src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_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_VdwLJSh_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_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     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256i          ewitab;
 109     __m128i          ewitab_lo,ewitab_hi;
 110     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m256           dummy_mask,cutoff_mask;
 114     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 115     __m256           one     = _mm256_set1_ps(1.0);
 116     __m256           two     = _mm256_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_ps(beta,beta);
 137     beta3            = _mm256_mul_ps(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 146     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 147     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 148     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 149
 150     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 151     rcutoff_scalar   = fr->rcoulomb;
 152     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 153     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 154
 155     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 156     rvdw             = _mm256_set1_ps(fr->rvdw);
 157
 158     /* Avoid stupid compiler warnings */
 159     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 160     j_coord_offsetA = 0;
 161     j_coord_offsetB = 0;
 162     j_coord_offsetC = 0;
 163     j_coord_offsetD = 0;
 164     j_coord_offsetE = 0;
 165     j_coord_offsetF = 0;
 166     j_coord_offsetG = 0;
 167     j_coord_offsetH = 0;
 168
 169     outeriter        = 0;
 170     inneriter        = 0;
 171
 172     for(iidx=0;iidx<4*DIM;iidx++)
 173     {
 174         scratch[iidx] = 0.0;
 175     }
 176
 177     /* Start outer loop over neighborlists */
 178     for(iidx=0; iidx<nri; iidx++)
 179     {
 180         /* Load shift vector for this list */
 181         i_shift_offset   = DIM*shiftidx[iidx];
 182
 183         /* Load limits for loop over neighbors */
 184         j_index_start    = jindex[iidx];
 185         j_index_end      = jindex[iidx+1];
 186
 187         /* Get outer coordinate index */
 188         inr              = iinr[iidx];
 189         i_coord_offset   = DIM*inr;
 190
 191         /* Load i particle coords and add shift vector */
 192         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 193                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 194
 195         fix0             = _mm256_setzero_ps();
 196         fiy0             = _mm256_setzero_ps();
 197         fiz0             = _mm256_setzero_ps();
 198         fix1             = _mm256_setzero_ps();
 199         fiy1             = _mm256_setzero_ps();
 200         fiz1             = _mm256_setzero_ps();
 201         fix2             = _mm256_setzero_ps();
 202         fiy2             = _mm256_setzero_ps();
 203         fiz2             = _mm256_setzero_ps();
 204         fix3             = _mm256_setzero_ps();
 205         fiy3             = _mm256_setzero_ps();
 206         fiz3             = _mm256_setzero_ps();
 207
 208         /* Reset potential sums */
 209         velecsum         = _mm256_setzero_ps();
 210         vvdwsum          = _mm256_setzero_ps();
 211
 212         /* Start inner kernel loop */
 213         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 214         {
 215
 216             /* Get j neighbor index, and coordinate index */
 217             jnrA             = jjnr[jidx];
 218             jnrB             = jjnr[jidx+1];
 219             jnrC             = jjnr[jidx+2];
 220             jnrD             = jjnr[jidx+3];
 221             jnrE             = jjnr[jidx+4];
 222             jnrF             = jjnr[jidx+5];
 223             jnrG             = jjnr[jidx+6];
 224             jnrH             = jjnr[jidx+7];
 225             j_coord_offsetA  = DIM*jnrA;
 226             j_coord_offsetB  = DIM*jnrB;
 227             j_coord_offsetC  = DIM*jnrC;
 228             j_coord_offsetD  = DIM*jnrD;
 229             j_coord_offsetE  = DIM*jnrE;
 230             j_coord_offsetF  = DIM*jnrF;
 231             j_coord_offsetG  = DIM*jnrG;
 232             j_coord_offsetH  = DIM*jnrH;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 237                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 238                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 239                                                  &jx0,&jy0,&jz0);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm256_sub_ps(ix0,jx0);
 243             dy00             = _mm256_sub_ps(iy0,jy0);
 244             dz00             = _mm256_sub_ps(iz0,jz0);
 245             dx10             = _mm256_sub_ps(ix1,jx0);
 246             dy10             = _mm256_sub_ps(iy1,jy0);
 247             dz10             = _mm256_sub_ps(iz1,jz0);
 248             dx20             = _mm256_sub_ps(ix2,jx0);
 249             dy20             = _mm256_sub_ps(iy2,jy0);
 250             dz20             = _mm256_sub_ps(iz2,jz0);
 251             dx30             = _mm256_sub_ps(ix3,jx0);
 252             dy30             = _mm256_sub_ps(iy3,jy0);
 253             dz30             = _mm256_sub_ps(iz3,jz0);
 254
 255             /* Calculate squared distance and things based on it */
 256             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 257             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 258             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 259             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 260
 261             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 262             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 263             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 264
 265             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 266             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 267             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 268             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 269
 270             /* Load parameters for j particles */
 271             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 272                                                                  charge+jnrC+0,charge+jnrD+0,
 273                                                                  charge+jnrE+0,charge+jnrF+0,
 274                                                                  charge+jnrG+0,charge+jnrH+0);
 275             vdwjidx0A        = 2*vdwtype[jnrA+0];
 276             vdwjidx0B        = 2*vdwtype[jnrB+0];
 277             vdwjidx0C        = 2*vdwtype[jnrC+0];
 278             vdwjidx0D        = 2*vdwtype[jnrD+0];
 279             vdwjidx0E        = 2*vdwtype[jnrE+0];
 280             vdwjidx0F        = 2*vdwtype[jnrF+0];
 281             vdwjidx0G        = 2*vdwtype[jnrG+0];
 282             vdwjidx0H        = 2*vdwtype[jnrH+0];
 283
 284             fjx0             = _mm256_setzero_ps();
 285             fjy0             = _mm256_setzero_ps();
 286             fjz0             = _mm256_setzero_ps();
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 293             {
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 297                                             vdwioffsetptr0+vdwjidx0B,
 298                                             vdwioffsetptr0+vdwjidx0C,
 299                                             vdwioffsetptr0+vdwjidx0D,
 300                                             vdwioffsetptr0+vdwjidx0E,
 301                                             vdwioffsetptr0+vdwjidx0F,
 302                                             vdwioffsetptr0+vdwjidx0G,
 303                                             vdwioffsetptr0+vdwjidx0H,
 304                                             &c6_00,&c12_00);
 305
 306             /* LENNARD-JONES DISPERSION/REPULSION */
 307
