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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     real *           vdwgridioffsetptr0;
  88     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  90     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  93     real             *charge;
  94     int              nvdwtype;
  95     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  96     int              *vdwtype;
  97     real             *vdwparam;
  98     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  99     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 100     __m256           c6grid_00;
 101     real             *vdwgridparam;
 102     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 103     __m256           one_half  = _mm256_set1_ps(0.5);
 104     __m256           minus_one = _mm256_set1_ps(-1.0);
 105     __m256i          ewitab;
 106     __m128i          ewitab_lo,ewitab_hi;
 107     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 109     real             *ewtab;
 110     __m256           dummy_mask,cutoff_mask;
 111     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 112     __m256           one     = _mm256_set1_ps(1.0);
 113     __m256           two     = _mm256_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130     vdwgridparam     = fr->ljpme_c6grid;
 131     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 132     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 133     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 134
 135     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 136     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 137     beta2            = _mm256_mul_ps(beta,beta);
 138     beta3            = _mm256_mul_ps(beta,beta2);
 139
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 143
 144     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 145     rcutoff_scalar   = fr->rcoulomb;
 146     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 147     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 148
 149     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 150     rvdw             = _mm256_set1_ps(fr->rvdw);
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158     j_coord_offsetE = 0;
 159     j_coord_offsetF = 0;
 160     j_coord_offsetG = 0;
 161     j_coord_offsetH = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     for(iidx=0;iidx<4*DIM;iidx++)
 167     {
 168         scratch[iidx] = 0.0;
 169     }
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 187
 188         fix0             = _mm256_setzero_ps();
 189         fiy0             = _mm256_setzero_ps();
 190         fiz0             = _mm256_setzero_ps();
 191
 192         /* Load parameters for i particles */
 193         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 194         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 195         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm256_setzero_ps();
 199         vvdwsum          = _mm256_setzero_ps();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             jnrE             = jjnr[jidx+4];
 211             jnrF             = jjnr[jidx+5];
 212             jnrG             = jjnr[jidx+6];
 213             jnrH             = jjnr[jidx+7];
 214             j_coord_offsetA  = DIM*jnrA;
 215             j_coord_offsetB  = DIM*jnrB;
 216             j_coord_offsetC  = DIM*jnrC;
 217             j_coord_offsetD  = DIM*jnrD;
 218             j_coord_offsetE  = DIM*jnrE;
 219             j_coord_offsetF  = DIM*jnrF;
 220             j_coord_offsetG  = DIM*jnrG;
 221             j_coord_offsetH  = DIM*jnrH;
 222
 223             /* load j atom coordinates */
 224             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 225                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 226                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 227                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 228                                                  &jx0,&jy0,&jz0);
 229
 230             /* Calculate displacement vector */
 231             dx00             = _mm256_sub_ps(ix0,jx0);
 232             dy00             = _mm256_sub_ps(iy0,jy0);
 233             dz00             = _mm256_sub_ps(iz0,jz0);
 234
 235             /* Calculate squared distance and things based on it */
 236             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 237
 238             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 239
 240             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                                  charge+jnrC+0,charge+jnrD+0,
 245                                                                  charge+jnrE+0,charge+jnrF+0,
 246                                                                  charge+jnrG+0,charge+jnrH+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251             vdwjidx0E        = 2*vdwtype[jnrE+0];
 252             vdwjidx0F        = 2*vdwtype[jnrF+0];
 253             vdwjidx0G        = 2*vdwtype[jnrG+0];
 254             vdwjidx0H        = 2*vdwtype[jnrH+0];
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 261             {
 262
 263             r00              = _mm256_mul_ps(rsq00,rinv00);
