src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwNone_GeomP1P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  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              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  88     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m256           dummy_mask,cutoff_mask;
  93     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  94     __m256           one     = _mm256_set1_ps(1.0);
  95     __m256           two     = _mm256_set1_ps(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_ps(fr->ic->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm256_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm256_set1_ps(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->ic->rcoulomb;
 115     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 116     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122     j_coord_offsetC = 0;
 123     j_coord_offsetD = 0;
 124     j_coord_offsetE = 0;
 125     j_coord_offsetF = 0;
 126     j_coord_offsetG = 0;
 127     j_coord_offsetH = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     for(iidx=0;iidx<4*DIM;iidx++)
 133     {
 134         scratch[iidx] = 0.0;
 135     }
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm256_setzero_ps();
 155         fiy0             = _mm256_setzero_ps();
 156         fiz0             = _mm256_setzero_ps();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm256_setzero_ps();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             jnrC             = jjnr[jidx+2];
 172             jnrD             = jjnr[jidx+3];
 173             jnrE             = jjnr[jidx+4];
 174             jnrF             = jjnr[jidx+5];
 175             jnrG             = jjnr[jidx+6];
 176             jnrH             = jjnr[jidx+7];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179             j_coord_offsetC  = DIM*jnrC;
 180             j_coord_offsetD  = DIM*jnrD;
 181             j_coord_offsetE  = DIM*jnrE;
 182             j_coord_offsetF  = DIM*jnrF;
 183             j_coord_offsetG  = DIM*jnrG;
 184             j_coord_offsetH  = DIM*jnrH;
 185
 186             /* load j atom coordinates */
 187             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 189                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 190                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 191                                                  &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm256_sub_ps(ix0,jx0);
 195             dy00             = _mm256_sub_ps(iy0,jy0);
 196             dz00             = _mm256_sub_ps(iz0,jz0);
 197
 198             /* Calculate squared distance and things based on it */
 199             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 200
 201             rinv00           = avx256_invsqrt_f(rsq00);
 202
 203             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 204
 205             /* Load parameters for j particles */
 206             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 207                                                                  charge+jnrC+0,charge+jnrD+0,
 208                                                                  charge+jnrE+0,charge+jnrF+0,
 209                                                                  charge+jnrG+0,charge+jnrH+0);
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 216             {
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm256_mul_ps(iq0,jq0);
 220
 221             /* REACTION-FIELD ELECTROSTATICS */
 222             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 223             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 224
 225             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velec            = _mm256_and_ps(velec,cutoff_mask);
 229             velecsum         = _mm256_add_ps(velecsum,velec);
 230
 231             fscal            = felec;
 232
 233             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 234
 235             /* Calculate temporary vectorial force */
 236             tx               = _mm256_mul_ps(fscal,dx00);
 237             ty               = _mm256_mul_ps(fscal,dy00);
 238             tz               = _mm256_mul_ps(fscal,dz00);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm256_add_ps(fix0,tx);
 242             fiy0             = _mm256_add_ps(fiy0,ty);
 243             fiz0             = _mm256_add_ps(fiz0,tz);
 244
 245             fjptrA             = f+j_coord_offsetA;
 246             fjptrB             = f+j_coord_offsetB;
 247             fjptrC             = f+j_coord_offsetC;
 248             fjptrD             = f+j_coord_offsetD;
 249             fjptrE             = f+j_coord_offsetE;
 250             fjptrF             = f+j_coord_offsetF;
 251             fjptrG             = f+j_coord_offsetG;
 252             fjptrH             = f+j_coord_offsetH;
 253             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 254
 255             }
 256
 257             /* Inner loop uses 36 flops */
 258         }
 259
 260         if(jidx<j_index_end)
 261         {
 262
 263             /* Get j neighbor index, and coordinate index */
 264             jnrlistA         = jjnr[jidx];
 265             jnrlistB         = jjnr[jidx+1];
 266             jnrlistC         = jjnr[jidx+2];
 267             jnrlistD         = jjnr[jidx+3];
 268             jnrlistE         = jjnr[jidx+4];
 269             jnrlistF         = jjnr[jidx+5];
 270             jnrlistG         = jjnr[jidx+6];
 271             jnrlistH         = jjnr[jidx+7];
 272             /* Sign of each element will be negative for non-real atoms.
