src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_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_ElecCoul_VdwLJ_GeomP1P1_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 102     vdwparam         = fr->nbfp;
 103     vdwtype          = mdatoms->typeA;
 104
 105     /* Avoid stupid compiler warnings */
 106     jnrA = jnrB = jnrC = jnrD = 0;
 107     j_coord_offsetA = 0;
 108     j_coord_offsetB = 0;
 109     j_coord_offsetC = 0;
 110     j_coord_offsetD = 0;
 111
 112     outeriter        = 0;
 113     inneriter        = 0;
 114
 115     for(iidx=0;iidx<4*DIM;iidx++)
 116     {
 117         scratch[iidx] = 0.0;
 118     }
 119
 120     /* Start outer loop over neighborlists */
 121     for(iidx=0; iidx<nri; iidx++)
 122     {
 123         /* Load shift vector for this list */
 124         i_shift_offset   = DIM*shiftidx[iidx];
 125
 126         /* Load limits for loop over neighbors */
 127         j_index_start    = jindex[iidx];
 128         j_index_end      = jindex[iidx+1];
 129
 130         /* Get outer coordinate index */
 131         inr              = iinr[iidx];
 132         i_coord_offset   = DIM*inr;
 133
 134         /* Load i particle coords and add shift vector */
 135         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 136
 137         fix0             = _mm256_setzero_pd();
 138         fiy0             = _mm256_setzero_pd();
 139         fiz0             = _mm256_setzero_pd();
 140
 141         /* Load parameters for i particles */
 142         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 143         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 144
 145         /* Reset potential sums */
 146         velecsum         = _mm256_setzero_pd();
 147         vvdwsum          = _mm256_setzero_pd();
 148
 149         /* Start inner kernel loop */
 150         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 151         {
 152
 153             /* Get j neighbor index, and coordinate index */
 154             jnrA             = jjnr[jidx];
 155             jnrB             = jjnr[jidx+1];
 156             jnrC             = jjnr[jidx+2];
 157             jnrD             = jjnr[jidx+3];
 158             j_coord_offsetA  = DIM*jnrA;
 159             j_coord_offsetB  = DIM*jnrB;
 160             j_coord_offsetC  = DIM*jnrC;
 161             j_coord_offsetD  = DIM*jnrD;
 162
 163             /* load j atom coordinates */
 164             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 166                                                  &jx0,&jy0,&jz0);
 167
 168             /* Calculate displacement vector */
 169             dx00             = _mm256_sub_pd(ix0,jx0);
 170             dy00             = _mm256_sub_pd(iy0,jy0);
 171             dz00             = _mm256_sub_pd(iz0,jz0);
 172
 173             /* Calculate squared distance and things based on it */
 174             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 175
 176             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 177
 178             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 179
 180             /* Load parameters for j particles */
 181             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 182                                                                  charge+jnrC+0,charge+jnrD+0);
 183             vdwjidx0A        = 2*vdwtype[jnrA+0];
 184             vdwjidx0B        = 2*vdwtype[jnrB+0];
 185             vdwjidx0C        = 2*vdwtype[jnrC+0];
 186             vdwjidx0D        = 2*vdwtype[jnrD+0];
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             qq00             = _mm256_mul_pd(iq0,jq0);
 194             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 195                                             vdwioffsetptr0+vdwjidx0B,
 196                                             vdwioffsetptr0+vdwjidx0C,
 197                                             vdwioffsetptr0+vdwjidx0D,
 198                                             &c6_00,&c12_00);
 199
 200             /* COULOMB ELECTROSTATICS */
 201             velec            = _mm256_mul_pd(qq00,rinv00);
 202             felec            = _mm256_mul_pd(velec,rinvsq00);
 203
 204             /* LENNARD-JONES DISPERSION/REPULSION */
 205
 206             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 207             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 208             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 209             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 210             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 211
 212             /* Update potential sum for this i atom from the interaction with this j atom. */
 213             velecsum         = _mm256_add_pd(velecsum,velec);
 214             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 215
 216             fscal            = _mm256_add_pd(felec,fvdw);
 217
 218             /* Calculate temporary vectorial force */
 219             tx               = _mm256_mul_pd(fscal,dx00);
 220             ty               = _mm256_mul_pd(fscal,dy00);
 221             tz               = _mm256_mul_pd(fscal,dz00);
 222
 223             /* Update vectorial force */
 224             fix0             = _mm256_add_pd(fix0,tx);
 225             fiy0             = _mm256_add_pd(fiy0,ty);
 226             fiz0             = _mm256_add_pd(fiz0,tz);
 227
 228             fjptrA             = f+j_coord_offsetA;
 229             fjptrB             = f+j_coord_offsetB;
 230             fjptrC             = f+j_coord_offsetC;
 231             fjptrD             = f+j_coord_offsetD;
 232             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 233
 234             /* Inner loop uses 39 flops */
 235         }
 236
 237         if(jidx<j_index_end)
 238         {
 239
 240             /* Get j neighbor index, and coordinate index */
 241             jnrlistA         = jjnr[jidx];
 242             jnrlistB         = jjnr[jidx+1];
 243             jnrlistC         = jjnr[jidx+2];
 244             jnrlistD         = jjnr[jidx+3];
 245             /* Sign of each element will be negative for non-real atoms.
