src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     real *           vdwgridioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  98     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  99     __m256           c6grid_00;
 100     real             *vdwgridparam;
 101     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 102     __m256           one_half  = _mm256_set1_ps(0.5);
 103     __m256           minus_one = _mm256_set1_ps(-1.0);
 104     __m256i          ewitab;
 105     __m128i          ewitab_lo,ewitab_hi;
 106     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 108     real             *ewtab;
 109     __m256           dummy_mask,cutoff_mask;
 110     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 111     __m256           one     = _mm256_set1_ps(1.0);
 112     __m256           two     = _mm256_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_ps(fr->ic->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129     vdwgridparam     = fr->ljpme_c6grid;
 130     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 131     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
 132     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 133
 134     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_ps(beta,beta);
 137     beta3            = _mm256_mul_ps(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 142
 143     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 144     rcutoff_scalar   = fr->ic->rcoulomb;
 145     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 146     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 147
 148     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 149     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157     j_coord_offsetE = 0;
 158     j_coord_offsetF = 0;
 159     j_coord_offsetG = 0;
 160     j_coord_offsetH = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     for(iidx=0;iidx<4*DIM;iidx++)
 166     {
 167         scratch[iidx] = 0.0;
 168     }
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 186
 187         fix0             = _mm256_setzero_ps();
 188         fiy0             = _mm256_setzero_ps();
 189         fiz0             = _mm256_setzero_ps();
 190
 191         /* Load parameters for i particles */
 192         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 193         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 194         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm256_setzero_ps();
 198         vvdwsum          = _mm256_setzero_ps();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             jnrE             = jjnr[jidx+4];
 210             jnrF             = jjnr[jidx+5];
 211             jnrG             = jjnr[jidx+6];
 212             jnrH             = jjnr[jidx+7];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217             j_coord_offsetE  = DIM*jnrE;
 218             j_coord_offsetF  = DIM*jnrF;
 219             j_coord_offsetG  = DIM*jnrG;
 220             j_coord_offsetH  = DIM*jnrH;
 221
 222             /* load j atom coordinates */
 223             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 224                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 225                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 226                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 227                                                  &jx0,&jy0,&jz0);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm256_sub_ps(ix0,jx0);
 231             dy00             = _mm256_sub_ps(iy0,jy0);
 232             dz00             = _mm256_sub_ps(iz0,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 236
 237             rinv00           = avx256_invsqrt_f(rsq00);
 238
 239             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                                  charge+jnrC+0,charge+jnrD+0,
 244                                                                  charge+jnrE+0,charge+jnrF+0,
 245                                                                  charge+jnrG+0,charge+jnrH+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248             vdwjidx0C        = 2*vdwtype[jnrC+0];
 249             vdwjidx0D        = 2*vdwtype[jnrD+0];
 250             vdwjidx0E        = 2*vdwtype[jnrE+0];
 251             vdwjidx0F        = 2*vdwtype[jnrF+0];
 252             vdwjidx0G        = 2*vdwtype[jnrG+0];
 253             vdwjidx0H        = 2*vdwtype[jnrH+0];
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 260             {
 261
 262             r00              = _mm256_mul_ps(rsq00,rinv00);
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             qq00             = _mm256_mul_ps(iq0,jq0);
 266             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 267                                             vdwioffsetptr0+vdwjidx0B,
 268                                             vdwioffsetptr0+vdwjidx0C,
 269                                             vdwioffsetptr0+vdwjidx0D,
 270                                             vdwioffsetptr0+vdwjidx0E,
 271                                             vdwioffsetptr0+vdwjidx0F,
 272                                             vdwioffsetptr0+vdwjidx0G,
 273                                             vdwioffsetptr0+vdwjidx0H,
 274                                             &c6_00,&c12_00);
 275
 276             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 277                                                                   vdwgridioffsetptr0+vdwjidx0B,
 278                                                                   vdwgridioffsetptr0+vdwjidx0C,
 279                                                                   vdwgridioffsetptr0+vdwjidx0D,
 280                                                                   vdwgridioffsetptr0+vdwjidx0E,
 281                                                                   vdwgridioffsetptr0+vdwjidx0F,
 282                                                                   vdwgridioffsetptr0+vdwjidx0G,
 283                                                                   vdwgridioffsetptr0+vdwjidx0H);
 284
 285             /* EWALD ELECTROSTATICS */
 286
 287             /* Analytical PME correction */
 288             zeta2            = _mm256_mul_ps(beta2,rsq00);
 289             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 290             pmecorrF         = avx256_pmecorrF_f(zeta2);
 291             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 292             felec            = _mm256_mul_ps(qq00,felec);
 293             pmecorrV         = avx256_pmecorrV_f(zeta2);
