src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m128i          vfitab;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m128i          ewitab;
 113     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 115     real             *ewtab;
 116     __m256d          dummy_mask,cutoff_mask;
 117     __m128           tmpmask0,tmpmask1;
 118     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 119     __m256d          one     = _mm256_set1_pd(1.0);
 120     __m256d          two     = _mm256_set1_pd(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm256_set1_pd(fr->epsfac);
 133     charge           = mdatoms->chargeA;
 134     nvdwtype         = fr->ntype;
 135     vdwparam         = fr->nbfp;
 136     vdwtype          = mdatoms->typeA;
 137
 138     vftab            = kernel_data->table_vdw->data;
 139     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 140
 141     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 142     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 143     beta2            = _mm256_mul_pd(beta,beta);
 144     beta3            = _mm256_mul_pd(beta,beta2);
 145
 146     ewtab            = fr->ic->tabq_coul_FDV0;
 147     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 148     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 149
 150     /* Setup water-specific parameters */
 151     inr              = nlist->iinr[0];
 152     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 153     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 154     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 155     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163
 164     outeriter        = 0;
 165     inneriter        = 0;
 166
 167     for(iidx=0;iidx<4*DIM;iidx++)
 168     {
 169         scratch[iidx] = 0.0;
 170     }
 171
 172     /* Start outer loop over neighborlists */
 173     for(iidx=0; iidx<nri; iidx++)
 174     {
 175         /* Load shift vector for this list */
 176         i_shift_offset   = DIM*shiftidx[iidx];
 177
 178         /* Load limits for loop over neighbors */
 179         j_index_start    = jindex[iidx];
 180         j_index_end      = jindex[iidx+1];
 181
 182         /* Get outer coordinate index */
 183         inr              = iinr[iidx];
 184         i_coord_offset   = DIM*inr;
 185
 186         /* Load i particle coords and add shift vector */
 187         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 188                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 189
 190         fix0             = _mm256_setzero_pd();
 191         fiy0             = _mm256_setzero_pd();
 192         fiz0             = _mm256_setzero_pd();
 193         fix1             = _mm256_setzero_pd();
 194         fiy1             = _mm256_setzero_pd();
 195         fiz1             = _mm256_setzero_pd();
 196         fix2             = _mm256_setzero_pd();
 197         fiy2             = _mm256_setzero_pd();
 198         fiz2             = _mm256_setzero_pd();
 199         fix3             = _mm256_setzero_pd();
 200         fiy3             = _mm256_setzero_pd();
 201         fiz3             = _mm256_setzero_pd();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm256_setzero_pd();
 205         vvdwsum          = _mm256_setzero_pd();
 206
 207         /* Start inner kernel loop */
 208         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrA             = jjnr[jidx];
 213             jnrB             = jjnr[jidx+1];
 214             jnrC             = jjnr[jidx+2];
 215             jnrD             = jjnr[jidx+3];
 216             j_coord_offsetA  = DIM*jnrA;
 217             j_coord_offsetB  = DIM*jnrB;
 218             j_coord_offsetC  = DIM*jnrC;
 219             j_coord_offsetD  = DIM*jnrD;
 220
 221             /* load j atom coordinates */
 222             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 224                                                  &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_pd(ix0,jx0);
 228             dy00             = _mm256_sub_pd(iy0,jy0);
 229             dz00             = _mm256_sub_pd(iz0,jz0);
 230             dx10             = _mm256_sub_pd(ix1,jx0);
 231             dy10             = _mm256_sub_pd(iy1,jy0);
 232             dz10             = _mm256_sub_pd(iz1,jz0);
 233             dx20             = _mm256_sub_pd(ix2,jx0);
 234             dy20             = _mm256_sub_pd(iy2,jy0);
 235             dz20             = _mm256_sub_pd(iz2,jz0);
 236             dx30             = _mm256_sub_pd(ix3,jx0);
 237             dy30             = _mm256_sub_pd(iy3,jy0);
 238             dz30             = _mm256_sub_pd(iz3,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 242             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 243             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 244             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 245
 246             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 247             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 248             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 249             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 250
 251             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 252             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 253             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 254
 255             /* Load parameters for j particles */
 256             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 257                                                                  charge+jnrC+0,charge+jnrD+0);
 258             vdwjidx0A        = 2*vdwtype[jnrA+0];
 259             vdwjidx0B        = 2*vdwtype[jnrB+0];
 260             vdwjidx0C        = 2*vdwtype[jnrC+0];
 261             vdwjidx0D        = 2*vdwtype[jnrD+0];
 262
 263             fjx0             = _mm256_setzero_pd();
 264             fjy0             = _mm256_setzero_pd();
 265             fjz0             = _mm256_setzero_pd();
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             r00              = _mm256_mul_pd(rsq00,rinv00);
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 275                                             vdwioffsetptr0+vdwjidx0B,
 276                                             vdwioffsetptr0+vdwjidx0C,
 277                                             vdwioffsetptr0+vdwjidx0D,
 278                                             &c6_00,&c12_00);
 279
 280             /* Calculate table index by multiplying r with table scale and truncate to integer */
 281             rt               = _mm256_mul_pd(r00,vftabscale);
 282             vfitab           = _mm256_cvttpd_epi32(rt);
 283             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 284             vfitab           = _mm_slli_epi32(vfitab,3);
 285
