src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecNone_VdwLJSh_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_VdwLJSh_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_VdwLJSh_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     rcutoff_scalar   = fr->rvdw;
 102     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 103     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 104
 105     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 106     rvdw             = _mm256_set1_pd(fr->rvdw);
 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         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 146
 147         /* Reset potential sums */
 148         vvdwsum          = _mm256_setzero_pd();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             jnrC             = jjnr[jidx+2];
 158             jnrD             = jjnr[jidx+3];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161             j_coord_offsetC  = DIM*jnrC;
 162             j_coord_offsetD  = DIM*jnrD;
 163
 164             /* load j atom coordinates */
 165             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 167                                                  &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx00             = _mm256_sub_pd(ix0,jx0);
 171             dy00             = _mm256_sub_pd(iy0,jy0);
 172             dz00             = _mm256_sub_pd(iz0,jz0);
 173
 174             /* Calculate squared distance and things based on it */
 175             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 176
 177             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 178
 179             /* Load parameters for j particles */
 180             vdwjidx0A        = 2*vdwtype[jnrA+0];
 181             vdwjidx0B        = 2*vdwtype[jnrB+0];
 182             vdwjidx0C        = 2*vdwtype[jnrC+0];
 183             vdwjidx0D        = 2*vdwtype[jnrD+0];
 184
 185             /**************************
 186              * CALCULATE INTERACTIONS *
 187              **************************/
 188
 189             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 190             {
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 194                                             vdwioffsetptr0+vdwjidx0B,
 195                                             vdwioffsetptr0+vdwjidx0C,
 196                                             vdwioffsetptr0+vdwjidx0D,
 197                                             &c6_00,&c12_00);
 198
 199             /* LENNARD-JONES DISPERSION/REPULSION */
 200
 201             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 202             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 203             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 204             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 205                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 206             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 207
 208             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 209
 210             /* Update potential sum for this i atom from the interaction with this j atom. */
 211             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 212             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 213
 214             fscal            = fvdw;
 215
 216             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 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             }
 235
 236             /* Inner loop uses 41 flops */
 237         }
 238
 239         if(jidx<j_index_end)
 240         {
 241
 242             /* Get j neighbor index, and coordinate index */
 243             jnrlistA         = jjnr[jidx];
 244             jnrlistB         = jjnr[jidx+1];
 245             jnrlistC         = jjnr[jidx+2];
 246             jnrlistD         = jjnr[jidx+3];
 247             /* Sign of each element will be negative for non-real atoms.
 248              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 249              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 250              */
 251             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 252
 253             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 254             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 255             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 256
 257             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 258             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 259             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 260             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 261             j_coord_offsetA  = DIM*jnrA;
 262             j_coord_offsetB  = DIM*jnrB;
 263             j_coord_offsetC  = DIM*jnrC;
 264             j_coord_offsetD  = DIM*jnrD;
 265
 266             /* load j atom coordinates */
 267             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 268                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 269                                                  &jx0,&jy0,&jz0);
 270
 271             /* Calculate displacement vector */
 272             dx00             = _mm256_sub_pd(ix0,jx0);
 273             dy00             = _mm256_sub_pd(iy0,jy0);
 274             dz00             = _mm256_sub_pd(iz0,jz0);
 275
 276             /* Calculate squared distance and things based on it */
 277             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 278
 279             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 280
 281             /* Load parameters for j particles */
 282             vdwjidx0A        = 2*vdwtype[jnrA+0];
 283             vdwjidx0B        = 2*vdwtype[jnrB+0];
 284             vdwjidx0C        = 2*vdwtype[jnrC+0];
 285             vdwjidx0D        = 2*vdwtype[jnrD+0];
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 292             {
 293
 294             /* Compute parameters for interactions between i and j atoms */
 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             /* LENNARD-JONES DISPERSION/REPULSION */
 302
 303             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 304             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 305             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 306             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 307                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 308             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 309
 310             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 314             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 315             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 316
 317             fscal            = fvdw;
 318
 319             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 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             }
 340
 341             /* Inner loop uses 41 flops */
 342         }
 343
 344         /* End of innermost loop */
 345
 346         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 347                                                  f+i_coord_offset,fshift+i_shift_offset);
 348
 349         ggid                        = gid[iidx];
 350         /* Update potential energies */
 351         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 352
 353         /* Increment number of inner iterations */
 354         inneriter                  += j_index_end - j_index_start;
 355
 356         /* Outer loop uses 7 flops */
 357     }
 358
 359     /* Increment number of outer iterations */
 360     outeriter        += nri;
 361
 362     /* Update outer/inner flops */
 363
 364     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*41);
 365 }
 366 /*
 367  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_256_double
 368  * Electrostatics interaction: None
 369  * VdW interaction:            LennardJones
 370  * Geometry:                   Particle-Particle
 371  * Calculate force/pot:        Force
 372  */
 373 void
 374 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_256_double
 375                     (t_nblist * gmx_restrict                nlist,
 376                      rvec * gmx_restrict                    xx,
 377                      rvec * gmx_restrict                    ff,
 378                      t_forcerec * gmx_restrict              fr,
 379                      t_mdatoms * gmx_restrict               mdatoms,
 380                      nb_kernel_data_t * gmx_restrict        kernel_data,
 381                      t_nrnb * gmx_restrict                  nrnb)
 382 {
 383     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 384      * just 0 for non-waters.