 308             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 309             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 310             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 311             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) ,
 312                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 313             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 314
 315             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 319             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 320
 321             fscal            = fvdw;
 322
 323             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm256_mul_ps(fscal,dx00);
 327             ty               = _mm256_mul_ps(fscal,dy00);
 328             tz               = _mm256_mul_ps(fscal,dz00);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm256_add_ps(fix0,tx);
 332             fiy0             = _mm256_add_ps(fiy0,ty);
 333             fiz0             = _mm256_add_ps(fiz0,tz);
 334
 335             fjx0             = _mm256_add_ps(fjx0,tx);
 336             fjy0             = _mm256_add_ps(fjy0,ty);
 337             fjz0             = _mm256_add_ps(fjz0,tz);
 338
 339             }
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 346             {
 347
 348             r10              = _mm256_mul_ps(rsq10,rinv10);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq10             = _mm256_mul_ps(iq1,jq0);
 352
 353             /* EWALD ELECTROSTATICS */
 354
 355             /* Analytical PME correction */
 356             zeta2            = _mm256_mul_ps(beta2,rsq10);
 357             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 358             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 359             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 360             felec            = _mm256_mul_ps(qq10,felec);
 361             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 362             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 363             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 364             velec            = _mm256_mul_ps(qq10,velec);
 365
 366             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _mm256_and_ps(velec,cutoff_mask);
 370             velecsum         = _mm256_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_ps(fscal,dx10);
 378             ty               = _mm256_mul_ps(fscal,dy10);
 379             tz               = _mm256_mul_ps(fscal,dz10);
 380
 381             /* Update vectorial force */
 382             fix1             = _mm256_add_ps(fix1,tx);
 383             fiy1             = _mm256_add_ps(fiy1,ty);
 384             fiz1             = _mm256_add_ps(fiz1,tz);
 385
 386             fjx0             = _mm256_add_ps(fjx0,tx);
 387             fjy0             = _mm256_add_ps(fjy0,ty);
 388             fjz0             = _mm256_add_ps(fjz0,tz);
 389
 390             }
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 397             {
 398
 399             r20              = _mm256_mul_ps(rsq20,rinv20);
 400
 401             /* Compute parameters for interactions between i and j atoms */
 402             qq20             = _mm256_mul_ps(iq2,jq0);
 403
 404             /* EWALD ELECTROSTATICS */
 405
 406             /* Analytical PME correction */
 407             zeta2            = _mm256_mul_ps(beta2,rsq20);
 408             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 409             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 410             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 411             felec            = _mm256_mul_ps(qq20,felec);
 412             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 413             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 414             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 415             velec            = _mm256_mul_ps(qq20,velec);
 416
 417             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm256_and_ps(velec,cutoff_mask);
 421             velecsum         = _mm256_add_ps(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm256_mul_ps(fscal,dx20);
 429             ty               = _mm256_mul_ps(fscal,dy20);
 430             tz               = _mm256_mul_ps(fscal,dz20);
 431
 432             /* Update vectorial force */
 433             fix2             = _mm256_add_ps(fix2,tx);
 434             fiy2             = _mm256_add_ps(fiy2,ty);
 435             fiz2             = _mm256_add_ps(fiz2,tz);
 436
 437             fjx0             = _mm256_add_ps(fjx0,tx);
 438             fjy0             = _mm256_add_ps(fjy0,ty);
 439             fjz0             = _mm256_add_ps(fjz0,tz);
 440
 441             }
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 448             {
 449
 450             r30              = _mm256_mul_ps(rsq30,rinv30);
 451
 452             /* Compute parameters for interactions between i and j atoms */
 453             qq30             = _mm256_mul_ps(iq3,jq0);
 454
 455             /* EWALD ELECTROSTATICS */
 456
 457             /* Analytical PME correction */
 458             zeta2            = _mm256_mul_ps(beta2,rsq30);
 459             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 460             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 461             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 462             felec            = _mm256_mul_ps(qq30,felec);
 463             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 464             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 465             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 466             velec            = _mm256_mul_ps(qq30,velec);
 467
 468             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             velec            = _mm256_and_ps(velec,cutoff_mask);
 472             velecsum         = _mm256_add_ps(velecsum,velec);
 473
 474             fscal            = felec;
 475
 476             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 477
 478             /* Calculate temporary vectorial force */
 479             tx               = _mm256_mul_ps(fscal,dx30);
 480             ty               = _mm256_mul_ps(fscal,dy30);
 481             tz               = _mm256_mul_ps(fscal,dz30);
 482
 483             /* Update vectorial force */
 484             fix3             = _mm256_add_ps(fix3,tx);
 485             fiy3             = _mm256_add_ps(fiy3,ty);
 486             fiz3             = _mm256_add_ps(fiz3,tz);
 487
 488             fjx0             = _mm256_add_ps(fjx0,tx);
 489             fjy0             = _mm256_add_ps(fjy0,ty);
 490             fjz0             = _mm256_add_ps(fjz0,tz);
 491
 492             }
 493
 494             fjptrA             = f+j_coord_offsetA;
 495             fjptrB             = f+j_coord_offsetB;
 496             fjptrC             = f+j_coord_offsetC;
 497             fjptrD             = f+j_coord_offsetD;
 498             fjptrE             = f+j_coord_offsetE;
 499             fjptrF             = f+j_coord_offsetF;
 500             fjptrG             = f+j_coord_offsetG;
 501             fjptrH             = f+j_coord_offsetH;
 502
 503             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 504
 505             /* Inner loop uses 371 flops */
 506         }
 507
 508         if(jidx<j_index_end)
 509         {
 510
 511             /* Get j neighbor index, and coordinate index */
 512             jnrlistA         = jjnr[jidx];
 513             jnrlistB         = jjnr[jidx+1];
 514             jnrlistC         = jjnr[jidx+2];
 515             jnrlistD         = jjnr[jidx+3];
 516             jnrlistE         = jjnr[jidx+4];
 517             jnrlistF         = jjnr[jidx+5];
 518             jnrlistG         = jjnr[jidx+6];
 519             jnrlistH         = jjnr[jidx+7];
 520             /* Sign of each element will be negative for non-real atoms.
