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

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  76     real             *charge;
  77     int              nvdwtype;
  78     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  79     int              *vdwtype;
  80     real             *vdwparam;
  81     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  82     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  83     __m256d          dummy_mask,cutoff_mask;
  84     __m128           tmpmask0,tmpmask1;
  85     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  86     __m256d          one     = _mm256_set1_pd(1.0);
  87     __m256d          two     = _mm256_set1_pd(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm256_set1_pd(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101     krf              = _mm256_set1_pd(fr->ic->k_rf);
 102     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 103     crf              = _mm256_set1_pd(fr->ic->c_rf);
 104     nvdwtype         = fr->ntype;
 105     vdwparam         = fr->nbfp;
 106     vdwtype          = mdatoms->typeA;
 107
 108     /* Avoid stupid compiler warnings */
 109     jnrA = jnrB = jnrC = jnrD = 0;
 110     j_coord_offsetA = 0;
 111     j_coord_offsetB = 0;
 112     j_coord_offsetC = 0;
 113     j_coord_offsetD = 0;
 114
 115     outeriter        = 0;
 116     inneriter        = 0;
 117
 118     for(iidx=0;iidx<4*DIM;iidx++)
 119     {
 120         scratch[iidx] = 0.0;
 121     }
 122
 123     /* Start outer loop over neighborlists */
 124     for(iidx=0; iidx<nri; iidx++)
 125     {
 126         /* Load shift vector for this list */
 127         i_shift_offset   = DIM*shiftidx[iidx];
 128
 129         /* Load limits for loop over neighbors */
 130         j_index_start    = jindex[iidx];
 131         j_index_end      = jindex[iidx+1];
 132
 133         /* Get outer coordinate index */
 134         inr              = iinr[iidx];
 135         i_coord_offset   = DIM*inr;
 136
 137         /* Load i particle coords and add shift vector */
 138         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 139
 140         fix0             = _mm256_setzero_pd();
 141         fiy0             = _mm256_setzero_pd();
 142         fiz0             = _mm256_setzero_pd();
 143
 144         /* Load parameters for i particles */
 145         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 146         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         velecsum         = _mm256_setzero_pd();
 150         vvdwsum          = _mm256_setzero_pd();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             jnrC             = jjnr[jidx+2];
 160             jnrD             = jjnr[jidx+3];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163             j_coord_offsetC  = DIM*jnrC;
 164             j_coord_offsetD  = DIM*jnrD;
 165
 166             /* load j atom coordinates */
 167             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 169                                                  &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm256_sub_pd(ix0,jx0);
 173             dy00             = _mm256_sub_pd(iy0,jy0);
 174             dz00             = _mm256_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 180
 181             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 185                                                                  charge+jnrC+0,charge+jnrD+0);
 186             vdwjidx0A        = 2*vdwtype[jnrA+0];
 187             vdwjidx0B        = 2*vdwtype[jnrB+0];
 188             vdwjidx0C        = 2*vdwtype[jnrC+0];
 189             vdwjidx0D        = 2*vdwtype[jnrD+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             /* Compute parameters for interactions between i and j atoms */
 196             qq00             = _mm256_mul_pd(iq0,jq0);
 197             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 198                                             vdwioffsetptr0+vdwjidx0B,
 199                                             vdwioffsetptr0+vdwjidx0C,
 200                                             vdwioffsetptr0+vdwjidx0D,
 201                                             &c6_00,&c12_00);
 202
 203             /* REACTION-FIELD ELECTROSTATICS */
 204             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 205             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 206
 207             /* LENNARD-JONES DISPERSION/REPULSION */
 208
 209             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 210             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 211             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 212             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 213             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 214
 215             /* Update potential sum for this i atom from the interaction with this j atom. */
 216             velecsum         = _mm256_add_pd(velecsum,velec);
 217             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 218
 219             fscal            = _mm256_add_pd(felec,fvdw);
 220
 221             /* Calculate temporary vectorial force */
 222             tx               = _mm256_mul_pd(fscal,dx00);
 223             ty               = _mm256_mul_pd(fscal,dy00);
 224             tz               = _mm256_mul_pd(fscal,dz00);
 225
 226             /* Update vectorial force */
 227             fix0             = _mm256_add_pd(fix0,tx);
 228             fiy0             = _mm256_add_pd(fiy0,ty);
 229             fiz0             = _mm256_add_pd(fiz0,tz);
 230
 231             fjptrA             = f+j_coord_offsetA;
 232             fjptrB             = f+j_coord_offsetB;
 233             fjptrC             = f+j_coord_offsetC;
 234             fjptrD             = f+j_coord_offsetD;
 235             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 236
 237             /* Inner loop uses 44 flops */
 238         }
 239
 240         if(jidx<j_index_end)
 241         {
 242
 243             /* Get j neighbor index, and coordinate index */
 244             jnrlistA         = jjnr[jidx];
 245             jnrlistB         = jjnr[jidx+1];
 246             jnrlistC         = jjnr[jidx+2];
 247             jnrlistD         = jjnr[jidx+3];
 248             /* Sign of each element will be negative for non-real atoms.
