src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecNone_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_ElecNone_VdwLJ_GeomP1P1_VF_avx_256_double
  38  * Electrostatics interaction: None
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_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     int              nvdwtype;
  76     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  77     int              *vdwtype;
  78     real             *vdwparam;
  79     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  80     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  81     __m256d          dummy_mask,cutoff_mask;
  82     __m128           tmpmask0,tmpmask1;
  83     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  84     __m256d          one     = _mm256_set1_pd(1.0);
  85     __m256d          two     = _mm256_set1_pd(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     nvdwtype         = fr->ntype;
  98     vdwparam         = fr->nbfp;
  99     vdwtype          = mdatoms->typeA;
 100
 101     /* Avoid stupid compiler warnings */
 102     jnrA = jnrB = jnrC = jnrD = 0;
 103     j_coord_offsetA = 0;
 104     j_coord_offsetB = 0;
 105     j_coord_offsetC = 0;
 106     j_coord_offsetD = 0;
 107
 108     outeriter        = 0;
 109     inneriter        = 0;
 110
 111     for(iidx=0;iidx<4*DIM;iidx++)
 112     {
 113         scratch[iidx] = 0.0;
 114     }
 115
 116     /* Start outer loop over neighborlists */
 117     for(iidx=0; iidx<nri; iidx++)
 118     {
 119         /* Load shift vector for this list */
 120         i_shift_offset   = DIM*shiftidx[iidx];
 121
 122         /* Load limits for loop over neighbors */
 123         j_index_start    = jindex[iidx];
 124         j_index_end      = jindex[iidx+1];
 125
 126         /* Get outer coordinate index */
 127         inr              = iinr[iidx];
 128         i_coord_offset   = DIM*inr;
 129
 130         /* Load i particle coords and add shift vector */
 131         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 132
 133         fix0             = _mm256_setzero_pd();
 134         fiy0             = _mm256_setzero_pd();
 135         fiz0             = _mm256_setzero_pd();
 136
 137         /* Load parameters for i particles */
 138         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140         /* Reset potential sums */
 141         vvdwsum          = _mm256_setzero_pd();
 142
 143         /* Start inner kernel loop */
 144         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 145         {
 146
 147             /* Get j neighbor index, and coordinate index */
 148             jnrA             = jjnr[jidx];
 149             jnrB             = jjnr[jidx+1];
 150             jnrC             = jjnr[jidx+2];
 151             jnrD             = jjnr[jidx+3];
 152             j_coord_offsetA  = DIM*jnrA;
 153             j_coord_offsetB  = DIM*jnrB;
 154             j_coord_offsetC  = DIM*jnrC;
 155             j_coord_offsetD  = DIM*jnrD;
 156
 157             /* load j atom coordinates */
 158             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 159                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 160                                                  &jx0,&jy0,&jz0);
 161
 162             /* Calculate displacement vector */
 163             dx00             = _mm256_sub_pd(ix0,jx0);
 164             dy00             = _mm256_sub_pd(iy0,jy0);
 165             dz00             = _mm256_sub_pd(iz0,jz0);
 166
 167             /* Calculate squared distance and things based on it */
 168             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 169
 170             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 171
 172             /* Load parameters for j particles */
 173             vdwjidx0A        = 2*vdwtype[jnrA+0];
 174             vdwjidx0B        = 2*vdwtype[jnrB+0];
 175             vdwjidx0C        = 2*vdwtype[jnrC+0];
 176             vdwjidx0D        = 2*vdwtype[jnrD+0];
 177
 178             /**************************
 179              * CALCULATE INTERACTIONS *
 180              **************************/
 181
 182             /* Compute parameters for interactions between i and j atoms */
 183             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 184                                             vdwioffsetptr0+vdwjidx0B,
 185                                             vdwioffsetptr0+vdwjidx0C,
 186                                             vdwioffsetptr0+vdwjidx0D,
 187                                             &c6_00,&c12_00);
 188
 189             /* LENNARD-JONES DISPERSION/REPULSION */
 190
 191             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 192             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 193             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 194             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 195             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 196
 197             /* Update potential sum for this i atom from the interaction with this j atom. */
 198             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 199
 200             fscal            = fvdw;
 201
 202             /* Calculate temporary vectorial force */
 203             tx               = _mm256_mul_pd(fscal,dx00);
 204             ty               = _mm256_mul_pd(fscal,dy00);
 205             tz               = _mm256_mul_pd(fscal,dz00);
 206
 207             /* Update vectorial force */
 208             fix0             = _mm256_add_pd(fix0,tx);
 209             fiy0             = _mm256_add_pd(fiy0,ty);
 210             fiz0             = _mm256_add_pd(fiz0,tz);
 211
 212             fjptrA             = f+j_coord_offsetA;
 213             fjptrB             = f+j_coord_offsetB;
 214             fjptrC             = f+j_coord_offsetC;
 215             fjptrD             = f+j_coord_offsetD;
 216             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 217
 218             /* Inner loop uses 32 flops */
 219         }
 220
 221         if(jidx<j_index_end)
 222         {
 223
 224             /* Get j neighbor index, and coordinate index */
 225             jnrlistA         = jjnr[jidx];
 226             jnrlistB         = jjnr[jidx+1];
 227             jnrlistC         = jjnr[jidx+2];
 228             jnrlistD         = jjnr[jidx+3];
 229             /* Sign of each element will be negative for non-real atoms.