 286             /* CUBIC SPLINE TABLE DISPERSION */
 287             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 288             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 289             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 290             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 291             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 292             Heps             = _mm256_mul_pd(vfeps,H);
 293             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 294             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 295             vvdw6            = _mm256_mul_pd(c6_00,VV);
 296             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 297             fvdw6            = _mm256_mul_pd(c6_00,FF);
 298
 299             /* CUBIC SPLINE TABLE REPULSION */
 300             vfitab           = _mm_add_epi32(vfitab,ifour);
 301             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 302             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 303             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 304             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 305             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 306             Heps             = _mm256_mul_pd(vfeps,H);
 307             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 308             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 309             vvdw12           = _mm256_mul_pd(c12_00,VV);
 310             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 311             fvdw12           = _mm256_mul_pd(c12_00,FF);
 312             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 313             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 317
 318             fscal            = fvdw;
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm256_mul_pd(fscal,dx00);
 322             ty               = _mm256_mul_pd(fscal,dy00);
 323             tz               = _mm256_mul_pd(fscal,dz00);
 324
 325             /* Update vectorial force */
 326             fix0             = _mm256_add_pd(fix0,tx);
 327             fiy0             = _mm256_add_pd(fiy0,ty);
 328             fiz0             = _mm256_add_pd(fiz0,tz);
 329
 330             fjx0             = _mm256_add_pd(fjx0,tx);
 331             fjy0             = _mm256_add_pd(fjy0,ty);
 332             fjz0             = _mm256_add_pd(fjz0,tz);
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             r10              = _mm256_mul_pd(rsq10,rinv10);
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq10             = _mm256_mul_pd(iq1,jq0);
 342
 343             /* EWALD ELECTROSTATICS */
 344
 345             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 346             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 347             ewitab           = _mm256_cvttpd_epi32(ewrt);
 348             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 349             ewitab           = _mm_slli_epi32(ewitab,2);
 350             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 351             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 352             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 353             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 354             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 355             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 356             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 357             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 358             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm256_add_pd(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm256_mul_pd(fscal,dx10);
 367             ty               = _mm256_mul_pd(fscal,dy10);
 368             tz               = _mm256_mul_pd(fscal,dz10);
 369
 370             /* Update vectorial force */
 371             fix1             = _mm256_add_pd(fix1,tx);
 372             fiy1             = _mm256_add_pd(fiy1,ty);
 373             fiz1             = _mm256_add_pd(fiz1,tz);
 374
 375             fjx0             = _mm256_add_pd(fjx0,tx);
 376             fjy0             = _mm256_add_pd(fjy0,ty);
 377             fjz0             = _mm256_add_pd(fjz0,tz);
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             r20              = _mm256_mul_pd(rsq20,rinv20);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq20             = _mm256_mul_pd(iq2,jq0);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 391             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 392             ewitab           = _mm256_cvttpd_epi32(ewrt);
 393             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 394             ewitab           = _mm_slli_epi32(ewitab,2);
 395             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 396             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 397             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 398             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 399             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 400             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 401             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 402             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 403             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velecsum         = _mm256_add_pd(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             /* Calculate temporary vectorial force */
 411             tx               = _mm256_mul_pd(fscal,dx20);
 412             ty               = _mm256_mul_pd(fscal,dy20);
 413             tz               = _mm256_mul_pd(fscal,dz20);
 414
 415             /* Update vectorial force */
 416             fix2             = _mm256_add_pd(fix2,tx);
 417             fiy2             = _mm256_add_pd(fiy2,ty);
 418             fiz2             = _mm256_add_pd(fiz2,tz);
 419
 420             fjx0             = _mm256_add_pd(fjx0,tx);
 421             fjy0             = _mm256_add_pd(fjy0,ty);
 422             fjz0             = _mm256_add_pd(fjz0,tz);
 423
 424             /**************************
 425              * CALCULATE INTERACTIONS *
 426              **************************/
 427
 428             r30              = _mm256_mul_pd(rsq30,rinv30);
 429
 430             /* Compute parameters for interactions between i and j atoms */
 431             qq30             = _mm256_mul_pd(iq3,jq0);
 432
 433             /* EWALD ELECTROSTATICS */
 434
 435             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 436             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 437             ewitab           = _mm256_cvttpd_epi32(ewrt);