 273              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 274              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 275              */
 276             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 277                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 278
 279             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 280             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 281             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 282             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 283             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 284             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 285             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 286             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 287             j_coord_offsetA  = DIM*jnrA;
 288             j_coord_offsetB  = DIM*jnrB;
 289             j_coord_offsetC  = DIM*jnrC;
 290             j_coord_offsetD  = DIM*jnrD;
 291             j_coord_offsetE  = DIM*jnrE;
 292             j_coord_offsetF  = DIM*jnrF;
 293             j_coord_offsetG  = DIM*jnrG;
 294             j_coord_offsetH  = DIM*jnrH;
 295
 296             /* load j atom coordinates */
 297             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 298                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 299                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 300                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 301                                                  &jx0,&jy0,&jz0);
 302
 303             /* Calculate displacement vector */
 304             dx00             = _mm256_sub_ps(ix0,jx0);
 305             dy00             = _mm256_sub_ps(iy0,jy0);
 306             dz00             = _mm256_sub_ps(iz0,jz0);
 307
 308             /* Calculate squared distance and things based on it */
 309             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 310
 311             rinv00           = avx256_invsqrt_f(rsq00);
 312
 313             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 314
 315             /* Load parameters for j particles */
 316             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 317                                                                  charge+jnrC+0,charge+jnrD+0,
 318                                                                  charge+jnrE+0,charge+jnrF+0,
 319                                                                  charge+jnrG+0,charge+jnrH+0);
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 326             {
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq00             = _mm256_mul_ps(iq0,jq0);
 330
 331             /* REACTION-FIELD ELECTROSTATICS */
 332             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 333             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 334
 335             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _mm256_and_ps(velec,cutoff_mask);
 339             velec            = _mm256_andnot_ps(dummy_mask,velec);
 340             velecsum         = _mm256_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 345
 346             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm256_mul_ps(fscal,dx00);
 350             ty               = _mm256_mul_ps(fscal,dy00);
 351             tz               = _mm256_mul_ps(fscal,dz00);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm256_add_ps(fix0,tx);
 355             fiy0             = _mm256_add_ps(fiy0,ty);
 356             fiz0             = _mm256_add_ps(fiz0,tz);
 357
 358             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 359             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 360             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 361             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 362             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 363             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 364             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 365             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 366             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 367
 368             }
 369
 370             /* Inner loop uses 36 flops */
 371         }
 372
 373         /* End of innermost loop */
 374
 375         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 376                                                  f+i_coord_offset,fshift+i_shift_offset);
 377
 378         ggid                        = gid[iidx];
 379         /* Update potential energies */
 380         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 381
 382         /* Increment number of inner iterations */
 383         inneriter                  += j_index_end - j_index_start;
 384
 385         /* Outer loop uses 8 flops */
 386     }
 387
 388     /* Increment number of outer iterations */
 389     outeriter        += nri;
 390
 391     /* Update outer/inner flops */
 392
 393     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 394 }
 395 /*
 396  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
 397  * Electrostatics interaction: ReactionField
 398  * VdW interaction:            None
 399  * Geometry:                   Particle-Particle
 400  * Calculate force/pot:        Force
 401  */
 402 void
 403 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
 404                     (t_nblist                    * gmx_restrict       nlist,
 405                      rvec                        * gmx_restrict          xx,
 406                      rvec                        * gmx_restrict          ff,
 407                      struct t_forcerec           * gmx_restrict          fr,
 408                      t_mdatoms                   * gmx_restrict     mdatoms,
 409                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 410                      t_nrnb                      * gmx_restrict        nrnb)
 411 {
 412     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 413      * just 0 for non-waters.
 414      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 415      * jnr indices corresponding to data put in the four positions in the SIMD register.
