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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m256           dummy_mask,cutoff_mask;
  96     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  97     __m256           one     = _mm256_set1_ps(1.0);
  98     __m256           two     = _mm256_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm256_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm256_set1_ps(fr->ic->k_rf);
 113     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 114     crf              = _mm256_set1_ps(fr->ic->c_rf);
 115
 116     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 117     rcutoff_scalar   = fr->rcoulomb;
 118     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 119     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127     j_coord_offsetE = 0;
 128     j_coord_offsetF = 0;
 129     j_coord_offsetG = 0;
 130     j_coord_offsetH = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     for(iidx=0;iidx<4*DIM;iidx++)
 136     {
 137         scratch[iidx] = 0.0;
 138     }
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 156
 157         fix0             = _mm256_setzero_ps();
 158         fiy0             = _mm256_setzero_ps();
 159         fiz0             = _mm256_setzero_ps();
 160
 161         /* Load parameters for i particles */
 162         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm256_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             jnrE             = jjnr[jidx+4];
 177             jnrF             = jjnr[jidx+5];
 178             jnrG             = jjnr[jidx+6];
 179             jnrH             = jjnr[jidx+7];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184             j_coord_offsetE  = DIM*jnrE;
 185             j_coord_offsetF  = DIM*jnrF;
 186             j_coord_offsetG  = DIM*jnrG;
 187             j_coord_offsetH  = DIM*jnrH;
 188
 189             /* load j atom coordinates */
 190             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 192                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 193                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 194                                                  &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm256_sub_ps(ix0,jx0);
 198             dy00             = _mm256_sub_ps(iy0,jy0);
 199             dz00             = _mm256_sub_ps(iz0,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 203
 204             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 205
 206             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 210                                                                  charge+jnrC+0,charge+jnrD+0,
 211                                                                  charge+jnrE+0,charge+jnrF+0,
 212                                                                  charge+jnrG+0,charge+jnrH+0);
 213
 214             /**************************
 215              * CALCULATE INTERACTIONS *
 216              **************************/
 217
 218             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 219             {
 220
 221             /* Compute parameters for interactions between i and j atoms */
 222             qq00             = _mm256_mul_ps(iq0,jq0);
 223
 224             /* REACTION-FIELD ELECTROSTATICS */
 225             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 226             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 227
 228             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             velec            = _mm256_and_ps(velec,cutoff_mask);
 232             velecsum         = _mm256_add_ps(velecsum,velec);
 233
 234             fscal            = felec;
 235
 236             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 237
 238             /* Calculate temporary vectorial force */
 239             tx               = _mm256_mul_ps(fscal,dx00);
 240             ty               = _mm256_mul_ps(fscal,dy00);
 241             tz               = _mm256_mul_ps(fscal,dz00);
 242
 243             /* Update vectorial force */
 244             fix0             = _mm256_add_ps(fix0,tx);
 245             fiy0             = _mm256_add_ps(fiy0,ty);
 246             fiz0             = _mm256_add_ps(fiz0,tz);
 247
 248             fjptrA             = f+j_coord_offsetA;
 249             fjptrB             = f+j_coord_offsetB;
 250             fjptrC             = f+j_coord_offsetC;
 251             fjptrD             = f+j_coord_offsetD;
 252             fjptrE             = f+j_coord_offsetE;
 253             fjptrF             = f+j_coord_offsetF;
 254             fjptrG             = f+j_coord_offsetG;
 255             fjptrH             = f+j_coord_offsetH;
 256             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 257
 258             }
 259
 260             /* Inner loop uses 36 flops */
 261         }
 262
 263         if(jidx<j_index_end)
 264         {
 265
 266             /* Get j neighbor index, and coordinate index */
 267             jnrlistA         = jjnr[jidx];
 268             jnrlistB         = jjnr[jidx+1];
 269             jnrlistC         = jjnr[jidx+2];
 270             jnrlistD         = jjnr[jidx+3];
 271             jnrlistE         = jjnr[jidx+4];
 272             jnrlistF         = jjnr[jidx+5];
 273             jnrlistG         = jjnr[jidx+6];
 274             jnrlistH         = jjnr[jidx+7];
 275             /* Sign of each element will be negative for non-real atoms.