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq00             = _mm256_mul_ps(iq0,jq0);
 267             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 268                                             vdwioffsetptr0+vdwjidx0B,
 269                                             vdwioffsetptr0+vdwjidx0C,
 270                                             vdwioffsetptr0+vdwjidx0D,
 271                                             vdwioffsetptr0+vdwjidx0E,
 272                                             vdwioffsetptr0+vdwjidx0F,
 273                                             vdwioffsetptr0+vdwjidx0G,
 274                                             vdwioffsetptr0+vdwjidx0H,
 275                                             &c6_00,&c12_00);
 276
 277             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 278                                                                   vdwgridioffsetptr0+vdwjidx0B,
 279                                                                   vdwgridioffsetptr0+vdwjidx0C,
 280                                                                   vdwgridioffsetptr0+vdwjidx0D,
 281                                                                   vdwgridioffsetptr0+vdwjidx0E,
 282                                                                   vdwgridioffsetptr0+vdwjidx0F,
 283                                                                   vdwgridioffsetptr0+vdwjidx0G,
 284                                                                   vdwgridioffsetptr0+vdwjidx0H);
 285
 286             /* EWALD ELECTROSTATICS */
 287
 288             /* Analytical PME correction */
 289             zeta2            = _mm256_mul_ps(beta2,rsq00);
 290             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 291             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 292             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 293             felec            = _mm256_mul_ps(qq00,felec);
 294             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 295             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 296             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 297             velec            = _mm256_mul_ps(qq00,velec);
 298
 299             /* Analytical LJ-PME */
 300             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 301             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 302             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 303             exponent         = gmx_simd_exp_r(ewcljrsq);
 304             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 305             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 306             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 307             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 308             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 309             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) ,
 310                                           _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));
 311             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 312             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);
 313
 314             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velec            = _mm256_and_ps(velec,cutoff_mask);
 318             velecsum         = _mm256_add_ps(velecsum,velec);
 319             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 320             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 321
 322             fscal            = _mm256_add_ps(felec,fvdw);
 323
 324             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm256_mul_ps(fscal,dx00);
 328             ty               = _mm256_mul_ps(fscal,dy00);
 329             tz               = _mm256_mul_ps(fscal,dz00);
 330
 331             /* Update vectorial force */
 332             fix0             = _mm256_add_ps(fix0,tx);
 333             fiy0             = _mm256_add_ps(fiy0,ty);
 334             fiz0             = _mm256_add_ps(fiz0,tz);
 335
 336             fjptrA             = f+j_coord_offsetA;
 337             fjptrB             = f+j_coord_offsetB;
 338             fjptrC             = f+j_coord_offsetC;
 339             fjptrD             = f+j_coord_offsetD;
 340             fjptrE             = f+j_coord_offsetE;
 341             fjptrF             = f+j_coord_offsetF;
 342             fjptrG             = f+j_coord_offsetG;
 343             fjptrH             = f+j_coord_offsetH;
 344             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 345
 346             }
 347
 348             /* Inner loop uses 145 flops */
 349         }
 350
 351         if(jidx<j_index_end)
 352         {
 353
 354             /* Get j neighbor index, and coordinate index */
 355             jnrlistA         = jjnr[jidx];
 356             jnrlistB         = jjnr[jidx+1];
 357             jnrlistC         = jjnr[jidx+2];
 358             jnrlistD         = jjnr[jidx+3];
 359             jnrlistE         = jjnr[jidx+4];
 360             jnrlistF         = jjnr[jidx+5];
 361             jnrlistG         = jjnr[jidx+6];
 362             jnrlistH         = jjnr[jidx+7];
 363             /* Sign of each element will be negative for non-real atoms.