 294             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 295             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 296             velec            = _mm256_mul_ps(qq00,velec);
 297
 298             /* Analytical LJ-PME */
 299             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 300             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 301             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 302             exponent         = avx256_exp_f(ewcljrsq);
 303             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 304             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 305             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 306             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 307             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 308             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 309                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 310             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 311             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
 312
 313             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm256_and_ps(velec,cutoff_mask);
 317             velecsum         = _mm256_add_ps(velecsum,velec);
 318             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 319             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 320
 321             fscal            = _mm256_add_ps(felec,fvdw);
 322
 323             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm256_mul_ps(fscal,dx00);
 327             ty               = _mm256_mul_ps(fscal,dy00);
 328             tz               = _mm256_mul_ps(fscal,dz00);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm256_add_ps(fix0,tx);
 332             fiy0             = _mm256_add_ps(fiy0,ty);
 333             fiz0             = _mm256_add_ps(fiz0,tz);
 334
 335             fjptrA             = f+j_coord_offsetA;
 336             fjptrB             = f+j_coord_offsetB;
 337             fjptrC             = f+j_coord_offsetC;
 338             fjptrD             = f+j_coord_offsetD;
 339             fjptrE             = f+j_coord_offsetE;
 340             fjptrF             = f+j_coord_offsetF;
 341             fjptrG             = f+j_coord_offsetG;
 342             fjptrH             = f+j_coord_offsetH;
 343             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 344
 345             }
 346
 347             /* Inner loop uses 145 flops */
 348         }
 349
 350         if(jidx<j_index_end)
 351         {
 352
 353             /* Get j neighbor index, and coordinate index */
 354             jnrlistA         = jjnr[jidx];
 355             jnrlistB         = jjnr[jidx+1];
 356             jnrlistC         = jjnr[jidx+2];
 357             jnrlistD         = jjnr[jidx+3];
 358             jnrlistE         = jjnr[jidx+4];
 359             jnrlistF         = jjnr[jidx+5];
 360             jnrlistG         = jjnr[jidx+6];
 361             jnrlistH         = jjnr[jidx+7];
 362             /* Sign of each element will be negative for non-real atoms.
 363              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 364              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 365              */
 366             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 367                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 368
 369             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 370             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 371             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 372             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 373             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 374             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 375             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 376             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 377             j_coord_offsetA  = DIM*jnrA;
 378             j_coord_offsetB  = DIM*jnrB;
 379             j_coord_offsetC  = DIM*jnrC;
 380             j_coord_offsetD  = DIM*jnrD;
 381             j_coord_offsetE  = DIM*jnrE;
 382             j_coord_offsetF  = DIM*jnrF;
 383             j_coord_offsetG  = DIM*jnrG;
 384             j_coord_offsetH  = DIM*jnrH;
 385
 386             /* load j atom coordinates */
 387             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 388                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 389                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 390                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 391                                                  &jx0,&jy0,&jz0);
 392
 393             /* Calculate displacement vector */
 394             dx00             = _mm256_sub_ps(ix0,jx0);
 395             dy00             = _mm256_sub_ps(iy0,jy0);
 396             dz00             = _mm256_sub_ps(iz0,jz0);
 397
 398             /* Calculate squared distance and things based on it */
 399             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 400
 401             rinv00           = avx256_invsqrt_f(rsq00);
 402
 403             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 404
 405             /* Load parameters for j particles */
 406             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 407                                                                  charge+jnrC+0,charge+jnrD+0,
 408                                                                  charge+jnrE+0,charge+jnrF+0,
 409                                                                  charge+jnrG+0,charge+jnrH+0);
 410             vdwjidx0A        = 2*vdwtype[jnrA+0];
 411             vdwjidx0B        = 2*vdwtype[jnrB+0];
 412             vdwjidx0C        = 2*vdwtype[jnrC+0];
 413             vdwjidx0D        = 2*vdwtype[jnrD+0];
 414             vdwjidx0E        = 2*vdwtype[jnrE+0];
 415             vdwjidx0F        = 2*vdwtype[jnrF+0];
 416             vdwjidx0G        = 2*vdwtype[jnrG+0];
 417             vdwjidx0H        = 2*vdwtype[jnrH+0];
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 424             {
 425
 426             r00              = _mm256_mul_ps(rsq00,rinv00);
 427             r00              = _mm256_andnot_ps(dummy_mask,r00);
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq00             = _mm256_mul_ps(iq0,jq0);
 431             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 432                                             vdwioffsetptr0+vdwjidx0B,
