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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256i          ewitab;
 111     __m128i          ewitab_lo,ewitab_hi;
 112     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 114     real             *ewtab;
 115     __m256           dummy_mask,cutoff_mask;
 116     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 117     __m256           one     = _mm256_set1_ps(1.0);
 118     __m256           two     = _mm256_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm256_set1_ps(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 137     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 138     beta2            = _mm256_mul_ps(beta,beta);
 139     beta3            = _mm256_mul_ps(beta,beta2);
 140
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 148     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 149     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 150     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 151
 152     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 153     rcutoff_scalar   = fr->rcoulomb;
 154     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 155     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 156
 157     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 158     rvdw             = _mm256_set1_ps(fr->rvdw);
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164     j_coord_offsetC = 0;
 165     j_coord_offsetD = 0;
 166     j_coord_offsetE = 0;
 167     j_coord_offsetF = 0;
 168     j_coord_offsetG = 0;
 169     j_coord_offsetH = 0;
 170
 171     outeriter        = 0;
 172     inneriter        = 0;
 173
 174     for(iidx=0;iidx<4*DIM;iidx++)
 175     {
 176         scratch[iidx] = 0.0;
 177     }
 178
 179     /* Start outer loop over neighborlists */
 180     for(iidx=0; iidx<nri; iidx++)
 181     {
 182         /* Load shift vector for this list */
 183         i_shift_offset   = DIM*shiftidx[iidx];
 184
 185         /* Load limits for loop over neighbors */
 186         j_index_start    = jindex[iidx];
 187         j_index_end      = jindex[iidx+1];
 188
 189         /* Get outer coordinate index */
 190         inr              = iinr[iidx];
 191         i_coord_offset   = DIM*inr;
 192
 193         /* Load i particle coords and add shift vector */
 194         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 195                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 196
 197         fix0             = _mm256_setzero_ps();
 198         fiy0             = _mm256_setzero_ps();
 199         fiz0             = _mm256_setzero_ps();
 200         fix1             = _mm256_setzero_ps();
 201         fiy1             = _mm256_setzero_ps();
 202         fiz1             = _mm256_setzero_ps();
 203         fix2             = _mm256_setzero_ps();
 204         fiy2             = _mm256_setzero_ps();
 205         fiz2             = _mm256_setzero_ps();
 206         fix3             = _mm256_setzero_ps();
 207         fiy3             = _mm256_setzero_ps();
 208         fiz3             = _mm256_setzero_ps();
 209
 210         /* Reset potential sums */
 211         velecsum         = _mm256_setzero_ps();
 212         vvdwsum          = _mm256_setzero_ps();
 213
 214         /* Start inner kernel loop */
 215         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 216         {
 217
 218             /* Get j neighbor index, and coordinate index */
 219             jnrA             = jjnr[jidx];
 220             jnrB             = jjnr[jidx+1];
 221             jnrC             = jjnr[jidx+2];
 222             jnrD             = jjnr[jidx+3];
 223             jnrE             = jjnr[jidx+4];
 224             jnrF             = jjnr[jidx+5];
 225             jnrG             = jjnr[jidx+6];
 226             jnrH             = jjnr[jidx+7];
 227             j_coord_offsetA  = DIM*jnrA;
 228             j_coord_offsetB  = DIM*jnrB;
 229             j_coord_offsetC  = DIM*jnrC;
 230             j_coord_offsetD  = DIM*jnrD;
 231             j_coord_offsetE  = DIM*jnrE;
 232             j_coord_offsetF  = DIM*jnrF;
 233             j_coord_offsetG  = DIM*jnrG;
 234             j_coord_offsetH  = DIM*jnrH;
 235
 236             /* load j atom coordinates */
 237             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 238                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 239                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 240                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 241                                                  &jx0,&jy0,&jz0);
 242
 243             /* Calculate displacement vector */
 244             dx00             = _mm256_sub_ps(ix0,jx0);
 245             dy00             = _mm256_sub_ps(iy0,jy0);
 246             dz00             = _mm256_sub_ps(iz0,jz0);
 247             dx10             = _mm256_sub_ps(ix1,jx0);
 248             dy10             = _mm256_sub_ps(iy1,jy0);
 249             dz10             = _mm256_sub_ps(iz1,jz0);
 250             dx20             = _mm256_sub_ps(ix2,jx0);
 251             dy20             = _mm256_sub_ps(iy2,jy0);
 252             dz20             = _mm256_sub_ps(iz2,jz0);
 253             dx30             = _mm256_sub_ps(ix3,jx0);
 254             dy30             = _mm256_sub_ps(iy3,jy0);
 255             dz30             = _mm256_sub_ps(iz3,jz0);
 256
 257             /* Calculate squared distance and things based on it */
 258             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 259             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 260             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 261             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 262
 263             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 264             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 265             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 266
 267             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 268             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 269             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 270             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 271
 272             /* Load parameters for j particles */
 273             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 274                                                                  charge+jnrC+0,charge+jnrD+0,
 275                                                                  charge+jnrE+0,charge+jnrF+0,
 276                                                                  charge+jnrG+0,charge+jnrH+0);
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278             vdwjidx0B        = 2*vdwtype[jnrB+0];
 279             vdwjidx0C        = 2*vdwtype[jnrC+0];
 280             vdwjidx0D        = 2*vdwtype[jnrD+0];
 281             vdwjidx0E        = 2*vdwtype[jnrE+0];
 282             vdwjidx0F        = 2*vdwtype[jnrF+0];
 283             vdwjidx0G        = 2*vdwtype[jnrG+0];
 284             vdwjidx0H        = 2*vdwtype[jnrH+0];
 285
 286             fjx0             = _mm256_setzero_ps();
 287             fjy0             = _mm256_setzero_ps();
 288             fjz0             = _mm256_setzero_ps();
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 295             {
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 299                                             vdwioffsetptr0+vdwjidx0B,
 300                                             vdwioffsetptr0+vdwjidx0C,
 301                                             vdwioffsetptr0+vdwjidx0D,
 302                                             vdwioffsetptr0+vdwjidx0E,
 303                                             vdwioffsetptr0+vdwjidx0F,
 304                                             vdwioffsetptr0+vdwjidx0G,
 305                                             vdwioffsetptr0+vdwjidx0H,
 306                                             &c6_00,&c12_00);