 521              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 522              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 523              */
 524             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 525                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 526
 527             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 528             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 529             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 530             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 531             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 532             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 533             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 534             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 535             j_coord_offsetA  = DIM*jnrA;
 536             j_coord_offsetB  = DIM*jnrB;
 537             j_coord_offsetC  = DIM*jnrC;
 538             j_coord_offsetD  = DIM*jnrD;
 539             j_coord_offsetE  = DIM*jnrE;
 540             j_coord_offsetF  = DIM*jnrF;
 541             j_coord_offsetG  = DIM*jnrG;
 542             j_coord_offsetH  = DIM*jnrH;
 543
 544             /* load j atom coordinates */
 545             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 546                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 547                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 548                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 549                                                  &jx0,&jy0,&jz0);
 550
 551             /* Calculate displacement vector */
 552             dx00             = _mm256_sub_ps(ix0,jx0);
 553             dy00             = _mm256_sub_ps(iy0,jy0);
 554             dz00             = _mm256_sub_ps(iz0,jz0);
 555             dx10             = _mm256_sub_ps(ix1,jx0);
 556             dy10             = _mm256_sub_ps(iy1,jy0);
 557             dz10             = _mm256_sub_ps(iz1,jz0);
 558             dx20             = _mm256_sub_ps(ix2,jx0);
 559             dy20             = _mm256_sub_ps(iy2,jy0);
 560             dz20             = _mm256_sub_ps(iz2,jz0);
 561             dx30             = _mm256_sub_ps(ix3,jx0);
 562             dy30             = _mm256_sub_ps(iy3,jy0);
 563             dz30             = _mm256_sub_ps(iz3,jz0);
 564
 565             /* Calculate squared distance and things based on it */
 566             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 567             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 568             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 569             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 570
 571             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 572             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 573             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 574
 575             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 576             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 577             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 578             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 579
 580             /* Load parameters for j particles */
 581             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 582                                                                  charge+jnrC+0,charge+jnrD+0,
 583                                                                  charge+jnrE+0,charge+jnrF+0,
 584                                                                  charge+jnrG+0,charge+jnrH+0);
 585             vdwjidx0A        = 2*vdwtype[jnrA+0];
 586             vdwjidx0B        = 2*vdwtype[jnrB+0];
 587             vdwjidx0C        = 2*vdwtype[jnrC+0];
 588             vdwjidx0D        = 2*vdwtype[jnrD+0];
 589             vdwjidx0E        = 2*vdwtype[jnrE+0];
 590             vdwjidx0F        = 2*vdwtype[jnrF+0];
 591             vdwjidx0G        = 2*vdwtype[jnrG+0];
 592             vdwjidx0H        = 2*vdwtype[jnrH+0];
 593
 594             fjx0             = _mm256_setzero_ps();
 595             fjy0             = _mm256_setzero_ps();
 596             fjz0             = _mm256_setzero_ps();
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 603             {
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 607                                             vdwioffsetptr0+vdwjidx0B,
 608                                             vdwioffsetptr0+vdwjidx0C,
 609                                             vdwioffsetptr0+vdwjidx0D,
 610                                             vdwioffsetptr0+vdwjidx0E,
 611                                             vdwioffsetptr0+vdwjidx0F,
 612                                             vdwioffsetptr0+vdwjidx0G,
 613                                             vdwioffsetptr0+vdwjidx0H,
 614                                             &c6_00,&c12_00);
 615
 616             /* LENNARD-JONES DISPERSION/REPULSION */
 617
 618             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 619             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 620             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 621             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) ,
 622                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 623             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 624
 625             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 629             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 630             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 631
 632             fscal            = fvdw;
 633
 634             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 635
 636             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 637
 638             /* Calculate temporary vectorial force */
 639             tx               = _mm256_mul_ps(fscal,dx00);
 640             ty               = _mm256_mul_ps(fscal,dy00);
 641             tz               = _mm256_mul_ps(fscal,dz00);
 642
 643             /* Update vectorial force */
 644             fix0             = _mm256_add_ps(fix0,tx);
 645             fiy0             = _mm256_add_ps(fiy0,ty);
 646             fiz0             = _mm256_add_ps(fiz0,tz);
 647
 648             fjx0             = _mm256_add_ps(fjx0,tx);
 649             fjy0             = _mm256_add_ps(fjy0,ty);
 650             fjz0             = _mm256_add_ps(fjz0,tz);
 651
 652             }
 653
 654             /**************************
 655              * CALCULATE INTERACTIONS *
 656              **************************/
 657
 658             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 659             {
 660
 661             r10              = _mm256_mul_ps(rsq10,rinv10);
 662             r10              = _mm256_andnot_ps(dummy_mask,r10);
 663
 664             /* Compute parameters for interactions between i and j atoms */
 665             qq10             = _mm256_mul_ps(iq1,jq0);
 666
 667             /* EWALD ELECTROSTATICS */
 668
 669             /* Analytical PME correction */
 670             zeta2            = _mm256_mul_ps(beta2,rsq10);
 671             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 672             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 673             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 674             felec            = _mm256_mul_ps(qq10,felec);
 675             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 676             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 677             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 678             velec            = _mm256_mul_ps(qq10,velec);
 679
 680             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 681
 682             /* Update potential sum for this i atom from the interaction with this j atom. */
 683             velec            = _mm256_and_ps(velec,cutoff_mask);
 684             velec            = _mm256_andnot_ps(dummy_mask,velec);