 364              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 365              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 366              */
 367             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 368                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 369
 370             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 371             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 372             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 373             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 374             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 375             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 376             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 377             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 378             j_coord_offsetA  = DIM*jnrA;
 379             j_coord_offsetB  = DIM*jnrB;
 380             j_coord_offsetC  = DIM*jnrC;
 381             j_coord_offsetD  = DIM*jnrD;
 382             j_coord_offsetE  = DIM*jnrE;
 383             j_coord_offsetF  = DIM*jnrF;
 384             j_coord_offsetG  = DIM*jnrG;
 385             j_coord_offsetH  = DIM*jnrH;
 386
 387             /* load j atom coordinates */
 388             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 389                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 390                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 391                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 392                                                  &jx0,&jy0,&jz0);
 393
 394             /* Calculate displacement vector */
 395             dx00             = _mm256_sub_ps(ix0,jx0);
 396             dy00             = _mm256_sub_ps(iy0,jy0);
 397             dz00             = _mm256_sub_ps(iz0,jz0);
 398
 399             /* Calculate squared distance and things based on it */
 400             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 401
 402             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 403
 404             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 405
 406             /* Load parameters for j particles */
 407             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 408                                                                  charge+jnrC+0,charge+jnrD+0,
 409                                                                  charge+jnrE+0,charge+jnrF+0,
 410                                                                  charge+jnrG+0,charge+jnrH+0);
 411             vdwjidx0A        = 2*vdwtype[jnrA+0];
 412             vdwjidx0B        = 2*vdwtype[jnrB+0];
 413             vdwjidx0C        = 2*vdwtype[jnrC+0];
 414             vdwjidx0D        = 2*vdwtype[jnrD+0];
 415             vdwjidx0E        = 2*vdwtype[jnrE+0];
 416             vdwjidx0F        = 2*vdwtype[jnrF+0];
 417             vdwjidx0G        = 2*vdwtype[jnrG+0];
 418             vdwjidx0H        = 2*vdwtype[jnrH+0];
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 425             {
 426
 427             r00              = _mm256_mul_ps(rsq00,rinv00);
 428             r00              = _mm256_andnot_ps(dummy_mask,r00);
 429
 430             /* Compute parameters for interactions between i and j atoms */
 431             qq00             = _mm256_mul_ps(iq0,jq0);
 432             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 433                                             vdwioffsetptr0+vdwjidx0B,
 434                                             vdwioffsetptr0+vdwjidx0C,
 435                                             vdwioffsetptr0+vdwjidx0D,
 436                                             vdwioffsetptr0+vdwjidx0E,
 437                                             vdwioffsetptr0+vdwjidx0F,
 438                                             vdwioffsetptr0+vdwjidx0G,
 439                                             vdwioffsetptr0+vdwjidx0H,
 440                                             &c6_00,&c12_00);
 441
 442             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 443                                                                   vdwgridioffsetptr0+vdwjidx0B,
 444                                                                   vdwgridioffsetptr0+vdwjidx0C,
 445                                                                   vdwgridioffsetptr0+vdwjidx0D,
 446                                                                   vdwgridioffsetptr0+vdwjidx0E,
 447                                                                   vdwgridioffsetptr0+vdwjidx0F,
 448                                                                   vdwgridioffsetptr0+vdwjidx0G,
 449                                                                   vdwgridioffsetptr0+vdwjidx0H);
 450
 451             /* EWALD ELECTROSTATICS */
 452
 453             /* Analytical PME correction */
 454             zeta2            = _mm256_mul_ps(beta2,rsq00);
 455             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 456             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 457             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 458             felec            = _mm256_mul_ps(qq00,felec);