 433                                             vdwioffsetptr0+vdwjidx0C,
 434                                             vdwioffsetptr0+vdwjidx0D,
 435                                             vdwioffsetptr0+vdwjidx0E,
 436                                             vdwioffsetptr0+vdwjidx0F,
 437                                             vdwioffsetptr0+vdwjidx0G,
 438                                             vdwioffsetptr0+vdwjidx0H,
 439                                             &c6_00,&c12_00);
 440
 441             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 442                                                                   vdwgridioffsetptr0+vdwjidx0B,
 443                                                                   vdwgridioffsetptr0+vdwjidx0C,
 444                                                                   vdwgridioffsetptr0+vdwjidx0D,
 445                                                                   vdwgridioffsetptr0+vdwjidx0E,
 446                                                                   vdwgridioffsetptr0+vdwjidx0F,
 447                                                                   vdwgridioffsetptr0+vdwjidx0G,
 448                                                                   vdwgridioffsetptr0+vdwjidx0H);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Analytical PME correction */
 453             zeta2            = _mm256_mul_ps(beta2,rsq00);
 454             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 455             pmecorrF         = avx256_pmecorrF_f(zeta2);
 456             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 457             felec            = _mm256_mul_ps(qq00,felec);
 458             pmecorrV         = avx256_pmecorrV_f(zeta2);
 459             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 460             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 461             velec            = _mm256_mul_ps(qq00,velec);
 462
 463             /* Analytical LJ-PME */
 464             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 465             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 466             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 467             exponent         = avx256_exp_f(ewcljrsq);
 468             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 469             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 470             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 471             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 472             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 473             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 474                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 475             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 476             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
 477
 478             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velec            = _mm256_and_ps(velec,cutoff_mask);
 482             velec            = _mm256_andnot_ps(dummy_mask,velec);
 483             velecsum         = _mm256_add_ps(velecsum,velec);
 484             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 485             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 486             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 487
 488             fscal            = _mm256_add_ps(felec,fvdw);
 489
 490             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 491
 492             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm256_mul_ps(fscal,dx00);
 496             ty               = _mm256_mul_ps(fscal,dy00);
 497             tz               = _mm256_mul_ps(fscal,dz00);
 498
 499             /* Update vectorial force */
 500             fix0             = _mm256_add_ps(fix0,tx);
 501             fiy0             = _mm256_add_ps(fiy0,ty);
 502             fiz0             = _mm256_add_ps(fiz0,tz);
 503
 504             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 505             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 506             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 507             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 508             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 509             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 510             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 511             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 512             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 513
 514             }
 515
 516             /* Inner loop uses 146 flops */
 517         }
 518
 519         /* End of innermost loop */
 520
 521         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 522                                                  f+i_coord_offset,fshift+i_shift_offset);
 523
 524         ggid                        = gid[iidx];
 525         /* Update potential energies */
 526         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 527         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 528
 529         /* Increment number of inner iterations */
 530         inneriter                  += j_index_end - j_index_start;
 531
 532         /* Outer loop uses 9 flops */
 533     }
 534
 535     /* Increment number of outer iterations */
 536     outeriter        += nri;
 537
 538     /* Update outer/inner flops */
 539
 540     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*146);
 541 }
 542 /*
 543  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_single
 544  * Electrostatics interaction: Ewald
 545  * VdW interaction:            LJEwald
 546  * Geometry:                   Particle-Particle
 547  * Calculate force/pot:        Force
 548  */
 549 void
 550 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_single
 551                     (t_nblist                    * gmx_restrict       nlist,
 552                      rvec                        * gmx_restrict          xx,
 553                      rvec                        * gmx_restrict          ff,
 554                      struct t_forcerec           * gmx_restrict          fr,
 555                      t_mdatoms                   * gmx_restrict     mdatoms,
 556                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 557                      t_nrnb                      * gmx_restrict        nrnb)
 558 {
 559     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 560      * just 0 for non-waters.