 246              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 247              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 248              */
 249             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 250
 251             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 252             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 253             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 254
 255             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 256             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 257             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 258             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 259             j_coord_offsetA  = DIM*jnrA;
 260             j_coord_offsetB  = DIM*jnrB;
 261             j_coord_offsetC  = DIM*jnrC;
 262             j_coord_offsetD  = DIM*jnrD;
 263
 264             /* load j atom coordinates */
 265             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 266                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 267                                                  &jx0,&jy0,&jz0);
 268
 269             /* Calculate displacement vector */
 270             dx00             = _mm256_sub_pd(ix0,jx0);
 271             dy00             = _mm256_sub_pd(iy0,jy0);
 272             dz00             = _mm256_sub_pd(iz0,jz0);
 273
 274             /* Calculate squared distance and things based on it */
 275             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 276
 277             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 278
 279             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 280
 281             /* Load parameters for j particles */
 282             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 283                                                                  charge+jnrC+0,charge+jnrD+0);
 284             vdwjidx0A        = 2*vdwtype[jnrA+0];
 285             vdwjidx0B        = 2*vdwtype[jnrB+0];
 286             vdwjidx0C        = 2*vdwtype[jnrC+0];
 287             vdwjidx0D        = 2*vdwtype[jnrD+0];
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq00             = _mm256_mul_pd(iq0,jq0);
 295             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 296                                             vdwioffsetptr0+vdwjidx0B,
 297                                             vdwioffsetptr0+vdwjidx0C,
 298                                             vdwioffsetptr0+vdwjidx0D,
 299                                             &c6_00,&c12_00);
 300
 301             /* COULOMB ELECTROSTATICS */
 302             velec            = _mm256_mul_pd(qq00,rinv00);
 303             felec            = _mm256_mul_pd(velec,rinvsq00);
 304
 305             /* LENNARD-JONES DISPERSION/REPULSION */
 306
 307             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 308             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 309             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 310             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 311             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velec            = _mm256_andnot_pd(dummy_mask,velec);
 315             velecsum         = _mm256_add_pd(velecsum,velec);
 316             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 317             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 318
 319             fscal            = _mm256_add_pd(felec,fvdw);
 320
 321             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_pd(fscal,dx00);
 325             ty               = _mm256_mul_pd(fscal,dy00);
 326             tz               = _mm256_mul_pd(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm256_add_pd(fix0,tx);
 330             fiy0             = _mm256_add_pd(fiy0,ty);
 331             fiz0             = _mm256_add_pd(fiz0,tz);
 332
 333             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 334             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 335             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 336             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 337             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 338
 339             /* Inner loop uses 39 flops */
 340         }
 341
 342         /* End of innermost loop */
 343
 344         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 345                                                  f+i_coord_offset,fshift+i_shift_offset);
 346
 347         ggid                        = gid[iidx];
 348         /* Update potential energies */
 349         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 350         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 351
 352         /* Increment number of inner iterations */
 353         inneriter                  += j_index_end - j_index_start;
 354
 355         /* Outer loop uses 9 flops */
 356     }
 357
 358     /* Increment number of outer iterations */
 359     outeriter        += nri;
 360
 361     /* Update outer/inner flops */
 362
 363     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*39);
 364 }
 365 /*
 366  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_double
 367  * Electrostatics interaction: Coulomb
 368  * VdW interaction:            LennardJones
 369  * Geometry:                   Particle-Particle
 370  * Calculate force/pot:        Force
 371  */
 372 void
 373 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_double
 374                     (t_nblist * gmx_restrict                nlist,
 375                      rvec * gmx_restrict                    xx,
 376                      rvec * gmx_restrict                    ff,
 377                      t_forcerec * gmx_restrict              fr,
 378                      t_mdatoms * gmx_restrict               mdatoms,
 379                      nb_kernel_data_t * gmx_restrict        kernel_data,
 380                      t_nrnb * gmx_restrict                  nrnb)
 381 {
 382     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 383      * just 0 for non-waters.