 459             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 460             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 461             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 462             velec            = _mm256_mul_ps(qq00,velec);
 463
 464             /* Analytical LJ-PME */
 465             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 466             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 467             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 468             exponent         = gmx_simd_exp_r(ewcljrsq);
 469             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 470             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 471             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 472             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 473             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 474             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) ,
 475                                           _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));
 476             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 477             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);
 478
 479             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velec            = _mm256_and_ps(velec,cutoff_mask);
 483             velec            = _mm256_andnot_ps(dummy_mask,velec);
 484             velecsum         = _mm256_add_ps(velecsum,velec);
 485             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 486             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 487             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 488
 489             fscal            = _mm256_add_ps(felec,fvdw);
 490
 491             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 492
 493             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 494
 495             /* Calculate temporary vectorial force */
 496             tx               = _mm256_mul_ps(fscal,dx00);
 497             ty               = _mm256_mul_ps(fscal,dy00);
 498             tz               = _mm256_mul_ps(fscal,dz00);
 499
 500             /* Update vectorial force */
 501             fix0             = _mm256_add_ps(fix0,tx);
 502             fiy0             = _mm256_add_ps(fiy0,ty);
 503             fiz0             = _mm256_add_ps(fiz0,tz);
 504
 505             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 506             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 507             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 508             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 509             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 510             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 511             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 512             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 513             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 514
 515             }
 516
 517             /* Inner loop uses 146 flops */
 518         }
 519
 520         /* End of innermost loop */
 521
 522         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 523                                                  f+i_coord_offset,fshift+i_shift_offset);
 524
 525         ggid                        = gid[iidx];
 526         /* Update potential energies */
 527         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 528         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 529
 530         /* Increment number of inner iterations */
 531         inneriter                  += j_index_end - j_index_start;
 532
 533         /* Outer loop uses 9 flops */
 534     }
 535
 536     /* Increment number of outer iterations */
 537     outeriter        += nri;
 538
 539     /* Update outer/inner flops */
 540
 541     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*146);
 542 }
 543 /*
 544  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_single
 545  * Electrostatics interaction: Ewald
 546  * VdW interaction:            LJEwald
 547  * Geometry:                   Particle-Particle
 548  * Calculate force/pot:        Force
 549  */
 550 void
 551 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_single
 552                     (t_nblist                    * gmx_restrict       nlist,
 553                      rvec                        * gmx_restrict          xx,
 554                      rvec                        * gmx_restrict          ff,
 555                      t_forcerec                  * gmx_restrict          fr,
 556                      t_mdatoms                   * gmx_restrict     mdatoms,
 557                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 558                      t_nrnb                      * gmx_restrict        nrnb)
 559 {
 560     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 561      * just 0 for non-waters.