 561      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 562      * jnr indices corresponding to data put in the four positions in the SIMD register.
 563      */
 564     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 565     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 566     int              jnrA,jnrB,jnrC,jnrD;
 567     int              jnrE,jnrF,jnrG,jnrH;
 568     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 569     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 570     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 571     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 572     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 573     real             rcutoff_scalar;
 574     real             *shiftvec,*fshift,*x,*f;
 575     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 576     real             scratch[4*DIM];
 577     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 578     real *           vdwioffsetptr0;
 579     real *           vdwgridioffsetptr0;
 580     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 581     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 582     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 583     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 584     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 585     real             *charge;
 586     int              nvdwtype;
 587     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 588     int              *vdwtype;
 589     real             *vdwparam;
 590     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 591     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 592     __m256           c6grid_00;
 593     real             *vdwgridparam;
 594     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 595     __m256           one_half  = _mm256_set1_ps(0.5);
 596     __m256           minus_one = _mm256_set1_ps(-1.0);
 597     __m256i          ewitab;
 598     __m128i          ewitab_lo,ewitab_hi;
 599     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 600     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 601     real             *ewtab;
 602     __m256           dummy_mask,cutoff_mask;
 603     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 604     __m256           one     = _mm256_set1_ps(1.0);
 605     __m256           two     = _mm256_set1_ps(2.0);
 606     x                = xx[0];
 607     f                = ff[0];
 608
 609     nri              = nlist->nri;
 610     iinr             = nlist->iinr;
 611     jindex           = nlist->jindex;
 612     jjnr             = nlist->jjnr;
 613     shiftidx         = nlist->shift;
 614     gid              = nlist->gid;
 615     shiftvec         = fr->shift_vec[0];
 616     fshift           = fr->fshift[0];
 617     facel            = _mm256_set1_ps(fr->ic->epsfac);
 618     charge           = mdatoms->chargeA;
 619     nvdwtype         = fr->ntype;
 620     vdwparam         = fr->nbfp;
 621     vdwtype          = mdatoms->typeA;
 622     vdwgridparam     = fr->ljpme_c6grid;
 623     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 624     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
 625     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 626
 627     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 628     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 629     beta2            = _mm256_mul_ps(beta,beta);
 630     beta3            = _mm256_mul_ps(beta,beta2);
 631
 632     ewtab            = fr->ic->tabq_coul_F;
 633     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 634     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 635
 636     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 637     rcutoff_scalar   = fr->ic->rcoulomb;
 638     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 639     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 640
 641     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 642     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 643
 644     /* Avoid stupid compiler warnings */
 645     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 646     j_coord_offsetA = 0;
 647     j_coord_offsetB = 0;
 648     j_coord_offsetC = 0;
 649     j_coord_offsetD = 0;
 650     j_coord_offsetE = 0;
 651     j_coord_offsetF = 0;
 652     j_coord_offsetG = 0;
 653     j_coord_offsetH = 0;
 654
 655     outeriter        = 0;