 384      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 385      * jnr indices corresponding to data put in the four positions in the SIMD register.
 386      */
 387     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 388     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 389     int              jnrA,jnrB,jnrC,jnrD;
 390     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 391     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 392     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 393     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 394     real             rcutoff_scalar;
 395     real             *shiftvec,*fshift,*x,*f;
 396     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 397     real             scratch[4*DIM];
 398     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 399     real *           vdwioffsetptr0;
 400     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 401     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 402     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 403     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 404     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 405     real             *charge;
 406     int              nvdwtype;
 407     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 408     int              *vdwtype;
 409     real             *vdwparam;
 410     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 411     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 412     __m256d          dummy_mask,cutoff_mask;
 413     __m128           tmpmask0,tmpmask1;
 414     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 415     __m256d          one     = _mm256_set1_pd(1.0);
 416     __m256d          two     = _mm256_set1_pd(2.0);
 417     x                = xx[0];
 418     f                = ff[0];
 419
 420     nri              = nlist->nri;
 421     iinr             = nlist->iinr;
 422     jindex           = nlist->jindex;
 423     jjnr             = nlist->jjnr;
 424     shiftidx         = nlist->shift;
 425     gid              = nlist->gid;
 426     shiftvec         = fr->shift_vec[0];
 427     fshift           = fr->fshift[0];
 428     facel            = _mm256_set1_pd(fr->epsfac);
 429     charge           = mdatoms->chargeA;
 430     nvdwtype         = fr->ntype;
 431     vdwparam         = fr->nbfp;
 432     vdwtype          = mdatoms->typeA;
 433
 434     /* Avoid stupid compiler warnings */
 435     jnrA = jnrB = jnrC = jnrD = 0;
 436     j_coord_offsetA = 0;
 437     j_coord_offsetB = 0;
 438     j_coord_offsetC = 0;
 439     j_coord_offsetD = 0;
 440
 441     outeriter        = 0;
 442     inneriter        = 0;
 443
 444     for(iidx=0;iidx<4*DIM;iidx++)
 445     {
 446         scratch[iidx] = 0.0;
 447     }
 448
 449     /* Start outer loop over neighborlists */
 450     for(iidx=0; iidx<nri; iidx++)
 451     {
 452         /* Load shift vector for this list */
 453         i_shift_offset   = DIM*shiftidx[iidx];
 454
 455         /* Load limits for loop over neighbors */
 456         j_index_start    = jindex[iidx];
 457         j_index_end      = jindex[iidx+1];
 458
 459         /* Get outer coordinate index */
 460         inr              = iinr[iidx];
 461         i_coord_offset   = DIM*inr;
 462
 463         /* Load i particle coords and add shift vector */
 464         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 465
 466         fix0             = _mm256_setzero_pd();
 467         fiy0             = _mm256_setzero_pd();
 468         fiz0             = _mm256_setzero_pd();
 469
 470         /* Load parameters for i particles */
 471         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 472         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 473
 474         /* Start inner kernel loop */
 475         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 476         {
 477
 478             /* Get j neighbor index, and coordinate index */
 479             jnrA             = jjnr[jidx];
 480             jnrB             = jjnr[jidx+1];
 481             jnrC             = jjnr[jidx+2];
 482             jnrD             = jjnr[jidx+3];
 483             j_coord_offsetA  = DIM*jnrA;
 484             j_coord_offsetB  = DIM*jnrB;
 485             j_coord_offsetC  = DIM*jnrC;
 486             j_coord_offsetD  = DIM*jnrD;
 487
 488             /* load j atom coordinates */
 489             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 490                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 491                                                  &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm256_sub_pd(ix0,jx0);
 495             dy00             = _mm256_sub_pd(iy0,jy0);
 496             dz00             = _mm256_sub_pd(iz0,jz0);
 497
 498             /* Calculate squared distance and things based on it */
 499             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 500
 501             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 502
 503             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 504
 505             /* Load parameters for j particles */
 506             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 507                                                                  charge+jnrC+0,charge+jnrD+0);
 508             vdwjidx0A        = 2*vdwtype[jnrA+0];
 509             vdwjidx0B        = 2*vdwtype[jnrB+0];
 510             vdwjidx0C        = 2*vdwtype[jnrC+0];
 511             vdwjidx0D        = 2*vdwtype[jnrD+0];
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             /* Compute parameters for interactions between i and j atoms */
 518             qq00             = _mm256_mul_pd(iq0,jq0);
 519             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 520                                             vdwioffsetptr0+vdwjidx0B,
 521                                             vdwioffsetptr0+vdwjidx0C,
 522                                             vdwioffsetptr0+vdwjidx0D,
 523                                             &c6_00,&c12_00);
 524
 525             /* COULOMB ELECTROSTATICS */