 249              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 250              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 251              */
 252             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 253
 254             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 255             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 256             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 257
 258             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 259             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 260             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 261             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 262             j_coord_offsetA  = DIM*jnrA;
 263             j_coord_offsetB  = DIM*jnrB;
 264             j_coord_offsetC  = DIM*jnrC;
 265             j_coord_offsetD  = DIM*jnrD;
 266
 267             /* load j atom coordinates */
 268             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 269                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 270                                                  &jx0,&jy0,&jz0);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _mm256_sub_pd(ix0,jx0);
 274             dy00             = _mm256_sub_pd(iy0,jy0);
 275             dz00             = _mm256_sub_pd(iz0,jz0);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 279
 280             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 281
 282             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 283
 284             /* Load parameters for j particles */
 285             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 286                                                                  charge+jnrC+0,charge+jnrD+0);
 287             vdwjidx0A        = 2*vdwtype[jnrA+0];
 288             vdwjidx0B        = 2*vdwtype[jnrB+0];
 289             vdwjidx0C        = 2*vdwtype[jnrC+0];
 290             vdwjidx0D        = 2*vdwtype[jnrD+0];
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq00             = _mm256_mul_pd(iq0,jq0);
 298             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 299                                             vdwioffsetptr0+vdwjidx0B,
 300                                             vdwioffsetptr0+vdwjidx0C,
 301                                             vdwioffsetptr0+vdwjidx0D,
 302                                             &c6_00,&c12_00);
 303
 304             /* REACTION-FIELD ELECTROSTATICS */
 305             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 306             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 307
 308             /* LENNARD-JONES DISPERSION/REPULSION */
 309
 310             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 311             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 312             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 313             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 314             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velec            = _mm256_andnot_pd(dummy_mask,velec);
 318             velecsum         = _mm256_add_pd(velecsum,velec);
 319             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 320             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 321
 322             fscal            = _mm256_add_pd(felec,fvdw);
 323
 324             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm256_mul_pd(fscal,dx00);
 328             ty               = _mm256_mul_pd(fscal,dy00);
 329             tz               = _mm256_mul_pd(fscal,dz00);
 330
 331             /* Update vectorial force */
 332             fix0             = _mm256_add_pd(fix0,tx);
 333             fiy0             = _mm256_add_pd(fiy0,ty);
 334             fiz0             = _mm256_add_pd(fiz0,tz);
 335
 336             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 337             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 338             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 339             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 340             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 341
 342             /* Inner loop uses 44 flops */
 343         }
 344
 345         /* End of innermost loop */
 346
 347         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 348                                                  f+i_coord_offset,fshift+i_shift_offset);
 349
 350         ggid                        = gid[iidx];
 351         /* Update potential energies */
 352         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 353         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 354
 355         /* Increment number of inner iterations */
 356         inneriter                  += j_index_end - j_index_start;
 357
 358         /* Outer loop uses 9 flops */
 359     }
 360
 361     /* Increment number of outer iterations */
 362     outeriter        += nri;
 363
 364     /* Update outer/inner flops */
 365
 366     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 367 }
 368 /*
 369  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_double
 370  * Electrostatics interaction: ReactionField
 371  * VdW interaction:            LennardJones
 372  * Geometry:                   Particle-Particle
 373  * Calculate force/pot:        Force
 374  */
 375 void
 376 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_double
 377                     (t_nblist * gmx_restrict                nlist,
 378                      rvec * gmx_restrict                    xx,
 379                      rvec * gmx_restrict                    ff,
 380                      t_forcerec * gmx_restrict              fr,
 381                      t_mdatoms * gmx_restrict               mdatoms,
 382                      nb_kernel_data_t * gmx_restrict        kernel_data,
 383                      t_nrnb * gmx_restrict                  nrnb)
 384 {
 385     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 386      * just 0 for non-waters.
 387      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 388      * jnr indices corresponding to data put in the four positions in the SIMD register.