 656     inneriter        = 0;
 657
 658     for(iidx=0;iidx<4*DIM;iidx++)
 659     {
 660         scratch[iidx] = 0.0;
 661     }
 662
 663     /* Start outer loop over neighborlists */
 664     for(iidx=0; iidx<nri; iidx++)
 665     {
 666         /* Load shift vector for this list */
 667         i_shift_offset   = DIM*shiftidx[iidx];
 668
 669         /* Load limits for loop over neighbors */
 670         j_index_start    = jindex[iidx];
 671         j_index_end      = jindex[iidx+1];
 672
 673         /* Get outer coordinate index */
 674         inr              = iinr[iidx];
 675         i_coord_offset   = DIM*inr;
 676
 677         /* Load i particle coords and add shift vector */
 678         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 679
 680         fix0             = _mm256_setzero_ps();
 681         fiy0             = _mm256_setzero_ps();
 682         fiz0             = _mm256_setzero_ps();
 683
 684         /* Load parameters for i particles */
 685         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 686         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 687         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 688
 689         /* Start inner kernel loop */
 690         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 691         {
 692
 693             /* Get j neighbor index, and coordinate index */
 694             jnrA             = jjnr[jidx];
 695             jnrB             = jjnr[jidx+1];
 696             jnrC             = jjnr[jidx+2];
 697             jnrD             = jjnr[jidx+3];
 698             jnrE             = jjnr[jidx+4];
 699             jnrF             = jjnr[jidx+5];
 700             jnrG             = jjnr[jidx+6];
 701             jnrH             = jjnr[jidx+7];
 702             j_coord_offsetA  = DIM*jnrA;
 703             j_coord_offsetB  = DIM*jnrB;
 704             j_coord_offsetC  = DIM*jnrC;
 705             j_coord_offsetD  = DIM*jnrD;
 706             j_coord_offsetE  = DIM*jnrE;
 707             j_coord_offsetF  = DIM*jnrF;
 708             j_coord_offsetG  = DIM*jnrG;
 709             j_coord_offsetH  = DIM*jnrH;
 710
 711             /* load j atom coordinates */
 712             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 713                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 714                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 715                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 716                                                  &jx0,&jy0,&jz0);
 717
 718             /* Calculate displacement vector */
 719             dx00             = _mm256_sub_ps(ix0,jx0);
 720             dy00             = _mm256_sub_ps(iy0,jy0);
 721             dz00             = _mm256_sub_ps(iz0,jz0);
 722
 723             /* Calculate squared distance and things based on it */
 724             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 725
 726             rinv00           = avx256_invsqrt_f(rsq00);
 727
 728             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 729
 730             /* Load parameters for j particles */
 731             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 732                                                                  charge+jnrC+0,charge+jnrD+0,
 733                                                                  charge+jnrE+0,charge+jnrF+0,
 734                                                                  charge+jnrG+0,charge+jnrH+0);
 735             vdwjidx0A        = 2*vdwtype[jnrA+0];
 736             vdwjidx0B        = 2*vdwtype[jnrB+0];
 737             vdwjidx0C        = 2*vdwtype[jnrC+0];
 738             vdwjidx0D        = 2*vdwtype[jnrD+0];
 739             vdwjidx0E        = 2*vdwtype[jnrE+0];
 740             vdwjidx0F        = 2*vdwtype[jnrF+0];
 741             vdwjidx0G        = 2*vdwtype[jnrG+0];
 742             vdwjidx0H        = 2*vdwtype[jnrH+0];
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 749             {
 750
 751             r00              = _mm256_mul_ps(rsq00,rinv00);
 752
 753             /* Compute parameters for interactions between i and j atoms */
 754             qq00             = _mm256_mul_ps(iq0,jq0);
 755             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 756                                             vdwioffsetptr0+vdwjidx0B,
 757                                             vdwioffsetptr0+vdwjidx0C,
