src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 115     rcutoff_scalar   = fr->rcoulomb;
 116     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 117     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125     j_coord_offsetE = 0;
 126     j_coord_offsetF = 0;
 127     j_coord_offsetG = 0;
 128     j_coord_offsetH = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm256_setzero_ps();
 156         fiy0             = _mm256_setzero_ps();
 157         fiz0             = _mm256_setzero_ps();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm256_setzero_ps();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174             jnrE             = jjnr[jidx+4];
 175             jnrF             = jjnr[jidx+5];
 176             jnrG             = jjnr[jidx+6];
 177             jnrH             = jjnr[jidx+7];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182             j_coord_offsetE  = DIM*jnrE;
 183             j_coord_offsetF  = DIM*jnrF;
 184             j_coord_offsetG  = DIM*jnrG;
 185             j_coord_offsetH  = DIM*jnrH;
 186
 187             /* load j atom coordinates */
 188             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 190                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 191                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 192                                                  &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm256_sub_ps(ix0,jx0);
 196             dy00             = _mm256_sub_ps(iy0,jy0);
 197             dz00             = _mm256_sub_ps(iz0,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 201
 202             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 203
 204             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 208                                                                  charge+jnrC+0,charge+jnrD+0,
 209                                                                  charge+jnrE+0,charge+jnrF+0,
 210                                                                  charge+jnrG+0,charge+jnrH+0);
 211
 212             /**************************
 213              * CALCULATE INTERACTIONS *
 214              **************************/
 215
 216             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 217             {
 218
 219             /* Compute parameters for interactions between i and j atoms */
 220             qq00             = _mm256_mul_ps(iq0,jq0);
 221
 222             /* REACTION-FIELD ELECTROSTATICS */
 223             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 224             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 225
 226             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             velec            = _mm256_and_ps(velec,cutoff_mask);
 230             velecsum         = _mm256_add_ps(velecsum,velec);
 231
 232             fscal            = felec;
 233
 234             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 235
 236             /* Calculate temporary vectorial force */
 237             tx               = _mm256_mul_ps(fscal,dx00);
 238             ty               = _mm256_mul_ps(fscal,dy00);
 239             tz               = _mm256_mul_ps(fscal,dz00);
 240
 241             /* Update vectorial force */
 242             fix0             = _mm256_add_ps(fix0,tx);
 243             fiy0             = _mm256_add_ps(fiy0,ty);
 244             fiz0             = _mm256_add_ps(fiz0,tz);
 245
 246             fjptrA             = f+j_coord_offsetA;
 247             fjptrB             = f+j_coord_offsetB;
 248             fjptrC             = f+j_coord_offsetC;
 249             fjptrD             = f+j_coord_offsetD;
 250             fjptrE             = f+j_coord_offsetE;
 251             fjptrF             = f+j_coord_offsetF;
 252             fjptrG             = f+j_coord_offsetG;
 253             fjptrH             = f+j_coord_offsetH;
 254             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 255
 256             }
 257
 258             /* Inner loop uses 36 flops */
 259         }
 260
 261         if(jidx<j_index_end)
 262         {
 263
 264             /* Get j neighbor index, and coordinate index */
 265             jnrlistA         = jjnr[jidx];
 266             jnrlistB         = jjnr[jidx+1];
 267             jnrlistC         = jjnr[jidx+2];
 268             jnrlistD         = jjnr[jidx+3];
 269             jnrlistE         = jjnr[jidx+4];
 270             jnrlistF         = jjnr[jidx+5];
 271             jnrlistG         = jjnr[jidx+6];
 272             jnrlistH         = jjnr[jidx+7];
 273             /* Sign of each element will be negative for non-real atoms.