 438             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 439             ewitab           = _mm_slli_epi32(ewitab,2);
 440             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 441             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 442             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 443             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 444             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 445             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 446             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 447             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 448             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velecsum         = _mm256_add_pd(velecsum,velec);
 452
 453             fscal            = felec;
 454
 455             /* Calculate temporary vectorial force */
 456             tx               = _mm256_mul_pd(fscal,dx30);
 457             ty               = _mm256_mul_pd(fscal,dy30);
 458             tz               = _mm256_mul_pd(fscal,dz30);
 459
 460             /* Update vectorial force */
 461             fix3             = _mm256_add_pd(fix3,tx);
 462             fiy3             = _mm256_add_pd(fiy3,ty);
 463             fiz3             = _mm256_add_pd(fiz3,tz);
 464
 465             fjx0             = _mm256_add_pd(fjx0,tx);
 466             fjy0             = _mm256_add_pd(fjy0,ty);
 467             fjz0             = _mm256_add_pd(fjz0,tz);
 468
 469             fjptrA             = f+j_coord_offsetA;
 470             fjptrB             = f+j_coord_offsetB;
 471             fjptrC             = f+j_coord_offsetC;
 472             fjptrD             = f+j_coord_offsetD;
 473
 474             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 475
 476             /* Inner loop uses 182 flops */
 477         }
 478
 479         if(jidx<j_index_end)
 480         {
 481
 482             /* Get j neighbor index, and coordinate index */
 483             jnrlistA         = jjnr[jidx];
 484             jnrlistB         = jjnr[jidx+1];
 485             jnrlistC         = jjnr[jidx+2];
 486             jnrlistD         = jjnr[jidx+3];
 487             /* Sign of each element will be negative for non-real atoms.
 488              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 489              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 490              */
 491             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 492
 493             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 494             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 495             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 496
 497             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 498             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 499             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 500             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 501             j_coord_offsetA  = DIM*jnrA;
 502             j_coord_offsetB  = DIM*jnrB;
 503             j_coord_offsetC  = DIM*jnrC;
 504             j_coord_offsetD  = DIM*jnrD;
 505
 506             /* load j atom coordinates */
 507             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 508                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 509                                                  &jx0,&jy0,&jz0);
 510
 511             /* Calculate displacement vector */
 512             dx00             = _mm256_sub_pd(ix0,jx0);
 513             dy00             = _mm256_sub_pd(iy0,jy0);
 514             dz00             = _mm256_sub_pd(iz0,jz0);
 515             dx10             = _mm256_sub_pd(ix1,jx0);
 516             dy10             = _mm256_sub_pd(iy1,jy0);
 517             dz10             = _mm256_sub_pd(iz1,jz0);
 518             dx20             = _mm256_sub_pd(ix2,jx0);
 519             dy20             = _mm256_sub_pd(iy2,jy0);
 520             dz20             = _mm256_sub_pd(iz2,jz0);
 521             dx30             = _mm256_sub_pd(ix3,jx0);
 522             dy30             = _mm256_sub_pd(iy3,jy0);
 523             dz30             = _mm256_sub_pd(iz3,jz0);
 524
 525             /* Calculate squared distance and things based on it */
 526             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 527             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 528             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 529             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 530
 531             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 532             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 533             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 534             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 535
 536             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 537             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 538             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 539
 540             /* Load parameters for j particles */
 541             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 542                                                                  charge+jnrC+0,charge+jnrD+0);
 543             vdwjidx0A        = 2*vdwtype[jnrA+0];
 544             vdwjidx0B        = 2*vdwtype[jnrB+0];
 545             vdwjidx0C        = 2*vdwtype[jnrC+0];
 546             vdwjidx0D        = 2*vdwtype[jnrD+0];
 547
 548             fjx0             = _mm256_setzero_pd();
 549             fjy0             = _mm256_setzero_pd();
 550             fjz0             = _mm256_setzero_pd();
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             r00              = _mm256_mul_pd(rsq00,rinv00);
 557             r00              = _mm256_andnot_pd(dummy_mask,r00);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 561                                             vdwioffsetptr0+vdwjidx0B,
 562                                             vdwioffsetptr0+vdwjidx0C,
 563                                             vdwioffsetptr0+vdwjidx0D,
 564                                             &c6_00,&c12_00);
 565
 566             /* Calculate table index by multiplying r with table scale and truncate to integer */
 567             rt               = _mm256_mul_pd(r00,vftabscale);
 568             vfitab           = _mm256_cvttpd_epi32(rt);
 569             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 570             vfitab           = _mm_slli_epi32(vfitab,3);
 571
 572             /* CUBIC SPLINE TABLE DISPERSION */
 573             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 574             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 575             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 576             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 577             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 578             Heps             = _mm256_mul_pd(vfeps,H);