 230              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 231              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 232              */
 233             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 234
 235             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 236             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 237             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 238
 239             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 240             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 241             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 242             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 243             j_coord_offsetA  = DIM*jnrA;
 244             j_coord_offsetB  = DIM*jnrB;
 245             j_coord_offsetC  = DIM*jnrC;
 246             j_coord_offsetD  = DIM*jnrD;
 247
 248             /* load j atom coordinates */
 249             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 250                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 251                                                  &jx0,&jy0,&jz0);
 252
 253             /* Calculate displacement vector */
 254             dx00             = _mm256_sub_pd(ix0,jx0);
 255             dy00             = _mm256_sub_pd(iy0,jy0);
 256             dz00             = _mm256_sub_pd(iz0,jz0);
 257
 258             /* Calculate squared distance and things based on it */
 259             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 260
 261             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 262
 263             /* Load parameters for j particles */
 264             vdwjidx0A        = 2*vdwtype[jnrA+0];
 265             vdwjidx0B        = 2*vdwtype[jnrB+0];
 266             vdwjidx0C        = 2*vdwtype[jnrC+0];
 267             vdwjidx0D        = 2*vdwtype[jnrD+0];
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 275                                             vdwioffsetptr0+vdwjidx0B,
 276                                             vdwioffsetptr0+vdwjidx0C,
 277                                             vdwioffsetptr0+vdwjidx0D,
 278                                             &c6_00,&c12_00);
 279
 280             /* LENNARD-JONES DISPERSION/REPULSION */
 281
 282             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 283             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 284             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 285             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 286             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 290             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 291
 292             fscal            = fvdw;
 293
 294             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm256_mul_pd(fscal,dx00);
 298             ty               = _mm256_mul_pd(fscal,dy00);
 299             tz               = _mm256_mul_pd(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm256_add_pd(fix0,tx);
 303             fiy0             = _mm256_add_pd(fiy0,ty);
 304             fiz0             = _mm256_add_pd(fiz0,tz);
 305
 306             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 307             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 308             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 309             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 310             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 311
 312             /* Inner loop uses 32 flops */
 313         }
 314
 315         /* End of innermost loop */
 316
 317         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 318                                                  f+i_coord_offset,fshift+i_shift_offset);
 319
 320         ggid                        = gid[iidx];
 321         /* Update potential energies */
 322         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 323
 324         /* Increment number of inner iterations */
 325         inneriter                  += j_index_end - j_index_start;
 326
 327         /* Outer loop uses 7 flops */
 328     }
 329
 330     /* Increment number of outer iterations */
 331     outeriter        += nri;
 332
 333     /* Update outer/inner flops */
 334
 335     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*32);
 336 }
 337 /*
 338  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_256_double
 339  * Electrostatics interaction: None
 340  * VdW interaction:            LennardJones
 341  * Geometry:                   Particle-Particle
 342  * Calculate force/pot:        Force
 343  */
 344 void
 345 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_256_double
 346                     (t_nblist * gmx_restrict                nlist,
 347                      rvec * gmx_restrict                    xx,
 348                      rvec * gmx_restrict                    ff,
 349                      t_forcerec * gmx_restrict              fr,
 350                      t_mdatoms * gmx_restrict               mdatoms,
 351                      nb_kernel_data_t * gmx_restrict        kernel_data,
 352                      t_nrnb * gmx_restrict                  nrnb)
 353 {
 354     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 355      * just 0 for non-waters.
