src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_avx_256_double.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_double 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_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              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              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     real *           vdwgridioffsetptr0;
  86     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     real *           vdwgridioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     real *           vdwgridioffsetptr2;
  92     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m256d           c6grid_00;
 107     __m256d           c6grid_10;
 108     __m256d           c6grid_20;
 109     real             *vdwgridparam;
 110     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 111     __m256d           one_half  = _mm256_set1_pd(0.5);
 112     __m256d           minus_one = _mm256_set1_pd(-1.0);
 113     __m128i          ewitab;
 114     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 116     real             *ewtab;
 117     __m256d          dummy_mask,cutoff_mask;
 118     __m128           tmpmask0,tmpmask1;
 119     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 120     __m256d          one     = _mm256_set1_pd(1.0);
 121     __m256d          two     = _mm256_set1_pd(2.0);
 122     x                = xx[0];
 123     f                = ff[0];
 124
 125     nri              = nlist->nri;
 126     iinr             = nlist->iinr;
 127     jindex           = nlist->jindex;
 128     jjnr             = nlist->jjnr;
 129     shiftidx         = nlist->shift;
 130     gid              = nlist->gid;
 131     shiftvec         = fr->shift_vec[0];
 132     fshift           = fr->fshift[0];
 133     facel            = _mm256_set1_pd(fr->epsfac);
 134     charge           = mdatoms->chargeA;
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138     vdwgridparam     = fr->ljpme_c6grid;
 139     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 140     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 141     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 142
 143     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 144     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 145     beta2            = _mm256_mul_pd(beta,beta);
 146     beta3            = _mm256_mul_pd(beta,beta2);
 147
 148     ewtab            = fr->ic->tabq_coul_FDV0;
 149     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 150     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 151
 152     /* Setup water-specific parameters */
 153     inr              = nlist->iinr[0];
 154     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 155     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 156     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 157     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 158     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 159
 160     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 161     rcutoff_scalar   = fr->rcoulomb;
 162     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 163     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 164
 165     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 166     rvdw             = _mm256_set1_pd(fr->rvdw);
 167
 168     /* Avoid stupid compiler warnings */
 169     jnrA = jnrB = jnrC = jnrD = 0;
 170     j_coord_offsetA = 0;
 171     j_coord_offsetB = 0;
 172     j_coord_offsetC = 0;
 173     j_coord_offsetD = 0;
 174
 175     outeriter        = 0;
 176     inneriter        = 0;
 177
 178     for(iidx=0;iidx<4*DIM;iidx++)
 179     {
 180         scratch[iidx] = 0.0;
 181     }
 182
 183     /* Start outer loop over neighborlists */
 184     for(iidx=0; iidx<nri; iidx++)
 185     {
 186         /* Load shift vector for this list */
 187         i_shift_offset   = DIM*shiftidx[iidx];
 188
 189         /* Load limits for loop over neighbors */
 190         j_index_start    = jindex[iidx];
 191         j_index_end      = jindex[iidx+1];
 192
 193         /* Get outer coordinate index */
 194         inr              = iinr[iidx];
 195         i_coord_offset   = DIM*inr;
 196
 197         /* Load i particle coords and add shift vector */
 198         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 199                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 200
 201         fix0             = _mm256_setzero_pd();
 202         fiy0             = _mm256_setzero_pd();
 203         fiz0             = _mm256_setzero_pd();
 204         fix1             = _mm256_setzero_pd();
 205         fiy1             = _mm256_setzero_pd();
 206         fiz1             = _mm256_setzero_pd();
 207         fix2             = _mm256_setzero_pd();
 208         fiy2             = _mm256_setzero_pd();
 209         fiz2             = _mm256_setzero_pd();
 210
 211         /* Reset potential sums */
 212         velecsum         = _mm256_setzero_pd();
 213         vvdwsum          = _mm256_setzero_pd();
 214
 215         /* Start inner kernel loop */
 216         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 217         {
 218
 219             /* Get j neighbor index, and coordinate index */
 220             jnrA             = jjnr[jidx];
 221             jnrB             = jjnr[jidx+1];
 222             jnrC             = jjnr[jidx+2];
 223             jnrD             = jjnr[jidx+3];
 224             j_coord_offsetA  = DIM*jnrA;
 225             j_coord_offsetB  = DIM*jnrB;
 226             j_coord_offsetC  = DIM*jnrC;
 227             j_coord_offsetD  = DIM*jnrD;
 228
 229             /* load j atom coordinates */
 230             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 231                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 232                                                  &jx0,&jy0,&jz0);
 233
 234             /* Calculate displacement vector */
 235             dx00             = _mm256_sub_pd(ix0,jx0);
 236             dy00             = _mm256_sub_pd(iy0,jy0);
 237             dz00             = _mm256_sub_pd(iz0,jz0);
 238             dx10             = _mm256_sub_pd(ix1,jx0);
 239             dy10             = _mm256_sub_pd(iy1,jy0);
 240             dz10             = _mm256_sub_pd(iz1,jz0);
 241             dx20             = _mm256_sub_pd(ix2,jx0);
 242             dy20             = _mm256_sub_pd(iy2,jy0);
 243             dz20             = _mm256_sub_pd(iz2,jz0);
 244
 245             /* Calculate squared distance and things based on it */
 246             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 247             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 248             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 249
 250             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 251             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 252             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 253
 254             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 255             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 256             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 260                                                                  charge+jnrC+0,charge+jnrD+0);
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262             vdwjidx0B        = 2*vdwtype[jnrB+0];
 263             vdwjidx0C        = 2*vdwtype[jnrC+0];
 264             vdwjidx0D        = 2*vdwtype[jnrD+0];
 265
 266             fjx0             = _mm256_setzero_pd();
 267             fjy0             = _mm256_setzero_pd();
 268             fjz0             = _mm256_setzero_pd();
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 275             {
 276
 277             r00              = _mm256_mul_pd(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq00             = _mm256_mul_pd(iq0,jq0);