 685             velecsum         = _mm256_add_ps(velecsum,velec);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 690
 691             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm256_mul_ps(fscal,dx10);
 695             ty               = _mm256_mul_ps(fscal,dy10);
 696             tz               = _mm256_mul_ps(fscal,dz10);
 697
 698             /* Update vectorial force */
 699             fix1             = _mm256_add_ps(fix1,tx);
 700             fiy1             = _mm256_add_ps(fiy1,ty);
 701             fiz1             = _mm256_add_ps(fiz1,tz);
 702
 703             fjx0             = _mm256_add_ps(fjx0,tx);
 704             fjy0             = _mm256_add_ps(fjy0,ty);
 705             fjz0             = _mm256_add_ps(fjz0,tz);
 706
 707             }
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 714             {
 715
 716             r20              = _mm256_mul_ps(rsq20,rinv20);
 717             r20              = _mm256_andnot_ps(dummy_mask,r20);
 718
 719             /* Compute parameters for interactions between i and j atoms */
 720             qq20             = _mm256_mul_ps(iq2,jq0);
 721
 722             /* EWALD ELECTROSTATICS */
 723
 724             /* Analytical PME correction */
 725             zeta2            = _mm256_mul_ps(beta2,rsq20);
 726             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 727             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 728             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 729             felec            = _mm256_mul_ps(qq20,felec);
 730             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 731             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 732             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 733             velec            = _mm256_mul_ps(qq20,velec);
 734
 735             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 736
 737             /* Update potential sum for this i atom from the interaction with this j atom. */
 738             velec            = _mm256_and_ps(velec,cutoff_mask);
 739             velec            = _mm256_andnot_ps(dummy_mask,velec);
 740             velecsum         = _mm256_add_ps(velecsum,velec);
 741
 742             fscal            = felec;
 743
 744             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 745
 746             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 747
 748             /* Calculate temporary vectorial force */
 749             tx               = _mm256_mul_ps(fscal,dx20);
 750             ty               = _mm256_mul_ps(fscal,dy20);
 751             tz               = _mm256_mul_ps(fscal,dz20);
 752
 753             /* Update vectorial force */
 754             fix2             = _mm256_add_ps(fix2,tx);
 755             fiy2             = _mm256_add_ps(fiy2,ty);
 756             fiz2             = _mm256_add_ps(fiz2,tz);
 757
 758             fjx0             = _mm256_add_ps(fjx0,tx);
 759             fjy0             = _mm256_add_ps(fjy0,ty);
 760             fjz0             = _mm256_add_ps(fjz0,tz);
 761
 762             }
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 769             {
 770
 771             r30              = _mm256_mul_ps(rsq30,rinv30);
 772             r30              = _mm256_andnot_ps(dummy_mask,r30);
 773
 774             /* Compute parameters for interactions between i and j atoms */
 775             qq30             = _mm256_mul_ps(iq3,jq0);
 776
 777             /* EWALD ELECTROSTATICS */
 778
 779             /* Analytical PME correction */
 780             zeta2            = _mm256_mul_ps(beta2,rsq30);
 781             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 782             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 783             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 784             felec            = _mm256_mul_ps(qq30,felec);
 785             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 786             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 787             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 788             velec            = _mm256_mul_ps(qq30,velec);
 789
 790             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 791
 792             /* Update potential sum for this i atom from the interaction with this j atom. */
 793             velec            = _mm256_and_ps(velec,cutoff_mask);
 794             velec            = _mm256_andnot_ps(dummy_mask,velec);
 795             velecsum         = _mm256_add_ps(velecsum,velec);
 796
 797             fscal            = felec;
 798
 799             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 800
 801             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 802
 803             /* Calculate temporary vectorial force */
 804             tx               = _mm256_mul_ps(fscal,dx30);
 805             ty               = _mm256_mul_ps(fscal,dy30);
 806             tz               = _mm256_mul_ps(fscal,dz30);
 807
 808             /* Update vectorial force */
 809             fix3             = _mm256_add_ps(fix3,tx);
 810             fiy3             = _mm256_add_ps(fiy3,ty);
 811             fiz3             = _mm256_add_ps(fiz3,tz);
 812
 813             fjx0             = _mm256_add_ps(fjx0,tx);
 814             fjy0             = _mm256_add_ps(fjy0,ty);
 815             fjz0             = _mm256_add_ps(fjz0,tz);
 816
 817             }
 818
 819             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 820             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 821             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 822             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 823             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 824             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 825             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 826             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 827
 828             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 829
 830             /* Inner loop uses 374 flops */
 831         }
 832
 833         /* End of innermost loop */
 834
 835         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 836                                                  f+i_coord_offset,fshift+i_shift_offset);
 837
 838         ggid                        = gid[iidx];
 839         /* Update potential energies */
 840         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 841         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 842
 843         /* Increment number of inner iterations */
 844         inneriter                  += j_index_end - j_index_start;
 845
 846         /* Outer loop uses 26 flops */
 847     }
 848
 849     /* Increment number of outer iterations */
 850     outeriter        += nri;
 851
 852     /* Update outer/inner flops */
 853
 854     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*374);
 855 }
 856 /*
 857  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_single
 858  * Electrostatics interaction: Ewald
 859  * VdW interaction:            LennardJones
 860  * Geometry:                   Water4-Particle
 861  * Calculate force/pot:        Force
 862  */
 863 void
 864 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_single
 865                     (t_nblist                    * gmx_restrict       nlist,
 866                      rvec                        * gmx_restrict          xx,
 867                      rvec                        * gmx_restrict          ff,
 868                      t_forcerec                  * gmx_restrict          fr,
 869                      t_mdatoms                   * gmx_restrict     mdatoms,
 870                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 871                      t_nrnb                      * gmx_restrict        nrnb)
 872 {
 873     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 874      * just 0 for non-waters.
 875      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 876      * jnr indices corresponding to data put in the four positions in the SIMD register.