 385      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 386      * jnr indices corresponding to data put in the four positions in the SIMD register.
 387      */
 388     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 389     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 390     int              jnrA,jnrB,jnrC,jnrD;
 391     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 392     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 393     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 394     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 395     real             rcutoff_scalar;
 396     real             *shiftvec,*fshift,*x,*f;
 397     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 398     real             scratch[4*DIM];
 399     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 400     real *           vdwioffsetptr0;
 401     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 402     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 403     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 404     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 405     int              nvdwtype;
 406     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 407     int              *vdwtype;
 408     real             *vdwparam;
 409     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 410     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 411     __m256d          dummy_mask,cutoff_mask;
 412     __m128           tmpmask0,tmpmask1;
 413     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 414     __m256d          one     = _mm256_set1_pd(1.0);
 415     __m256d          two     = _mm256_set1_pd(2.0);
 416     x                = xx[0];
 417     f                = ff[0];
 418
 419     nri              = nlist->nri;
 420     iinr             = nlist->iinr;
 421     jindex           = nlist->jindex;
 422     jjnr             = nlist->jjnr;
 423     shiftidx         = nlist->shift;
 424     gid              = nlist->gid;
 425     shiftvec         = fr->shift_vec[0];
 426     fshift           = fr->fshift[0];
 427     nvdwtype         = fr->ntype;
 428     vdwparam         = fr->nbfp;
 429     vdwtype          = mdatoms->typeA;
 430
 431     rcutoff_scalar   = fr->rvdw;
 432     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 433     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 434
 435     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 436     rvdw             = _mm256_set1_pd(fr->rvdw);
 437
 438     /* Avoid stupid compiler warnings */
 439     jnrA = jnrB = jnrC = jnrD = 0;
 440     j_coord_offsetA = 0;
 441     j_coord_offsetB = 0;
 442     j_coord_offsetC = 0;
 443     j_coord_offsetD = 0;
 444
 445     outeriter        = 0;
 446     inneriter        = 0;
 447
 448     for(iidx=0;iidx<4*DIM;iidx++)
 449     {
 450         scratch[iidx] = 0.0;
 451     }
 452
 453     /* Start outer loop over neighborlists */
 454     for(iidx=0; iidx<nri; iidx++)
 455     {
 456         /* Load shift vector for this list */
 457         i_shift_offset   = DIM*shiftidx[iidx];
 458
 459         /* Load limits for loop over neighbors */
 460         j_index_start    = jindex[iidx];
 461         j_index_end      = jindex[iidx+1];
 462
 463         /* Get outer coordinate index */
 464         inr              = iinr[iidx];
 465         i_coord_offset   = DIM*inr;
 466
 467         /* Load i particle coords and add shift vector */
 468         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 469
 470         fix0             = _mm256_setzero_pd();
 471         fiy0             = _mm256_setzero_pd();
 472         fiz0             = _mm256_setzero_pd();
 473
 474         /* Load parameters for i particles */
 475         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 476
 477         /* Start inner kernel loop */
 478         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 479         {
 480
 481             /* Get j neighbor index, and coordinate index */
 482             jnrA             = jjnr[jidx];
 483             jnrB             = jjnr[jidx+1];
 484             jnrC             = jjnr[jidx+2];
 485             jnrD             = jjnr[jidx+3];
 486             j_coord_offsetA  = DIM*jnrA;
 487             j_coord_offsetB  = DIM*jnrB;
 488             j_coord_offsetC  = DIM*jnrC;
 489             j_coord_offsetD  = DIM*jnrD;
 490
 491             /* load j atom coordinates */
 492             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 493                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 494                                                  &jx0,&jy0,&jz0);
 495
 496             /* Calculate displacement vector */
 497             dx00             = _mm256_sub_pd(ix0,jx0);
 498             dy00             = _mm256_sub_pd(iy0,jy0);
 499             dz00             = _mm256_sub_pd(iz0,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 503
 504             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 505
 506             /* Load parameters for j particles */
 507             vdwjidx0A        = 2*vdwtype[jnrA+0];
 508             vdwjidx0B        = 2*vdwtype[jnrB+0];
 509             vdwjidx0C        = 2*vdwtype[jnrC+0];
 510             vdwjidx0D        = 2*vdwtype[jnrD+0];
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 517             {
 518
 519             /* Compute parameters for interactions between i and j atoms */
 520             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 521                                             vdwioffsetptr0+vdwjidx0B,
 522                                             vdwioffsetptr0+vdwjidx0C,
 523                                             vdwioffsetptr0+vdwjidx0D,
 524                                             &c6_00,&c12_00);
 525