 416      */
 417     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 418     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 419     int              jnrA,jnrB,jnrC,jnrD;
 420     int              jnrE,jnrF,jnrG,jnrH;
 421     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 422     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 423     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 424     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 425     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 426     real             rcutoff_scalar;
 427     real             *shiftvec,*fshift,*x,*f;
 428     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 429     real             scratch[4*DIM];
 430     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 431     real *           vdwioffsetptr0;
 432     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 433     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 434     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 435     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 436     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 437     real             *charge;
 438     __m256           dummy_mask,cutoff_mask;
 439     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 440     __m256           one     = _mm256_set1_ps(1.0);
 441     __m256           two     = _mm256_set1_ps(2.0);
 442     x                = xx[0];
 443     f                = ff[0];
 444
 445     nri              = nlist->nri;
 446     iinr             = nlist->iinr;
 447     jindex           = nlist->jindex;
 448     jjnr             = nlist->jjnr;
 449     shiftidx         = nlist->shift;
 450     gid              = nlist->gid;
 451     shiftvec         = fr->shift_vec[0];
 452     fshift           = fr->fshift[0];
 453     facel            = _mm256_set1_ps(fr->ic->epsfac);
 454     charge           = mdatoms->chargeA;
 455     krf              = _mm256_set1_ps(fr->ic->k_rf);
 456     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 457     crf              = _mm256_set1_ps(fr->ic->c_rf);
 458
 459     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 460     rcutoff_scalar   = fr->ic->rcoulomb;
 461     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 462     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 463
 464     /* Avoid stupid compiler warnings */
 465     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 466     j_coord_offsetA = 0;
 467     j_coord_offsetB = 0;
 468     j_coord_offsetC = 0;
 469     j_coord_offsetD = 0;
 470     j_coord_offsetE = 0;
 471     j_coord_offsetF = 0;
 472     j_coord_offsetG = 0;
 473     j_coord_offsetH = 0;
 474
 475     outeriter        = 0;
 476     inneriter        = 0;
 477
 478     for(iidx=0;iidx<4*DIM;iidx++)
 479     {
 480         scratch[iidx] = 0.0;
 481     }
 482
 483     /* Start outer loop over neighborlists */
 484     for(iidx=0; iidx<nri; iidx++)
 485     {
 486         /* Load shift vector for this list */
 487         i_shift_offset   = DIM*shiftidx[iidx];
 488
 489         /* Load limits for loop over neighbors */
 490         j_index_start    = jindex[iidx];
 491         j_index_end      = jindex[iidx+1];
 492
 493         /* Get outer coordinate index */
 494         inr              = iinr[iidx];
 495         i_coord_offset   = DIM*inr;
 496
 497         /* Load i particle coords and add shift vector */
 498         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 499
 500         fix0             = _mm256_setzero_ps();
 501         fiy0             = _mm256_setzero_ps();
 502         fiz0             = _mm256_setzero_ps();
 503
 504         /* Load parameters for i particles */
 505         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 506
 507         /* Start inner kernel loop */
 508         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 509         {
 510
 511             /* Get j neighbor index, and coordinate index */
 512             jnrA             = jjnr[jidx];
 513             jnrB             = jjnr[jidx+1];
 514             jnrC             = jjnr[jidx+2];
 515             jnrD             = jjnr[jidx+3];
 516             jnrE             = jjnr[jidx+4];
 517             jnrF             = jjnr[jidx+5];
 518             jnrG             = jjnr[jidx+6];
 519             jnrH             = jjnr[jidx+7];
 520             j_coord_offsetA  = DIM*jnrA;
 521             j_coord_offsetB  = DIM*jnrB;
 522             j_coord_offsetC  = DIM*jnrC;
 523             j_coord_offsetD  = DIM*jnrD;
 524             j_coord_offsetE  = DIM*jnrE;
 525             j_coord_offsetF  = DIM*jnrF;
 526             j_coord_offsetG  = DIM*jnrG;
 527             j_coord_offsetH  = DIM*jnrH;
 528
 529             /* load j atom coordinates */
 530             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 531                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 532                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 533                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 534                                                  &jx0,&jy0,&jz0);
 535
 536             /* Calculate displacement vector */
 537             dx00             = _mm256_sub_ps(ix0,jx0);
 538             dy00             = _mm256_sub_ps(iy0,jy0);
 539             dz00             = _mm256_sub_ps(iz0,jz0);
 540
 541             /* Calculate squared distance and things based on it */
 542             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 543
 544             rinv00           = avx256_invsqrt_f(rsq00);
 545
 546             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 547
 548             /* Load parameters for j particles */
 549             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 550                                                                  charge+jnrC+0,charge+jnrD+0,
 551                                                                  charge+jnrE+0,charge+jnrF+0,
 552                                                                  charge+jnrG+0,charge+jnrH+0);
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 559             {
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq00             = _mm256_mul_ps(iq0,jq0);
 563
 564             /* REACTION-FIELD ELECTROSTATICS */
 565             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 566
 567             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm256_mul_ps(fscal,dx00);
