4*DIM];
 406     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 407     real *           vdwioffsetptr0;
 408     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 409     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 410     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 411     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 412     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 413     real             *charge;
 414     __m256           dummy_mask,cutoff_mask;
 415     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 416     __m256           one     = _mm256_set1_ps(1.0);
 417     __m256           two     = _mm256_set1_ps(2.0);
 418     x                = xx[0];
 419     f                = ff[0];
 420
 421     nri              = nlist->nri;
 422     iinr             = nlist->iinr;
 423     jindex           = nlist->jindex;
 424     jjnr             = nlist->jjnr;
 425     shiftidx         = nlist->shift;
 426     gid              = nlist->gid;
 427     shiftvec         = fr->shift_vec[0];
 428     fshift           = fr->fshift[0];
 429     facel            = _mm256_set1_ps(fr->epsfac);
 430     charge           = mdatoms->chargeA;
 431     krf              = _mm256_set1_ps(fr->ic->k_rf);
 432     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 433     crf              = _mm256_set1_ps(fr->ic->c_rf);
 434
 435     /* Avoid stupid compiler warnings */
 436     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 437     j_coord_offsetA = 0;
 438     j_coord_offsetB = 0;
 439     j_coord_offsetC = 0;
 440     j_coord_offsetD = 0;
 441     j_coord_offsetE = 0;
 442     j_coord_offsetF = 0;
 443     j_coord_offsetG = 0;
 444     j_coord_offsetH = 0;
 445
 446     outeriter        = 0;
 447     inneriter        = 0;
 448
 449     for(iidx=0;iidx<4*DIM;iidx++)
 450     {
 451         scratch[iidx] = 0.0;
 452     }
 453
 454     /* Start outer loop over neighborlists */
 455     for(iidx=0; iidx<nri; iidx++)
 456     {
 457         /* Load shift vector for this list */
 458         i_shift_offset   = DIM*shiftidx[iidx];
 459
 460         /* Load limits for loop over neighbors */
 461         j_index_start    = jindex[iidx];
 462         j_index_end      = jindex[iidx+1];
 463
 464         /* Get outer coordinate index */
 465         inr              = iinr[iidx];
 466         i_coord_offset   = DIM*inr;
 467
 468         /* Load i particle coords and add shift vector */
 469         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 470
 471         fix0             = _mm256_setzero_ps();
 472         fiy0             = _mm256_setzero_ps();
 473         fiz0             = _mm256_setzero_ps();
 474
 475         /* Load parameters for i particles */
 476         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 477
 478         /* Start inner kernel loop */
 479         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 480         {
 481
 482             /* Get j neighbor index, and coordinate index */
 483             jnrA             = jjnr[jidx];
 484             jnrB             = jjnr[jidx+1];
 485             jnrC             = jjnr[jidx+2];
 486             jnrD             = jjnr[jidx+3];
 487             jnrE             = jjnr[jidx+4];
 488             jnrF             = jjnr[jidx+5];
 489             jnrG             = jjnr[jidx+6];
 490             jnrH             = jjnr[jidx+7];
 491             j_coord_offsetA  = DIM*jnrA;
 492             j_coord_offsetB  = DIM*jnrB;
 493             j_coord_offsetC  = DIM*jnrC;
 494             j_coord_offsetD  = DIM*jnrD;
 495             j_coord_offsetE  = DIM*jnrE;
 496             j_coord_offsetF  = DIM*jnrF;
 497             j_coord_offsetG  = DIM*jnrG;
 498             j_coord_offsetH  = DIM*jnrH;
 499
 500             /* load j atom coordinates */
 501             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 502                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 503                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 504                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 505                                                  &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm256_sub_ps(ix0,jx0);
 509             dy00             = _mm256_sub_ps(iy0,jy0);
 510             dz00             = _mm256_sub_ps(iz0,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 514
 515             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 516
 517             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 521                                                                  charge+jnrC+0,charge+jnrD+0,
 522                                                                  charge+jnrE+0,charge+jnrF+0,
 523                                                                  charge+jnrG+0,charge+jnrH+0);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq00             = _mm256_mul_ps(iq0,jq0);
 531
 532             /* REACTION-FIELD ELECTROSTATICS */
 533             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 534
 535             fscal            = felec;
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm256_mul_ps(fscal,dx00);
 539             ty               = _mm256_mul_ps(fscal,dy00);
 540             tz               = _mm256_mul_ps(fscal,dz00);
 541
 542             /* Update vectorial force */
 543             fix0             = _mm256_add_ps(fix0,tx);
 544             fiy0             = _mm256_add_ps(fiy0,ty);
 545             fiz0             = _mm256_add_ps(fiz0,tz);
 546
 547             fjptrA             = f+j_coord_offsetA;