 281             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 282                                             vdwioffsetptr0+vdwjidx0B,
 283                                             vdwioffsetptr0+vdwjidx0C,
 284                                             vdwioffsetptr0+vdwjidx0D,
 285                                             &c6_00,&c12_00);
 286
 287             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 288                                                                   vdwgridioffsetptr0+vdwjidx0B,
 289                                                                   vdwgridioffsetptr0+vdwjidx0C,
 290                                                                   vdwgridioffsetptr0+vdwjidx0D);
 291
 292             /* EWALD ELECTROSTATICS */
 293
 294             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 295             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 296             ewitab           = _mm256_cvttpd_epi32(ewrt);
 297             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 298             ewitab           = _mm_slli_epi32(ewitab,2);
 299             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 300             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 301             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 302             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 303             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 304             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 305             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 306             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 307             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 308
 309             /* Analytical LJ-PME */
 310             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 311             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 312             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 313             exponent         = gmx_simd_exp_d(ewcljrsq);
 314             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 315             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 316             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 317             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 318             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 319             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 320                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 321             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 322             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 323
 324             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velec            = _mm256_and_pd(velec,cutoff_mask);
 328             velecsum         = _mm256_add_pd(velecsum,velec);
 329             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 330             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 331
 332             fscal            = _mm256_add_pd(felec,fvdw);
 333
 334             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 335
 336             /* Calculate temporary vectorial force */
 337             tx               = _mm256_mul_pd(fscal,dx00);
 338             ty               = _mm256_mul_pd(fscal,dy00);
 339             tz               = _mm256_mul_pd(fscal,dz00);
 340
 341             /* Update vectorial force */
 342             fix0             = _mm256_add_pd(fix0,tx);
 343             fiy0             = _mm256_add_pd(fiy0,ty);
 344             fiz0             = _mm256_add_pd(fiz0,tz);
 345
 346             fjx0             = _mm256_add_pd(fjx0,tx);
 347             fjy0             = _mm256_add_pd(fjy0,ty);
 348             fjz0             = _mm256_add_pd(fjz0,tz);
 349
 350             }
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 357             {
 358
 359             r10              = _mm256_mul_pd(rsq10,rinv10);
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq10             = _mm256_mul_pd(iq1,jq0);
 363
 364             /* EWALD ELECTROSTATICS */
 365
 366             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 367             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 368             ewitab           = _mm256_cvttpd_epi32(ewrt);
 369             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 370             ewitab           = _mm_slli_epi32(ewitab,2);
 371             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 372             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 373             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 374             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 375             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 376             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 377             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 378             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 379             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 380
 381             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velec            = _mm256_and_pd(velec,cutoff_mask);
 385             velecsum         = _mm256_add_pd(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm256_mul_pd(fscal,dx10);
 393             ty               = _mm256_mul_pd(fscal,dy10);
 394             tz               = _mm256_mul_pd(fscal,dz10);
 395
 396             /* Update vectorial force */
 397             fix1             = _mm256_add_pd(fix1,tx);
 398             fiy1             = _mm256_add_pd(fiy1,ty);
 399             fiz1             = _mm256_add_pd(fiz1,tz);
 400
 401             fjx0             = _mm256_add_pd(fjx0,tx);
 402             fjy0             = _mm256_add_pd(fjy0,ty);
 403             fjz0             = _mm256_add_pd(fjz0,tz);
 404
 405             }
 406
 407             /**************************
 408              * CALCULATE INTERACTIONS *
 409              **************************/
 410
 411             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 412             {
 413
 414             r20              = _mm256_mul_pd(rsq20,rinv20);
 415
 416             /* Compute parameters for interactions between i and j atoms */
 417             qq20             = _mm256_mul_pd(iq2,jq0);
 418
 419             /* EWALD ELECTROSTATICS */
 420
 421             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 422             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 423             ewitab           = _mm256_cvttpd_epi32(ewrt);
 424             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 425             ewitab           = _mm_slli_epi32(ewitab,2);
 426             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 427             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 428             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 429             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 430             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 431             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 432             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 433             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 434             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 435
 436             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velec            = _mm256_and_pd(velec,cutoff_mask);
 440             velecsum         = _mm256_add_pd(velecsum,velec);
 441
 442             fscal            = felec;
 443
 444             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm256_mul_pd(fscal,dx20);
 448             ty               = _mm256_mul_pd(fscal,dy20);
 449             tz               = _mm256_mul_pd(fscal,dz20);
 450
 451             /* Update vectorial force */
 452             fix2             = _mm256_add_pd(fix2,tx);