 276              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 277              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 278              */
 279             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 280                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 281
 282             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 283             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 284             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 285             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 286             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 287             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 288             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 289             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 290             j_coord_offsetA  = DIM*jnrA;
 291             j_coord_offsetB  = DIM*jnrB;
 292             j_coord_offsetC  = DIM*jnrC;
 293             j_coord_offsetD  = DIM*jnrD;
 294             j_coord_offsetE  = DIM*jnrE;
 295             j_coord_offsetF  = DIM*jnrF;
 296             j_coord_offsetG  = DIM*jnrG;
 297             j_coord_offsetH  = DIM*jnrH;
 298
 299             /* load j atom coordinates */
 300             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 301                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 302                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 303                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 304                                                  &jx0,&jy0,&jz0);
 305
 306             /* Calculate displacement vector */
 307             dx00             = _mm256_sub_ps(ix0,jx0);
 308             dy00             = _mm256_sub_ps(iy0,jy0);
 309             dz00             = _mm256_sub_ps(iz0,jz0);
 310
 311             /* Calculate squared distance and things based on it */
 312             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 313
 314             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 315
 316             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 317
 318             /* Load parameters for j particles */
 319             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 320                                                                  charge+jnrC+0,charge+jnrD+0,
 321                                                                  charge+jnrE+0,charge+jnrF+0,
 322                                                                  charge+jnrG+0,charge+jnrH+0);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 329             {
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq00             = _mm256_mul_ps(iq0,jq0);
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 336             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 337
 338             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm256_and_ps(velec,cutoff_mask);
 342             velec            = _mm256_andnot_ps(dummy_mask,velec);
 343             velecsum         = _mm256_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 348
 349             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_ps(fscal,dx00);
 353             ty               = _mm256_mul_ps(fscal,dy00);
 354             tz               = _mm256_mul_ps(fscal,dz00);
 355
 356             /* Update vectorial force */
 357             fix0             = _mm256_add_ps(fix0,tx);
 358             fiy0             = _mm256_add_ps(fiy0,ty);
 359             fiz0             = _mm256_add_ps(fiz0,tz);
 360
 361             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 362             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 363             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 364             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 365             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 366             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 367             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 368             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 369             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 370
 371             }
 372
 373             /* Inner loop uses 36 flops */
 374         }
 375
 376         /* End of innermost loop */
 377
 378         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 379                                                  f+i_coord_offset,fshift+i_shift_offset);
 380
 381         ggid                        = gid[iidx];
 382         /* Update potential energies */
 383         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 384
 385         /* Increment number of inner iterations */
 386         inneriter                  += j_index_end - j_index_start;
 387
 388         /* Outer loop uses 8 flops */
 389     }
 390
 391     /* Increment number of outer iterations */
 392     outeriter        += nri;
 393
 394     /* Update outer/inner flops */
 395
 396     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 397 }
 398 /*
 399  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
 400  * Electrostatics interaction: ReactionField
 401  * VdW interaction:            None
 402  * Geometry:                   Particle-Particle
 403  * Calculate force/pot:        Force
 404  */
 405 void
 406 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
 407                     (t_nblist                    * gmx_restrict       nlist,
 408                      rvec                        * gmx_restrict          xx,
 409                      rvec                        * gmx_restrict          ff,
 410                      t_forcerec                  * gmx_restrict          fr,
 411                      t_mdatoms                   * gmx_restrict     mdatoms,
 412                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 413                      t_nrnb                      * gmx_restrict        nrnb)
 414 {
 415     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 416      * just 0 for non-waters.
 417      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 418      * jnr indices corresponding to data put in the four positions in the SIMD register.