 877      */
 878     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 879     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 880     int              jnrA,jnrB,jnrC,jnrD;
 881     int              jnrE,jnrF,jnrG,jnrH;
 882     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 883     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 884     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 885     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 886     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 887     real             rcutoff_scalar;
 888     real             *shiftvec,*fshift,*x,*f;
 889     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 890     real             scratch[4*DIM];
 891     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 892     real *           vdwioffsetptr0;
 893     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 894     real *           vdwioffsetptr1;
 895     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 896     real *           vdwioffsetptr2;
 897     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 898     real *           vdwioffsetptr3;
 899     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 900     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 901     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 902     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 903     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 904     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 905     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 906     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 907     real             *charge;
 908     int              nvdwtype;
 909     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 910     int              *vdwtype;
 911     real             *vdwparam;
 912     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 913     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 914     __m256i          ewitab;
 915     __m128i          ewitab_lo,ewitab_hi;
 916     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 917     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 918     real             *ewtab;
 919     __m256           dummy_mask,cutoff_mask;
 920     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 921     __m256           one     = _mm256_set1_ps(1.0);
 922     __m256           two     = _mm256_set1_ps(2.0);
 923     x                = xx[0];
 924     f                = ff[0];
 925
 926     nri              = nlist->nri;
 927     iinr             = nlist->iinr;
 928     jindex           = nlist->jindex;
 929     jjnr             = nlist->jjnr;
 930     shiftidx         = nlist->shift;
 931     gid              = nlist->gid;
 932     shiftvec         = fr->shift_vec[0];
 933     fshift           = fr->fshift[0];
 934     facel            = _mm256_set1_ps(fr->epsfac);
 935     charge           = mdatoms->chargeA;
 936     nvdwtype         = fr->ntype;
 937     vdwparam         = fr->nbfp;
 938     vdwtype          = mdatoms->typeA;
 939
 940     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 941     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 942     beta2            = _mm256_mul_ps(beta,beta);
 943     beta3            = _mm256_mul_ps(beta,beta2);
 944
 945     ewtab            = fr->ic->tabq_coul_F;
 946     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 947     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 948
 949     /* Setup water-specific parameters */
 950     inr              = nlist->iinr[0];
 951     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 952     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 953     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 954     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 955
 956     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 957     rcutoff_scalar   = fr->rcoulomb;
 958     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 959     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 960
 961     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 962     rvdw             = _mm256_set1_ps(fr->rvdw);
 963
 964     /* Avoid stupid compiler warnings */
 965     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 966     j_coord_offsetA = 0;
 967     j_coord_offsetB = 0;
 968     j_coord_offsetC = 0;
 969     j_coord_offsetD = 0;
 970     j_coord_offsetE = 0;
 971     j_coord_offsetF = 0;
 972     j_coord_offsetG = 0;
 973     j_coord_offsetH = 0;
 974
 975     outeriter        = 0;
 976     inneriter        = 0;
 977
 978     for(iidx=0;iidx<4*DIM;iidx++)
 979     {
 980         scratch[iidx] = 0.0;
 981     }
 982
 983     /* Start outer loop over neighborlists */
 984     for(iidx=0; iidx<nri; iidx++)
 985     {
 986         /* Load shift vector for this list */
 987         i_shift_offset   = DIM*shiftidx[iidx];
 988
 989         /* Load limits for loop over neighbors */
 990         j_index_start    = jindex[iidx];
 991         j_index_end      = jindex[iidx+1];
 992
 993         /* Get outer coordinate index */
 994         inr              = iinr[iidx];
 995         i_coord_offset   = DIM*inr;
 996
 997         /* Load i particle coords and add shift vector */
 998         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 999                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1000
1001         fix0             = _mm256_setzero_ps();
1002         fiy0             = _mm256_setzero_ps();
1003         fiz0             = _mm256_setzero_ps();
1004         fix1             = _mm256_setzero_ps();
1005         fiy1             = _mm256_setzero_ps();
1006         fiz1             = _mm256_setzero_ps();
1007         fix2             = _mm256_setzero_ps();
1008         fiy2             = _mm256_setzero_ps();
1009         fiz2             = _mm256_setzero_ps();
1010         fix3             = _mm256_setzero_ps();
1011         fiy3             = _mm256_setzero_ps();
1012         fiz3             = _mm256_setzero_ps();
1013
1014         /* Start inner kernel loop */
1015         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1016         {
1017
1018             /* Get j neighbor index, and coordinate index */
1019             jnrA             = jjnr[jidx];
1020             jnrB             = jjnr[jidx+1];
1021             jnrC             = jjnr[jidx+2];
1022             jnrD             = jjnr[jidx+3];
1023             jnrE             = jjnr[jidx+4];
1024             jnrF             = jjnr[jidx+5];
1025             jnrG             = jjnr[jidx+6];
1026             jnrH             = jjnr[jidx+7];
1027             j_coord_offsetA  = DIM*jnrA;
1028             j_coord_offsetB  = DIM*jnrB;
1029             j_coord_offsetC  = DIM*jnrC;
1030             j_coord_offsetD  = DIM*jnrD;
1031             j_coord_offsetE  = DIM*jnrE;
1032             j_coord_offsetF  = DIM*jnrF;
1033             j_coord_offsetG  = DIM*jnrG;
1034             j_coord_offsetH  = DIM*jnrH;
1035
1036             /* load j atom coordinates */
1037             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1038                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1039                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1040                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1041                                                  &jx0,&jy0,&jz0);
1042
1043             /* Calculate displacement vector */
1044             dx00             = _mm256_sub_ps(ix0,jx0);
1045             dy00             = _mm256_sub_ps(iy0,jy0);
1046             dz00             = _mm256_sub_ps(iz0,jz0);
1047             dx10             = _mm256_sub_ps(ix1,jx0);
1048             dy10             = _mm256_sub_ps(iy1,jy0);
1049             dz10             = _mm256_sub_ps(iz1,jz0);
1050             dx20             = _mm256_sub_ps(ix2,jx0);
1051             dy20             = _mm256_sub_ps(iy2,jy0);
