src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecNone_VdwLJ_GeomP1P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single 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_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_avx_256_single
  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_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  75     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     int              nvdwtype;
  78     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  79     int              *vdwtype;
  80     real             *vdwparam;
  81     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  82     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  83     __m256           dummy_mask,cutoff_mask;
  84     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  85     __m256           one     = _mm256_set1_ps(1.0);
  86     __m256           two     = _mm256_set1_ps(2.0);
  87     x                = xx[0];
  88     f                = ff[0];
  89
  90     nri              = nlist->nri;
  91     iinr             = nlist->iinr;
  92     jindex           = nlist->jindex;
  93     jjnr             = nlist->jjnr;
  94     shiftidx         = nlist->shift;
  95     gid              = nlist->gid;
  96     shiftvec         = fr->shift_vec[0];
  97     fshift           = fr->fshift[0];
  98     nvdwtype         = fr->ntype;
  99     vdwparam         = fr->nbfp;
 100     vdwtype          = mdatoms->typeA;
 101
 102     /* Avoid stupid compiler warnings */
 103     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 104     j_coord_offsetA = 0;
 105     j_coord_offsetB = 0;
 106     j_coord_offsetC = 0;
 107     j_coord_offsetD = 0;
 108     j_coord_offsetE = 0;
 109     j_coord_offsetF = 0;
 110     j_coord_offsetG = 0;
 111     j_coord_offsetH = 0;
 112
 113     outeriter        = 0;
 114     inneriter        = 0;
 115
 116     for(iidx=0;iidx<4*DIM;iidx++)
 117     {
 118         scratch[iidx] = 0.0;
 119     }
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 137
 138         fix0             = _mm256_setzero_ps();
 139         fiy0             = _mm256_setzero_ps();
 140         fiz0             = _mm256_setzero_ps();
 141
 142         /* Load parameters for i particles */
 143         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 144
 145         /* Reset potential sums */
 146         vvdwsum          = _mm256_setzero_ps();
 147
 148         /* Start inner kernel loop */
 149         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 150         {
 151
 152             /* Get j neighbor index, and coordinate index */
 153             jnrA             = jjnr[jidx];
 154             jnrB             = jjnr[jidx+1];
 155             jnrC             = jjnr[jidx+2];
 156             jnrD             = jjnr[jidx+3];
 157             jnrE             = jjnr[jidx+4];
 158             jnrF             = jjnr[jidx+5];
 159             jnrG             = jjnr[jidx+6];
 160             jnrH             = jjnr[jidx+7];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163             j_coord_offsetC  = DIM*jnrC;
 164             j_coord_offsetD  = DIM*jnrD;
 165             j_coord_offsetE  = DIM*jnrE;
 166             j_coord_offsetF  = DIM*jnrF;
 167             j_coord_offsetG  = DIM*jnrG;
 168             j_coord_offsetH  = DIM*jnrH;
 169
 170             /* load j atom coordinates */
 171             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 173                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 174                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 175                                                  &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm256_sub_ps(ix0,jx0);
 179             dy00             = _mm256_sub_ps(iy0,jy0);
 180             dz00             = _mm256_sub_ps(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 184
 185             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 186
 187             /* Load parameters for j particles */
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190             vdwjidx0C        = 2*vdwtype[jnrC+0];
 191             vdwjidx0D        = 2*vdwtype[jnrD+0];
 192             vdwjidx0E        = 2*vdwtype[jnrE+0];
 193             vdwjidx0F        = 2*vdwtype[jnrF+0];
 194             vdwjidx0G        = 2*vdwtype[jnrG+0];
 195             vdwjidx0H        = 2*vdwtype[jnrH+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             /* Compute parameters for interactions between i and j atoms */
 202             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 203                                             vdwioffsetptr0+vdwjidx0B,
 204                                             vdwioffsetptr0+vdwjidx0C,
 205                                             vdwioffsetptr0+vdwjidx0D,
 206                                             vdwioffsetptr0+vdwjidx0E,
 207                                             vdwioffsetptr0+vdwjidx0F,
 208                                             vdwioffsetptr0+vdwjidx0G,
 209                                             vdwioffsetptr0+vdwjidx0H,
 210                                             &c6_00,&c12_00);
 211
 212             /* LENNARD-JONES DISPERSION/REPULSION */
 213
 214             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 215             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 216             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 217             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 218             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 219
 220             /* Update potential sum for this i atom from the interaction with this j atom. */