 307
 308             /* LENNARD-JONES DISPERSION/REPULSION */
 309
 310             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 311             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 312             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 313             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) ,
 314                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 315             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 316
 317             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 321             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 322
 323             fscal            = fvdw;
 324
 325             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm256_mul_ps(fscal,dx00);
 329             ty               = _mm256_mul_ps(fscal,dy00);
 330             tz               = _mm256_mul_ps(fscal,dz00);
 331
 332             /* Update vectorial force */
 333             fix0             = _mm256_add_ps(fix0,tx);
 334             fiy0             = _mm256_add_ps(fiy0,ty);
 335             fiz0             = _mm256_add_ps(fiz0,tz);
 336
 337             fjx0             = _mm256_add_ps(fjx0,tx);
 338             fjy0             = _mm256_add_ps(fjy0,ty);
 339             fjz0             = _mm256_add_ps(fjz0,tz);
 340
 341             }
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 348             {
 349
 350             r10              = _mm256_mul_ps(rsq10,rinv10);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq10             = _mm256_mul_ps(iq1,jq0);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Analytical PME correction */
 358             zeta2            = _mm256_mul_ps(beta2,rsq10);
 359             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 360             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 361             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 362             felec            = _mm256_mul_ps(qq10,felec);
 363             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 364             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 365             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 366             velec            = _mm256_mul_ps(qq10,velec);
 367
 368             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm256_and_ps(velec,cutoff_mask);
 372             velecsum         = _mm256_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm256_mul_ps(fscal,dx10);
 380             ty               = _mm256_mul_ps(fscal,dy10);
 381             tz               = _mm256_mul_ps(fscal,dz10);
 382
 383             /* Update vectorial force */
 384             fix1             = _mm256_add_ps(fix1,tx);
 385             fiy1             = _mm256_add_ps(fiy1,ty);
 386             fiz1             = _mm256_add_ps(fiz1,tz);
 387
 388             fjx0             = _mm256_add_ps(fjx0,tx);
 389             fjy0             = _mm256_add_ps(fjy0,ty);
 390             fjz0             = _mm256_add_ps(fjz0,tz);
 391
 392             }
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 399             {
 400
 401             r20              = _mm256_mul_ps(rsq20,rinv20);
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq20             = _mm256_mul_ps(iq2,jq0);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Analytical PME correction */
 409             zeta2            = _mm256_mul_ps(beta2,rsq20);
 410             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 411             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 412             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 413             felec            = _mm256_mul_ps(qq20,felec);
 414             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 415             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 416             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 417             velec            = _mm256_mul_ps(qq20,velec);
 418
 419             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 420
 421             /* Update potential sum for this i atom from the interaction with this j atom. */
 422             velec            = _mm256_and_ps(velec,cutoff_mask);
 423             velecsum         = _mm256_add_ps(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm256_mul_ps(fscal,dx20);
 431             ty               = _mm256_mul_ps(fscal,dy20);
 432             tz               = _mm256_mul_ps(fscal,dz20);
 433
 434             /* Update vectorial force */
 435             fix2             = _mm256_add_ps(fix2,tx);
 436             fiy2             = _mm256_add_ps(fiy2,ty);
 437             fiz2             = _mm256_add_ps(fiz2,tz);
 438
 439             fjx0             = _mm256_add_ps(fjx0,tx);
 440             fjy0             = _mm256_add_ps(fjy0,ty);
 441             fjz0             = _mm256_add_ps(fjz0,tz);
 442
 443             }
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 450             {
 451
 452             r30              = _mm256_mul_ps(rsq30,rinv30);
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq30             = _mm256_mul_ps(iq3,jq0);
 456
 457             /* EWALD ELECTROSTATICS */
 458
 459             /* Analytical PME correction */
 460             zeta2            = _mm256_mul_ps(beta2,rsq30);
 461             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 462             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 463             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 464             felec            = _mm256_mul_ps(qq30,felec);
 465             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 466             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 467             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 468             velec            = _mm256_mul_ps(qq30,velec);
 469
 470             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velec            = _mm256_and_ps(velec,cutoff_mask);
 474             velecsum         = _mm256_add_ps(velecsum,velec);
 475
 476             fscal            = felec;
 477
 478             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm256_mul_ps(fscal,dx30);
 482             ty               = _mm256_mul_ps(fscal,dy30);
 483             tz               = _mm256_mul_ps(fscal,dz30);
 484
 485             /* Update vectorial force */
 486             fix3             = _mm256_add_ps(fix3,tx);
 487             fiy3             = _mm256_add_ps(fiy3,ty);
 488             fiz3             = _mm256_add_ps(fiz3,tz);
 489
 490             fjx0             = _mm256_add_ps(fjx0,tx);
 491             fjy0             = _mm256_add_ps(fjy0,ty);
 492             fjz0             = _mm256_add_ps(fjz0,tz);
 493
 494             }
 495
 496             fjptrA             = f+j_coord_offsetA;
 497             fjptrB             = f+j_coord_offsetB;
 498             fjptrC             = f+j_coord_offsetC;
 499             fjptrD             = f+j_coord_offsetD;
 500             fjptrE             = f+j_coord_offsetE;
 501             fjptrF             = f+j_coord_offsetF;
 502             fjptrG             = f+j_coord_offsetG;
 503             fjptrH             = f+j_coord_offsetH;
 504
 505             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 506
 507             /* Inner loop uses 371 flops */
 508         }
 509
 510         if(jidx<j_index_end)
 511         {
 512
 513             /* Get j neighbor index, and coordinate index */
 514             jnrlistA         = jjnr[jidx];
 515             jnrlistB         = jjnr[jidx+1];
 516             jnrlistC         = jjnr[jidx+2];
 517             jnrlistD         = jjnr[jidx+3];
 518             jnrlistE         = jjnr[jidx+4];
 519             jnrlistF         = jjnr[jidx+5];
 520             jnrlistG         = jjnr[jidx+6];
 521             jnrlistH         = jjnr[jidx+7];
 522             /* Sign of each element will be negative for non-real atoms.