 453             fiy2             = _mm256_add_pd(fiy2,ty);
 454             fiz2             = _mm256_add_pd(fiz2,tz);
 455
 456             fjx0             = _mm256_add_pd(fjx0,tx);
 457             fjy0             = _mm256_add_pd(fjy0,ty);
 458             fjz0             = _mm256_add_pd(fjz0,tz);
 459
 460             }
 461
 462             fjptrA             = f+j_coord_offsetA;
 463             fjptrB             = f+j_coord_offsetB;
 464             fjptrC             = f+j_coord_offsetC;
 465             fjptrD             = f+j_coord_offsetD;
 466
 467             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 468
 469             /* Inner loop uses 177 flops */
 470         }
 471
 472         if(jidx<j_index_end)
 473         {
 474
 475             /* Get j neighbor index, and coordinate index */
 476             jnrlistA         = jjnr[jidx];
 477             jnrlistB         = jjnr[jidx+1];
 478             jnrlistC         = jjnr[jidx+2];
 479             jnrlistD         = jjnr[jidx+3];
 480             /* Sign of each element will be negative for non-real atoms.
 481              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 482              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 483              */
 484             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 485
 486             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 487             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 488             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 489
 490             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 491             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 492             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 493             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 494             j_coord_offsetA  = DIM*jnrA;
 495             j_coord_offsetB  = DIM*jnrB;
 496             j_coord_offsetC  = DIM*jnrC;
 497             j_coord_offsetD  = DIM*jnrD;
 498
 499             /* load j atom coordinates */
 500             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 501                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 502                                                  &jx0,&jy0,&jz0);
 503
 504             /* Calculate displacement vector */
 505             dx00             = _mm256_sub_pd(ix0,jx0);
 506             dy00             = _mm256_sub_pd(iy0,jy0);
 507             dz00             = _mm256_sub_pd(iz0,jz0);
 508             dx10             = _mm256_sub_pd(ix1,jx0);
 509             dy10             = _mm256_sub_pd(iy1,jy0);
 510             dz10             = _mm256_sub_pd(iz1,jz0);
 511             dx20             = _mm256_sub_pd(ix2,jx0);
 512             dy20             = _mm256_sub_pd(iy2,jy0);
 513             dz20             = _mm256_sub_pd(iz2,jz0);
 514
 515             /* Calculate squared distance and things based on it */
 516             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 517             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 518             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 519
 520             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 521             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 522             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 523
 524             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 525             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 526             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 527
 528             /* Load parameters for j particles */
 529             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 530                                                                  charge+jnrC+0,charge+jnrD+0);
 531             vdwjidx0A        = 2*vdwtype[jnrA+0];
 532             vdwjidx0B        = 2*vdwtype[jnrB+0];
 533             vdwjidx0C        = 2*vdwtype[jnrC+0];
 534             vdwjidx0D        = 2*vdwtype[jnrD+0];
 535
 536             fjx0             = _mm256_setzero_pd();
 537             fjy0             = _mm256_setzero_pd();
 538             fjz0             = _mm256_setzero_pd();
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 545             {
 546
 547             r00              = _mm256_mul_pd(rsq00,rinv00);
 548             r00              = _mm256_andnot_pd(dummy_mask,r00);
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq00             = _mm256_mul_pd(iq0,jq0);
 552             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 553                                             vdwioffsetptr0+vdwjidx0B,
 554                                             vdwioffsetptr0+vdwjidx0C,
 555                                             vdwioffsetptr0+vdwjidx0D,
 556                                             &c6_00,&c12_00);
 557
 558             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 559                                                                   vdwgridioffsetptr0+vdwjidx0B,
 560                                                                   vdwgridioffsetptr0+vdwjidx0C,
 561                                                                   vdwgridioffsetptr0+vdwjidx0D);
 562
 563             /* EWALD ELECTROSTATICS */
 564
 565             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 566             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 567             ewitab           = _mm256_cvttpd_epi32(ewrt);
 568             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 569             ewitab           = _mm_slli_epi32(ewitab,2);
 570             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 571             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 572             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 573             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 574             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 575             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 576             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 577             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 578             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 579
 580             /* Analytical LJ-PME */
 581             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 582             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 583             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 584             exponent         = gmx_simd_exp_d(ewcljrsq);
 585             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 586             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 587             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 588             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 589             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 590             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 591                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 592             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 593             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 594
 595             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 596
 597             /* Update potential sum for this i atom from the interaction with this j atom. */
 598             velec            = _mm256_and_pd(velec,cutoff_mask);
 599             velec            = _mm256_andnot_pd(dummy_mask,velec);
 600             velecsum         = _mm256_add_pd(velecsum,velec);
 601             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 602             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 603             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 604
 605             fscal            = _mm256_add_pd(felec,fvdw);
 606
 607             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 608