 356      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 357      * jnr indices corresponding to data put in the four positions in the SIMD register.
 358      */
 359     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 360     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 361     int              jnrA,jnrB,jnrC,jnrD;
 362     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 363     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 364     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 365     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 366     real             rcutoff_scalar;
 367     real             *shiftvec,*fshift,*x,*f;
 368     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 369     real             scratch[4*DIM];
 370     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 371     real *           vdwioffsetptr0;
 372     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 373     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 374     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 375     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 376     int              nvdwtype;
 377     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 378     int              *vdwtype;
 379     real             *vdwparam;
 380     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 381     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 382     __m256d          dummy_mask,cutoff_mask;
 383     __m128           tmpmask0,tmpmask1;
 384     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 385     __m256d          one     = _mm256_set1_pd(1.0);
 386     __m256d          two     = _mm256_set1_pd(2.0);
 387     x                = xx[0];
 388     f                = ff[0];
 389
 390     nri              = nlist->nri;
 391     iinr             = nlist->iinr;
 392     jindex           = nlist->jindex;
 393     jjnr             = nlist->jjnr;
 394     shiftidx         = nlist->shift;
 395     gid              = nlist->gid;
 396     shiftvec         = fr->shift_vec[0];
 397     fshift           = fr->fshift[0];
 398     nvdwtype         = fr->ntype;
 399     vdwparam         = fr->nbfp;
 400     vdwtype          = mdatoms->typeA;
 401
 402     /* Avoid stupid compiler warnings */
 403     jnrA = jnrB = jnrC = jnrD = 0;
 404     j_coord_offsetA = 0;
 405     j_coord_offsetB = 0;
 406     j_coord_offsetC = 0;
 407     j_coord_offsetD = 0;
 408
 409     outeriter        = 0;
 410     inneriter        = 0;
 411
 412     for(iidx=0;iidx<4*DIM;iidx++)
 413     {
 414         scratch[iidx] = 0.0;
 415     }
 416
 417     /* Start outer loop over neighborlists */
 418     for(iidx=0; iidx<nri; iidx++)
 419     {
 420         /* Load shift vector for this list */
 421         i_shift_offset   = DIM*shiftidx[iidx];
 422
 423         /* Load limits for loop over neighbors */
 424         j_index_start    = jindex[iidx];
 425         j_index_end      = jindex[iidx+1];
 426
 427         /* Get outer coordinate index */
 428         inr              = iinr[iidx];
 429         i_coord_offset   = DIM*inr;
 430
 431         /* Load i particle coords and add shift vector */
 432         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 433
 434         fix0             = _mm256_setzero_pd();
 435         fiy0             = _mm256_setzero_pd();
 436         fiz0             = _mm256_setzero_pd();
 437
 438         /* Load parameters for i particles */
 439         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 440
 441         /* Start inner kernel loop */
 442         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 443         {
 444
 445             /* Get j neighbor index, and coordinate index */
 446             jnrA             = jjnr[jidx];
 447             jnrB             = jjnr[jidx+1];
 448             jnrC             = jjnr[jidx+2];
 449             jnrD             = jjnr[jidx+3];
 450             j_coord_offsetA  = DIM*jnrA;
 451             j_coord_offsetB  = DIM*jnrB;
 452             j_coord_offsetC  = DIM*jnrC;
 453             j_coord_offsetD  = DIM*jnrD;
 454
 455             /* load j atom coordinates */
 456             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 457                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 458                                                  &jx0,&jy0,&jz0);
 459
 460             /* Calculate displacement vector */
 461             dx00             = _mm256_sub_pd(ix0,jx0);
 462             dy00             = _mm256_sub_pd(iy0,jy0);
 463             dz00             = _mm256_sub_pd(iz0,jz0);
 464
 465             /* Calculate squared distance and things based on it */
 466             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 467
 468             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 469
 470             /* Load parameters for j particles */
 471             vdwjidx0A        = 2*vdwtype[jnrA+0];
 472             vdwjidx0B        = 2*vdwtype[jnrB+0];
 473             vdwjidx0C        = 2*vdwtype[jnrC+0];
 474             vdwjidx0D        = 2*vdwtype[jnrD+0];
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             /* Compute parameters for interactions between i and j atoms */
 481             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 482                                             vdwioffsetptr0+vdwjidx0B,
 483                                             vdwioffsetptr0+vdwjidx0C,
 484                                             vdwioffsetptr0+vdwjidx0D,