 523              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 524              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 525              */
 526             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 527                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 528
 529             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 530             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 531             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 532             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 533             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 534             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 535             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 536             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 537             j_coord_offsetA  = DIM*jnrA;
 538             j_coord_offsetB  = DIM*jnrB;
 539             j_coord_offsetC  = DIM*jnrC;
 540             j_coord_offsetD  = DIM*jnrD;
 541             j_coord_offsetE  = DIM*jnrE;
 542             j_coord_offsetF  = DIM*jnrF;
 543             j_coord_offsetG  = DIM*jnrG;
 544             j_coord_offsetH  = DIM*jnrH;
 545
 546             /* load j atom coordinates */
 547             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 548                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 549                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 550                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 551                                                  &jx0,&jy0,&jz0);
 552
 553             /* Calculate displacement vector */
 554             dx00             = _mm256_sub_ps(ix0,jx0);
 555             dy00             = _mm256_sub_ps(iy0,jy0);
 556             dz00             = _mm256_sub_ps(iz0,jz0);
 557             dx10             = _mm256_sub_ps(ix1,jx0);
 558             dy10             = _mm256_sub_ps(iy1,jy0);
 559             dz10             = _mm256_sub_ps(iz1,jz0);
 560             dx20             = _mm256_sub_ps(ix2,jx0);
 561             dy20             = _mm256_sub_ps(iy2,jy0);
 562             dz20             = _mm256_sub_ps(iz2,jz0);
 563             dx30             = _mm256_sub_ps(ix3,jx0);
 564             dy30             = _mm256_sub_ps(iy3,jy0);
 565             dz30             = _mm256_sub_ps(iz3,jz0);
 566
 567             /* Calculate squared distance and things based on it */
 568             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 569             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 570             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 571             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 572
 573             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 574             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 575             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 576
 577             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 578             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 579             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 580             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 581
 582             /* Load parameters for j particles */
 583             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 584                                                                  charge+jnrC+0,charge+jnrD+0,
 585                                                                  charge+jnrE+0,charge+jnrF+0,
 586                                                                  charge+jnrG+0,charge+jnrH+0);
 587             vdwjidx0A        = 2*vdwtype[jnrA+0];
 588             vdwjidx0B        = 2*vdwtype[jnrB+0];
 589             vdwjidx0C        = 2*vdwtype[jnrC+0];
 590             vdwjidx0D        = 2*vdwtype[jnrD+0];
 591             vdwjidx0E        = 2*vdwtype[jnrE+0];
 592             vdwjidx0F        = 2*vdwtype[jnrF+0];
 593             vdwjidx0G        = 2*vdwtype[jnrG+0];
 594             vdwjidx0H        = 2*vdwtype[jnrH+0];
 595
 596             fjx0             = _mm256_setzero_ps();
 597             fjy0             = _mm256_setzero_ps();
 598             fjz0             = _mm256_setzero_ps();
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 605             {
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 609                                             vdwioffsetptr0+vdwjidx0B,
 610                                             vdwioffsetptr0+vdwjidx0C,
 611                                             vdwioffsetptr0+vdwjidx0D,
 612                                             vdwioffsetptr0+vdwjidx0E,
 613                                             vdwioffsetptr0+vdwjidx0F,
 614                                             vdwioffsetptr0+vdwjidx0G,
 615                                             vdwioffsetptr0+vdwjidx0H,
 616                                             &c6_00,&c12_00);
 617
 618             /* LENNARD-JONES DISPERSION/REPULSION */
 619
 620             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 621             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 622             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 623             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) ,
 624                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 625             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 626
 627             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 631             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 632             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 633
 634             fscal            = fvdw;
 635
 636             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 637
 638             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 639
 640             /* Calculate temporary vectorial force */
 641             tx               = _mm256_mul_ps(fscal,dx00);
 642             ty               = _mm256_mul_ps(fscal,dy00);
 643             tz               = _mm256_mul_ps(fscal,dz00);
 644
 645             /* Update vectorial force */
 646             fix0             = _mm256_add_ps(fix0,tx);
 647             fiy0             = _mm256_add_ps(fiy0,ty);
 648             fiz0             = _mm256_add_ps(fiz0,tz);
 649
 650             fjx0             = _mm256_add_ps(fjx0,tx);
 651             fjy0             = _mm256_add_ps(fjy0,ty);
 652             fjz0             = _mm256_add_ps(fjz0,tz);
 653
 654             }
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 661             {
 662
 663             r10              = _mm256_mul_ps(rsq10,rinv10);
 664             r10              = _mm256_andnot_ps(dummy_mask,r10);
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq10             = _mm256_mul_ps(iq1,jq0);
 668
 669             /* EWALD ELECTROSTATICS */
 670
 671             /* Analytical PME correction */
 672             zeta2            = _mm256_mul_ps(beta2,rsq10);
 673             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 674             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 675             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 676             felec            = _mm256_mul_ps(qq10,felec);
 677             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 678             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 679             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 680             velec            = _mm256_mul_ps(qq10,velec);
 681
 682             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 683
 684             /* Update potential sum for this i atom from the interaction with this j atom. */
 685             velec            = _mm256_and_ps(velec,cutoff_mask);
 686             velec            = _mm256_andnot_ps(dummy_mask,velec);
 687             velecsum         = _mm256_add_ps(velecsum,velec);
 688
 689             fscal            = felec;
 690