 609             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm256_mul_pd(fscal,dx00);
 613             ty               = _mm256_mul_pd(fscal,dy00);
 614             tz               = _mm256_mul_pd(fscal,dz00);
 615
 616             /* Update vectorial force */
 617             fix0             = _mm256_add_pd(fix0,tx);
 618             fiy0             = _mm256_add_pd(fiy0,ty);
 619             fiz0             = _mm256_add_pd(fiz0,tz);
 620
 621             fjx0             = _mm256_add_pd(fjx0,tx);
 622             fjy0             = _mm256_add_pd(fjy0,ty);
 623             fjz0             = _mm256_add_pd(fjz0,tz);
 624
 625             }
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 632             {
 633
 634             r10              = _mm256_mul_pd(rsq10,rinv10);
 635             r10              = _mm256_andnot_pd(dummy_mask,r10);
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             qq10             = _mm256_mul_pd(iq1,jq0);
 639
 640             /* EWALD ELECTROSTATICS */
 641
 642             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 643             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 644             ewitab           = _mm256_cvttpd_epi32(ewrt);
 645             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 646             ewitab           = _mm_slli_epi32(ewitab,2);
 647             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 648             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 649             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 650             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 651             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 652             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 653             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 654             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 655             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 656
 657             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 658
 659             /* Update potential sum for this i atom from the interaction with this j atom. */
 660             velec            = _mm256_and_pd(velec,cutoff_mask);
 661             velec            = _mm256_andnot_pd(dummy_mask,velec);
 662             velecsum         = _mm256_add_pd(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 667
 668             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm256_mul_pd(fscal,dx10);
 672             ty               = _mm256_mul_pd(fscal,dy10);
 673             tz               = _mm256_mul_pd(fscal,dz10);
 674
 675             /* Update vectorial force */
 676             fix1             = _mm256_add_pd(fix1,tx);
 677             fiy1             = _mm256_add_pd(fiy1,ty);
 678             fiz1             = _mm256_add_pd(fiz1,tz);
 679
 680             fjx0             = _mm256_add_pd(fjx0,tx);
 681             fjy0             = _mm256_add_pd(fjy0,ty);
 682             fjz0             = _mm256_add_pd(fjz0,tz);
 683
 684             }
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 691             {
 692
 693             r20              = _mm256_mul_pd(rsq20,rinv20);
 694             r20              = _mm256_andnot_pd(dummy_mask,r20);
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq20             = _mm256_mul_pd(iq2,jq0);
 698
 699             /* EWALD ELECTROSTATICS */
 700
 701             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 702             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 703             ewitab           = _mm256_cvttpd_epi32(ewrt);
 704             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 705             ewitab           = _mm_slli_epi32(ewitab,2);
 706             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 707             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 708             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 709             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 710             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 711             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 712             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 713             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 714             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 715
 716             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 717
 718             /* Update potential sum for this i atom from the interaction with this j atom. */
 719             velec            = _mm256_and_pd(velec,cutoff_mask);
 720             velec            = _mm256_andnot_pd(dummy_mask,velec);
 721             velecsum         = _mm256_add_pd(velecsum,velec);
 722
 723             fscal            = felec;
 724
 725             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 726
 727             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 728
 729             /* Calculate temporary vectorial force */
 730             tx               = _mm256_mul_pd(fscal,dx20);
 731             ty               = _mm256_mul_pd(fscal,dy20);
 732             tz               = _mm256_mul_pd(fscal,dz20);
 733
 734             /* Update vectorial force */
 735             fix2             = _mm256_add_pd(fix2,tx);
 736             fiy2             = _mm256_add_pd(fiy2,ty);
 737             fiz2             = _mm256_add_pd(fiz2,tz);
 738
 739             fjx0             = _mm256_add_pd(fjx0,tx);
 740             fjy0             = _mm256_add_pd(fjy0,ty);
 741             fjz0             = _mm256_add_pd(fjz0,tz);
 742
 743             }
 744
 745             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 746             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 747             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 748             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 749
 750             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 751
 752             /* Inner loop uses 180 flops */
 753         }
 754
 755         /* End of innermost loop */
 756
 757         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 758                                                  f+i_coord_offset,fshift+i_shift_offset);
 759
 760         ggid                        = gid[iidx];
 761         /* Update potential energies */
 762         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 763         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 764
 765         /* Increment number of inner iterations */
 766         inneriter                  += j_index_end - j_index_start;
 767
 768         /* Outer loop uses 20 flops */
 769     }
 770
 771     /* Increment number of outer iterations */
 772     outeriter        += nri;
 773
 774     /* Update outer/inner flops */
 775
 776     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*180);
 777 }
 778 /*
 779  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_double
 780  * Electrostatics interaction: Ewald
 781  * VdW interaction:            LJEwald
 782  * Geometry:                   Water3-Particle
 783  * Calculate force/pot:        Force
 784  */
 785 void
 786 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_double
 787                     (t_nblist                    * gmx_restrict       nlist,
 788                      rvec                        * gmx_restrict          xx,
 789                      rvec                        * gmx_restrict          ff,
 790                      t_forcerec                  * gmx_restrict          fr,
 791                      t_mdatoms                   * gmx_restrict     mdatoms,
 792                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 793                      t_nrnb                      * gmx_restrict        nrnb)
 794 {
 795     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 796      * just 0 for non-waters.