 221             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 222
 223             fscal            = fvdw;
 224
 225             /* Calculate temporary vectorial force */
 226             tx               = _mm256_mul_ps(fscal,dx00);
 227             ty               = _mm256_mul_ps(fscal,dy00);
 228             tz               = _mm256_mul_ps(fscal,dz00);
 229
 230             /* Update vectorial force */
 231             fix0             = _mm256_add_ps(fix0,tx);
 232             fiy0             = _mm256_add_ps(fiy0,ty);
 233             fiz0             = _mm256_add_ps(fiz0,tz);
 234
 235             fjptrA             = f+j_coord_offsetA;
 236             fjptrB             = f+j_coord_offsetB;
 237             fjptrC             = f+j_coord_offsetC;
 238             fjptrD             = f+j_coord_offsetD;
 239             fjptrE             = f+j_coord_offsetE;
 240             fjptrF             = f+j_coord_offsetF;
 241             fjptrG             = f+j_coord_offsetG;
 242             fjptrH             = f+j_coord_offsetH;
 243             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 244
 245             /* Inner loop uses 32 flops */
 246         }
 247
 248         if(jidx<j_index_end)
 249         {
 250
 251             /* Get j neighbor index, and coordinate index */
 252             jnrlistA         = jjnr[jidx];
 253             jnrlistB         = jjnr[jidx+1];
 254             jnrlistC         = jjnr[jidx+2];
 255             jnrlistD         = jjnr[jidx+3];
 256             jnrlistE         = jjnr[jidx+4];
 257             jnrlistF         = jjnr[jidx+5];
 258             jnrlistG         = jjnr[jidx+6];
 259             jnrlistH         = jjnr[jidx+7];
 260             /* Sign of each element will be negative for non-real atoms.
 261              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 262              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 263              */
 264             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 265                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 266
 267             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 268             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 269             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 270             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 271             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 272             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 273             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 274             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 275             j_coord_offsetA  = DIM*jnrA;
 276             j_coord_offsetB  = DIM*jnrB;
 277             j_coord_offsetC  = DIM*jnrC;
 278             j_coord_offsetD  = DIM*jnrD;
 279             j_coord_offsetE  = DIM*jnrE;
 280             j_coord_offsetF  = DIM*jnrF;
 281             j_coord_offsetG  = DIM*jnrG;
 282             j_coord_offsetH  = DIM*jnrH;
 283
 284             /* load j atom coordinates */
 285             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 286                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 287                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 288                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 289                                                  &jx0,&jy0,&jz0);
 290
 291             /* Calculate displacement vector */
 292             dx00             = _mm256_sub_ps(ix0,jx0);
 293             dy00             = _mm256_sub_ps(iy0,jy0);
 294             dz00             = _mm256_sub_ps(iz0,jz0);
 295
 296             /* Calculate squared distance and things based on it */
 297             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 298
 299             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 300
 301             /* Load parameters for j particles */
 302             vdwjidx0A        = 2*vdwtype[jnrA+0];
 303             vdwjidx0B        = 2*vdwtype[jnrB+0];
 304             vdwjidx0C        = 2*vdwtype[jnrC+0];
 305             vdwjidx0D        = 2*vdwtype[jnrD+0];
 306             vdwjidx0E        = 2*vdwtype[jnrE+0];
 307             vdwjidx0F        = 2*vdwtype[jnrF+0];
 308             vdwjidx0G        = 2*vdwtype[jnrG+0];
 309             vdwjidx0H        = 2*vdwtype[jnrH+0];
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 317                                             vdwioffsetptr0+vdwjidx0B,
 318                                             vdwioffsetptr0+vdwjidx0C,
 319                                             vdwioffsetptr0+vdwjidx0D,
 320                                             vdwioffsetptr0+vdwjidx0E,
 321                                             vdwioffsetptr0+vdwjidx0F,
 322                                             vdwioffsetptr0+vdwjidx0G,
 323                                             vdwioffsetptr0+vdwjidx0H,
 324                                             &c6_00,&c12_00);
 325
 326             /* LENNARD-JONES DISPERSION/REPULSION */
 327
 328             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 329             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 330             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 331             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 332             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 336             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 337
 338             fscal            = fvdw;
 339
 340             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm256_mul_ps(fscal,dx00);