 274              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 275              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 276              */
 277             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 278                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 279
 280             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 281             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 282             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 283             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 284             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 285             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 286             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 287             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 288             j_coord_offsetA  = DIM*jnrA;
 289             j_coord_offsetB  = DIM*jnrB;
 290             j_coord_offsetC  = DIM*jnrC;
 291             j_coord_offsetD  = DIM*jnrD;
 292             j_coord_offsetE  = DIM*jnrE;
 293             j_coord_offsetF  = DIM*jnrF;
 294             j_coord_offsetG  = DIM*jnrG;
 295             j_coord_offsetH  = DIM*jnrH;
 296
 297             /* load j atom coordinates */
 298             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 299                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 300                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 301                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 302                                                  &jx0,&jy0,&jz0);
 303
 304             /* Calculate displacement vector */
 305             dx00             = _mm256_sub_ps(ix0,jx0);
 306             dy00             = _mm256_sub_ps(iy0,jy0);
 307             dz00             = _mm256_sub_ps(iz0,jz0);
 308
 309             /* Calculate squared distance and things based on it */
 310             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 311
 312             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 313
 314             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 315
 316             /* Load parameters for j particles */
 317             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 318                                                                  charge+jnrC+0,charge+jnrD+0,
 319                                                                  charge+jnrE+0,charge+jnrF+0,
 320                                                                  charge+jnrG+0,charge+jnrH+0);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 327             {
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq00             = _mm256_mul_ps(iq0,jq0);
 331
 332             /* REACTION-FIELD ELECTROSTATICS */
 333             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 334             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 335
 336             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm256_and_ps(velec,cutoff_mask);
 340             velec            = _mm256_andnot_ps(dummy_mask,velec);
 341             velecsum         = _mm256_add_ps(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 346
 347             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_ps(fscal,dx00);
 351             ty               = _mm256_mul_ps(fscal,dy00);
 352             tz               = _mm256_mul_ps(fscal,dz00);
 353
 354             /* Update vectorial force */
 355             fix0             = _mm256_add_ps(fix0,tx);
 356             fiy0             = _mm256_add_ps(fiy0,ty);
 357             fiz0             = _mm256_add_ps(fiz0,tz);
 358
 359             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 360             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 361             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 362             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 363             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 364             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 365             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 366             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 367             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 368
 369             }
 370
 371             /* Inner loop uses 36 flops */
 372         }
 373
 374         /* End of innermost loop */
 375
 376         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 377                                                  f+i_coord_offset,fshift+i_shift_offset);
 378
 379         ggid                        = gid[iidx];
 380         /* Update potential energies */
 381         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 382
 383         /* Increment number of inner iterations */
 384         inneriter                  += j_index_end - j_index_start;
 385
 386         /* Outer loop uses 8 flops */
 387     }
 388
 389     /* Increment number of outer iterations */
 390     outeriter        += nri;
 391
 392     /* Update outer/inner flops */
 393
 394     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 395 }
 396 /*
 397  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
 398  * Electrostatics interaction: ReactionField
 399  * VdW interaction:            None
 400  * Geometry:                   Particle-Particle
 401  * Calculate force/pot:        Force
 402  */
 403 void
 404 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
 405                     (t_nblist                    * gmx_restrict       nlist,
 406                      rvec                        * gmx_restrict          xx,
 407                      rvec                        * gmx_restrict          ff,
 408                      t_forcerec                  * gmx_restrict          fr,
 409                      t_mdatoms                   * gmx_restrict     mdatoms,
 410                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 411                      t_nrnb                      * gmx_restrict        nrnb)
 412 {
 413     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 414      * just 0 for non-waters.
 415      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 416      * jnr indices corresponding to data put in the four positions in the SIMD register.