 419      */
 420     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 421     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 422     int              jnrA,jnrB,jnrC,jnrD;
 423     int              jnrE,jnrF,jnrG,jnrH;
 424     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 425     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 426     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 427     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 428     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 429     real             rcutoff_scalar;
 430     real             *shiftvec,*fshift,*x,*f;
 431     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 432     real             scratch[4*DIM];
 433     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 434     real *           vdwioffsetptr0;
 435     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 436     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 437     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 438     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 439     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 440     real             *charge;
 441     __m256           dummy_mask,cutoff_mask;
 442     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 443     __m256           one     = _mm256_set1_ps(1.0);
 444     __m256           two     = _mm256_set1_ps(2.0);
 445     x                = xx[0];
 446     f                = ff[0];
 447
 448     nri              = nlist->nri;
 449     iinr             = nlist->iinr;
 450     jindex           = nlist->jindex;
 451     jjnr             = nlist->jjnr;
 452     shiftidx         = nlist->shift;
 453     gid              = nlist->gid;
 454     shiftvec         = fr->shift_vec[0];
 455     fshift           = fr->fshift[0];
 456     facel            = _mm256_set1_ps(fr->epsfac);
 457     charge           = mdatoms->chargeA;
 458     krf              = _mm256_set1_ps(fr->ic->k_rf);
 459     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 460     crf              = _mm256_set1_ps(fr->ic->c_rf);
 461
 462     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 463     rcutoff_scalar   = fr->rcoulomb;
 464     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 465     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 466
 467     /* Avoid stupid compiler warnings */
 468     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 469     j_coord_offsetA = 0;
 470     j_coord_offsetB = 0;
 471     j_coord_offsetC = 0;
 472     j_coord_offsetD = 0;
 473     j_coord_offsetE = 0;
 474     j_coord_offsetF = 0;
 475     j_coord_offsetG = 0;
 476     j_coord_offsetH = 0;
 477
 478     outeriter        = 0;
 479     inneriter        = 0;
 480
 481     for(iidx=0;iidx<4*DIM;iidx++)
 482     {
 483         scratch[iidx] = 0.0;
 484     }
 485
 486     /* Start outer loop over neighborlists */
 487     for(iidx=0; iidx<nri; iidx++)
 488     {
 489         /* Load shift vector for this list */
 490         i_shift_offset   = DIM*shiftidx[iidx];
 491
 492         /* Load limits for loop over neighbors */
 493         j_index_start    = jindex[iidx];
 494         j_index_end      = jindex[iidx+1];
 495
 496         /* Get outer coordinate index */
 497         inr              = iinr[iidx];
 498         i_coord_offset   = DIM*inr;
 499
 500         /* Load i particle coords and add shift vector */
 501         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 502
 503         fix0             = _mm256_setzero_ps();
 504         fiy0             = _mm256_setzero_ps();
 505         fiz0             = _mm256_setzero_ps();
 506
 507         /* Load parameters for i particles */
 508         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 509
 510         /* Start inner kernel loop */
 511         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 512         {
 513
 514             /* Get j neighbor index, and coordinate index */
 515             jnrA             = jjnr[jidx];
 516             jnrB             = jjnr[jidx+1];
 517             jnrC             = jjnr[jidx+2];
 518             jnrD             = jjnr[jidx+3];
 519             jnrE             = jjnr[jidx+4];
 520             jnrF             = jjnr[jidx+5];
 521             jnrG             = jjnr[jidx+6];
 522             jnrH             = jjnr[jidx+7];
 523             j_coord_offsetA  = DIM*jnrA;
 524             j_coord_offsetB  = DIM*jnrB;
 525             j_coord_offsetC  = DIM*jnrC;
 526             j_coord_offsetD  = DIM*jnrD;
 527             j_coord_offsetE  = DIM*jnrE;
 528             j_coord_offsetF  = DIM*jnrF;
 529             j_coord_offsetG  = DIM*jnrG;
 530             j_coord_offsetH  = DIM*jnrH;
 531
 532             /* load j atom coordinates */
 533             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 534                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 535                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 536                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 537                                                  &jx0,&jy0,&jz0);
 538
 539             /* Calculate displacement vector */
 540             dx00             = _mm256_sub_ps(ix0,jx0);
 541             dy00             = _mm256_sub_ps(iy0,jy0);
 542             dz00             = _mm256_sub_ps(iz0,jz0);
 543
 544             /* Calculate squared distance and things based on it */
 545             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 546
 547             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 548
 549             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 550
 551             /* Load parameters for j particles */
 552             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 553                                                                  charge+jnrC+0,charge+jnrD+0,
 554                                                                  charge+jnrE+0,charge+jnrF+0,
 555                                                                  charge+jnrG+0,charge+jnrH+0);
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 562             {
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq00             = _mm256_mul_ps(iq0,jq0);
 566
 567             /* REACTION-FIELD ELECTROSTATICS */
 568             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 569
 570             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 575
 576             /* Calculate temporary vectorial force */