 344             ty               = _mm256_mul_ps(fscal,dy00);
 345             tz               = _mm256_mul_ps(fscal,dz00);
 346
 347             /* Update vectorial force */
 348             fix0             = _mm256_add_ps(fix0,tx);
 349             fiy0             = _mm256_add_ps(fiy0,ty);
 350             fiz0             = _mm256_add_ps(fiz0,tz);
 351
 352             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 353             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 354             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 355             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 356             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 357             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 358             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 359             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 360             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 361
 362             /* Inner loop uses 32 flops */
 363         }
 364
 365         /* End of innermost loop */
 366
 367         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 368                                                  f+i_coord_offset,fshift+i_shift_offset);
 369
 370         ggid                        = gid[iidx];
 371         /* Update potential energies */
 372         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 373
 374         /* Increment number of inner iterations */
 375         inneriter                  += j_index_end - j_index_start;
 376
 377         /* Outer loop uses 7 flops */
 378     }
 379
 380     /* Increment number of outer iterations */
 381     outeriter        += nri;
 382
 383     /* Update outer/inner flops */
 384
 385     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*32);
 386 }
 387 /*
 388  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_256_single
 389  * Electrostatics interaction: None
 390  * VdW interaction:            LennardJones
 391  * Geometry:                   Particle-Particle
 392  * Calculate force/pot:        Force
 393  */
 394 void
 395 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_256_single
 396                     (t_nblist * gmx_restrict                nlist,
 397                      rvec * gmx_restrict                    xx,
 398                      rvec * gmx_restrict                    ff,
 399                      t_forcerec * gmx_restrict              fr,
 400                      t_mdatoms * gmx_restrict               mdatoms,
 401                      nb_kernel_data_t * gmx_restrict        kernel_data,
 402                      t_nrnb * gmx_restrict                  nrnb)
 403 {
 404     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 405      * just 0 for non-waters.
 406      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 407      * jnr indices corresponding to data put in the four positions in the SIMD register.
 408      */
 409     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 410     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 411     int              jnrA,jnrB,jnrC,jnrD;
 412     int              jnrE,jnrF,jnrG,jnrH;
 413     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 414     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 415     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 416     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 417     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 418     real             rcutoff_scalar;
 419     real             *shiftvec,*fshift,*x,*f;
 420     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 421     real             scratch[4*DIM];
 422     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 423     real *           vdwioffsetptr0;
 424     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 425     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 426     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 427     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 428     int              nvdwtype;
 429     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 430     int              *vdwtype;
 431     real             *vdwparam;
 432     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 433     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 434     __m256           dummy_mask,cutoff_mask;
 435     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 436     __m256           one     = _mm256_set1_ps(1.0);
 437     __m256           two     = _mm256_set1_ps(2.0);
 438     x                = xx[0];
 439     f                = ff[0];
 440
 441     nri              = nlist->nri;
 442     iinr             = nlist->iinr;
 443     jindex           = nlist->jindex;
 444     jjnr             = nlist->jjnr;
 445     shiftidx         = nlist->shift;
 446     gid              = nlist->gid;
 447     shiftvec         = fr->shift_vec[0];
 448     fshift           = fr->fshift[0];
 449     nvdwtype         = fr->ntype;
 450     vdwparam         = fr->nbfp;
 451     vdwtype          = mdatoms->typeA;
 452
 453     /* Avoid stupid compiler warnings */
 454     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 455     j_coord_offsetA = 0;
 456     j_coord_offsetB = 0;
 457     j_coord_offsetC = 0;
 458     j_coord_offsetD = 0;
 459     j_coord_offsetE = 0;
 460     j_coord_offsetF = 0;
 461     j_coord_offsetG = 0;
 462     j_coord_offsetH = 0;
 463
 464     outeriter        = 0;
 465     inneriter        = 0;
 466
 467     for(iidx=0;iidx<4*DIM;iidx++)
 468     {
 469         scratch[iidx] = 0.0;
 470     }
 471
 472     /* Start outer loop over neighborlists */
 473     for(iidx=0; iidx<nri; iidx++)
 474     {