 691             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 692
 693             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 694
 695             /* Calculate temporary vectorial force */
 696             tx               = _mm256_mul_ps(fscal,dx10);
 697             ty               = _mm256_mul_ps(fscal,dy10);
 698             tz               = _mm256_mul_ps(fscal,dz10);
 699
 700             /* Update vectorial force */
 701             fix1             = _mm256_add_ps(fix1,tx);
 702             fiy1             = _mm256_add_ps(fiy1,ty);
 703             fiz1             = _mm256_add_ps(fiz1,tz);
 704
 705             fjx0             = _mm256_add_ps(fjx0,tx);
 706             fjy0             = _mm256_add_ps(fjy0,ty);
 707             fjz0             = _mm256_add_ps(fjz0,tz);
 708
 709             }
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 716             {
 717
 718             r20              = _mm256_mul_ps(rsq20,rinv20);
 719             r20              = _mm256_andnot_ps(dummy_mask,r20);
 720
 721             /* Compute parameters for interactions between i and j atoms */
 722             qq20             = _mm256_mul_ps(iq2,jq0);
 723
 724             /* EWALD ELECTROSTATICS */
 725
 726             /* Analytical PME correction */
 727             zeta2            = _mm256_mul_ps(beta2,rsq20);
 728             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 729             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 730             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 731             felec            = _mm256_mul_ps(qq20,felec);
 732             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 733             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 734             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 735             velec            = _mm256_mul_ps(qq20,velec);
 736
 737             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 738
 739             /* Update potential sum for this i atom from the interaction with this j atom. */
 740             velec            = _mm256_and_ps(velec,cutoff_mask);
 741             velec            = _mm256_andnot_ps(dummy_mask,velec);
 742             velecsum         = _mm256_add_ps(velecsum,velec);
 743
 744             fscal            = felec;
 745
 746             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 747
 748             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm256_mul_ps(fscal,dx20);
 752             ty               = _mm256_mul_ps(fscal,dy20);
 753             tz               = _mm256_mul_ps(fscal,dz20);
 754
 755             /* Update vectorial force */
 756             fix2             = _mm256_add_ps(fix2,tx);
 757             fiy2             = _mm256_add_ps(fiy2,ty);
 758             fiz2             = _mm256_add_ps(fiz2,tz);
 759
 760             fjx0             = _mm256_add_ps(fjx0,tx);
 761             fjy0             = _mm256_add_ps(fjy0,ty);
 762             fjz0             = _mm256_add_ps(fjz0,tz);
 763
 764             }
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 771             {
 772
 773             r30              = _mm256_mul_ps(rsq30,rinv30);
 774             r30              = _mm256_andnot_ps(dummy_mask,r30);
 775
 776             /* Compute parameters for interactions between i and j atoms */
 777             qq30             = _mm256_mul_ps(iq3,jq0);
 778
 779             /* EWALD ELECTROSTATICS */
 780
 781             /* Analytical PME correction */
 782             zeta2            = _mm256_mul_ps(beta2,rsq30);
 783             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 784             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 785             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 786             felec            = _mm256_mul_ps(qq30,felec);
 787             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 788             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 789             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 790             velec            = _mm256_mul_ps(qq30,velec);
 791
 792             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 793
 794             /* Update potential sum for this i atom from the interaction with this j atom. */
 795             velec            = _mm256_and_ps(velec,cutoff_mask);
 796             velec            = _mm256_andnot_ps(dummy_mask,velec);
 797             velecsum         = _mm256_add_ps(velecsum,velec);
 798
 799             fscal            = felec;
 800
 801             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 802
 803             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 804
 805             /* Calculate temporary vectorial force */
 806             tx               = _mm256_mul_ps(fscal,dx30);
 807             ty               = _mm256_mul_ps(fscal,dy30);
 808             tz               = _mm256_mul_ps(fscal,dz30);
 809
 810             /* Update vectorial force */
 811             fix3             = _mm256_add_ps(fix3,tx);
 812             fiy3             = _mm256_add_ps(fiy3,ty);
 813             fiz3             = _mm256_add_ps(fiz3,tz);
 814
 815             fjx0             = _mm256_add_ps(fjx0,tx);
 816             fjy0             = _mm256_add_ps(fjy0,ty);
 817             fjz0             = _mm256_add_ps(fjz0,tz);
 818
 819             }
 820
 821             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 822             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 823             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 824             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 825             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 826             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 827             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 828             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 829
 830             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 831
 832             /* Inner loop uses 374 flops */
 833         }
 834
 835         /* End of innermost loop */
 836
 837         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 838                                                  f+i_coord_offset,fshift+i_shift_offset);
 839
 840         ggid                        = gid[iidx];
 841         /* Update potential energies */
 842         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 843         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 844
 845         /* Increment number of inner iterations */
 846         inneriter                  += j_index_end - j_index_start;
 847
 848         /* Outer loop uses 26 flops */
 849     }
 850
 851     /* Increment number of outer iterations */
 852     outeriter        += nri;
 853
 854     /* Update outer/inner flops */
 855
 856     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*374);
 857 }
 858 /*
 859  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_single
 860  * Electrostatics interaction: Ewald
 861  * VdW interaction:            LennardJones
 862  * Geometry:                   Water4-Particle
 863  * Calculate force/pot:        Force
 864  */
 865 void
 866 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_single
 867                     (t_nblist                    * gmx_restrict       nlist,
 868                      rvec                        * gmx_restrict          xx,
 869                      rvec                        * gmx_restrict          ff,
 870                      t_forcerec                  * gmx_restrict          fr,
 871                      t_mdatoms                   * gmx_restrict     mdatoms,
 872                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 873                      t_nrnb                      * gmx_restrict        nrnb)
 874 {
 875     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 876      * just 0 for non-waters.
 877      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 878      * jnr indices corresponding to data put in the four positions in the SIMD register.