 485                                             &c6_00,&c12_00);
 486
 487             /* LENNARD-JONES DISPERSION/REPULSION */
 488
 489             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 490             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 491
 492             fscal            = fvdw;
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm256_mul_pd(fscal,dx00);
 496             ty               = _mm256_mul_pd(fscal,dy00);
 497             tz               = _mm256_mul_pd(fscal,dz00);
 498
 499             /* Update vectorial force */
 500             fix0             = _mm256_add_pd(fix0,tx);
 501             fiy0             = _mm256_add_pd(fiy0,ty);
 502             fiz0             = _mm256_add_pd(fiz0,tz);
 503
 504             fjptrA             = f+j_coord_offsetA;
 505             fjptrB             = f+j_coord_offsetB;
 506             fjptrC             = f+j_coord_offsetC;
 507             fjptrD             = f+j_coord_offsetD;
 508             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 509
 510             /* Inner loop uses 27 flops */
 511         }
 512
 513         if(jidx<j_index_end)
 514         {
 515
 516             /* Get j neighbor index, and coordinate index */
 517             jnrlistA         = jjnr[jidx];
 518             jnrlistB         = jjnr[jidx+1];
 519             jnrlistC         = jjnr[jidx+2];
 520             jnrlistD         = jjnr[jidx+3];
 521             /* Sign of each element will be negative for non-real atoms.
 522              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 523              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 524              */
 525             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 526
 527             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 528             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 529             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 530
 531             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 532             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 533             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 534             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 535             j_coord_offsetA  = DIM*jnrA;
 536             j_coord_offsetB  = DIM*jnrB;
 537             j_coord_offsetC  = DIM*jnrC;
 538             j_coord_offsetD  = DIM*jnrD;
 539
 540             /* load j atom coordinates */
 541             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 542                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 543                                                  &jx0,&jy0,&jz0);
 544
 545             /* Calculate displacement vector */
 546             dx00             = _mm256_sub_pd(ix0,jx0);
 547             dy00             = _mm256_sub_pd(iy0,jy0);
 548             dz00             = _mm256_sub_pd(iz0,jz0);
 549
 550             /* Calculate squared distance and things based on it */
 551             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 552
 553             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 554
 555             /* Load parameters for j particles */
 556             vdwjidx0A        = 2*vdwtype[jnrA+0];
 557             vdwjidx0B        = 2*vdwtype[jnrB+0];
 558             vdwjidx0C        = 2*vdwtype[jnrC+0];
 559             vdwjidx0D        = 2*vdwtype[jnrD+0];
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 567                                             vdwioffsetptr0+vdwjidx0B,
 568                                             vdwioffsetptr0+vdwjidx0C,
 569                                             vdwioffsetptr0+vdwjidx0D,
 570                                             &c6_00,&c12_00);
 571
 572             /* LENNARD-JONES DISPERSION/REPULSION */
 573
 574             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 575             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 576
 577             fscal            = fvdw;
 578
 579             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm256_mul_pd(fscal,dx00);
 583             ty               = _mm256_mul_pd(fscal,dy00);
 584             tz               = _mm256_mul_pd(fscal,dz00);
 585
 586             /* Update vectorial force */
 587             fix0             = _mm256_add_pd(fix0,tx);
 588             fiy0             = _mm256_add_pd(fiy0,ty);
 589             fiz0             = _mm256_add_pd(fiz0,tz);
 590
 591             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 592             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 593             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 594             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 595             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 596
 597             /* Inner loop uses 27 flops */
 598         }
 599
 600         /* End of innermost loop */
 601
 602         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 603                                                  f+i_coord_offset,fshift+i_shift_offset);
 604
 605         /* Increment number of inner iterations */
 606         inneriter                  += j_index_end - j_index_start;
 607
 608         /* Outer loop uses 6 flops */
 609     }
 610
 611     /* Increment number of outer iterations */
 612     outeriter        += nri;
 613
 614     /* Update outer/inner flops */
 615
 616     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*27);
 617 }