 475         /* Load shift vector for this list */
 476         i_shift_offset   = DIM*shiftidx[iidx];
 477
 478         /* Load limits for loop over neighbors */
 479         j_index_start    = jindex[iidx];
 480         j_index_end      = jindex[iidx+1];
 481
 482         /* Get outer coordinate index */
 483         inr              = iinr[iidx];
 484         i_coord_offset   = DIM*inr;
 485
 486         /* Load i particle coords and add shift vector */
 487         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 488
 489         fix0             = _mm256_setzero_ps();
 490         fiy0             = _mm256_setzero_ps();
 491         fiz0             = _mm256_setzero_ps();
 492
 493         /* Load parameters for i particles */
 494         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 495
 496         /* Start inner kernel loop */
 497         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 498         {
 499
 500             /* Get j neighbor index, and coordinate index */
 501             jnrA             = jjnr[jidx];
 502             jnrB             = jjnr[jidx+1];
 503             jnrC             = jjnr[jidx+2];
 504             jnrD             = jjnr[jidx+3];
 505             jnrE             = jjnr[jidx+4];
 506             jnrF             = jjnr[jidx+5];
 507             jnrG             = jjnr[jidx+6];
 508             jnrH             = jjnr[jidx+7];
 509             j_coord_offsetA  = DIM*jnrA;
 510             j_coord_offsetB  = DIM*jnrB;
 511             j_coord_offsetC  = DIM*jnrC;
 512             j_coord_offsetD  = DIM*jnrD;
 513             j_coord_offsetE  = DIM*jnrE;
 514             j_coord_offsetF  = DIM*jnrF;
 515             j_coord_offsetG  = DIM*jnrG;
 516             j_coord_offsetH  = DIM*jnrH;
 517
 518             /* load j atom coordinates */
 519             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 520                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 521                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 522                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 523                                                  &jx0,&jy0,&jz0);
 524
 525             /* Calculate displacement vector */
 526             dx00             = _mm256_sub_ps(ix0,jx0);
 527             dy00             = _mm256_sub_ps(iy0,jy0);
 528             dz00             = _mm256_sub_ps(iz0,jz0);
 529
 530             /* Calculate squared distance and things based on it */
 531             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 532
 533             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 534
 535             /* Load parameters for j particles */
 536             vdwjidx0A        = 2*vdwtype[jnrA+0];
 537             vdwjidx0B        = 2*vdwtype[jnrB+0];
 538             vdwjidx0C        = 2*vdwtype[jnrC+0];
 539             vdwjidx0D        = 2*vdwtype[jnrD+0];
 540             vdwjidx0E        = 2*vdwtype[jnrE+0];
 541             vdwjidx0F        = 2*vdwtype[jnrF+0];
 542             vdwjidx0G        = 2*vdwtype[jnrG+0];
 543             vdwjidx0H        = 2*vdwtype[jnrH+0];
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 551                                             vdwioffsetptr0+vdwjidx0B,
 552                                             vdwioffsetptr0+vdwjidx0C,
 553                                             vdwioffsetptr0+vdwjidx0D,
 554                                             vdwioffsetptr0+vdwjidx0E,
 555                                             vdwioffsetptr0+vdwjidx0F,
 556                                             vdwioffsetptr0+vdwjidx0G,
 557                                             vdwioffsetptr0+vdwjidx0H,
 558                                             &c6_00,&c12_00);
 559
 560             /* LENNARD-JONES DISPERSION/REPULSION */
 561
 562             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 563             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 564
 565             fscal            = fvdw;
 566
 567             /* Calculate temporary vectorial force */
 568             tx               = _mm256_mul_ps(fscal,dx00);
 569             ty               = _mm256_mul_ps(fscal,dy00);
 570             tz               = _mm256_mul_ps(fscal,dz00);
 571
 572             /* Update vectorial force */
 573             fix0             = _mm256_add_ps(fix0,tx);
 574             fiy0             = _mm256_add_ps(fiy0,ty);
 575             fiz0             = _mm256_add_ps(fiz0,tz);
 576
 577             fjptrA             = f+j_coord_offsetA;
 578             fjptrB             = f+j_coord_offsetB;
 579             fjptrC             = f+j_coord_offsetC;
 580             fjptrD             = f+j_coord_offsetD;
 581             fjptrE             = f+j_coord_offsetE;
 582             fjptrF             = f+j_coord_offsetF;
 583             fjptrG             = f+j_coord_offsetG;
 584             fjptrH             = f+j_coord_offsetH;
 585             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 586
 587             /* Inner loop uses 27 flops */
 588         }
 589
 590         if(jidx<j_index_end)
 591         {
 592
 593             /* Get j neighbor index, and coordinate index */
 594             jnrlistA         = jjnr[jidx];
 595             jnrlistB         = jjnr[jidx+1];
 596             jnrlistC         = jjnr[jidx+2];
 597             jnrlistD         = jjnr[jidx+3];
 598             jnrlistE         = jjnr[jidx+4];
 599             jnrlistF         = jjnr[jidx+5];
 600             jnrlistG         = jjnr[jidx+6];
 601             jnrlistH         = jjnr[jidx+7];
 602             /* Sign of each element will be negative for non-real atoms.