 797      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 798      * jnr indices corresponding to data put in the four positions in the SIMD register.
 799      */
 800     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 801     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 802     int              jnrA,jnrB,jnrC,jnrD;
 803     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 804     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 805     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 806     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 807     real             rcutoff_scalar;
 808     real             *shiftvec,*fshift,*x,*f;
 809     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 810     real             scratch[4*DIM];
 811     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 812     real *           vdwioffsetptr0;
 813     real *           vdwgridioffsetptr0;
 814     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 815     real *           vdwioffsetptr1;
 816     real *           vdwgridioffsetptr1;
 817     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 818     real *           vdwioffsetptr2;
 819     real *           vdwgridioffsetptr2;
 820     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 821     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 822     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 823     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 824     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 825     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 826     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 827     real             *charge;
 828     int              nvdwtype;
 829     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 830     int              *vdwtype;
 831     real             *vdwparam;
 832     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 833     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 834     __m256d           c6grid_00;
 835     __m256d           c6grid_10;
 836     __m256d           c6grid_20;
 837     real             *vdwgridparam;
 838     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 839     __m256d           one_half  = _mm256_set1_pd(0.5);
 840     __m256d           minus_one = _mm256_set1_pd(-1.0);
 841     __m128i          ewitab;
 842     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 843     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 844     real             *ewtab;
 845     __m256d          dummy_mask,cutoff_mask;
 846     __m128           tmpmask0,tmpmask1;
 847     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 848     __m256d          one     = _mm256_set1_pd(1.0);
 849     __m256d          two     = _mm256_set1_pd(2.0);
 850     x                = xx[0];
 851     f                = ff[0];
 852
 853     nri              = nlist->nri;
 854     iinr             = nlist->iinr;
 855     jindex           = nlist->jindex;
 856     jjnr             = nlist->jjnr;
 857     shiftidx         = nlist->shift;
 858     gid              = nlist->gid;
 859     shiftvec         = fr->shift_vec[0];
 860     fshift           = fr->fshift[0];
 861     facel            = _mm256_set1_pd(fr->epsfac);
 862     charge           = mdatoms->chargeA;
 863     nvdwtype         = fr->ntype;
 864     vdwparam         = fr->nbfp;
 865     vdwtype          = mdatoms->typeA;
 866     vdwgridparam     = fr->ljpme_c6grid;
 867     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 868     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 869     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 870
 871     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 872     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 873     beta2            = _mm256_mul_pd(beta,beta);
 874     beta3            = _mm256_mul_pd(beta,beta2);
 875
 876     ewtab            = fr->ic->tabq_coul_F;
 877     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 878     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 879
 880     /* Setup water-specific parameters */
 881     inr              = nlist->iinr[0];
 882     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 883     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 884     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 885     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 886     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 887
 888     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 889     rcutoff_scalar   = fr->rcoulomb;
 890     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 891     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 892
 893     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 894     rvdw             = _mm256_set1_pd(fr->rvdw);
 895
 896     /* Avoid stupid compiler warnings */
 897     jnrA = jnrB = jnrC = jnrD = 0;
 898     j_coord_offsetA = 0;
 899     j_coord_offsetB = 0;
 900     j_coord_offsetC = 0;
 901     j_coord_offsetD = 0;
 902
 903     outeriter        = 0;
 904     inneriter        = 0;
 905
 906     for(iidx=0;iidx<4*DIM;iidx++)
 907     {
 908         scratch[iidx] = 0.0;
 909     }
 910
 911     /* Start outer loop over neighborlists */
 912     for(iidx=0; iidx<nri; iidx++)
 913     {
 914         /* Load shift vector for this list */
 915         i_shift_offset   = DIM*shiftidx[iidx];
 916
 917         /* Load limits for loop over neighbors */
 918         j_index_start    = jindex[iidx];
 919         j_index_end      = jindex[iidx+1];
 920
 921         /* Get outer coordinate index */
 922         inr              = iinr[iidx];
 923         i_coord_offset   = DIM*inr;
 924
 925         /* Load i particle coords and add shift vector */
 926         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 927                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 928
 929         fix0             = _mm256_setzero_pd();
 930         fiy0             = _mm256_setzero_pd();
 931         fiz0             = _mm256_setzero_pd();
 932         fix1             = _mm256_setzero_pd();
 933         fiy1             = _mm256_setzero_pd();
 934         fiz1             = _mm256_setzero_pd();
 935         fix2             = _mm256_setzero_pd();
 936         fiy2             = _mm256_setzero_pd();
 937         fiz2             = _mm256_setzero_pd();
 938
 939         /* Start inner kernel loop */
 940         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 941         {
 942
 943             /* Get j neighbor index, and coordinate index */
 944             jnrA             = jjnr[jidx];
 945             jnrB             = jjnr[jidx+1];
 946             jnrC             = jjnr[jidx+2];
 947             jnrD             = jjnr[jidx+3];
 948             j_coord_offsetA  = DIM*jnrA;
 949             j_coord_offsetB  = DIM*jnrB;
 950             j_coord_offsetC  = DIM*jnrC;
 951             j_coord_offsetD  = DIM*jnrD;
 952
 953             /* load j atom coordinates */
 954             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 955                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 956                                                  &jx0,&jy0,&jz0);