 758                                             vdwioffsetptr0+vdwjidx0D,
 759                                             vdwioffsetptr0+vdwjidx0E,
 760                                             vdwioffsetptr0+vdwjidx0F,
 761                                             vdwioffsetptr0+vdwjidx0G,
 762                                             vdwioffsetptr0+vdwjidx0H,
 763                                             &c6_00,&c12_00);
 764
 765             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 766                                                                   vdwgridioffsetptr0+vdwjidx0B,
 767                                                                   vdwgridioffsetptr0+vdwjidx0C,
 768                                                                   vdwgridioffsetptr0+vdwjidx0D,
 769                                                                   vdwgridioffsetptr0+vdwjidx0E,
 770                                                                   vdwgridioffsetptr0+vdwjidx0F,
 771                                                                   vdwgridioffsetptr0+vdwjidx0G,
 772                                                                   vdwgridioffsetptr0+vdwjidx0H);
 773
 774             /* EWALD ELECTROSTATICS */
 775
 776             /* Analytical PME correction */
 777             zeta2            = _mm256_mul_ps(beta2,rsq00);
 778             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 779             pmecorrF         = avx256_pmecorrF_f(zeta2);
 780             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 781             felec            = _mm256_mul_ps(qq00,felec);
 782
 783             /* Analytical LJ-PME */
 784             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 785             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 786             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 787             exponent         = avx256_exp_f(ewcljrsq);
 788             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 789             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 790             /* f6A = 6 * C6grid * (1 - poly) */
 791             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
 792             /* f6B = C6grid * exponent * beta^6 */
 793             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
 794             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 795             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 796
 797             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 798
 799             fscal            = _mm256_add_ps(felec,fvdw);
 800
 801             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 802
 803             /* Calculate temporary vectorial force */
 804             tx               = _mm256_mul_ps(fscal,dx00);
 805             ty               = _mm256_mul_ps(fscal,dy00);
 806             tz               = _mm256_mul_ps(fscal,dz00);
 807
 808             /* Update vectorial force */
 809             fix0             = _mm256_add_ps(fix0,tx);
 810             fiy0             = _mm256_add_ps(fiy0,ty);
 811             fiz0             = _mm256_add_ps(fiz0,tz);
 812
 813             fjptrA             = f+j_coord_offsetA;
 814             fjptrB             = f+j_coord_offsetB;
 815             fjptrC             = f+j_coord_offsetC;
 816             fjptrD             = f+j_coord_offsetD;
 817             fjptrE             = f+j_coord_offsetE;
 818             fjptrF             = f+j_coord_offsetF;
 819             fjptrG             = f+j_coord_offsetG;
 820             fjptrH             = f+j_coord_offsetH;
 821             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 822
 823             }
 824
 825             /* Inner loop uses 82 flops */
 826         }
 827
 828         if(jidx<j_index_end)
 829         {
 830
 831             /* Get j neighbor index, and coordinate index */
 832             jnrlistA         = jjnr[jidx];
 833             jnrlistB         = jjnr[jidx+1];
 834             jnrlistC         = jjnr[jidx+2];
 835             jnrlistD         = jjnr[jidx+3];
 836             jnrlistE         = jjnr[jidx+4];
 837             jnrlistF         = jjnr[jidx+5];
 838             jnrlistG         = jjnr[jidx+6];
 839             jnrlistH         = jjnr[jidx+7];
 840             /* Sign of each element will be negative for non-real atoms.