 562      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 563      * jnr indices corresponding to data put in the four positions in the SIMD register.
 564      */
 565     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 566     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 567     int              jnrA,jnrB,jnrC,jnrD;
 568     int              jnrE,jnrF,jnrG,jnrH;
 569     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 570     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 571     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 572     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 573     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 574     real             rcutoff_scalar;
 575     real             *shiftvec,*fshift,*x,*f;
 576     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 577     real             scratch[4*DIM];
 578     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 579     real *           vdwioffsetptr0;
 580     real *           vdwgridioffsetptr0;
 581     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 582     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 583     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 584     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 585     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 586     real             *charge;
 587     int              nvdwtype;
 588     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 589     int              *vdwtype;
 590     real             *vdwparam;
 591     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 592     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 593     __m256           c6grid_00;
 594     real             *vdwgridparam;
 595     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 596     __m256           one_half  = _mm256_set1_ps(0.5);
 597     __m256           minus_one = _mm256_set1_ps(-1.0);
 598     __m256i          ewitab;
 599     __m128i          ewitab_lo,ewitab_hi;
 600     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 601     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 602     real             *ewtab;
 603     __m256           dummy_mask,cutoff_mask;
 604     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 605     __m256           one     = _mm256_set1_ps(1.0);
 606     __m256           two     = _mm256_set1_ps(2.0);
 607     x                = xx[0];
 608     f                = ff[0];
 609
 610     nri              = nlist->nri;
 611     iinr             = nlist->iinr;
 612     jindex           = nlist->jindex;
 613     jjnr             = nlist->jjnr;
 614     shiftidx         = nlist->shift;
 615     gid              = nlist->gid;
 616     shiftvec         = fr->shift_vec[0];
 617     fshift           = fr->fshift[0];
 618     facel            = _mm256_set1_ps(fr->epsfac);
 619     charge           = mdatoms->chargeA;
 620     nvdwtype         = fr->ntype;
 621     vdwparam         = fr->nbfp;
 622     vdwtype          = mdatoms->typeA;
 623     vdwgridparam     = fr->ljpme_c6grid;
 624     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 625     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 626     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 627
 628     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 629     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 630     beta2            = _mm256_mul_ps(beta,beta);
 631     beta3            = _mm256_mul_ps(beta,beta2);
 632
 633     ewtab            = fr->ic->tabq_coul_F;
 634     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 635     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 636
 637     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 638     rcutoff_scalar   = fr->rcoulomb;
 639     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 640     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 641
 642     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 643     rvdw             = _mm256_set1_ps(fr->rvdw);
 644
 645     /* Avoid stupid compiler warnings */
 646     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 647     j_coord_offsetA = 0;
 648     j_coord_offsetB = 0;
 649     j_coord_offsetC = 0;
 650     j_coord_offsetD = 0;
 651     j_coord_offsetE = 0;
 652     j_coord_offsetF = 0;
 653     j_coord_offsetG = 0;
 654     j_coord_offsetH = 0;
 655
 656     outeriter        = 0;
 657     inneriter        = 0;
 658
 659     for(iidx=0;iidx<4*DIM;iidx++)
 660     {
 661         scratch[iidx] = 0.0;
 662     }
 663
 664     /* Start outer loop over neighborlists */
 665     for(iidx=0; iidx<nri; iidx++)
 666     {
 667         /* Load shift vector for this list */
 668         i_shift_offset   = DIM*shiftidx[iidx];
 669
 670         /* Load limits for loop over neighbors */
 671         j_index_start    = jindex[iidx];
 672         j_index_end      = jindex[iidx+1];
 673
 674         /* Get outer coordinate index */
 675         inr              = iinr[iidx];
 676         i_coord_offset   = DIM*inr;
 677
 678         /* Load i particle coords and add shift vector */
 679         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 680
 681         fix0             = _mm256_setzero_ps();
 682         fiy0             = _mm256_setzero_ps();
 683         fiz0             = _mm256_setzero_ps();
 684
 685         /* Load parameters for i particles */
 686         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 687         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 688         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 689
 690         /* Start inner kernel loop */
 691         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 692         {
 693
 694             /* Get j neighbor index, and coordinate index */
 695             jnrA             = jjnr[jidx];
 696             jnrB             = jjnr[jidx+1];
 697             jnrC             = jjnr[jidx+2];
 698             jnrD             = jjnr[jidx+3];
 699             jnrE             = jjnr[jidx+4];
 700             jnrF             = jjnr[jidx+5];
 701             jnrG             = jjnr[jidx+6];
 702             jnrH             = jjnr[jidx+7];
 703             j_coord_offsetA  = DIM*jnrA;
 704             j_coord_offsetB  = DIM*jnrB;
 705             j_coord_offsetC  = DIM*jnrC;
 706             j_coord_offsetD  = DIM*jnrD;
 707             j_coord_offsetE  = DIM*jnrE;
 708             j_coord_offsetF  = DIM*jnrF;
 709             j_coord_offsetG  = DIM*jnrG;
 710             j_coord_offsetH  = DIM*jnrH;
 711
 712             /* load j atom coordinates */