1052             dz20             = _mm256_sub_ps(iz2,jz0);
1053             dx30             = _mm256_sub_ps(ix3,jx0);
1054             dy30             = _mm256_sub_ps(iy3,jy0);
1055             dz30             = _mm256_sub_ps(iz3,jz0);
1056
1057             /* Calculate squared distance and things based on it */
1058             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1059             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1060             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1061             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1062
1063             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1064             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1065             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1066
1067             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1068             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1069             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1070             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1071
1072             /* Load parameters for j particles */
1073             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1074                                                                  charge+jnrC+0,charge+jnrD+0,
1075                                                                  charge+jnrE+0,charge+jnrF+0,
1076                                                                  charge+jnrG+0,charge+jnrH+0);
1077             vdwjidx0A        = 2*vdwtype[jnrA+0];
1078             vdwjidx0B        = 2*vdwtype[jnrB+0];
1079             vdwjidx0C        = 2*vdwtype[jnrC+0];
1080             vdwjidx0D        = 2*vdwtype[jnrD+0];
1081             vdwjidx0E        = 2*vdwtype[jnrE+0];
1082             vdwjidx0F        = 2*vdwtype[jnrF+0];
1083             vdwjidx0G        = 2*vdwtype[jnrG+0];
1084             vdwjidx0H        = 2*vdwtype[jnrH+0];
1085
1086             fjx0             = _mm256_setzero_ps();
1087             fjy0             = _mm256_setzero_ps();
1088             fjz0             = _mm256_setzero_ps();
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
1092              **************************/
1093
1094             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1095             {
1096
1097             /* Compute parameters for interactions between i and j atoms */
1098             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1099                                             vdwioffsetptr0+vdwjidx0B,
1100                                             vdwioffsetptr0+vdwjidx0C,
1101                                             vdwioffsetptr0+vdwjidx0D,
1102                                             vdwioffsetptr0+vdwjidx0E,
1103                                             vdwioffsetptr0+vdwjidx0F,
1104                                             vdwioffsetptr0+vdwjidx0G,
1105                                             vdwioffsetptr0+vdwjidx0H,
1106                                             &c6_00,&c12_00);
1107
1108             /* LENNARD-JONES DISPERSION/REPULSION */
1109
1110             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1111             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1112
1113             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1114
1115             fscal            = fvdw;
1116
1117             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm256_mul_ps(fscal,dx00);
1121             ty               = _mm256_mul_ps(fscal,dy00);
1122             tz               = _mm256_mul_ps(fscal,dz00);
1123
1124             /* Update vectorial force */
1125             fix0             = _mm256_add_ps(fix0,tx);
1126             fiy0             = _mm256_add_ps(fiy0,ty);
1127             fiz0             = _mm256_add_ps(fiz0,tz);
1128
1129             fjx0             = _mm256_add_ps(fjx0,tx);
1130             fjy0             = _mm256_add_ps(fjy0,ty);
1131             fjz0             = _mm256_add_ps(fjz0,tz);
1132
1133             }
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1140             {
1141
1142             r10              = _mm256_mul_ps(rsq10,rinv10);
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             qq10             = _mm256_mul_ps(iq1,jq0);
1146
1147             /* EWALD ELECTROSTATICS */
1148
1149             /* Analytical PME correction */
1150             zeta2            = _mm256_mul_ps(beta2,rsq10);
1151             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1152             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1153             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1154             felec            = _mm256_mul_ps(qq10,felec);
1155
1156             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1157
1158             fscal            = felec;
1159
1160             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1161
1162             /* Calculate temporary vectorial force */
1163             tx               = _mm256_mul_ps(fscal,dx10);
1164             ty               = _mm256_mul_ps(fscal,dy10);
1165             tz               = _mm256_mul_ps(fscal,dz10);
1166
1167             /* Update vectorial force */
1168             fix1             = _mm256_add_ps(fix1,tx);
1169             fiy1             = _mm256_add_ps(fiy1,ty);
1170             fiz1             = _mm256_add_ps(fiz1,tz);
1171
1172             fjx0             = _mm256_add_ps(fjx0,tx);
1173             fjy0             = _mm256_add_ps(fjy0,ty);
1174             fjz0             = _mm256_add_ps(fjz0,tz);
1175
1176             }
1177
1178             /**************************
1179              * CALCULATE INTERACTIONS *
1180              **************************/
1181
1182             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1183             {
1184
1185             r20              = _mm256_mul_ps(rsq20,rinv20);
1186
1187             /* Compute parameters for interactions between i and j atoms */
1188             qq20             = _mm256_mul_ps(iq2,jq0);
1189
1190             /* EWALD ELECTROSTATICS */
1191
1192             /* Analytical PME correction */
1193             zeta2            = _mm256_mul_ps(beta2,rsq20);
1194             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1195             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1196             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1197             felec            = _mm256_mul_ps(qq20,felec);
1198
1199             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1200
1201             fscal            = felec;
1202
1203             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1204
1205             /* Calculate temporary vectorial force */
1206             tx               = _mm256_mul_ps(fscal,dx20);
1207             ty               = _mm256_mul_ps(fscal,dy20);
1208             tz               = _mm256_mul_ps(fscal,dz20);
1209
1210             /* Update vectorial force */
1211             fix2             = _mm256_add_ps(fix2,tx);
1212             fiy2             = _mm256_add_ps(fiy2,ty);
1213             fiz2             = _mm256_add_ps(fiz2,tz);
1214
1215             fjx0             = _mm256_add_ps(fjx0,tx);
1216             fjy0             = _mm256_add_ps(fjy0,ty);
1217             fjz0             = _mm256_add_ps(fjz0,tz);
1218
1219             }
1220
1221             /**************************
1222              * CALCULATE INTERACTIONS *
1223              **************************/
1224
1225             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1226             {
1227
1228             r30              = _mm256_mul_ps(rsq30,rinv30);
1229
1230             /* Compute parameters for interactions between i and j atoms */
1231             qq30             = _mm256_mul_ps(iq3,jq0);
1232
1233             /* EWALD ELECTROSTATICS */
1234
1235             /* Analytical PME correction */