 879      */
 880     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 881     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 882     int              jnrA,jnrB,jnrC,jnrD;
 883     int              jnrE,jnrF,jnrG,jnrH;
 884     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 885     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 886     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 887     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 888     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 889     real             rcutoff_scalar;
 890     real             *shiftvec,*fshift,*x,*f;
 891     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 892     real             scratch[4*DIM];
 893     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 894     real *           vdwioffsetptr0;
 895     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 896     real *           vdwioffsetptr1;
 897     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 898     real *           vdwioffsetptr2;
 899     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 900     real *           vdwioffsetptr3;
 901     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 902     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 903     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 904     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 905     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 906     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 907     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 908     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 909     real             *charge;
 910     int              nvdwtype;
 911     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 912     int              *vdwtype;
 913     real             *vdwparam;
 914     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 915     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 916     __m256i          ewitab;
 917     __m128i          ewitab_lo,ewitab_hi;
 918     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 919     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 920     real             *ewtab;
 921     __m256           dummy_mask,cutoff_mask;
 922     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 923     __m256           one     = _mm256_set1_ps(1.0);
 924     __m256           two     = _mm256_set1_ps(2.0);
 925     x                = xx[0];
 926     f                = ff[0];
 927
 928     nri              = nlist->nri;
 929     iinr             = nlist->iinr;
 930     jindex           = nlist->jindex;
 931     jjnr             = nlist->jjnr;
 932     shiftidx         = nlist->shift;
 933     gid              = nlist->gid;
 934     shiftvec         = fr->shift_vec[0];
 935     fshift           = fr->fshift[0];
 936     facel            = _mm256_set1_ps(fr->epsfac);
 937     charge           = mdatoms->chargeA;
 938     nvdwtype         = fr->ntype;
 939     vdwparam         = fr->nbfp;
 940     vdwtype          = mdatoms->typeA;
 941
 942     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 943     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 944     beta2            = _mm256_mul_ps(beta,beta);
 945     beta3            = _mm256_mul_ps(beta,beta2);
 946
 947     ewtab            = fr->ic->tabq_coul_F;
 948     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 949     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 950
 951     /* Setup water-specific parameters */
 952     inr              = nlist->iinr[0];
 953     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 954     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 955     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 956     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 957
 958     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 959     rcutoff_scalar   = fr->rcoulomb;
 960     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 961     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 962
 963     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 964     rvdw             = _mm256_set1_ps(fr->rvdw);
 965
 966     /* Avoid stupid compiler warnings */
 967     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 968     j_coord_offsetA = 0;
 969     j_coord_offsetB = 0;
 970     j_coord_offsetC = 0;
 971     j_coord_offsetD = 0;
 972     j_coord_offsetE = 0;
 973     j_coord_offsetF = 0;
 974     j_coord_offsetG = 0;
 975     j_coord_offsetH = 0;
 976
 977     outeriter        = 0;
 978     inneriter        = 0;
 979
 980     for(iidx=0;iidx<4*DIM;iidx++)
 981     {
 982         scratch[iidx] = 0.0;
 983     }
 984
 985     /* Start outer loop over neighborlists */
 986     for(iidx=0; iidx<nri; iidx++)
 987     {
 988         /* Load shift vector for this list */
 989         i_shift_offset   = DIM*shiftidx[iidx];
 990
 991         /* Load limits for loop over neighbors */
 992         j_index_start    = jindex[iidx];
 993         j_index_end      = jindex[iidx+1];
 994
 995         /* Get outer coordinate index */
 996         inr              = iinr[iidx];
 997         i_coord_offset   = DIM*inr;
 998
 999         /* Load i particle coords and add shift vector */
1000         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1001                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1002
1003         fix0             = _mm256_setzero_ps();
1004         fiy0             = _mm256_setzero_ps();
1005         fiz0             = _mm256_setzero_ps();
1006         fix1             = _mm256_setzero_ps();
1007         fiy1             = _mm256_setzero_ps();
1008         fiz1             = _mm256_setzero_ps();
1009         fix2             = _mm256_setzero_ps();
1010         fiy2             = _mm256_setzero_ps();
1011         fiz2             = _mm256_setzero_ps();
1012         fix3             = _mm256_setzero_ps();
1013         fiy3             = _mm256_setzero_ps();
1014         fiz3             = _mm256_setzero_ps();
1015
1016         /* Start inner kernel loop */
1017         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1018         {
1019
1020             /* Get j neighbor index, and coordinate index */
1021             jnrA             = jjnr[jidx];
1022             jnrB             = jjnr[jidx+1];
1023             jnrC             = jjnr[jidx+2];
1024             jnrD             = jjnr[jidx+3];
1025             jnrE             = jjnr[jidx+4];
1026             jnrF             = jjnr[jidx+5];
1027             jnrG             = jjnr[jidx+6];
1028             jnrH             = jjnr[jidx+7];
1029             j_coord_offsetA  = DIM*jnrA;
1030             j_coord_offsetB  = DIM*jnrB;
1031             j_coord_offsetC  = DIM*jnrC;
1032             j_coord_offsetD  = DIM*jnrD;
1033             j_coord_offsetE  = DIM*jnrE;
1034             j_coord_offsetF  = DIM*jnrF;
1035             j_coord_offsetG  = DIM*jnrG;
1036             j_coord_offsetH  = DIM*jnrH;
1037
1038             /* load j atom coordinates */
1039             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1040                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1041                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1042                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1043                                                  &jx0,&jy0,&jz0);
1044
1045             /* Calculate displacement vector */
1046             dx00             = _mm256_sub_ps(ix0,jx0);
1047             dy00             = _mm256_sub_ps(iy0,jy0);
1048             dz00             = _mm256_sub_ps(iz0,jz0);
1049             dx10             = _mm256_sub_ps(ix1,jx0);
1050             dy10             = _mm256_sub_ps(iy1,jy0);
1051             dz10             = _mm256_sub_ps(iz1,jz0);