 713             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 714                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 715                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 716                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 717                                                  &jx0,&jy0,&jz0);
 718
 719             /* Calculate displacement vector */
 720             dx00             = _mm256_sub_ps(ix0,jx0);
 721             dy00             = _mm256_sub_ps(iy0,jy0);
 722             dz00             = _mm256_sub_ps(iz0,jz0);
 723
 724             /* Calculate squared distance and things based on it */
 725             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 726
 727             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 728
 729             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 730
 731             /* Load parameters for j particles */
 732             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 733                                                                  charge+jnrC+0,charge+jnrD+0,
 734                                                                  charge+jnrE+0,charge+jnrF+0,
 735                                                                  charge+jnrG+0,charge+jnrH+0);
 736             vdwjidx0A        = 2*vdwtype[jnrA+0];
 737             vdwjidx0B        = 2*vdwtype[jnrB+0];
 738             vdwjidx0C        = 2*vdwtype[jnrC+0];
 739             vdwjidx0D        = 2*vdwtype[jnrD+0];
 740             vdwjidx0E        = 2*vdwtype[jnrE+0];
 741             vdwjidx0F        = 2*vdwtype[jnrF+0];
 742             vdwjidx0G        = 2*vdwtype[jnrG+0];
 743             vdwjidx0H        = 2*vdwtype[jnrH+0];
 744
 745             /**************************
 746              * CALCULATE INTERACTIONS *
 747              **************************/
 748
 749             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 750             {
 751
 752             r00              = _mm256_mul_ps(rsq00,rinv00);
 753
 754             /* Compute parameters for interactions between i and j atoms */
 755             qq00             = _mm256_mul_ps(iq0,jq0);
 756             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 757                                             vdwioffsetptr0+vdwjidx0B,
 758                                             vdwioffsetptr0+vdwjidx0C,
 759                                             vdwioffsetptr0+vdwjidx0D,
 760                                             vdwioffsetptr0+vdwjidx0E,
 761                                             vdwioffsetptr0+vdwjidx0F,
 762                                             vdwioffsetptr0+vdwjidx0G,
 763                                             vdwioffsetptr0+vdwjidx0H,
 764                                             &c6_00,&c12_00);
 765
 766             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 767                                                                   vdwgridioffsetptr0+vdwjidx0B,
 768                                                                   vdwgridioffsetptr0+vdwjidx0C,
 769                                                                   vdwgridioffsetptr0+vdwjidx0D,
 770                                                                   vdwgridioffsetptr0+vdwjidx0E,
 771                                                                   vdwgridioffsetptr0+vdwjidx0F,
 772                                                                   vdwgridioffsetptr0+vdwjidx0G,
 773                                                                   vdwgridioffsetptr0+vdwjidx0H);
 774
 775             /* EWALD ELECTROSTATICS */
 776
 777             /* Analytical PME correction */
 778             zeta2            = _mm256_mul_ps(beta2,rsq00);
 779             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 780             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 781             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 782             felec            = _mm256_mul_ps(qq00,felec);
 783
 784             /* Analytical LJ-PME */
 785             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 786             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 787             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 788             exponent         = gmx_simd_exp_r(ewcljrsq);
 789             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 790             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 791             /* f6A = 6 * C6grid * (1 - poly) */
 792             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
 793             /* f6B = C6grid * exponent * beta^6 */
 794             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
 795             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 796             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);
 797
 798             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 799
 800             fscal            = _mm256_add_ps(felec,fvdw);
 801
 802             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 803
 804             /* Calculate temporary vectorial force */
 805             tx               = _mm256_mul_ps(fscal,dx00);
 806             ty               = _mm256_mul_ps(fscal,dy00);
 807             tz               = _mm256_mul_ps(fscal,dz00);
 808
 809             /* Update vectorial force */
 810             fix0             = _mm256_add_ps(fix0,tx);
 811             fiy0             = _mm256_add_ps(fiy0,ty);
 812             fiz0             = _mm256_add_ps(fiz0,tz);
 813
 814             fjptrA             = f+j_coord_offsetA;
 815             fjptrB             = f+j_coord_offsetB;
 816             fjptrC             = f+j_coord_offsetC;
 817             fjptrD             = f+j_coord_offsetD;
 818             fjptrE             = f+j_coord_offsetE;
 819             fjptrF             = f+j_coord_offsetF;
 820             fjptrG             = f+j_coord_offsetG;
 821             fjptrH             = f+j_coord_offsetH;
 822             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 823
 824             }
 825
 826             /* Inner loop uses 82 flops */
 827         }
 828
 829         if(jidx<j_index_end)
 830         {
 831
 832             /* Get j neighbor index, and coordinate index */
 833             jnrlistA         = jjnr[jidx];
 834             jnrlistB         = jjnr[jidx+1];
 835             jnrlistC         = jjnr[jidx+2];
 836             jnrlistD         = jjnr[jidx+3];
 837             jnrlistE         = jjnr[jidx+4];
 838             jnrlistF         = jjnr[jidx+5];
 839             jnrlistG         = jjnr[jidx+6];
 840             jnrlistH         = jjnr[jidx+7];
 841             /* Sign of each element will be negative for non-real atoms.