 957
 958             /* Calculate displacement vector */
 959             dx00             = _mm256_sub_pd(ix0,jx0);
 960             dy00             = _mm256_sub_pd(iy0,jy0);
 961             dz00             = _mm256_sub_pd(iz0,jz0);
 962             dx10             = _mm256_sub_pd(ix1,jx0);
 963             dy10             = _mm256_sub_pd(iy1,jy0);
 964             dz10             = _mm256_sub_pd(iz1,jz0);
 965             dx20             = _mm256_sub_pd(ix2,jx0);
 966             dy20             = _mm256_sub_pd(iy2,jy0);
 967             dz20             = _mm256_sub_pd(iz2,jz0);
 968
 969             /* Calculate squared distance and things based on it */
 970             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 971             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 972             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 973
 974             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 975             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 976             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 977
 978             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 979             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 980             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 981
 982             /* Load parameters for j particles */
 983             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 984                                                                  charge+jnrC+0,charge+jnrD+0);
 985             vdwjidx0A        = 2*vdwtype[jnrA+0];
 986             vdwjidx0B        = 2*vdwtype[jnrB+0];
 987             vdwjidx0C        = 2*vdwtype[jnrC+0];
 988             vdwjidx0D        = 2*vdwtype[jnrD+0];
 989
 990             fjx0             = _mm256_setzero_pd();
 991             fjy0             = _mm256_setzero_pd();
 992             fjz0             = _mm256_setzero_pd();
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 999             {
1000
1001             r00              = _mm256_mul_pd(rsq00,rinv00);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq00             = _mm256_mul_pd(iq0,jq0);
1005             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1006                                             vdwioffsetptr0+vdwjidx0B,
1007                                             vdwioffsetptr0+vdwjidx0C,
1008                                             vdwioffsetptr0+vdwjidx0D,
1009                                             &c6_00,&c12_00);
1010
1011             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1012                                                                   vdwgridioffsetptr0+vdwjidx0B,
1013                                                                   vdwgridioffsetptr0+vdwjidx0C,
1014                                                                   vdwgridioffsetptr0+vdwjidx0D);
1015
1016             /* EWALD ELECTROSTATICS */
1017
1018             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1019             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1020             ewitab           = _mm256_cvttpd_epi32(ewrt);
1021             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1022             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1023                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1024                                             &ewtabF,&ewtabFn);
1025             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1026             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1027
1028             /* Analytical LJ-PME */
1029             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1030             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1031             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1032             exponent         = gmx_simd_exp_d(ewcljrsq);
1033             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1034             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1035             /* f6A = 6 * C6grid * (1 - poly) */
1036             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1037             /* f6B = C6grid * exponent * beta^6 */
1038             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1039             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1040             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1041
1042             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1043
1044             fscal            = _mm256_add_pd(felec,fvdw);
1045
1046             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm256_mul_pd(fscal,dx00);
1050             ty               = _mm256_mul_pd(fscal,dy00);
1051             tz               = _mm256_mul_pd(fscal,dz00);
1052
1053             /* Update vectorial force */
1054             fix0             = _mm256_add_pd(fix0,tx);
1055             fiy0             = _mm256_add_pd(fiy0,ty);
1056             fiz0             = _mm256_add_pd(fiz0,tz);
1057
1058             fjx0             = _mm256_add_pd(fjx0,tx);
1059             fjy0             = _mm256_add_pd(fjy0,ty);
1060             fjz0             = _mm256_add_pd(fjz0,tz);
1061
1062             }
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1069             {
1070
1071             r10              = _mm256_mul_pd(rsq10,rinv10);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq10             = _mm256_mul_pd(iq1,jq0);
1075
1076             /* EWALD ELECTROSTATICS */
1077
1078             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1079             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1080             ewitab           = _mm256_cvttpd_epi32(ewrt);
1081             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1082             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1083                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1084                                             &ewtabF,&ewtabFn);
1085             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1086             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1087
1088             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm256_mul_pd(fscal,dx10);
1096             ty               = _mm256_mul_pd(fscal,dy10);
1097             tz               = _mm256_mul_pd(fscal,dz10);
1098
1099             /* Update vectorial force */
1100             fix1             = _mm256_add_pd(fix1,tx);
1101             fiy1             = _mm256_add_pd(fiy1,ty);
1102             fiz1             = _mm256_add_pd(fiz1,tz);
1103
1104             fjx0             = _mm256_add_pd(fjx0,tx);
1105             fjy0             = _mm256_add_pd(fjy0,ty);
1106             fjz0             = _mm256_add_pd(fjz0,tz);
1107
1108             }
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1115             {
1116
1117             r20              = _mm256_mul_pd(rsq20,rinv20);
1118
1119             /* Compute parameters for interactions between i and j atoms */
1120             qq20             = _mm256_mul_pd(iq2,jq0);
1121
1122             /* EWALD ELECTROSTATICS */
1123