 841              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 842              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 843              */
 844             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 845                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 846
 847             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 848             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 849             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 850             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 851             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 852             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 853             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 854             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 855             j_coord_offsetA  = DIM*jnrA;
 856             j_coord_offsetB  = DIM*jnrB;
 857             j_coord_offsetC  = DIM*jnrC;
 858             j_coord_offsetD  = DIM*jnrD;
 859             j_coord_offsetE  = DIM*jnrE;
 860             j_coord_offsetF  = DIM*jnrF;
 861             j_coord_offsetG  = DIM*jnrG;
 862             j_coord_offsetH  = DIM*jnrH;
 863
 864             /* load j atom coordinates */
 865             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 866                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 867                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 868                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 869                                                  &jx0,&jy0,&jz0);
 870
 871             /* Calculate displacement vector */
 872             dx00             = _mm256_sub_ps(ix0,jx0);
 873             dy00             = _mm256_sub_ps(iy0,jy0);
 874             dz00             = _mm256_sub_ps(iz0,jz0);
 875
 876             /* Calculate squared distance and things based on it */
 877             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 878
 879             rinv00           = avx256_invsqrt_f(rsq00);
 880
 881             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 882
 883             /* Load parameters for j particles */
 884             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 885                                                                  charge+jnrC+0,charge+jnrD+0,
 886                                                                  charge+jnrE+0,charge+jnrF+0,
 887                                                                  charge+jnrG+0,charge+jnrH+0);
 888             vdwjidx0A        = 2*vdwtype[jnrA+0];
 889             vdwjidx0B        = 2*vdwtype[jnrB+0];
 890             vdwjidx0C        = 2*vdwtype[jnrC+0];
 891             vdwjidx0D        = 2*vdwtype[jnrD+0];
 892             vdwjidx0E        = 2*vdwtype[jnrE+0];
 893             vdwjidx0F        = 2*vdwtype[jnrF+0];
 894             vdwjidx0G        = 2*vdwtype[jnrG+0];
 895             vdwjidx0H        = 2*vdwtype[jnrH+0];
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 902             {
 903
 904             r00              = _mm256_mul_ps(rsq00,rinv00);
 905             r00              = _mm256_andnot_ps(dummy_mask,r00);
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq00             = _mm256_mul_ps(iq0,jq0);
 909             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 910                                             vdwioffsetptr0+vdwjidx0B,
 911                                             vdwioffsetptr0+vdwjidx0C,
 912                                             vdwioffsetptr0+vdwjidx0D,
 913                                             vdwioffsetptr0+vdwjidx0E,
 914                                             vdwioffsetptr0+vdwjidx0F,
 915                                             vdwioffsetptr0+vdwjidx0G,
 916                                             vdwioffsetptr0+vdwjidx0H,
 917                                             &c6_00,&c12_00);
 918
 919             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 920                                                                   vdwgridioffsetptr0+vdwjidx0B,
 921                                                                   vdwgridioffsetptr0+vdwjidx0C,
 922                                                                   vdwgridioffsetptr0+vdwjidx0D,
 923                                                                   vdwgridioffsetptr0+vdwjidx0E,
 924                                                                   vdwgridioffsetptr0+vdwjidx0F,
 925                                                                   vdwgridioffsetptr0+vdwjidx0G,
 926                                                                   vdwgridioffsetptr0+vdwjidx0H);
 927
 928             /* EWALD ELECTROSTATICS */
 929
 930             /* Analytical PME correction */
 931             zeta2            = _mm256_mul_ps(beta2,rsq00);
 932             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 933             pmecorrF         = avx256_pmecorrF_f(zeta2);
 934             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 935             felec            = _mm256_mul_ps(qq00,felec);
 936
 937             /* Analytical LJ-PME */
 938             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 939             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 940             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 941             exponent         = avx256_exp_f(ewcljrsq);
 942             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 943             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 944             /* f6A = 6 * C6grid * (1 - poly) */
 945             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
 946             /* f6B = C6grid * exponent * beta^6 */
 947             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
 948             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 949             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 950
 951             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 952
 953             fscal            = _mm256_add_ps(felec,fvdw);
 954
 955             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 956
 957             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 958
 959             /* Calculate temporary vectorial force */
 960             tx               = _mm256_mul_ps(fscal,dx00);
 961             ty               = _mm256_mul_ps(fscal,dy00);
 962             tz               = _mm256_mul_ps(fscal,dz00);
 963
 964             /* Update vectorial force */
 965             fix0             = _mm256_add_ps(fix0,tx);
 966             fiy0             = _mm256_add_ps(fiy0,ty);
 967             fiz0             = _mm256_add_ps(fiz0,tz);
 968
 969             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 970             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 971             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 972             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 973             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 974             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 975             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 976             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 977             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 978
 979             }
 980
 981             /* Inner loop uses 83 flops */
 982         }
 983
 984         /* End of innermost loop */
 985
 986         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 987                                                  f+i_coord_offset,fshift+i_shift_offset);
 988
 989         /* Increment number of inner iterations */
 990         inneriter                  += j_index_end - j_index_start;
 991
 992         /* Outer loop uses 7 flops */
 993     }
 994
 995     /* Increment number of outer iterations */
 996     outeriter        += nri;
 997
 998     /* Update outer/inner flops */
 999
1000     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
1001 }