 389      */
 390     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 391     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 392     int              jnrA,jnrB,jnrC,jnrD;
 393     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 394     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 395     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 396     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 397     real             rcutoff_scalar;
 398     real             *shiftvec,*fshift,*x,*f;
 399     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 400     real             scratch[4*DIM];
 401     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 402     real *           vdwioffsetptr0;
 403     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 404     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 405     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 406     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 407     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 408     real             *charge;
 409     int              nvdwtype;
 410     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 411     int              *vdwtype;
 412     real             *vdwparam;
 413     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 414     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 415     __m256d          dummy_mask,cutoff_mask;
 416     __m128           tmpmask0,tmpmask1;
 417     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 418     __m256d          one     = _mm256_set1_pd(1.0);
 419     __m256d          two     = _mm256_set1_pd(2.0);
 420     x                = xx[0];
 421     f                = ff[0];
 422
 423     nri              = nlist->nri;
 424     iinr             = nlist->iinr;
 425     jindex           = nlist->jindex;
 426     jjnr             = nlist->jjnr;
 427     shiftidx         = nlist->shift;
 428     gid              = nlist->gid;
 429     shiftvec         = fr->shift_vec[0];
 430     fshift           = fr->fshift[0];
 431     facel            = _mm256_set1_pd(fr->epsfac);
 432     charge           = mdatoms->chargeA;
 433     krf              = _mm256_set1_pd(fr->ic->k_rf);
 434     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 435     crf              = _mm256_set1_pd(fr->ic->c_rf);
 436     nvdwtype         = fr->ntype;
 437     vdwparam         = fr->nbfp;
 438     vdwtype          = mdatoms->typeA;
 439
 440     /* Avoid stupid compiler warnings */
 441     jnrA = jnrB = jnrC = jnrD = 0;
 442     j_coord_offsetA = 0;
 443     j_coord_offsetB = 0;
 444     j_coord_offsetC = 0;
 445     j_coord_offsetD = 0;
 446
 447     outeriter        = 0;
 448     inneriter        = 0;
 449
 450     for(iidx=0;iidx<4*DIM;iidx++)
 451     {
 452         scratch[iidx] = 0.0;
 453     }
 454
 455     /* Start outer loop over neighborlists */
 456     for(iidx=0; iidx<nri; iidx++)
 457     {
 458         /* Load shift vector for this list */
 459         i_shift_offset   = DIM*shiftidx[iidx];
 460
 461         /* Load limits for loop over neighbors */
 462         j_index_start    = jindex[iidx];
 463         j_index_end      = jindex[iidx+1];
 464
 465         /* Get outer coordinate index */
 466         inr              = iinr[iidx];
 467         i_coord_offset   = DIM*inr;
 468
 469         /* Load i particle coords and add shift vector */
 470         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 471
 472         fix0             = _mm256_setzero_pd();
 473         fiy0             = _mm256_setzero_pd();
 474         fiz0             = _mm256_setzero_pd();
 475
 476         /* Load parameters for i particles */
 477         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 478         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 479
 480         /* Start inner kernel loop */
 481         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 482         {
 483
 484             /* Get j neighbor index, and coordinate index */
 485             jnrA             = jjnr[jidx];
 486             jnrB             = jjnr[jidx+1];
 487             jnrC             = jjnr[jidx+2];
 488             jnrD             = jjnr[jidx+3];
 489             j_coord_offsetA  = DIM*jnrA;
 490             j_coord_offsetB  = DIM*jnrB;
 491             j_coord_offsetC  = DIM*jnrC;
 492             j_coord_offsetD  = DIM*jnrD;
 493
 494             /* load j atom coordinates */
 495             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 496                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 497                                                  &jx0,&jy0,&jz0);
 498
 499             /* Calculate displacement vector */
 500             dx00             = _mm256_sub_pd(ix0,jx0);
 501             dy00             = _mm256_sub_pd(iy0,jy0);
 502             dz00             = _mm256_sub_pd(iz0,jz0);
 503
 504             /* Calculate squared distance and things based on it */
 505             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 506
 507             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 508
 509             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 510
 511             /* Load parameters for j particles */
 512             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 513                                                                  charge+jnrC+0,charge+jnrD+0);
 514             vdwjidx0A        = 2*vdwtype[jnrA+0];
 515             vdwjidx0B        = 2*vdwtype[jnrB+0];
 516             vdwjidx0C        = 2*vdwtype[jnrC+0];
 517             vdwjidx0D        = 2*vdwtype[jnrD+0];
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq00             = _mm256_mul_pd(iq0,jq0);
 525             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 526                                             vdwioffsetptr0+vdwjidx0B,
 527                                             vdwioffsetptr0+vdwjidx0C,
 528                                             vdwioffsetptr0+vdwjidx0D,
 529                                             &c6_00,&c12_00);
 530
 531             /* REACTION-FIELD ELECTROSTATICS */