 603              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 604              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 605              */
 606             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 607                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 608
 609             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 610             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 611             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 612             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 613             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 614             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 615             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 616             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 617             j_coord_offsetA  = DIM*jnrA;
 618             j_coord_offsetB  = DIM*jnrB;
 619             j_coord_offsetC  = DIM*jnrC;
 620             j_coord_offsetD  = DIM*jnrD;
 621             j_coord_offsetE  = DIM*jnrE;
 622             j_coord_offsetF  = DIM*jnrF;
 623             j_coord_offsetG  = DIM*jnrG;
 624             j_coord_offsetH  = DIM*jnrH;
 625
 626             /* load j atom coordinates */
 627             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 628                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 629                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 630                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 631                                                  &jx0,&jy0,&jz0);
 632
 633             /* Calculate displacement vector */
 634             dx00             = _mm256_sub_ps(ix0,jx0);
 635             dy00             = _mm256_sub_ps(iy0,jy0);
 636             dz00             = _mm256_sub_ps(iz0,jz0);
 637
 638             /* Calculate squared distance and things based on it */
 639             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 640
 641             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 642
 643             /* Load parameters for j particles */
 644             vdwjidx0A        = 2*vdwtype[jnrA+0];
 645             vdwjidx0B        = 2*vdwtype[jnrB+0];
 646             vdwjidx0C        = 2*vdwtype[jnrC+0];
 647             vdwjidx0D        = 2*vdwtype[jnrD+0];
 648             vdwjidx0E        = 2*vdwtype[jnrE+0];
 649             vdwjidx0F        = 2*vdwtype[jnrF+0];
 650             vdwjidx0G        = 2*vdwtype[jnrG+0];
 651             vdwjidx0H        = 2*vdwtype[jnrH+0];
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 659                                             vdwioffsetptr0+vdwjidx0B,
 660                                             vdwioffsetptr0+vdwjidx0C,
 661                                             vdwioffsetptr0+vdwjidx0D,
 662                                             vdwioffsetptr0+vdwjidx0E,
 663                                             vdwioffsetptr0+vdwjidx0F,
 664                                             vdwioffsetptr0+vdwjidx0G,
 665                                             vdwioffsetptr0+vdwjidx0H,
 666                                             &c6_00,&c12_00);
 667
 668             /* LENNARD-JONES DISPERSION/REPULSION */
 669
 670             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 671             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 672
 673             fscal            = fvdw;
 674
 675             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 676
 677             /* Calculate temporary vectorial force */
 678             tx               = _mm256_mul_ps(fscal,dx00);
 679             ty               = _mm256_mul_ps(fscal,dy00);
 680             tz               = _mm256_mul_ps(fscal,dz00);
 681
 682             /* Update vectorial force */
 683             fix0             = _mm256_add_ps(fix0,tx);
 684             fiy0             = _mm256_add_ps(fiy0,ty);
 685             fiz0             = _mm256_add_ps(fiz0,tz);
 686
 687             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 688             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 689             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 690             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 691             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 692             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 693             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 694             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 695             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 696
 697             /* Inner loop uses 27 flops */
 698         }
 699
 700         /* End of innermost loop */
 701
 702         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 703                                                  f+i_coord_offset,fshift+i_shift_offset);
 704
 705         /* Increment number of inner iterations */
 706         inneriter                  += j_index_end - j_index_start;
 707
 708         /* Outer loop uses 6 flops */
 709     }
 710
 711     /* Increment number of outer iterations */
 712     outeriter        += nri;
 713
 714     /* Update outer/inner flops */
 715
 716     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*27);
 717 }