 548             fjptrB             = f+j_coord_offsetB;
 549             fjptrC             = f+j_coord_offsetC;
 550             fjptrD             = f+j_coord_offsetD;
 551             fjptrE             = f+j_coord_offsetE;
 552             fjptrF             = f+j_coord_offsetF;
 553             fjptrG             = f+j_coord_offsetG;
 554             fjptrH             = f+j_coord_offsetH;
 555             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 556
 557             /* Inner loop uses 27 flops */
 558         }
 559
 560         if(jidx<j_index_end)
 561         {
 562
 563             /* Get j neighbor index, and coordinate index */
 564             jnrlistA         = jjnr[jidx];
 565             jnrlistB         = jjnr[jidx+1];
 566             jnrlistC         = jjnr[jidx+2];
 567             jnrlistD         = jjnr[jidx+3];
 568             jnrlistE         = jjnr[jidx+4];
 569             jnrlistF         = jjnr[jidx+5];
 570             jnrlistG         = jjnr[jidx+6];
 571             jnrlistH         = jjnr[jidx+7];
 572             /* Sign of each element will be negative for non-real atoms.
 573              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 574              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 575              */
 576             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 577                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 578
 579             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 580             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 581             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 582             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 583             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 584             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 585             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 586             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 587             j_coord_offsetA  = DIM*jnrA;
 588             j_coord_offsetB  = DIM*jnrB;
 589             j_coord_offsetC  = DIM*jnrC;
 590             j_coord_offsetD  = DIM*jnrD;
 591             j_coord_offsetE  = DIM*jnrE;
 592             j_coord_offsetF  = DIM*jnrF;
 593             j_coord_offsetG  = DIM*jnrG;
 594             j_coord_offsetH  = DIM*jnrH;
 595
 596             /* load j atom coordinates */
 597             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 598                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 599                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 600                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 601                                                  &jx0,&jy0,&jz0);
 602
 603             /* Calculate displacement vector */
 604             dx00             = _mm256_sub_ps(ix0,jx0);
 605             dy00             = _mm256_sub_ps(iy0,jy0);
 606             dz00             = _mm256_sub_ps(iz0,jz0);
 607
 608             /* Calculate squared distance and things based on it */
 609             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 610
 611             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 612
 613             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 614
 615             /* Load parameters for j particles */
 616             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 617                                                                  charge+jnrC+0,charge+jnrD+0,
 618                                                                  charge+jnrE+0,charge+jnrF+0,
 619                                                                  charge+jnrG+0,charge+jnrH+0);
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq00             = _mm256_mul_ps(iq0,jq0);
 627
 628             /* REACTION-FIELD ELECTROSTATICS */
 629             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm256_mul_ps(fscal,dx00);
 637             ty               = _mm256_mul_ps(fscal,dy00);
 638             tz               = _mm256_mul_ps(fscal,dz00);
 639
 640             /* Update vectorial force */
 641             fix0             = _mm256_add_ps(fix0,tx);
 642             fiy0             = _mm256_add_ps(fiy0,ty);
 643             fiz0             = _mm256_add_ps(fiz0,tz);
 644
 645             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 646             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 647             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 648             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 649             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 650             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 651             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 652             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 653             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 654
 655             /* Inner loop uses 27 flops */
 656         }
 657
 658         /* End of innermost loop */
 659
 660         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 661                                                  f+i_coord_offset,fshift+i_shift_offset);
 662
 663         /* Increment number of inner iterations */
 664         inneriter                  += j_index_end - j_index_start;
 665
 666         /* Outer loop uses 7 flops */
 667     }
 668
 669     /* Increment number of outer iterations */
 670     outeriter        += nri;
 671
 672     /* Update outer/inner flops */
 673
 674     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 675 }