 417      */
 418     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 419     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 420     int              jnrA,jnrB,jnrC,jnrD;
 421     int              jnrE,jnrF,jnrG,jnrH;
 422     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 423     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 424     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 425     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 426     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 427     real             rcutoff_scalar;
 428     real             *shiftvec,*fshift,*x,*f;
 429     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 430     real             scratch[4*DIM];
 431     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 432     real *           vdwioffsetptr0;
 433     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 434     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 435     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 436     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 437     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 438     real             *charge;
 439     __m256           dummy_mask,cutoff_mask;
 440     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 441     __m256           one     = _mm256_set1_ps(1.0);
 442     __m256           two     = _mm256_set1_ps(2.0);
 443     x                = xx[0];
 444     f                = ff[0];
 445
 446     nri              = nlist->nri;
 447     iinr             = nlist->iinr;
 448     jindex           = nlist->jindex;
 449     jjnr             = nlist->jjnr;
 450     shiftidx         = nlist->shift;
 451     gid              = nlist->gid;
 452     shiftvec         = fr->shift_vec[0];
 453     fshift           = fr->fshift[0];
 454     facel            = _mm256_set1_ps(fr->epsfac);
 455     charge           = mdatoms->chargeA;
 456     krf              = _mm256_set1_ps(fr->ic->k_rf);
 457     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 458     crf              = _mm256_set1_ps(fr->ic->c_rf);
 459
 460     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 461     rcutoff_scalar   = fr->rcoulomb;
 462     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 463     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 464
 465     /* Avoid stupid compiler warnings */
 466     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 467     j_coord_offsetA = 0;
 468     j_coord_offsetB = 0;
 469     j_coord_offsetC = 0;
 470     j_coord_offsetD = 0;
 471     j_coord_offsetE = 0;
 472     j_coord_offsetF = 0;
 473     j_coord_offsetG = 0;
 474     j_coord_offsetH = 0;
 475
 476     outeriter        = 0;
 477     inneriter        = 0;
 478
 479     for(iidx=0;iidx<4*DIM;iidx++)
 480     {
 481         scratch[iidx] = 0.0;
 482     }
 483
 484     /* Start outer loop over neighborlists */
 485     for(iidx=0; iidx<nri; iidx++)
 486     {
 487         /* Load shift vector for this list */
 488         i_shift_offset   = DIM*shiftidx[iidx];
 489
 490         /* Load limits for loop over neighbors */
 491         j_index_start    = jindex[iidx];
 492         j_index_end      = jindex[iidx+1];
 493
 494         /* Get outer coordinate index */
 495         inr              = iinr[iidx];
 496         i_coord_offset   = DIM*inr;
 497
 498         /* Load i particle coords and add shift vector */
 499         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 500
 501         fix0             = _mm256_setzero_ps();
 502         fiy0             = _mm256_setzero_ps();
 503         fiz0             = _mm256_setzero_ps();
 504
 505         /* Load parameters for i particles */
 506         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 507
 508         /* Start inner kernel loop */
 509         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 510         {
 511
 512             /* Get j neighbor index, and coordinate index */
 513             jnrA             = jjnr[jidx];
 514             jnrB             = jjnr[jidx+1];
 515             jnrC             = jjnr[jidx+2];
 516             jnrD             = jjnr[jidx+3];
 517             jnrE             = jjnr[jidx+4];
 518             jnrF             = jjnr[jidx+5];
 519             jnrG             = jjnr[jidx+6];
 520             jnrH             = jjnr[jidx+7];
 521             j_coord_offsetA  = DIM*jnrA;
 522             j_coord_offsetB  = DIM*jnrB;
 523             j_coord_offsetC  = DIM*jnrC;
 524             j_coord_offsetD  = DIM*jnrD;
 525             j_coord_offsetE  = DIM*jnrE;
 526             j_coord_offsetF  = DIM*jnrF;
 527             j_coord_offsetG  = DIM*jnrG;
 528             j_coord_offsetH  = DIM*jnrH;
 529
 530             /* load j atom coordinates */
 531             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 533                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 534                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 535                                                  &jx0,&jy0,&jz0);
 536
 537             /* Calculate displacement vector */
 538             dx00             = _mm256_sub_ps(ix0,jx0);
 539             dy00             = _mm256_sub_ps(iy0,jy0);
 540             dz00             = _mm256_sub_ps(iz0,jz0);
 541
 542             /* Calculate squared distance and things based on it */
 543             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 544
 545             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 546
 547             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 548
 549             /* Load parameters for j particles */
 550             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 551                                                                  charge+jnrC+0,charge+jnrD+0,
 552                                                                  charge+jnrE+0,charge+jnrF+0,
 553                                                                  charge+jnrG+0,charge+jnrH+0);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 560             {
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq00             = _mm256_mul_ps(iq0,jq0);
 564
 565             /* REACTION-FIELD ELECTROSTATICS */
 566             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 567
 568             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm256_mul_ps(fscal,dx00);
 576             ty               = _mm256_mul_ps(fscal,dy00);