 842              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 843              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 844              */
 845             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 846                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 847
 848             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 849             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 850             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 851             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 852             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 853             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 854             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 855             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 856             j_coord_offsetA  = DIM*jnrA;
 857             j_coord_offsetB  = DIM*jnrB;
 858             j_coord_offsetC  = DIM*jnrC;
 859             j_coord_offsetD  = DIM*jnrD;
 860             j_coord_offsetE  = DIM*jnrE;
 861             j_coord_offsetF  = DIM*jnrF;
 862             j_coord_offsetG  = DIM*jnrG;
 863             j_coord_offsetH  = DIM*jnrH;
 864
 865             /* load j atom coordinates */
 866             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 867                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 868                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 869                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 870                                                  &jx0,&jy0,&jz0);
 871
 872             /* Calculate displacement vector */
 873             dx00             = _mm256_sub_ps(ix0,jx0);
 874             dy00             = _mm256_sub_ps(iy0,jy0);
 875             dz00             = _mm256_sub_ps(iz0,jz0);
 876
 877             /* Calculate squared distance and things based on it */
 878             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 879
 880             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 881
 882             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 883
 884             /* Load parameters for j particles */
 885             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 886                                                                  charge+jnrC+0,charge+jnrD+0,
 887                                                                  charge+jnrE+0,charge+jnrF+0,
 888                                                                  charge+jnrG+0,charge+jnrH+0);
 889             vdwjidx0A        = 2*vdwtype[jnrA+0];
 890             vdwjidx0B        = 2*vdwtype[jnrB+0];
 891             vdwjidx0C        = 2*vdwtype[jnrC+0];
 892             vdwjidx0D        = 2*vdwtype[jnrD+0];
 893             vdwjidx0E        = 2*vdwtype[jnrE+0];
 894             vdwjidx0F        = 2*vdwtype[jnrF+0];
 895             vdwjidx0G        = 2*vdwtype[jnrG+0];
 896             vdwjidx0H        = 2*vdwtype[jnrH+0];
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 903             {
 904
 905             r00              = _mm256_mul_ps(rsq00,rinv00);
 906             r00              = _mm256_andnot_ps(dummy_mask,r00);
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             qq00             = _mm256_mul_ps(iq0,jq0);
 910             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 911                                             vdwioffsetptr0+vdwjidx0B,
 912                                             vdwioffsetptr0+vdwjidx0C,
 913                                             vdwioffsetptr0+vdwjidx0D,
 914                                             vdwioffsetptr0+vdwjidx0E,
 915                                             vdwioffsetptr0+vdwjidx0F,
 916                                             vdwioffsetptr0+vdwjidx0G,
 917                                             vdwioffsetptr0+vdwjidx0H,
 918                                             &c6_00,&c12_00);
 919
 920             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 921                                                                   vdwgridioffsetptr0+vdwjidx0B,
 922                                                                   vdwgridioffsetptr0+vdwjidx0C,
 923                                                                   vdwgridioffsetptr0+vdwjidx0D,
 924                                                                   vdwgridioffsetptr0+vdwjidx0E,
 925                                                                   vdwgridioffsetptr0+vdwjidx0F,
 926                                                                   vdwgridioffsetptr0+vdwjidx0G,
 927                                                                   vdwgridioffsetptr0+vdwjidx0H);
 928
 929             /* EWALD ELECTROSTATICS */
 930
 931             /* Analytical PME correction */
 932             zeta2            = _mm256_mul_ps(beta2,rsq00);
 933             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 934             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 935             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 936             felec            = _mm256_mul_ps(qq00,felec);
 937
 938             /* Analytical LJ-PME */
 939             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 940             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 941             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 942             exponent         = gmx_simd_exp_r(ewcljrsq);
 943             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 944             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 945             /* f6A = 6 * C6grid * (1 - poly) */
 946             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
 947             /* f6B = C6grid * exponent * beta^6 */
 948             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
 949             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 950             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);
 951
 952             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 953
 954             fscal            = _mm256_add_ps(felec,fvdw);
 955
 956             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 957
 958             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 959
 960             /* Calculate temporary vectorial force */
 961             tx               = _mm256_mul_ps(fscal,dx00);
 962             ty               = _mm256_mul_ps(fscal,dy00);
 963             tz               = _mm256_mul_ps(fscal,dz00);
 964
 965             /* Update vectorial force */
 966             fix0             = _mm256_add_ps(fix0,tx);
 967             fiy0             = _mm256_add_ps(fiy0,ty);
 968             fiz0             = _mm256_add_ps(fiz0,tz);
 969
 970             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 971             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 972             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 973             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 974             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 975             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 976             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 977             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 978             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 979
 980             }
 981
 982             /* Inner loop uses 83 flops */
 983         }
 984
 985         /* End of innermost loop */
 986
 987         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 988                                                  f+i_coord_offset,fshift+i_shift_offset);
 989
 990         /* Increment number of inner iterations */
 991         inneriter                  += j_index_end - j_index_start;
 992
 993         /* Outer loop uses 7 flops */
 994     }
 995
 996     /* Increment number of outer iterations */
 997     outeriter        += nri;
 998
 999     /* Update outer/inner flops */
1000
1001     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
1002 }