src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwLJ_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  90     real             *charge;
  91     int              nvdwtype;
  92     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  93     int              *vdwtype;
  94     real             *vdwparam;
  95     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  96     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  97     __m256d          dummy_mask,cutoff_mask;
  98     __m128           tmpmask0,tmpmask1;
  99     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 100     __m256d          one     = _mm256_set1_pd(1.0);
 101     __m256d          two     = _mm256_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm256_set1_pd(fr->ic->k_rf);
 116     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 117     crf              = _mm256_set1_pd(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 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_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm256_setzero_pd();
 155         fiy0             = _mm256_setzero_pd();
 156         fiz0             = _mm256_setzero_pd();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 160         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm256_setzero_pd();
 164         vvdwsum          = _mm256_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             jnrC             = jjnr[jidx+2];
 174             jnrD             = jjnr[jidx+3];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179
 180             /* load j atom coordinates */
 181             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 183                                                  &jx0,&jy0,&jz0);
 184
 185             /* Calculate displacement vector */
 186             dx00             = _mm256_sub_pd(ix0,jx0);
 187             dy00             = _mm256_sub_pd(iy0,jy0);
 188             dz00             = _mm256_sub_pd(iz0,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 192
 193             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 194
 195             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 196
 197             /* Load parameters for j particles */
 198             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 199                                                                  charge+jnrC+0,charge+jnrD+0);
 200             vdwjidx0A        = 2*vdwtype[jnrA+0];
 201             vdwjidx0B        = 2*vdwtype[jnrB+0];
 202             vdwjidx0C        = 2*vdwtype[jnrC+0];
 203             vdwjidx0D        = 2*vdwtype[jnrD+0];
 204
 205             /**************************
 206              * CALCULATE INTERACTIONS *
 207              **************************/
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq00             = _mm256_mul_pd(iq0,jq0);
 211             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 212                                             vdwioffsetptr0+vdwjidx0B,
 213                                             vdwioffsetptr0+vdwjidx0C,
 214                                             vdwioffsetptr0+vdwjidx0D,
 215                                             &c6_00,&c12_00);
 216
 217             /* REACTION-FIELD ELECTROSTATICS */
 218             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 219             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 220
 221             /* LENNARD-JONES DISPERSION/REPULSION */
 222
 223             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 224             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 225             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 226             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 227             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             velecsum         = _mm256_add_pd(velecsum,velec);
 231             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 232
 233             fscal            = _mm256_add_pd(felec,fvdw);
 234
 235             /* Calculate temporary vectorial force */
 236             tx               = _mm256_mul_pd(fscal,dx00);
 237             ty               = _mm256_mul_pd(fscal,dy00);
 238             tz               = _mm256_mul_pd(fscal,dz00);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm256_add_pd(fix0,tx);
 242             fiy0             = _mm256_add_pd(fiy0,ty);
 243             fiz0             = _mm256_add_pd(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             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 250
 251             /* Inner loop uses 44 flops */
 252         }
 253
 254         if(jidx<j_index_end)
 255         {
 256
 257             /* Get j neighbor index, and coordinate index */
 258             jnrlistA         = jjnr[jidx];
 259             jnrlistB         = jjnr[jidx+1];
 260             jnrlistC         = jjnr[jidx+2];
 261             jnrlistD         = jjnr[jidx+3];
 262             /* Sign of each element will be negative for non-real atoms.