 579             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 580             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 581             vvdw6            = _mm256_mul_pd(c6_00,VV);
 582             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 583             fvdw6            = _mm256_mul_pd(c6_00,FF);
 584
 585             /* CUBIC SPLINE TABLE REPULSION */
 586             vfitab           = _mm_add_epi32(vfitab,ifour);
 587             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 588             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 589             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 590             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 591             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 592             Heps             = _mm256_mul_pd(vfeps,H);
 593             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 594             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 595             vvdw12           = _mm256_mul_pd(c12_00,VV);
 596             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 597             fvdw12           = _mm256_mul_pd(c12_00,FF);
 598             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 599             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 600
 601             /* Update potential sum for this i atom from the interaction with this j atom. */
 602             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 603             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 604
 605             fscal            = fvdw;
 606
 607             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm256_mul_pd(fscal,dx00);
 611             ty               = _mm256_mul_pd(fscal,dy00);
 612             tz               = _mm256_mul_pd(fscal,dz00);
 613
 614             /* Update vectorial force */
 615             fix0             = _mm256_add_pd(fix0,tx);
 616             fiy0             = _mm256_add_pd(fiy0,ty);
 617             fiz0             = _mm256_add_pd(fiz0,tz);
 618
 619             fjx0             = _mm256_add_pd(fjx0,tx);
 620             fjy0             = _mm256_add_pd(fjy0,ty);
 621             fjz0             = _mm256_add_pd(fjz0,tz);
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             r10              = _mm256_mul_pd(rsq10,rinv10);
 628             r10              = _mm256_andnot_pd(dummy_mask,r10);
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq10             = _mm256_mul_pd(iq1,jq0);
 632
 633             /* EWALD ELECTROSTATICS */
 634
 635             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 636             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 637             ewitab           = _mm256_cvttpd_epi32(ewrt);
 638             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 639             ewitab           = _mm_slli_epi32(ewitab,2);
 640             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 641             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 642             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 643             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 644             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 645             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 646             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 647             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 648             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm256_andnot_pd(dummy_mask,velec);
 652             velecsum         = _mm256_add_pd(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm256_mul_pd(fscal,dx10);
 660             ty               = _mm256_mul_pd(fscal,dy10);
 661             tz               = _mm256_mul_pd(fscal,dz10);
 662
 663             /* Update vectorial force */
 664             fix1             = _mm256_add_pd(fix1,tx);
 665             fiy1             = _mm256_add_pd(fiy1,ty);
 666             fiz1             = _mm256_add_pd(fiz1,tz);
 667
 668             fjx0             = _mm256_add_pd(fjx0,tx);
 669             fjy0             = _mm256_add_pd(fjy0,ty);
 670             fjz0             = _mm256_add_pd(fjz0,tz);
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             r20              = _mm256_mul_pd(rsq20,rinv20);
 677             r20              = _mm256_andnot_pd(dummy_mask,r20);
 678
 679             /* Compute parameters for interactions between i and j atoms */
 680             qq20             = _mm256_mul_pd(iq2,jq0);
 681
 682             /* EWALD ELECTROSTATICS */
 683
 684             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 685             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 686             ewitab           = _mm256_cvttpd_epi32(ewrt);
 687             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 688             ewitab           = _mm_slli_epi32(ewitab,2);
 689             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 690             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 691             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 692             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 693             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 694             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 695             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 696             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 697             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 698
 699             /* Update potential sum for this i atom from the interaction with this j atom. */
 700             velec            = _mm256_andnot_pd(dummy_mask,velec);
 701             velecsum         = _mm256_add_pd(velecsum,velec);
 702
 703             fscal            = felec;
 704
 705             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 706
 707             /* Calculate temporary vectorial force */
 708             tx               = _mm256_mul_pd(fscal,dx20);
 709             ty               = _mm256_mul_pd(fscal,dy20);
 710             tz               = _mm256_mul_pd(fscal,dz20);
 711
 712             /* Update vectorial force */
 713             fix2             = _mm256_add_pd(fix2,tx);
 714             fiy2             = _mm256_add_pd(fiy2,ty);
 715             fiz2             = _mm256_add_pd(fiz2,tz);
 716
 717             fjx0             = _mm256_add_pd(fjx0,tx);
 718             fjy0             = _mm256_add_pd(fjy0,ty);
 719             fjz0             = _mm256_add_pd(fjz0,tz);
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             r30              = _mm256_mul_pd(rsq30,rinv30);