 577             tz               = _mm256_mul_ps(fscal,dz00);
 578
 579             /* Update vectorial force */
 580             fix0             = _mm256_add_ps(fix0,tx);
 581             fiy0             = _mm256_add_ps(fiy0,ty);
 582             fiz0             = _mm256_add_ps(fiz0,tz);
 583
 584             fjptrA             = f+j_coord_offsetA;
 585             fjptrB             = f+j_coord_offsetB;
 586             fjptrC             = f+j_coord_offsetC;
 587             fjptrD             = f+j_coord_offsetD;
 588             fjptrE             = f+j_coord_offsetE;
 589             fjptrF             = f+j_coord_offsetF;
 590             fjptrG             = f+j_coord_offsetG;
 591             fjptrH             = f+j_coord_offsetH;
 592             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 593
 594             }
 595
 596             /* Inner loop uses 30 flops */
 597         }
 598
 599         if(jidx<j_index_end)
 600         {
 601
 602             /* Get j neighbor index, and coordinate index */
 603             jnrlistA         = jjnr[jidx];
 604             jnrlistB         = jjnr[jidx+1];
 605             jnrlistC         = jjnr[jidx+2];
 606             jnrlistD         = jjnr[jidx+3];
 607             jnrlistE         = jjnr[jidx+4];
 608             jnrlistF         = jjnr[jidx+5];
 609             jnrlistG         = jjnr[jidx+6];
 610             jnrlistH         = jjnr[jidx+7];
 611             /* Sign of each element will be negative for non-real atoms.
 612              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 613              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 614              */
 615             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 616                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 617
 618             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 619             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 620             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 621             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 622             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 623             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 624             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 625             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 626             j_coord_offsetA  = DIM*jnrA;
 627             j_coord_offsetB  = DIM*jnrB;
 628             j_coord_offsetC  = DIM*jnrC;
 629             j_coord_offsetD  = DIM*jnrD;
 630             j_coord_offsetE  = DIM*jnrE;
 631             j_coord_offsetF  = DIM*jnrF;
 632             j_coord_offsetG  = DIM*jnrG;
 633             j_coord_offsetH  = DIM*jnrH;
 634
 635             /* load j atom coordinates */
 636             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 637                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 638                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 639                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 640                                                  &jx0,&jy0,&jz0);
 641
 642             /* Calculate displacement vector */
 643             dx00             = _mm256_sub_ps(ix0,jx0);
 644             dy00             = _mm256_sub_ps(iy0,jy0);
 645             dz00             = _mm256_sub_ps(iz0,jz0);
 646
 647             /* Calculate squared distance and things based on it */
 648             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 649
 650             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 651
 652             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 653
 654             /* Load parameters for j particles */
 655             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 656                                                                  charge+jnrC+0,charge+jnrD+0,
 657                                                                  charge+jnrE+0,charge+jnrF+0,
 658                                                                  charge+jnrG+0,charge+jnrH+0);
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 665             {
 666
 667             /* Compute parameters for interactions between i and j atoms */
 668             qq00             = _mm256_mul_ps(iq0,jq0);
 669
 670             /* REACTION-FIELD ELECTROSTATICS */
 671             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 672
 673             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 674
 675             fscal            = felec;
 676
 677             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 678
 679             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm256_mul_ps(fscal,dx00);
 683             ty               = _mm256_mul_ps(fscal,dy00);
 684             tz               = _mm256_mul_ps(fscal,dz00);
 685
 686             /* Update vectorial force */
 687             fix0             = _mm256_add_ps(fix0,tx);
 688             fiy0             = _mm256_add_ps(fiy0,ty);
 689             fiz0             = _mm256_add_ps(fiz0,tz);
 690
 691             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 692             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 693             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 694             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 695             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 696             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 697             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 698             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 699             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 700
 701             }
 702
 703             /* Inner loop uses 30 flops */
 704         }
 705
 706         /* End of innermost loop */
 707
 708         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 709                                                  f+i_coord_offset,fshift+i_shift_offset);
 710
 711         /* Increment number of inner iterations */
 712         inneriter                  += j_index_end - j_index_start;
 713
 714         /* Outer loop uses 7 flops */
 715     }
 716
 717     /* Increment number of outer iterations */
 718     outeriter        += nri;
 719
 720     /* Update outer/inner flops */
 721
 722     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 723 }