1236             zeta2            = _mm256_mul_ps(beta2,rsq30);
1237             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1238             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1239             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1240             felec            = _mm256_mul_ps(qq30,felec);
1241
1242             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1243
1244             fscal            = felec;
1245
1246             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1247
1248             /* Calculate temporary vectorial force */
1249             tx               = _mm256_mul_ps(fscal,dx30);
1250             ty               = _mm256_mul_ps(fscal,dy30);
1251             tz               = _mm256_mul_ps(fscal,dz30);
1252
1253             /* Update vectorial force */
1254             fix3             = _mm256_add_ps(fix3,tx);
1255             fiy3             = _mm256_add_ps(fiy3,ty);
1256             fiz3             = _mm256_add_ps(fiz3,tz);
1257
1258             fjx0             = _mm256_add_ps(fjx0,tx);
1259             fjy0             = _mm256_add_ps(fjy0,ty);
1260             fjz0             = _mm256_add_ps(fjz0,tz);
1261
1262             }
1263
1264             fjptrA             = f+j_coord_offsetA;
1265             fjptrB             = f+j_coord_offsetB;
1266             fjptrC             = f+j_coord_offsetC;
1267             fjptrD             = f+j_coord_offsetD;
1268             fjptrE             = f+j_coord_offsetE;
1269             fjptrF             = f+j_coord_offsetF;
1270             fjptrG             = f+j_coord_offsetG;
1271             fjptrH             = f+j_coord_offsetH;
1272
1273             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1274
1275             /* Inner loop uses 210 flops */
1276         }
1277
1278         if(jidx<j_index_end)
1279         {
1280
1281             /* Get j neighbor index, and coordinate index */
1282             jnrlistA         = jjnr[jidx];
1283             jnrlistB         = jjnr[jidx+1];
1284             jnrlistC         = jjnr[jidx+2];
1285             jnrlistD         = jjnr[jidx+3];
1286             jnrlistE         = jjnr[jidx+4];
1287             jnrlistF         = jjnr[jidx+5];
1288             jnrlistG         = jjnr[jidx+6];
1289             jnrlistH         = jjnr[jidx+7];
1290             /* Sign of each element will be negative for non-real atoms.
1291              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1292              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1293              */
1294             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1295                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1296
1297             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1298             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1299             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1300             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1301             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1302             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1303             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1304             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1305             j_coord_offsetA  = DIM*jnrA;
1306             j_coord_offsetB  = DIM*jnrB;
1307             j_coord_offsetC  = DIM*jnrC;
1308             j_coord_offsetD  = DIM*jnrD;
1309             j_coord_offsetE  = DIM*jnrE;
1310             j_coord_offsetF  = DIM*jnrF;
1311             j_coord_offsetG  = DIM*jnrG;
1312             j_coord_offsetH  = DIM*jnrH;
1313
1314             /* load j atom coordinates */
1315             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1316                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1317                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1318                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1319                                                  &jx0,&jy0,&jz0);
1320
1321             /* Calculate displacement vector */
1322             dx00             = _mm256_sub_ps(ix0,jx0);
1323             dy00             = _mm256_sub_ps(iy0,jy0);
1324             dz00             = _mm256_sub_ps(iz0,jz0);
1325             dx10             = _mm256_sub_ps(ix1,jx0);
1326             dy10             = _mm256_sub_ps(iy1,jy0);
1327             dz10             = _mm256_sub_ps(iz1,jz0);
1328             dx20             = _mm256_sub_ps(ix2,jx0);
1329             dy20             = _mm256_sub_ps(iy2,jy0);
1330             dz20             = _mm256_sub_ps(iz2,jz0);
1331             dx30             = _mm256_sub_ps(ix3,jx0);
1332             dy30             = _mm256_sub_ps(iy3,jy0);
1333             dz30             = _mm256_sub_ps(iz3,jz0);
1334
1335             /* Calculate squared distance and things based on it */
1336             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1337             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1338             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1339             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1340
1341             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1342             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1343             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1344
1345             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1346             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1347             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1348             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1349
1350             /* Load parameters for j particles */
1351             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1352                                                                  charge+jnrC+0,charge+jnrD+0,
1353                                                                  charge+jnrE+0,charge+jnrF+0,
1354                                                                  charge+jnrG+0,charge+jnrH+0);
1355             vdwjidx0A        = 2*vdwtype[jnrA+0];
1356             vdwjidx0B        = 2*vdwtype[jnrB+0];
1357             vdwjidx0C        = 2*vdwtype[jnrC+0];
1358             vdwjidx0D        = 2*vdwtype[jnrD+0];
1359             vdwjidx0E        = 2*vdwtype[jnrE+0];
1360             vdwjidx0F        = 2*vdwtype[jnrF+0];
1361             vdwjidx0G        = 2*vdwtype[jnrG+0];
1362             vdwjidx0H        = 2*vdwtype[jnrH+0];
1363
1364             fjx0             = _mm256_setzero_ps();
1365             fjy0             = _mm256_setzero_ps();
1366             fjz0             = _mm256_setzero_ps();
1367
1368             /**************************
1369              * CALCULATE INTERACTIONS *
1370              **************************/
1371
1372             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1373             {
1374
1375             /* Compute parameters for interactions between i and j atoms */
1376             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1377                                             vdwioffsetptr0+vdwjidx0B,
1378                                             vdwioffsetptr0+vdwjidx0C,
1379                                             vdwioffsetptr0+vdwjidx0D,
1380                                             vdwioffsetptr0+vdwjidx0E,
1381                                             vdwioffsetptr0+vdwjidx0F,
1382                                             vdwioffsetptr0+vdwjidx0G,
1383                                             vdwioffsetptr0+vdwjidx0H,
1384                                             &c6_00,&c12_00);
1385
1386             /* LENNARD-JONES DISPERSION/REPULSION */
1387
1388             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1389             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1390
1391             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1392