 726             r30              = _mm256_andnot_pd(dummy_mask,r30);
 727
 728             /* Compute parameters for interactions between i and j atoms */
 729             qq30             = _mm256_mul_pd(iq3,jq0);
 730
 731             /* EWALD ELECTROSTATICS */
 732
 733             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 734             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 735             ewitab           = _mm256_cvttpd_epi32(ewrt);
 736             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 737             ewitab           = _mm_slli_epi32(ewitab,2);
 738             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 739             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 740             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 741             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 742             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 743             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 744             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 745             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 746             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 747
 748             /* Update potential sum for this i atom from the interaction with this j atom. */
 749             velec            = _mm256_andnot_pd(dummy_mask,velec);
 750             velecsum         = _mm256_add_pd(velecsum,velec);
 751
 752             fscal            = felec;
 753
 754             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 755
 756             /* Calculate temporary vectorial force */
 757             tx               = _mm256_mul_pd(fscal,dx30);
 758             ty               = _mm256_mul_pd(fscal,dy30);
 759             tz               = _mm256_mul_pd(fscal,dz30);
 760
 761             /* Update vectorial force */
 762             fix3             = _mm256_add_pd(fix3,tx);
 763             fiy3             = _mm256_add_pd(fiy3,ty);
 764             fiz3             = _mm256_add_pd(fiz3,tz);
 765
 766             fjx0             = _mm256_add_pd(fjx0,tx);
 767             fjy0             = _mm256_add_pd(fjy0,ty);
 768             fjz0             = _mm256_add_pd(fjz0,tz);
 769
 770             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 771             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 772             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 773             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 774
 775             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 776
 777             /* Inner loop uses 186 flops */
 778         }
 779
 780         /* End of innermost loop */
 781
 782         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 783                                                  f+i_coord_offset,fshift+i_shift_offset);
 784
 785         ggid                        = gid[iidx];
 786         /* Update potential energies */
 787         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 788         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 789
 790         /* Increment number of inner iterations */
 791         inneriter                  += j_index_end - j_index_start;
 792
 793         /* Outer loop uses 26 flops */
 794     }
 795
 796     /* Increment number of outer iterations */
 797     outeriter        += nri;
 798
 799     /* Update outer/inner flops */
 800
 801     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*186);
 802 }
 803 /*
 804  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_double
 805  * Electrostatics interaction: Ewald
 806  * VdW interaction:            CubicSplineTable
 807  * Geometry:                   Water4-Particle
 808  * Calculate force/pot:        Force
 809  */
 810 void
 811 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_double
 812                     (t_nblist                    * gmx_restrict       nlist,
 813                      rvec                        * gmx_restrict          xx,
 814                      rvec                        * gmx_restrict          ff,
 815                      t_forcerec                  * gmx_restrict          fr,
 816                      t_mdatoms                   * gmx_restrict     mdatoms,
 817                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 818                      t_nrnb                      * gmx_restrict        nrnb)
 819 {
 820     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 821      * just 0 for non-waters.
 822      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 823      * jnr indices corresponding to data put in the four positions in the SIMD register.
 824      */
 825     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 826     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 827     int              jnrA,jnrB,jnrC,jnrD;
 828     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 829     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 830     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 831     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 832     real             rcutoff_scalar;
 833     real             *shiftvec,*fshift,*x,*f;
 834     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 835     real             scratch[4*DIM];
 836     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 837     real *           vdwioffsetptr0;
 838     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 839     real *           vdwioffsetptr1;
 840     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 841     real *           vdwioffsetptr2;
 842     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 843     real *           vdwioffsetptr3;
 844     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 845     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 846     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 847     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 848     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 849     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 850     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 851     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 852     real             *charge;
 853     int              nvdwtype;
 854     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 855     int              *vdwtype;
 856     real             *vdwparam;
 857     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 858     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 859     __m128i          vfitab;
 860     __m128i          ifour       = _mm_set1_epi32(4);
 861     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 862     real             *vftab;
 863     __m128i          ewitab;