 526             velec            = _mm256_mul_pd(qq00,rinv00);
 527             felec            = _mm256_mul_pd(velec,rinvsq00);
 528
 529             /* LENNARD-JONES DISPERSION/REPULSION */
 530
 531             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 532             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 533
 534             fscal            = _mm256_add_pd(felec,fvdw);
 535
 536             /* Calculate temporary vectorial force */
 537             tx               = _mm256_mul_pd(fscal,dx00);
 538             ty               = _mm256_mul_pd(fscal,dy00);
 539             tz               = _mm256_mul_pd(fscal,dz00);
 540
 541             /* Update vectorial force */
 542             fix0             = _mm256_add_pd(fix0,tx);
 543             fiy0             = _mm256_add_pd(fiy0,ty);
 544             fiz0             = _mm256_add_pd(fiz0,tz);
 545
 546             fjptrA             = f+j_coord_offsetA;
 547             fjptrB             = f+j_coord_offsetB;
 548             fjptrC             = f+j_coord_offsetC;
 549             fjptrD             = f+j_coord_offsetD;
 550             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 551
 552             /* Inner loop uses 33 flops */
 553         }
 554
 555         if(jidx<j_index_end)
 556         {
 557
 558             /* Get j neighbor index, and coordinate index */
 559             jnrlistA         = jjnr[jidx];
 560             jnrlistB         = jjnr[jidx+1];
 561             jnrlistC         = jjnr[jidx+2];
 562             jnrlistD         = jjnr[jidx+3];
 563             /* Sign of each element will be negative for non-real atoms.
 564              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 565              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 566              */
 567             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 568
 569             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 570             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 571             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 572
 573             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 574             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 575             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 576             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 577             j_coord_offsetA  = DIM*jnrA;
 578             j_coord_offsetB  = DIM*jnrB;
 579             j_coord_offsetC  = DIM*jnrC;
 580             j_coord_offsetD  = DIM*jnrD;
 581
 582             /* load j atom coordinates */
 583             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 584                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 585                                                  &jx0,&jy0,&jz0);
 586
 587             /* Calculate displacement vector */
 588             dx00             = _mm256_sub_pd(ix0,jx0);
 589             dy00             = _mm256_sub_pd(iy0,jy0);
 590             dz00             = _mm256_sub_pd(iz0,jz0);
 591
 592             /* Calculate squared distance and things based on it */
 593             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 594
 595             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 596
 597             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 598
 599             /* Load parameters for j particles */
 600             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 601                                                                  charge+jnrC+0,charge+jnrD+0);
 602             vdwjidx0A        = 2*vdwtype[jnrA+0];
 603             vdwjidx0B        = 2*vdwtype[jnrB+0];
 604             vdwjidx0C        = 2*vdwtype[jnrC+0];
 605             vdwjidx0D        = 2*vdwtype[jnrD+0];
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _mm256_mul_pd(iq0,jq0);
 613             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 614                                             vdwioffsetptr0+vdwjidx0B,
 615                                             vdwioffsetptr0+vdwjidx0C,
 616                                             vdwioffsetptr0+vdwjidx0D,
 617                                             &c6_00,&c12_00);
 618
 619             /* COULOMB ELECTROSTATICS */
 620             velec            = _mm256_mul_pd(qq00,rinv00);
 621             felec            = _mm256_mul_pd(velec,rinvsq00);
 622
 623             /* LENNARD-JONES DISPERSION/REPULSION */
 624
 625             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 626             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 627
 628             fscal            = _mm256_add_pd(felec,fvdw);
 629
 630             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm256_mul_pd(fscal,dx00);
 634             ty               = _mm256_mul_pd(fscal,dy00);
 635             tz               = _mm256_mul_pd(fscal,dz00);
 636
 637             /* Update vectorial force */
 638             fix0             = _mm256_add_pd(fix0,tx);
 639             fiy0             = _mm256_add_pd(fiy0,ty);
 640             fiz0             = _mm256_add_pd(fiz0,tz);
 641
 642             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 643             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 644             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 645             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 646             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 647
 648             /* Inner loop uses 33 flops */
 649         }
 650
 651         /* End of innermost loop */
 652
 653         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 654                                                  f+i_coord_offset,fshift+i_shift_offset);
 655
 656         /* Increment number of inner iterations */
 657         inneriter                  += j_index_end - j_index_start;
 658
 659         /* Outer loop uses 7 flops */
 660     }
 661
 662     /* Increment number of outer iterations */
 663     outeriter        += nri;
 664
 665     /* Update outer/inner flops */
 666
 667     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*33);
 668 }