 263              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 264              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 265              */
 266             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 267
 268             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 269             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 270             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 271
 272             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 273             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 274             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 275             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 276             j_coord_offsetA  = DIM*jnrA;
 277             j_coord_offsetB  = DIM*jnrB;
 278             j_coord_offsetC  = DIM*jnrC;
 279             j_coord_offsetD  = DIM*jnrD;
 280
 281             /* load j atom coordinates */
 282             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 283                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 284                                                  &jx0,&jy0,&jz0);
 285
 286             /* Calculate displacement vector */
 287             dx00             = _mm256_sub_pd(ix0,jx0);
 288             dy00             = _mm256_sub_pd(iy0,jy0);
 289             dz00             = _mm256_sub_pd(iz0,jz0);
 290
 291             /* Calculate squared distance and things based on it */
 292             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 293
 294             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 295
 296             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 297
 298             /* Load parameters for j particles */
 299             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 300                                                                  charge+jnrC+0,charge+jnrD+0);
 301             vdwjidx0A        = 2*vdwtype[jnrA+0];
 302             vdwjidx0B        = 2*vdwtype[jnrB+0];
 303             vdwjidx0C        = 2*vdwtype[jnrC+0];
 304             vdwjidx0D        = 2*vdwtype[jnrD+0];
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq00             = _mm256_mul_pd(iq0,jq0);
 312             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 313                                             vdwioffsetptr0+vdwjidx0B,
 314                                             vdwioffsetptr0+vdwjidx0C,
 315                                             vdwioffsetptr0+vdwjidx0D,
 316                                             &c6_00,&c12_00);
 317
 318             /* REACTION-FIELD ELECTROSTATICS */
 319             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 320             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 321
 322             /* LENNARD-JONES DISPERSION/REPULSION */
 323
 324             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 325             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 326             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 327             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 328             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velec            = _mm256_andnot_pd(dummy_mask,velec);
 332             velecsum         = _mm256_add_pd(velecsum,velec);
 333             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 334             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 335
 336             fscal            = _mm256_add_pd(felec,fvdw);
 337
 338             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm256_mul_pd(fscal,dx00);
 342             ty               = _mm256_mul_pd(fscal,dy00);
 343             tz               = _mm256_mul_pd(fscal,dz00);
 344
 345             /* Update vectorial force */
 346             fix0             = _mm256_add_pd(fix0,tx);
 347             fiy0             = _mm256_add_pd(fiy0,ty);
 348             fiz0             = _mm256_add_pd(fiz0,tz);
 349
 350             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 351             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 352             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 353             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 354             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 355
 356             /* Inner loop uses 44 flops */
 357         }
 358
 359         /* End of innermost loop */
 360
 361         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 362                                                  f+i_coord_offset,fshift+i_shift_offset);
 363
 364         ggid                        = gid[iidx];
 365         /* Update potential energies */
 366         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 367         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 368
 369         /* Increment number of inner iterations */
 370         inneriter                  += j_index_end - j_index_start;
 371
 372         /* Outer loop uses 9 flops */
 373     }
 374
 375     /* Increment number of outer iterations */
 376     outeriter        += nri;
 377
 378     /* Update outer/inner flops */
 379
 380     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 381 }
 382 /*
 383  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_double
 384  * Electrostatics interaction: ReactionField
 385  * VdW interaction:            LennardJones
 386  * Geometry:                   Particle-Particle
 387  * Calculate force/pot:        Force
 388  */
 389 void
 390 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_double
 391                     (t_nblist                    * gmx_restrict       nlist,
 392                      rvec                        * gmx_restrict          xx,
 393                      rvec                        * gmx_restrict          ff,
 394                      t_forcerec                  * gmx_restrict          fr,
 395                      t_mdatoms                   * gmx_restrict     mdatoms,
 396                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 397                      t_nrnb                      * gmx_restrict        nrnb)
 398 {
 399     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 400      * just 0 for non-waters.
 401      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 402      * jnr indices corresponding to data put in the four positions in the SIMD register.