1052             dx20             = _mm256_sub_ps(ix2,jx0);
1053             dy20             = _mm256_sub_ps(iy2,jy0);
1054             dz20             = _mm256_sub_ps(iz2,jz0);
1055             dx30             = _mm256_sub_ps(ix3,jx0);
1056             dy30             = _mm256_sub_ps(iy3,jy0);
1057             dz30             = _mm256_sub_ps(iz3,jz0);
1058
1059             /* Calculate squared distance and things based on it */
1060             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1061             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1062             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1063             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1064
1065             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1066             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1067             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1068
1069             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1070             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1071             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1072             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1073
1074             /* Load parameters for j particles */
1075             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1076                                                                  charge+jnrC+0,charge+jnrD+0,
1077                                                                  charge+jnrE+0,charge+jnrF+0,
1078                                                                  charge+jnrG+0,charge+jnrH+0);
1079             vdwjidx0A        = 2*vdwtype[jnrA+0];
1080             vdwjidx0B        = 2*vdwtype[jnrB+0];
1081             vdwjidx0C        = 2*vdwtype[jnrC+0];
1082             vdwjidx0D        = 2*vdwtype[jnrD+0];
1083             vdwjidx0E        = 2*vdwtype[jnrE+0];
1084             vdwjidx0F        = 2*vdwtype[jnrF+0];
1085             vdwjidx0G        = 2*vdwtype[jnrG+0];
1086             vdwjidx0H        = 2*vdwtype[jnrH+0];
1087
1088             fjx0             = _mm256_setzero_ps();
1089             fjy0             = _mm256_setzero_ps();
1090             fjz0             = _mm256_setzero_ps();
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1097             {
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1101                                             vdwioffsetptr0+vdwjidx0B,
1102                                             vdwioffsetptr0+vdwjidx0C,
1103                                             vdwioffsetptr0+vdwjidx0D,
1104                                             vdwioffsetptr0+vdwjidx0E,
1105                                             vdwioffsetptr0+vdwjidx0F,
1106                                             vdwioffsetptr0+vdwjidx0G,
1107                                             vdwioffsetptr0+vdwjidx0H,
1108                                             &c6_00,&c12_00);
1109
1110             /* LENNARD-JONES DISPERSION/REPULSION */
1111
1112             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1113             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1114
1115             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1116
1117             fscal            = fvdw;
1118
1119             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm256_mul_ps(fscal,dx00);
1123             ty               = _mm256_mul_ps(fscal,dy00);
1124             tz               = _mm256_mul_ps(fscal,dz00);
1125
1126             /* Update vectorial force */
1127             fix0             = _mm256_add_ps(fix0,tx);
1128             fiy0             = _mm256_add_ps(fiy0,ty);
1129             fiz0             = _mm256_add_ps(fiz0,tz);
1130
1131             fjx0             = _mm256_add_ps(fjx0,tx);
1132             fjy0             = _mm256_add_ps(fjy0,ty);
1133             fjz0             = _mm256_add_ps(fjz0,tz);
1134
1135             }
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1142             {
1143
1144             r10              = _mm256_mul_ps(rsq10,rinv10);
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             qq10             = _mm256_mul_ps(iq1,jq0);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Analytical PME correction */
1152             zeta2            = _mm256_mul_ps(beta2,rsq10);
1153             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1154             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1155             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1156             felec            = _mm256_mul_ps(qq10,felec);
1157
1158             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1159
1160             fscal            = felec;
1161
1162             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1163
1164             /* Calculate temporary vectorial force */
1165             tx               = _mm256_mul_ps(fscal,dx10);
1166             ty               = _mm256_mul_ps(fscal,dy10);
1167             tz               = _mm256_mul_ps(fscal,dz10);
1168
1169             /* Update vectorial force */
1170             fix1             = _mm256_add_ps(fix1,tx);
1171             fiy1             = _mm256_add_ps(fiy1,ty);
1172             fiz1             = _mm256_add_ps(fiz1,tz);
1173
1174             fjx0             = _mm256_add_ps(fjx0,tx);
1175             fjy0             = _mm256_add_ps(fjy0,ty);
1176             fjz0             = _mm256_add_ps(fjz0,tz);
1177
1178             }
1179
1180             /**************************
1181              * CALCULATE INTERACTIONS *
1182              **************************/
1183
1184             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1185             {
1186
1187             r20              = _mm256_mul_ps(rsq20,rinv20);
1188
1189             /* Compute parameters for interactions between i and j atoms */
1190             qq20             = _mm256_mul_ps(iq2,jq0);
1191
1192             /* EWALD ELECTROSTATICS */
1193
1194             /* Analytical PME correction */
1195             zeta2            = _mm256_mul_ps(beta2,rsq20);
1196             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1197             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1198             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1199             felec            = _mm256_mul_ps(qq20,felec);
1200
1201             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1202
1203             fscal            = felec;
1204
1205             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm256_mul_ps(fscal,dx20);
1209             ty               = _mm256_mul_ps(fscal,dy20);
1210             tz               = _mm256_mul_ps(fscal,dz20);
1211
1212             /* Update vectorial force */
1213             fix2             = _mm256_add_ps(fix2,tx);
1214             fiy2             = _mm256_add_ps(fiy2,ty);
1215             fiz2             = _mm256_add_ps(fiz2,tz);
1216
1217             fjx0             = _mm256_add_ps(fjx0,tx);
1218             fjy0             = _mm256_add_ps(fjy0,ty);
1219             fjz0             = _mm256_add_ps(fjz0,tz);
1220
1221             }
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1228             {
1229
1230             r30              = _mm256_mul_ps(rsq30,rinv30);
1231
1232             /* Compute parameters for interactions between i and j atoms */
1233             qq30             = _mm256_mul_ps(iq3,jq0);
1234
1235             /* EWALD ELECTROSTATICS */
1236
1237             /* Analytical PME correction */
1238             zeta2            = _mm256_mul_ps(beta2,rsq30);
1239             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1240             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1241             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1242             felec            = _mm256_mul_ps(qq30,felec);