 575             ty               = _mm256_mul_ps(fscal,dy00);
 576             tz               = _mm256_mul_ps(fscal,dz00);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm256_add_ps(fix0,tx);
 580             fiy0             = _mm256_add_ps(fiy0,ty);
 581             fiz0             = _mm256_add_ps(fiz0,tz);
 582
 583             fjptrA             = f+j_coord_offsetA;
 584             fjptrB             = f+j_coord_offsetB;
 585             fjptrC             = f+j_coord_offsetC;
 586             fjptrD             = f+j_coord_offsetD;
 587             fjptrE             = f+j_coord_offsetE;
 588             fjptrF             = f+j_coord_offsetF;
 589             fjptrG             = f+j_coord_offsetG;
 590             fjptrH             = f+j_coord_offsetH;
 591             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 592
 593             }
 594
 595             /* Inner loop uses 30 flops */
 596         }
 597
 598         if(jidx<j_index_end)
 599         {
 600
 601             /* Get j neighbor index, and coordinate index */
 602             jnrlistA         = jjnr[jidx];
 603             jnrlistB         = jjnr[jidx+1];
 604             jnrlistC         = jjnr[jidx+2];
 605             jnrlistD         = jjnr[jidx+3];
 606             jnrlistE         = jjnr[jidx+4];
 607             jnrlistF         = jjnr[jidx+5];
 608             jnrlistG         = jjnr[jidx+6];
 609             jnrlistH         = jjnr[jidx+7];
 610             /* Sign of each element will be negative for non-real atoms.
 611              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 612              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 613              */
 614             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 615                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 616
 617             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 618             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 619             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 620             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 621             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 622             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 623             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 624             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 625             j_coord_offsetA  = DIM*jnrA;
 626             j_coord_offsetB  = DIM*jnrB;
 627             j_coord_offsetC  = DIM*jnrC;
 628             j_coord_offsetD  = DIM*jnrD;
 629             j_coord_offsetE  = DIM*jnrE;
 630             j_coord_offsetF  = DIM*jnrF;
 631             j_coord_offsetG  = DIM*jnrG;
 632             j_coord_offsetH  = DIM*jnrH;
 633
 634             /* load j atom coordinates */
 635             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 636                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 637                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 638                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 639                                                  &jx0,&jy0,&jz0);
 640
 641             /* Calculate displacement vector */
 642             dx00             = _mm256_sub_ps(ix0,jx0);
 643             dy00             = _mm256_sub_ps(iy0,jy0);
 644             dz00             = _mm256_sub_ps(iz0,jz0);
 645
 646             /* Calculate squared distance and things based on it */
 647             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 648
 649             rinv00           = avx256_invsqrt_f(rsq00);
 650
 651             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 652
 653             /* Load parameters for j particles */
 654             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 655                                                                  charge+jnrC+0,charge+jnrD+0,
 656                                                                  charge+jnrE+0,charge+jnrF+0,
 657                                                                  charge+jnrG+0,charge+jnrH+0);
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 664             {
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq00             = _mm256_mul_ps(iq0,jq0);
 668
 669             /* REACTION-FIELD ELECTROSTATICS */
 670             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 671
 672             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 673
 674             fscal            = felec;
 675
 676             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 677
 678             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 679
 680             /* Calculate temporary vectorial force */
 681             tx               = _mm256_mul_ps(fscal,dx00);
 682             ty               = _mm256_mul_ps(fscal,dy00);
 683             tz               = _mm256_mul_ps(fscal,dz00);
 684
 685             /* Update vectorial force */
 686             fix0             = _mm256_add_ps(fix0,tx);
 687             fiy0             = _mm256_add_ps(fiy0,ty);
 688             fiz0             = _mm256_add_ps(fiz0,tz);
 689
 690             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 691             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 692             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 693             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 694             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 695             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 696             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 697             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 698             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 699
 700             }
 701
 702             /* Inner loop uses 30 flops */
 703         }
 704
 705         /* End of innermost loop */
 706
 707         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 708                                                  f+i_coord_offset,fshift+i_shift_offset);
 709
 710         /* Increment number of inner iterations */
 711         inneriter                  += j_index_end - j_index_start;
 712
 713         /* Outer loop uses 7 flops */
 714     }
 715
 716     /* Increment number of outer iterations */
 717     outeriter        += nri;
 718
 719     /* Update outer/inner flops */
 720
 721     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 722 }