1124             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1125             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1126             ewitab           = _mm256_cvttpd_epi32(ewrt);
1127             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1128             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1129                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1130                                             &ewtabF,&ewtabFn);
1131             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1132             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1133
1134             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1135
1136             fscal            = felec;
1137
1138             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm256_mul_pd(fscal,dx20);
1142             ty               = _mm256_mul_pd(fscal,dy20);
1143             tz               = _mm256_mul_pd(fscal,dz20);
1144
1145             /* Update vectorial force */
1146             fix2             = _mm256_add_pd(fix2,tx);
1147             fiy2             = _mm256_add_pd(fiy2,ty);
1148             fiz2             = _mm256_add_pd(fiz2,tz);
1149
1150             fjx0             = _mm256_add_pd(fjx0,tx);
1151             fjy0             = _mm256_add_pd(fjy0,ty);
1152             fjz0             = _mm256_add_pd(fjz0,tz);
1153
1154             }
1155
1156             fjptrA             = f+j_coord_offsetA;
1157             fjptrB             = f+j_coord_offsetB;
1158             fjptrC             = f+j_coord_offsetC;
1159             fjptrD             = f+j_coord_offsetD;
1160
1161             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1162
1163             /* Inner loop uses 143 flops */
1164         }
1165
1166         if(jidx<j_index_end)
1167         {
1168
1169             /* Get j neighbor index, and coordinate index */
1170             jnrlistA         = jjnr[jidx];
1171             jnrlistB         = jjnr[jidx+1];
1172             jnrlistC         = jjnr[jidx+2];
1173             jnrlistD         = jjnr[jidx+3];
1174             /* Sign of each element will be negative for non-real atoms.
1175              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1176              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1177              */
1178             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1179
1180             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1181             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1182             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1183
1184             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1185             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1186             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1187             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1188             j_coord_offsetA  = DIM*jnrA;
1189             j_coord_offsetB  = DIM*jnrB;
1190             j_coord_offsetC  = DIM*jnrC;
1191             j_coord_offsetD  = DIM*jnrD;
1192
1193             /* load j atom coordinates */
1194             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1195                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1196                                                  &jx0,&jy0,&jz0);
1197
1198             /* Calculate displacement vector */
1199             dx00             = _mm256_sub_pd(ix0,jx0);
1200             dy00             = _mm256_sub_pd(iy0,jy0);
1201             dz00             = _mm256_sub_pd(iz0,jz0);
1202             dx10             = _mm256_sub_pd(ix1,jx0);
1203             dy10             = _mm256_sub_pd(iy1,jy0);
1204             dz10             = _mm256_sub_pd(iz1,jz0);
1205             dx20             = _mm256_sub_pd(ix2,jx0);
1206             dy20             = _mm256_sub_pd(iy2,jy0);
1207             dz20             = _mm256_sub_pd(iz2,jz0);
1208
1209             /* Calculate squared distance and things based on it */
1210             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1211             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1212             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1213
1214             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1215             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1216             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1217
1218             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1219             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1220             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1221
1222             /* Load parameters for j particles */
1223             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1224                                                                  charge+jnrC+0,charge+jnrD+0);
1225             vdwjidx0A        = 2*vdwtype[jnrA+0];
1226             vdwjidx0B        = 2*vdwtype[jnrB+0];
1227             vdwjidx0C        = 2*vdwtype[jnrC+0];
1228             vdwjidx0D        = 2*vdwtype[jnrD+0];
1229
1230             fjx0             = _mm256_setzero_pd();
1231             fjy0             = _mm256_setzero_pd();
1232             fjz0             = _mm256_setzero_pd();
1233
1234             /**************************
1235              * CALCULATE INTERACTIONS *
1236              **************************/
1237
1238             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1239             {
1240
1241             r00              = _mm256_mul_pd(rsq00,rinv00);
1242             r00              = _mm256_andnot_pd(dummy_mask,r00);
1243
1244             /* Compute parameters for interactions between i and j atoms */
1245             qq00             = _mm256_mul_pd(iq0,jq0);
1246             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1247                                             vdwioffsetptr0+vdwjidx0B,
1248                                             vdwioffsetptr0+vdwjidx0C,
1249                                             vdwioffsetptr0+vdwjidx0D,
1250                                             &c6_00,&c12_00);
1251
1252             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1253                                                                   vdwgridioffsetptr0+vdwjidx0B,
1254                                                                   vdwgridioffsetptr0+vdwjidx0C,
1255                                                                   vdwgridioffsetptr0+vdwjidx0D);
1256
1257             /* EWALD ELECTROSTATICS */
1258
1259             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1260             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1261             ewitab           = _mm256_cvttpd_epi32(ewrt);
1262             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1263             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1264                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1265                                             &ewtabF,&ewtabFn);
1266             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1267             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1268