 403      */
 404     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 405     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 406     int              jnrA,jnrB,jnrC,jnrD;
 407     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 408     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 409     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 410     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 411     real             rcutoff_scalar;
 412     real             *shiftvec,*fshift,*x,*f;
 413     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 414     real             scratch[4*DIM];
 415     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 416     real *           vdwioffsetptr0;
 417     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 418     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 419     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 420     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 421     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 422     real             *charge;
 423     int              nvdwtype;
 424     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 425     int              *vdwtype;
 426     real             *vdwparam;
 427     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 428     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 429     __m256d          dummy_mask,cutoff_mask;
 430     __m128           tmpmask0,tmpmask1;
 431     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 432     __m256d          one     = _mm256_set1_pd(1.0);
 433     __m256d          two     = _mm256_set1_pd(2.0);
 434     x                = xx[0];
 435     f                = ff[0];
 436
 437     nri              = nlist->nri;
 438     iinr             = nlist->iinr;
 439     jindex           = nlist->jindex;
 440     jjnr             = nlist->jjnr;
 441     shiftidx         = nlist->shift;
 442     gid              = nlist->gid;
 443     shiftvec         = fr->shift_vec[0];
 444     fshift           = fr->fshift[0];
 445     facel            = _mm256_set1_pd(fr->epsfac);
 446     charge           = mdatoms->chargeA;
 447     krf              = _mm256_set1_pd(fr->ic->k_rf);
 448     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 449     crf              = _mm256_set1_pd(fr->ic->c_rf);
 450     nvdwtype         = fr->ntype;
 451     vdwparam         = fr->nbfp;
 452     vdwtype          = mdatoms->typeA;
 453
 454     /* Avoid stupid compiler warnings */
 455     jnrA = jnrB = jnrC = jnrD = 0;
 456     j_coord_offsetA = 0;
 457     j_coord_offsetB = 0;
 458     j_coord_offsetC = 0;
 459     j_coord_offsetD = 0;
 460
 461     outeriter        = 0;
 462     inneriter        = 0;
 463
 464     for(iidx=0;iidx<4*DIM;iidx++)
 465     {
 466         scratch[iidx] = 0.0;
 467     }
 468
 469     /* Start outer loop over neighborlists */
 470     for(iidx=0; iidx<nri; iidx++)
 471     {
 472         /* Load shift vector for this list */
 473         i_shift_offset   = DIM*shiftidx[iidx];
 474
 475         /* Load limits for loop over neighbors */
 476         j_index_start    = jindex[iidx];
 477         j_index_end      = jindex[iidx+1];
 478
 479         /* Get outer coordinate index */
 480         inr              = iinr[iidx];
 481         i_coord_offset   = DIM*inr;
 482
 483         /* Load i particle coords and add shift vector */
 484         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 485
 486         fix0             = _mm256_setzero_pd();
 487         fiy0             = _mm256_setzero_pd();
 488         fiz0             = _mm256_setzero_pd();
 489
 490         /* Load parameters for i particles */
 491         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 492         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 493
 494         /* Start inner kernel loop */
 495         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 496         {
 497
 498             /* Get j neighbor index, and coordinate index */
 499             jnrA             = jjnr[jidx];
 500             jnrB             = jjnr[jidx+1];
 501             jnrC             = jjnr[jidx+2];
 502             jnrD             = jjnr[jidx+3];
 503             j_coord_offsetA  = DIM*jnrA;
 504             j_coord_offsetB  = DIM*jnrB;
 505             j_coord_offsetC  = DIM*jnrC;
 506             j_coord_offsetD  = DIM*jnrD;
 507
 508             /* load j atom coordinates */
 509             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 510                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 511                                                  &jx0,&jy0,&jz0);
 512
 513             /* Calculate displacement vector */
 514             dx00             = _mm256_sub_pd(ix0,jx0);
 515             dy00             = _mm256_sub_pd(iy0,jy0);
 516             dz00             = _mm256_sub_pd(iz0,jz0);
 517
 518             /* Calculate squared distance and things based on it */
 519             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 520
 521             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 522
 523             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 524
 525             /* Load parameters for j particles */
 526             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 527                                                                  charge+jnrC+0,charge+jnrD+0);
 528             vdwjidx0A        = 2*vdwtype[jnrA+0];
 529             vdwjidx0B        = 2*vdwtype[jnrB+0];
 530             vdwjidx0C        = 2*vdwtype[jnrC+0];
 531             vdwjidx0D        = 2*vdwtype[jnrD+0];
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq00             = _mm256_mul_pd(iq0,jq0);
 539             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 540                                             vdwioffsetptr0+vdwjidx0B,
 541                                             vdwioffsetptr0+vdwjidx0C,
 542                                             vdwioffsetptr0+vdwjidx0D,
 543                                             &c6_00,&c12_00);
 544
 545             /* REACTION-FIELD ELECTROSTATICS */