 864     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 865     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 866     real             *ewtab;
 867     __m256d          dummy_mask,cutoff_mask;
 868     __m128           tmpmask0,tmpmask1;
 869     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 870     __m256d          one     = _mm256_set1_pd(1.0);
 871     __m256d          two     = _mm256_set1_pd(2.0);
 872     x                = xx[0];
 873     f                = ff[0];
 874
 875     nri              = nlist->nri;
 876     iinr             = nlist->iinr;
 877     jindex           = nlist->jindex;
 878     jjnr             = nlist->jjnr;
 879     shiftidx         = nlist->shift;
 880     gid              = nlist->gid;
 881     shiftvec         = fr->shift_vec[0];
 882     fshift           = fr->fshift[0];
 883     facel            = _mm256_set1_pd(fr->epsfac);
 884     charge           = mdatoms->chargeA;
 885     nvdwtype         = fr->ntype;
 886     vdwparam         = fr->nbfp;
 887     vdwtype          = mdatoms->typeA;
 888
 889     vftab            = kernel_data->table_vdw->data;
 890     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 891
 892     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 893     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 894     beta2            = _mm256_mul_pd(beta,beta);
 895     beta3            = _mm256_mul_pd(beta,beta2);
 896
 897     ewtab            = fr->ic->tabq_coul_F;
 898     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 899     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 900
 901     /* Setup water-specific parameters */
 902     inr              = nlist->iinr[0];
 903     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 904     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 905     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 906     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 907
 908     /* Avoid stupid compiler warnings */
 909     jnrA = jnrB = jnrC = jnrD = 0;
 910     j_coord_offsetA = 0;
 911     j_coord_offsetB = 0;
 912     j_coord_offsetC = 0;
 913     j_coord_offsetD = 0;
 914
 915     outeriter        = 0;
 916     inneriter        = 0;
 917
 918     for(iidx=0;iidx<4*DIM;iidx++)
 919     {
 920         scratch[iidx] = 0.0;
 921     }
 922
 923     /* Start outer loop over neighborlists */
 924     for(iidx=0; iidx<nri; iidx++)
 925     {
 926         /* Load shift vector for this list */
 927         i_shift_offset   = DIM*shiftidx[iidx];
 928
 929         /* Load limits for loop over neighbors */
 930         j_index_start    = jindex[iidx];
 931         j_index_end      = jindex[iidx+1];
 932
 933         /* Get outer coordinate index */
 934         inr              = iinr[iidx];
 935         i_coord_offset   = DIM*inr;
 936
 937         /* Load i particle coords and add shift vector */
 938         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 939                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 940
 941         fix0             = _mm256_setzero_pd();
 942         fiy0             = _mm256_setzero_pd();
 943         fiz0             = _mm256_setzero_pd();
 944         fix1             = _mm256_setzero_pd();
 945         fiy1             = _mm256_setzero_pd();
 946         fiz1             = _mm256_setzero_pd();
 947         fix2             = _mm256_setzero_pd();
 948         fiy2             = _mm256_setzero_pd();
 949         fiz2             = _mm256_setzero_pd();
 950         fix3             = _mm256_setzero_pd();
 951         fiy3             = _mm256_setzero_pd();
 952         fiz3             = _mm256_setzero_pd();
 953
 954         /* Start inner kernel loop */
 955         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 956         {
 957
 958             /* Get j neighbor index, and coordinate index */
 959             jnrA             = jjnr[jidx];
 960             jnrB             = jjnr[jidx+1];
 961             jnrC             = jjnr[jidx+2];
 962             jnrD             = jjnr[jidx+3];
 963             j_coord_offsetA  = DIM*jnrA;
 964             j_coord_offsetB  = DIM*jnrB;
 965             j_coord_offsetC  = DIM*jnrC;
 966             j_coord_offsetD  = DIM*jnrD;
 967
 968             /* load j atom coordinates */
 969             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 970                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 971                                                  &jx0,&jy0,&jz0);
 972
 973             /* Calculate displacement vector */
 974             dx00             = _mm256_sub_pd(ix0,jx0);
 975             dy00             = _mm256_sub_pd(iy0,jy0);
 976             dz00             = _mm256_sub_pd(iz0,jz0);
 977             dx10             = _mm256_sub_pd(ix1,jx0);
 978             dy10             = _mm256_sub_pd(iy1,jy0);
 979             dz10             = _mm256_sub_pd(iz1,jz0);
 980             dx20             = _mm256_sub_pd(ix2,jx0);
 981             dy20             = _mm256_sub_pd(iy2,jy0);
 982             dz20             = _mm256_sub_pd(iz2,jz0);
 983             dx30             = _mm256_sub_pd(ix3,jx0);
 984             dy30             = _mm256_sub_pd(iy3,jy0);
 985             dz30             = _mm256_sub_pd(iz3,jz0);
 986
 987             /* Calculate squared distance and things based on it */
 988             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 989             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 990             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 991             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 992
 993             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 994             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 995             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 996             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 997
 998             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 999             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1000             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1001
1002             /* Load parameters for j particles */
1003             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1004                                                                  charge+jnrC+0,charge+jnrD+0);
1005             vdwjidx0A        = 2*vdwtype[jnrA+0];
1006             vdwjidx0B        = 2*vdwtype[jnrB+0];
1007             vdwjidx0C        = 2*vdwtype[jnrC+0];
1008             vdwjidx0D        = 2*vdwtype[jnrD+0];
1009
1010             fjx0             = _mm256_setzero_pd();
1011             fjy0             = _mm256_setzero_pd();
1012             fjz0             = _mm256_setzero_pd();
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             r00              = _mm256_mul_pd(rsq00,rinv00);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1022                                             vdwioffsetptr0+vdwjidx0B,