1269             /* Analytical LJ-PME */
1270             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1271             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1272             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1273             exponent         = gmx_simd_exp_d(ewcljrsq);
1274             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1275             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1276             /* f6A = 6 * C6grid * (1 - poly) */
1277             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1278             /* f6B = C6grid * exponent * beta^6 */
1279             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1280             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1281             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1282
1283             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1284
1285             fscal            = _mm256_add_pd(felec,fvdw);
1286
1287             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1288
1289             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1290
1291             /* Calculate temporary vectorial force */
1292             tx               = _mm256_mul_pd(fscal,dx00);
1293             ty               = _mm256_mul_pd(fscal,dy00);
1294             tz               = _mm256_mul_pd(fscal,dz00);
1295
1296             /* Update vectorial force */
1297             fix0             = _mm256_add_pd(fix0,tx);
1298             fiy0             = _mm256_add_pd(fiy0,ty);
1299             fiz0             = _mm256_add_pd(fiz0,tz);
1300
1301             fjx0             = _mm256_add_pd(fjx0,tx);
1302             fjy0             = _mm256_add_pd(fjy0,ty);
1303             fjz0             = _mm256_add_pd(fjz0,tz);
1304
1305             }
1306
1307             /**************************
1308              * CALCULATE INTERACTIONS *
1309              **************************/
1310
1311             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1312             {
1313
1314             r10              = _mm256_mul_pd(rsq10,rinv10);
1315             r10              = _mm256_andnot_pd(dummy_mask,r10);
1316
1317             /* Compute parameters for interactions between i and j atoms */
1318             qq10             = _mm256_mul_pd(iq1,jq0);
1319
1320             /* EWALD ELECTROSTATICS */
1321
1322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1323             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1324             ewitab           = _mm256_cvttpd_epi32(ewrt);
1325             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1326             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1327                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1328                                             &ewtabF,&ewtabFn);
1329             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1330             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1331
1332             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1333
1334             fscal            = felec;
1335
1336             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1337
1338             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1339
1340             /* Calculate temporary vectorial force */
1341             tx               = _mm256_mul_pd(fscal,dx10);
1342             ty               = _mm256_mul_pd(fscal,dy10);
1343             tz               = _mm256_mul_pd(fscal,dz10);
1344
1345             /* Update vectorial force */
1346             fix1             = _mm256_add_pd(fix1,tx);
1347             fiy1             = _mm256_add_pd(fiy1,ty);
1348             fiz1             = _mm256_add_pd(fiz1,tz);
1349
1350             fjx0             = _mm256_add_pd(fjx0,tx);
1351             fjy0             = _mm256_add_pd(fjy0,ty);
1352             fjz0             = _mm256_add_pd(fjz0,tz);
1353
1354             }
1355
1356             /**************************
1357              * CALCULATE INTERACTIONS *
1358              **************************/
1359
1360             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1361             {
1362
1363             r20              = _mm256_mul_pd(rsq20,rinv20);
1364             r20              = _mm256_andnot_pd(dummy_mask,r20);
1365
1366             /* Compute parameters for interactions between i and j atoms */
1367             qq20             = _mm256_mul_pd(iq2,jq0);
1368
1369             /* EWALD ELECTROSTATICS */
1370
1371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1372             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1373             ewitab           = _mm256_cvttpd_epi32(ewrt);
1374             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1375             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1376                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1377                                             &ewtabF,&ewtabFn);
1378             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1379             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1380
1381             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1382
1383             fscal            = felec;
1384
1385             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1386
1387             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1388
1389             /* Calculate temporary vectorial force */
1390             tx               = _mm256_mul_pd(fscal,dx20);
1391             ty               = _mm256_mul_pd(fscal,dy20);
1392             tz               = _mm256_mul_pd(fscal,dz20);
1393
1394             /* Update vectorial force */
1395             fix2             = _mm256_add_pd(fix2,tx);
1396             fiy2             = _mm256_add_pd(fiy2,ty);
1397             fiz2             = _mm256_add_pd(fiz2,tz);
1398
1399             fjx0             = _mm256_add_pd(fjx0,tx);
1400             fjy0             = _mm256_add_pd(fjy0,ty);
1401             fjz0             = _mm256_add_pd(fjz0,tz);
1402
1403             }
1404
1405             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1406             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1407             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1408             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1409
1410             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1411
1412             /* Inner loop uses 146 flops */
1413         }
1414
1415         /* End of innermost loop */
1416
1417         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1418                                                  f+i_coord_offset,fshift+i_shift_offset);
1419
1420         /* Increment number of inner iterations */
1421         inneriter                  += j_index_end - j_index_start;
1422
1423         /* Outer loop uses 18 flops */
1424     }
1425
1426     /* Increment number of outer iterations */
1427     outeriter        += nri;
1428
1429     /* Update outer/inner flops */
1430
1431     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*146);
1432 }