1243
1244             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1245
1246             fscal            = felec;
1247
1248             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm256_mul_ps(fscal,dx30);
1252             ty               = _mm256_mul_ps(fscal,dy30);
1253             tz               = _mm256_mul_ps(fscal,dz30);
1254
1255             /* Update vectorial force */
1256             fix3             = _mm256_add_ps(fix3,tx);
1257             fiy3             = _mm256_add_ps(fiy3,ty);
1258             fiz3             = _mm256_add_ps(fiz3,tz);
1259
1260             fjx0             = _mm256_add_ps(fjx0,tx);
1261             fjy0             = _mm256_add_ps(fjy0,ty);
1262             fjz0             = _mm256_add_ps(fjz0,tz);
1263
1264             }
1265
1266             fjptrA             = f+j_coord_offsetA;
1267             fjptrB             = f+j_coord_offsetB;
1268             fjptrC             = f+j_coord_offsetC;
1269             fjptrD             = f+j_coord_offsetD;
1270             fjptrE             = f+j_coord_offsetE;
1271             fjptrF             = f+j_coord_offsetF;
1272             fjptrG             = f+j_coord_offsetG;
1273             fjptrH             = f+j_coord_offsetH;
1274
1275             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1276
1277             /* Inner loop uses 210 flops */
1278         }
1279
1280         if(jidx<j_index_end)
1281         {
1282
1283             /* Get j neighbor index, and coordinate index */
1284             jnrlistA         = jjnr[jidx];
1285             jnrlistB         = jjnr[jidx+1];
1286             jnrlistC         = jjnr[jidx+2];
1287             jnrlistD         = jjnr[jidx+3];
1288             jnrlistE         = jjnr[jidx+4];
1289             jnrlistF         = jjnr[jidx+5];
1290             jnrlistG         = jjnr[jidx+6];
1291             jnrlistH         = jjnr[jidx+7];
1292             /* Sign of each element will be negative for non-real atoms.
1293              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1294              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1295              */
1296             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1297                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1298
1299             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1300             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1301             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1302             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1303             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1304             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1305             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1306             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1307             j_coord_offsetA  = DIM*jnrA;
1308             j_coord_offsetB  = DIM*jnrB;
1309             j_coord_offsetC  = DIM*jnrC;
1310             j_coord_offsetD  = DIM*jnrD;
1311             j_coord_offsetE  = DIM*jnrE;
1312             j_coord_offsetF  = DIM*jnrF;
1313             j_coord_offsetG  = DIM*jnrG;
1314             j_coord_offsetH  = DIM*jnrH;
1315
1316             /* load j atom coordinates */
1317             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1318                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1319                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1320                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1321                                                  &jx0,&jy0,&jz0);
1322
1323             /* Calculate displacement vector */
1324             dx00             = _mm256_sub_ps(ix0,jx0);
1325             dy00             = _mm256_sub_ps(iy0,jy0);
1326             dz00             = _mm256_sub_ps(iz0,jz0);
1327             dx10             = _mm256_sub_ps(ix1,jx0);
1328             dy10             = _mm256_sub_ps(iy1,jy0);
1329             dz10             = _mm256_sub_ps(iz1,jz0);
1330             dx20             = _mm256_sub_ps(ix2,jx0);
1331             dy20             = _mm256_sub_ps(iy2,jy0);
1332             dz20             = _mm256_sub_ps(iz2,jz0);
1333             dx30             = _mm256_sub_ps(ix3,jx0);
1334             dy30             = _mm256_sub_ps(iy3,jy0);
1335             dz30             = _mm256_sub_ps(iz3,jz0);
1336
1337             /* Calculate squared distance and things based on it */
1338             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1339             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1340             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1341             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1342
1343             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1344             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1345             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1346
1347             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1348             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1349             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1350             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1351
1352             /* Load parameters for j particles */
1353             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1354                                                                  charge+jnrC+0,charge+jnrD+0,
1355                                                                  charge+jnrE+0,charge+jnrF+0,
1356                                                                  charge+jnrG+0,charge+jnrH+0);
1357             vdwjidx0A        = 2*vdwtype[jnrA+0];
1358             vdwjidx0B        = 2*vdwtype[jnrB+0];
1359             vdwjidx0C        = 2*vdwtype[jnrC+0];
1360             vdwjidx0D        = 2*vdwtype[jnrD+0];
1361             vdwjidx0E        = 2*vdwtype[jnrE+0];
1362             vdwjidx0F        = 2*vdwtype[jnrF+0];
1363             vdwjidx0G        = 2*vdwtype[jnrG+0];
1364             vdwjidx0H        = 2*vdwtype[jnrH+0];
1365
1366             fjx0             = _mm256_setzero_ps();
1367             fjy0             = _mm256_setzero_ps();
1368             fjz0             = _mm256_setzero_ps();
1369
1370             /**************************
1371              * CALCULATE INTERACTIONS *
1372              **************************/
1373
1374             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1375             {
1376
1377             /* Compute parameters for interactions between i and j atoms */
1378             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1379                                             vdwioffsetptr0+vdwjidx0B,
1380                                             vdwioffsetptr0+vdwjidx0C,
1381                                             vdwioffsetptr0+vdwjidx0D,
1382                                             vdwioffsetptr0+vdwjidx0E,
1383                                             vdwioffsetptr0+vdwjidx0F,
1384                                             vdwioffsetptr0+vdwjidx0G,
1385                                             vdwioffsetptr0+vdwjidx0H,
1386                                             &c6_00,&c12_00);
1387
1388             /* LENNARD-JONES DISPERSION/REPULSION */
1389
1390             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1391             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1392
1393             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1394
1395             fscal            = fvdw;
1396
1397             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1398