1023                                             vdwioffsetptr0+vdwjidx0C,
1024                                             vdwioffsetptr0+vdwjidx0D,
1025                                             &c6_00,&c12_00);
1026
1027             /* Calculate table index by multiplying r with table scale and truncate to integer */
1028             rt               = _mm256_mul_pd(r00,vftabscale);
1029             vfitab           = _mm256_cvttpd_epi32(rt);
1030             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1031             vfitab           = _mm_slli_epi32(vfitab,3);
1032
1033             /* CUBIC SPLINE TABLE DISPERSION */
1034             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1035             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1036             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1037             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1038             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1039             Heps             = _mm256_mul_pd(vfeps,H);
1040             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1041             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1042             fvdw6            = _mm256_mul_pd(c6_00,FF);
1043
1044             /* CUBIC SPLINE TABLE REPULSION */
1045             vfitab           = _mm_add_epi32(vfitab,ifour);
1046             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1047             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1048             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1049             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1050             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1051             Heps             = _mm256_mul_pd(vfeps,H);
1052             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1053             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1054             fvdw12           = _mm256_mul_pd(c12_00,FF);
1055             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1056
1057             fscal            = fvdw;
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm256_mul_pd(fscal,dx00);
1061             ty               = _mm256_mul_pd(fscal,dy00);
1062             tz               = _mm256_mul_pd(fscal,dz00);
1063
1064             /* Update vectorial force */
1065             fix0             = _mm256_add_pd(fix0,tx);
1066             fiy0             = _mm256_add_pd(fiy0,ty);
1067             fiz0             = _mm256_add_pd(fiz0,tz);
1068
1069             fjx0             = _mm256_add_pd(fjx0,tx);
1070             fjy0             = _mm256_add_pd(fjy0,ty);
1071             fjz0             = _mm256_add_pd(fjz0,tz);
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             r10              = _mm256_mul_pd(rsq10,rinv10);
1078
1079             /* Compute parameters for interactions between i and j atoms */
1080             qq10             = _mm256_mul_pd(iq1,jq0);
1081
1082             /* EWALD ELECTROSTATICS */
1083
1084             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1085             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1086             ewitab           = _mm256_cvttpd_epi32(ewrt);
1087             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1088             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1089                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1090                                             &ewtabF,&ewtabFn);
1091             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1092             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1093
1094             fscal            = felec;
1095
1096             /* Calculate temporary vectorial force */
1097             tx               = _mm256_mul_pd(fscal,dx10);
1098             ty               = _mm256_mul_pd(fscal,dy10);
1099             tz               = _mm256_mul_pd(fscal,dz10);
1100
1101             /* Update vectorial force */
1102             fix1             = _mm256_add_pd(fix1,tx);
1103             fiy1             = _mm256_add_pd(fiy1,ty);
1104             fiz1             = _mm256_add_pd(fiz1,tz);
1105
1106             fjx0             = _mm256_add_pd(fjx0,tx);
1107             fjy0             = _mm256_add_pd(fjy0,ty);
1108             fjz0             = _mm256_add_pd(fjz0,tz);
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             r20              = _mm256_mul_pd(rsq20,rinv20);
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq20             = _mm256_mul_pd(iq2,jq0);
1118
1119             /* EWALD ELECTROSTATICS */
1120
1121             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1122             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1123             ewitab           = _mm256_cvttpd_epi32(ewrt);
1124             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1125             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1126                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1127                                             &ewtabF,&ewtabFn);
1128             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1129             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1130
1131             fscal            = felec;
1132
1133             /* Calculate temporary vectorial force */
1134             tx               = _mm256_mul_pd(fscal,dx20);
1135             ty               = _mm256_mul_pd(fscal,dy20);
1136             tz               = _mm256_mul_pd(fscal,dz20);
1137
1138             /* Update vectorial force */
1139             fix2             = _mm256_add_pd(fix2,tx);
1140             fiy2             = _mm256_add_pd(fiy2,ty);
1141             fiz2             = _mm256_add_pd(fiz2,tz);
1142
1143             fjx0             = _mm256_add_pd(fjx0,tx);
1144             fjy0             = _mm256_add_pd(fjy0,ty);
1145             fjz0             = _mm256_add_pd(fjz0,tz);
1146
1147             /**************************
1148              * CALCULATE INTERACTIONS *
1149              **************************/
1150
1151             r30              = _mm256_mul_pd(rsq30,rinv30);
1152
1153             /* Compute parameters for interactions between i and j atoms */
1154             qq30             = _mm256_mul_pd(iq3,jq0);
1155
1156             /* EWALD ELECTROSTATICS */
1157
1158             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1159             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1160             ewitab           = _mm256_cvttpd_epi32(ewrt);
1161             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1162             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1163                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1164                                             &ewtabF,&ewtabFn);