 532             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 533
 534             /* LENNARD-JONES DISPERSION/REPULSION */
 535
 536             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 537             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 538
 539             fscal            = _mm256_add_pd(felec,fvdw);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm256_mul_pd(fscal,dx00);
 543             ty               = _mm256_mul_pd(fscal,dy00);
 544             tz               = _mm256_mul_pd(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm256_add_pd(fix0,tx);
 548             fiy0             = _mm256_add_pd(fiy0,ty);
 549             fiz0             = _mm256_add_pd(fiz0,tz);
 550
 551             fjptrA             = f+j_coord_offsetA;
 552             fjptrB             = f+j_coord_offsetB;
 553             fjptrC             = f+j_coord_offsetC;
 554             fjptrD             = f+j_coord_offsetD;
 555             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 556
 557             /* Inner loop uses 34 flops */
 558         }
 559
 560         if(jidx<j_index_end)
 561         {
 562
 563             /* Get j neighbor index, and coordinate index */
 564             jnrlistA         = jjnr[jidx];
 565             jnrlistB         = jjnr[jidx+1];
 566             jnrlistC         = jjnr[jidx+2];
 567             jnrlistD         = jjnr[jidx+3];
 568             /* Sign of each element will be negative for non-real atoms.
 569              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 570              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 571              */
 572             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 573
 574             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 575             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 576             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 577
 578             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 579             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 580             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 581             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 582             j_coord_offsetA  = DIM*jnrA;
 583             j_coord_offsetB  = DIM*jnrB;
 584             j_coord_offsetC  = DIM*jnrC;
 585             j_coord_offsetD  = DIM*jnrD;
 586
 587             /* load j atom coordinates */
 588             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 589                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 590                                                  &jx0,&jy0,&jz0);
 591
 592             /* Calculate displacement vector */
 593             dx00             = _mm256_sub_pd(ix0,jx0);
 594             dy00             = _mm256_sub_pd(iy0,jy0);
 595             dz00             = _mm256_sub_pd(iz0,jz0);
 596
 597             /* Calculate squared distance and things based on it */
 598             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 599
 600             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 601
 602             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 603
 604             /* Load parameters for j particles */
 605             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 606                                                                  charge+jnrC+0,charge+jnrD+0);
 607             vdwjidx0A        = 2*vdwtype[jnrA+0];
 608             vdwjidx0B        = 2*vdwtype[jnrB+0];
 609             vdwjidx0C        = 2*vdwtype[jnrC+0];
 610             vdwjidx0D        = 2*vdwtype[jnrD+0];
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq00             = _mm256_mul_pd(iq0,jq0);
 618             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 619                                             vdwioffsetptr0+vdwjidx0B,
 620                                             vdwioffsetptr0+vdwjidx0C,
 621                                             vdwioffsetptr0+vdwjidx0D,
 622                                             &c6_00,&c12_00);
 623
 624             /* REACTION-FIELD ELECTROSTATICS */
 625             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 626
 627             /* LENNARD-JONES DISPERSION/REPULSION */
 628
 629             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 630             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 631
 632             fscal            = _mm256_add_pd(felec,fvdw);
 633
 634             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 635
 636             /* Calculate temporary vectorial force */
 637             tx               = _mm256_mul_pd(fscal,dx00);
 638             ty               = _mm256_mul_pd(fscal,dy00);
 639             tz               = _mm256_mul_pd(fscal,dz00);
 640
 641             /* Update vectorial force */
 642             fix0             = _mm256_add_pd(fix0,tx);
 643             fiy0             = _mm256_add_pd(fiy0,ty);
 644             fiz0             = _mm256_add_pd(fiz0,tz);
 645
 646             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 647             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 648             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 649             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 650             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 651
 652             /* Inner loop uses 34 flops */
 653         }
 654
 655         /* End of innermost loop */
 656
 657         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 658                                                  f+i_coord_offset,fshift+i_shift_offset);
 659
 660         /* Increment number of inner iterations */
 661         inneriter                  += j_index_end - j_index_start;
 662
 663         /* Outer loop uses 7 flops */
 664     }
 665
 666     /* Increment number of outer iterations */
 667     outeriter        += nri;
 668
 669     /* Update outer/inner flops */
 670
 671     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 672 }