 546             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 547
 548             /* LENNARD-JONES DISPERSION/REPULSION */
 549
 550             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 551             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 552
 553             fscal            = _mm256_add_pd(felec,fvdw);
 554
 555             /* Calculate temporary vectorial force */
 556             tx               = _mm256_mul_pd(fscal,dx00);
 557             ty               = _mm256_mul_pd(fscal,dy00);
 558             tz               = _mm256_mul_pd(fscal,dz00);
 559
 560             /* Update vectorial force */
 561             fix0             = _mm256_add_pd(fix0,tx);
 562             fiy0             = _mm256_add_pd(fiy0,ty);
 563             fiz0             = _mm256_add_pd(fiz0,tz);
 564
 565             fjptrA             = f+j_coord_offsetA;
 566             fjptrB             = f+j_coord_offsetB;
 567             fjptrC             = f+j_coord_offsetC;
 568             fjptrD             = f+j_coord_offsetD;
 569             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 570
 571             /* Inner loop uses 34 flops */
 572         }
 573
 574         if(jidx<j_index_end)
 575         {
 576
 577             /* Get j neighbor index, and coordinate index */
 578             jnrlistA         = jjnr[jidx];
 579             jnrlistB         = jjnr[jidx+1];
 580             jnrlistC         = jjnr[jidx+2];
 581             jnrlistD         = jjnr[jidx+3];
 582             /* Sign of each element will be negative for non-real atoms.
 583              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 584              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 585              */
 586             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 587
 588             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 589             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 590             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 591
 592             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 593             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 594             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 595             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 596             j_coord_offsetA  = DIM*jnrA;
 597             j_coord_offsetB  = DIM*jnrB;
 598             j_coord_offsetC  = DIM*jnrC;
 599             j_coord_offsetD  = DIM*jnrD;
 600
 601             /* load j atom coordinates */
 602             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 603                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 604                                                  &jx0,&jy0,&jz0);
 605
 606             /* Calculate displacement vector */
 607             dx00             = _mm256_sub_pd(ix0,jx0);
 608             dy00             = _mm256_sub_pd(iy0,jy0);
 609             dz00             = _mm256_sub_pd(iz0,jz0);
 610
 611             /* Calculate squared distance and things based on it */
 612             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 613
 614             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 615
 616             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 617
 618             /* Load parameters for j particles */
 619             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 620                                                                  charge+jnrC+0,charge+jnrD+0);
 621             vdwjidx0A        = 2*vdwtype[jnrA+0];
 622             vdwjidx0B        = 2*vdwtype[jnrB+0];
 623             vdwjidx0C        = 2*vdwtype[jnrC+0];
 624             vdwjidx0D        = 2*vdwtype[jnrD+0];
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq00             = _mm256_mul_pd(iq0,jq0);
 632             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 633                                             vdwioffsetptr0+vdwjidx0B,
 634                                             vdwioffsetptr0+vdwjidx0C,
 635                                             vdwioffsetptr0+vdwjidx0D,
 636                                             &c6_00,&c12_00);
 637
 638             /* REACTION-FIELD ELECTROSTATICS */
 639             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 640
 641             /* LENNARD-JONES DISPERSION/REPULSION */
 642
 643             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 644             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 645
 646             fscal            = _mm256_add_pd(felec,fvdw);
 647
 648             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm256_mul_pd(fscal,dx00);
 652             ty               = _mm256_mul_pd(fscal,dy00);
 653             tz               = _mm256_mul_pd(fscal,dz00);
 654
 655             /* Update vectorial force */
 656             fix0             = _mm256_add_pd(fix0,tx);
 657             fiy0             = _mm256_add_pd(fiy0,ty);
 658             fiz0             = _mm256_add_pd(fiz0,tz);
 659
 660             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 661             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 662             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 663             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 664             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 665
 666             /* Inner loop uses 34 flops */
 667         }
 668
 669         /* End of innermost loop */
 670
 671         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 672                                                  f+i_coord_offset,fshift+i_shift_offset);
 673
 674         /* Increment number of inner iterations */
 675         inneriter                  += j_index_end - j_index_start;
 676
 677         /* Outer loop uses 7 flops */
 678     }
 679
 680     /* Increment number of outer iterations */
 681     outeriter        += nri;
 682
 683     /* Update outer/inner flops */
 684
 685     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 686 }