 526             /* LENNARD-JONES DISPERSION/REPULSION */
 527
 528             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 529             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 530
 531             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 532
 533             fscal            = fvdw;
 534
 535             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm256_mul_pd(fscal,dx00);
 539             ty               = _mm256_mul_pd(fscal,dy00);
 540             tz               = _mm256_mul_pd(fscal,dz00);
 541
 542             /* Update vectorial force */
 543             fix0             = _mm256_add_pd(fix0,tx);
 544             fiy0             = _mm256_add_pd(fiy0,ty);
 545             fiz0             = _mm256_add_pd(fiz0,tz);
 546
 547             fjptrA             = f+j_coord_offsetA;
 548             fjptrB             = f+j_coord_offsetB;
 549             fjptrC             = f+j_coord_offsetC;
 550             fjptrD             = f+j_coord_offsetD;
 551             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 552
 553             }
 554
 555             /* Inner loop uses 30 flops */
 556         }
 557
 558         if(jidx<j_index_end)
 559         {
 560
 561             /* Get j neighbor index, and coordinate index */
 562             jnrlistA         = jjnr[jidx];
 563             jnrlistB         = jjnr[jidx+1];
 564             jnrlistC         = jjnr[jidx+2];
 565             jnrlistD         = jjnr[jidx+3];
 566             /* Sign of each element will be negative for non-real atoms.
 567              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 568              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 569              */
 570             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 571
 572             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 573             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 574             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 575
 576             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 577             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 578             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 579             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 580             j_coord_offsetA  = DIM*jnrA;
 581             j_coord_offsetB  = DIM*jnrB;
 582             j_coord_offsetC  = DIM*jnrC;
 583             j_coord_offsetD  = DIM*jnrD;
 584
 585             /* load j atom coordinates */
 586             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 587                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 588                                                  &jx0,&jy0,&jz0);
 589
 590             /* Calculate displacement vector */
 591             dx00             = _mm256_sub_pd(ix0,jx0);
 592             dy00             = _mm256_sub_pd(iy0,jy0);
 593             dz00             = _mm256_sub_pd(iz0,jz0);
 594
 595             /* Calculate squared distance and things based on it */
 596             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 597
 598             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 599
 600             /* Load parameters for j particles */
 601             vdwjidx0A        = 2*vdwtype[jnrA+0];
 602             vdwjidx0B        = 2*vdwtype[jnrB+0];
 603             vdwjidx0C        = 2*vdwtype[jnrC+0];
 604             vdwjidx0D        = 2*vdwtype[jnrD+0];
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 611             {
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 615                                             vdwioffsetptr0+vdwjidx0B,
 616                                             vdwioffsetptr0+vdwjidx0C,
 617                                             vdwioffsetptr0+vdwjidx0D,
 618                                             &c6_00,&c12_00);
 619
 620             /* LENNARD-JONES DISPERSION/REPULSION */
 621
 622             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 623             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 624
 625             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 626
 627             fscal            = fvdw;
 628
 629             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 630
 631             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm256_mul_pd(fscal,dx00);
 635             ty               = _mm256_mul_pd(fscal,dy00);
 636             tz               = _mm256_mul_pd(fscal,dz00);
 637
 638             /* Update vectorial force */
 639             fix0             = _mm256_add_pd(fix0,tx);
 640             fiy0             = _mm256_add_pd(fiy0,ty);
 641             fiz0             = _mm256_add_pd(fiz0,tz);
 642
 643             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 644             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 645             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 646             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 647             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 648
 649             }
 650
 651             /* Inner loop uses 30 flops */
 652         }
 653
 654         /* End of innermost loop */
 655
 656         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 657                                                  f+i_coord_offset,fshift+i_shift_offset);
 658
 659         /* Increment number of inner iterations */
 660         inneriter                  += j_index_end - j_index_start;
 661
 662         /* Outer loop uses 6 flops */
 663     }
 664
 665     /* Increment number of outer iterations */
 666     outeriter        += nri;
 667
 668     /* Update outer/inner flops */
 669
 670     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 671 }