 577             tx               = _mm256_mul_ps(fscal,dx00);
 578             ty               = _mm256_mul_ps(fscal,dy00);
 579             tz               = _mm256_mul_ps(fscal,dz00);
 580
 581             /* Update vectorial force */
 582             fix0             = _mm256_add_ps(fix0,tx);
 583             fiy0             = _mm256_add_ps(fiy0,ty);
 584             fiz0             = _mm256_add_ps(fiz0,tz);
 585
 586             fjptrA             = f+j_coord_offsetA;
 587             fjptrB             = f+j_coord_offsetB;
 588             fjptrC             = f+j_coord_offsetC;
 589             fjptrD             = f+j_coord_offsetD;
 590             fjptrE             = f+j_coord_offsetE;
 591             fjptrF             = f+j_coord_offsetF;
 592             fjptrG             = f+j_coord_offsetG;
 593             fjptrH             = f+j_coord_offsetH;
 594             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 595
 596             }
 597
 598             /* Inner loop uses 30 flops */
 599         }
 600
 601         if(jidx<j_index_end)
 602         {
 603
 604             /* Get j neighbor index, and coordinate index */
 605             jnrlistA         = jjnr[jidx];
 606             jnrlistB         = jjnr[jidx+1];
 607             jnrlistC         = jjnr[jidx+2];
 608             jnrlistD         = jjnr[jidx+3];
 609             jnrlistE         = jjnr[jidx+4];
 610             jnrlistF         = jjnr[jidx+5];
 611             jnrlistG         = jjnr[jidx+6];
 612             jnrlistH         = jjnr[jidx+7];
 613             /* Sign of each element will be negative for non-real atoms.
 614              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 615              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 616              */
 617             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 618                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 619
 620             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 621             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 622             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 623             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 624             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 625             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 626             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 627             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 628             j_coord_offsetA  = DIM*jnrA;
 629             j_coord_offsetB  = DIM*jnrB;
 630             j_coord_offsetC  = DIM*jnrC;
 631             j_coord_offsetD  = DIM*jnrD;
 632             j_coord_offsetE  = DIM*jnrE;
 633             j_coord_offsetF  = DIM*jnrF;
 634             j_coord_offsetG  = DIM*jnrG;
 635             j_coord_offsetH  = DIM*jnrH;
 636
 637             /* load j atom coordinates */
 638             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 639                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 640                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 641                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 642                                                  &jx0,&jy0,&jz0);
 643
 644             /* Calculate displacement vector */
 645             dx00             = _mm256_sub_ps(ix0,jx0);
 646             dy00             = _mm256_sub_ps(iy0,jy0);
 647             dz00             = _mm256_sub_ps(iz0,jz0);
 648
 649             /* Calculate squared distance and things based on it */
 650             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 651
 652             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 653
 654             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 655
 656             /* Load parameters for j particles */
 657             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 658                                                                  charge+jnrC+0,charge+jnrD+0,
 659                                                                  charge+jnrE+0,charge+jnrF+0,
 660                                                                  charge+jnrG+0,charge+jnrH+0);
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 667             {
 668
 669             /* Compute parameters for interactions between i and j atoms */
 670             qq00             = _mm256_mul_ps(iq0,jq0);
 671
 672             /* REACTION-FIELD ELECTROSTATICS */
 673             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 674
 675             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 676
 677             fscal            = felec;
 678
 679             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 680
 681             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm256_mul_ps(fscal,dx00);
 685             ty               = _mm256_mul_ps(fscal,dy00);
 686             tz               = _mm256_mul_ps(fscal,dz00);
 687
 688             /* Update vectorial force */
 689             fix0             = _mm256_add_ps(fix0,tx);
 690             fiy0             = _mm256_add_ps(fiy0,ty);
 691             fiz0             = _mm256_add_ps(fiz0,tz);
 692
 693             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 694             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 695             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 696             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 697             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 698             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 699             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 700             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 701             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 702
 703             }
 704
 705             /* Inner loop uses 30 flops */
 706         }
 707
 708         /* End of innermost loop */
 709
 710         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 711                                                  f+i_coord_offset,fshift+i_shift_offset);
 712
 713         /* Increment number of inner iterations */
 714         inneriter                  += j_index_end - j_index_start;
 715
 716         /* Outer loop uses 7 flops */
 717     }
 718
 719     /* Increment number of outer iterations */
 720     outeriter        += nri;
 721
 722     /* Update outer/inner flops */
 723
 724     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 725 }