1393             fscal            = fvdw;
1394
1395             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1396
1397             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1398
1399             /* Calculate temporary vectorial force */
1400             tx               = _mm256_mul_ps(fscal,dx00);
1401             ty               = _mm256_mul_ps(fscal,dy00);
1402             tz               = _mm256_mul_ps(fscal,dz00);
1403
1404             /* Update vectorial force */
1405             fix0             = _mm256_add_ps(fix0,tx);
1406             fiy0             = _mm256_add_ps(fiy0,ty);
1407             fiz0             = _mm256_add_ps(fiz0,tz);
1408
1409             fjx0             = _mm256_add_ps(fjx0,tx);
1410             fjy0             = _mm256_add_ps(fjy0,ty);
1411             fjz0             = _mm256_add_ps(fjz0,tz);
1412
1413             }
1414
1415             /**************************
1416              * CALCULATE INTERACTIONS *
1417              **************************/
1418
1419             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1420             {
1421
1422             r10              = _mm256_mul_ps(rsq10,rinv10);
1423             r10              = _mm256_andnot_ps(dummy_mask,r10);
1424
1425             /* Compute parameters for interactions between i and j atoms */
1426             qq10             = _mm256_mul_ps(iq1,jq0);
1427
1428             /* EWALD ELECTROSTATICS */
1429
1430             /* Analytical PME correction */
1431             zeta2            = _mm256_mul_ps(beta2,rsq10);
1432             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1433             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1434             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1435             felec            = _mm256_mul_ps(qq10,felec);
1436
1437             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1438
1439             fscal            = felec;
1440
1441             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1442
1443             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1444
1445             /* Calculate temporary vectorial force */
1446             tx               = _mm256_mul_ps(fscal,dx10);
1447             ty               = _mm256_mul_ps(fscal,dy10);
1448             tz               = _mm256_mul_ps(fscal,dz10);
1449
1450             /* Update vectorial force */
1451             fix1             = _mm256_add_ps(fix1,tx);
1452             fiy1             = _mm256_add_ps(fiy1,ty);
1453             fiz1             = _mm256_add_ps(fiz1,tz);
1454
1455             fjx0             = _mm256_add_ps(fjx0,tx);
1456             fjy0             = _mm256_add_ps(fjy0,ty);
1457             fjz0             = _mm256_add_ps(fjz0,tz);
1458
1459             }
1460
1461             /**************************
1462              * CALCULATE INTERACTIONS *
1463              **************************/
1464
1465             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1466             {
1467
1468             r20              = _mm256_mul_ps(rsq20,rinv20);
1469             r20              = _mm256_andnot_ps(dummy_mask,r20);
1470
1471             /* Compute parameters for interactions between i and j atoms */
1472             qq20             = _mm256_mul_ps(iq2,jq0);
1473
1474             /* EWALD ELECTROSTATICS */
1475
1476             /* Analytical PME correction */
1477             zeta2            = _mm256_mul_ps(beta2,rsq20);
1478             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1479             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1480             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1481             felec            = _mm256_mul_ps(qq20,felec);
1482
1483             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1484
1485             fscal            = felec;
1486
1487             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1488
1489             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1490
1491             /* Calculate temporary vectorial force */
1492             tx               = _mm256_mul_ps(fscal,dx20);
1493             ty               = _mm256_mul_ps(fscal,dy20);
1494             tz               = _mm256_mul_ps(fscal,dz20);
1495
1496             /* Update vectorial force */
1497             fix2             = _mm256_add_ps(fix2,tx);
1498             fiy2             = _mm256_add_ps(fiy2,ty);
1499             fiz2             = _mm256_add_ps(fiz2,tz);
1500
1501             fjx0             = _mm256_add_ps(fjx0,tx);
1502             fjy0             = _mm256_add_ps(fjy0,ty);
1503             fjz0             = _mm256_add_ps(fjz0,tz);
1504
1505             }
1506
1507             /**************************
1508              * CALCULATE INTERACTIONS *
1509              **************************/
1510
1511             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1512             {
1513
1514             r30              = _mm256_mul_ps(rsq30,rinv30);
1515             r30              = _mm256_andnot_ps(dummy_mask,r30);
1516
1517             /* Compute parameters for interactions between i and j atoms */
1518             qq30             = _mm256_mul_ps(iq3,jq0);
1519
1520             /* EWALD ELECTROSTATICS */
1521
1522             /* Analytical PME correction */
1523             zeta2            = _mm256_mul_ps(beta2,rsq30);
1524             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1525             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1526             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1527             felec            = _mm256_mul_ps(qq30,felec);
1528
1529             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1530
1531             fscal            = felec;
1532
1533             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1534
1535             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1536
1537             /* Calculate temporary vectorial force */
1538             tx               = _mm256_mul_ps(fscal,dx30);
1539             ty               = _mm256_mul_ps(fscal,dy30);
1540             tz               = _mm256_mul_ps(fscal,dz30);
1541
1542             /* Update vectorial force */
1543             fix3             = _mm256_add_ps(fix3,tx);
1544             fiy3             = _mm256_add_ps(fiy3,ty);
1545             fiz3             = _mm256_add_ps(fiz3,tz);
1546
1547             fjx0             = _mm256_add_ps(fjx0,tx);
1548             fjy0             = _mm256_add_ps(fjy0,ty);
1549             fjz0             = _mm256_add_ps(fjz0,tz);
1550
1551             }
1552
1553             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1554             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1555             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1556             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1557             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1558             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1559             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1560             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1561
1562             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1563
1564             /* Inner loop uses 213 flops */
1565         }
1566
1567         /* End of innermost loop */
1568
1569         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1570                                                  f+i_coord_offset,fshift+i_shift_offset);
1571
1572         /* Increment number of inner iterations */
1573         inneriter                  += j_index_end - j_index_start;
1574
1575         /* Outer loop uses 24 flops */
1576     }
1577
1578     /* Increment number of outer iterations */
1579     outeriter        += nri;
1580
1581     /* Update outer/inner flops */
1582
1583     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*213);
1584 }