1399             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1400
1401             /* Calculate temporary vectorial force */
1402             tx               = _mm256_mul_ps(fscal,dx00);
1403             ty               = _mm256_mul_ps(fscal,dy00);
1404             tz               = _mm256_mul_ps(fscal,dz00);
1405
1406             /* Update vectorial force */
1407             fix0             = _mm256_add_ps(fix0,tx);
1408             fiy0             = _mm256_add_ps(fiy0,ty);
1409             fiz0             = _mm256_add_ps(fiz0,tz);
1410
1411             fjx0             = _mm256_add_ps(fjx0,tx);
1412             fjy0             = _mm256_add_ps(fjy0,ty);
1413             fjz0             = _mm256_add_ps(fjz0,tz);
1414
1415             }
1416
1417             /**************************
1418              * CALCULATE INTERACTIONS *
1419              **************************/
1420
1421             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1422             {
1423
1424             r10              = _mm256_mul_ps(rsq10,rinv10);
1425             r10              = _mm256_andnot_ps(dummy_mask,r10);
1426
1427             /* Compute parameters for interactions between i and j atoms */
1428             qq10             = _mm256_mul_ps(iq1,jq0);
1429
1430             /* EWALD ELECTROSTATICS */
1431
1432             /* Analytical PME correction */
1433             zeta2            = _mm256_mul_ps(beta2,rsq10);
1434             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1435             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1436             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1437             felec            = _mm256_mul_ps(qq10,felec);
1438
1439             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1440
1441             fscal            = felec;
1442
1443             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1444
1445             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1446
1447             /* Calculate temporary vectorial force */
1448             tx               = _mm256_mul_ps(fscal,dx10);
1449             ty               = _mm256_mul_ps(fscal,dy10);
1450             tz               = _mm256_mul_ps(fscal,dz10);
1451
1452             /* Update vectorial force */
1453             fix1             = _mm256_add_ps(fix1,tx);
1454             fiy1             = _mm256_add_ps(fiy1,ty);
1455             fiz1             = _mm256_add_ps(fiz1,tz);
1456
1457             fjx0             = _mm256_add_ps(fjx0,tx);
1458             fjy0             = _mm256_add_ps(fjy0,ty);
1459             fjz0             = _mm256_add_ps(fjz0,tz);
1460
1461             }
1462
1463             /**************************
1464              * CALCULATE INTERACTIONS *
1465              **************************/
1466
1467             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1468             {
1469
1470             r20              = _mm256_mul_ps(rsq20,rinv20);
1471             r20              = _mm256_andnot_ps(dummy_mask,r20);
1472
1473             /* Compute parameters for interactions between i and j atoms */
1474             qq20             = _mm256_mul_ps(iq2,jq0);
1475
1476             /* EWALD ELECTROSTATICS */
1477
1478             /* Analytical PME correction */
1479             zeta2            = _mm256_mul_ps(beta2,rsq20);
1480             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1481             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1482             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1483             felec            = _mm256_mul_ps(qq20,felec);
1484
1485             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1486
1487             fscal            = felec;
1488
1489             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1490
1491             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1492
1493             /* Calculate temporary vectorial force */
1494             tx               = _mm256_mul_ps(fscal,dx20);
1495             ty               = _mm256_mul_ps(fscal,dy20);
1496             tz               = _mm256_mul_ps(fscal,dz20);
1497
1498             /* Update vectorial force */
1499             fix2             = _mm256_add_ps(fix2,tx);
1500             fiy2             = _mm256_add_ps(fiy2,ty);
1501             fiz2             = _mm256_add_ps(fiz2,tz);
1502
1503             fjx0             = _mm256_add_ps(fjx0,tx);
1504             fjy0             = _mm256_add_ps(fjy0,ty);
1505             fjz0             = _mm256_add_ps(fjz0,tz);
1506
1507             }
1508
1509             /**************************
1510              * CALCULATE INTERACTIONS *
1511              **************************/
1512
1513             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1514             {
1515
1516             r30              = _mm256_mul_ps(rsq30,rinv30);
1517             r30              = _mm256_andnot_ps(dummy_mask,r30);
1518
1519             /* Compute parameters for interactions between i and j atoms */
1520             qq30             = _mm256_mul_ps(iq3,jq0);
1521
1522             /* EWALD ELECTROSTATICS */
1523
1524             /* Analytical PME correction */
1525             zeta2            = _mm256_mul_ps(beta2,rsq30);
1526             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1527             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1528             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1529             felec            = _mm256_mul_ps(qq30,felec);
1530
1531             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1532
1533             fscal            = felec;
1534
1535             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1536
1537             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1538
1539             /* Calculate temporary vectorial force */
1540             tx               = _mm256_mul_ps(fscal,dx30);
1541             ty               = _mm256_mul_ps(fscal,dy30);
1542             tz               = _mm256_mul_ps(fscal,dz30);
1543
1544             /* Update vectorial force */
1545             fix3             = _mm256_add_ps(fix3,tx);
1546             fiy3             = _mm256_add_ps(fiy3,ty);
1547             fiz3             = _mm256_add_ps(fiz3,tz);
1548
1549             fjx0             = _mm256_add_ps(fjx0,tx);
1550             fjy0             = _mm256_add_ps(fjy0,ty);
1551             fjz0             = _mm256_add_ps(fjz0,tz);
1552
1553             }
1554
1555             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1556             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1557             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1558             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1559             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1560             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1561             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1562             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1563
1564             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1565
1566             /* Inner loop uses 213 flops */
1567         }
1568
1569         /* End of innermost loop */
1570
1571         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1572                                                  f+i_coord_offset,fshift+i_shift_offset);
1573
1574         /* Increment number of inner iterations */
1575         inneriter                  += j_index_end - j_index_start;
1576
1577         /* Outer loop uses 24 flops */
1578     }
1579
1580     /* Increment number of outer iterations */
1581     outeriter        += nri;
1582
1583     /* Update outer/inner flops */
1584
1585     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*213);
1586 }