1165             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1166             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1167
1168             fscal            = felec;
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm256_mul_pd(fscal,dx30);
1172             ty               = _mm256_mul_pd(fscal,dy30);
1173             tz               = _mm256_mul_pd(fscal,dz30);
1174
1175             /* Update vectorial force */
1176             fix3             = _mm256_add_pd(fix3,tx);
1177             fiy3             = _mm256_add_pd(fiy3,ty);
1178             fiz3             = _mm256_add_pd(fiz3,tz);
1179
1180             fjx0             = _mm256_add_pd(fjx0,tx);
1181             fjy0             = _mm256_add_pd(fjy0,ty);
1182             fjz0             = _mm256_add_pd(fjz0,tz);
1183
1184             fjptrA             = f+j_coord_offsetA;
1185             fjptrB             = f+j_coord_offsetB;
1186             fjptrC             = f+j_coord_offsetC;
1187             fjptrD             = f+j_coord_offsetD;
1188
1189             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1190
1191             /* Inner loop uses 159 flops */
1192         }
1193
1194         if(jidx<j_index_end)
1195         {
1196
1197             /* Get j neighbor index, and coordinate index */
1198             jnrlistA         = jjnr[jidx];
1199             jnrlistB         = jjnr[jidx+1];
1200             jnrlistC         = jjnr[jidx+2];
1201             jnrlistD         = jjnr[jidx+3];
1202             /* Sign of each element will be negative for non-real atoms.
1203              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1204              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1205              */
1206             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1207
1208             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1209             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1210             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1211
1212             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1213             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1214             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1215             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1216             j_coord_offsetA  = DIM*jnrA;
1217             j_coord_offsetB  = DIM*jnrB;
1218             j_coord_offsetC  = DIM*jnrC;
1219             j_coord_offsetD  = DIM*jnrD;
1220
1221             /* load j atom coordinates */
1222             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1223                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1224                                                  &jx0,&jy0,&jz0);
1225
1226             /* Calculate displacement vector */
1227             dx00             = _mm256_sub_pd(ix0,jx0);
1228             dy00             = _mm256_sub_pd(iy0,jy0);
1229             dz00             = _mm256_sub_pd(iz0,jz0);
1230             dx10             = _mm256_sub_pd(ix1,jx0);
1231             dy10             = _mm256_sub_pd(iy1,jy0);
1232             dz10             = _mm256_sub_pd(iz1,jz0);
1233             dx20             = _mm256_sub_pd(ix2,jx0);
1234             dy20             = _mm256_sub_pd(iy2,jy0);
1235             dz20             = _mm256_sub_pd(iz2,jz0);
1236             dx30             = _mm256_sub_pd(ix3,jx0);
1237             dy30             = _mm256_sub_pd(iy3,jy0);
1238             dz30             = _mm256_sub_pd(iz3,jz0);
1239
1240             /* Calculate squared distance and things based on it */
1241             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1242             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1243             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1244             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1245
1246             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1247             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1248             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1249             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1250
1251             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1252             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1253             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1254
1255             /* Load parameters for j particles */
1256             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1257                                                                  charge+jnrC+0,charge+jnrD+0);
1258             vdwjidx0A        = 2*vdwtype[jnrA+0];
1259             vdwjidx0B        = 2*vdwtype[jnrB+0];
1260             vdwjidx0C        = 2*vdwtype[jnrC+0];
1261             vdwjidx0D        = 2*vdwtype[jnrD+0];
1262
1263             fjx0             = _mm256_setzero_pd();
1264             fjy0             = _mm256_setzero_pd();
1265             fjz0             = _mm256_setzero_pd();
1266
1267             /**************************
1268              * CALCULATE INTERACTIONS *
1269              **************************/
1270
1271             r00              = _mm256_mul_pd(rsq00,rinv00);
1272             r00              = _mm256_andnot_pd(dummy_mask,r00);
1273
1274             /* Compute parameters for interactions between i and j atoms */
1275             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1276                                             vdwioffsetptr0+vdwjidx0B,
1277                                             vdwioffsetptr0+vdwjidx0C,
1278                                             vdwioffsetptr0+vdwjidx0D,
1279                                             &c6_00,&c12_00);
1280
1281             /* Calculate table index by multiplying r with table scale and truncate to integer */
1282             rt               = _mm256_mul_pd(r00,vftabscale);
1283             vfitab           = _mm256_cvttpd_epi32(rt);
1284             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1285             vfitab           = _mm_slli_epi32(vfitab,3);
1286
1287             /* CUBIC SPLINE TABLE DISPERSION */
1288             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1289             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1290             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1291             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1292             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1293             Heps             = _mm256_mul_pd(vfeps,H);
1294             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1295             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1296             fvdw6            = _mm256_mul_pd(c6_00,FF);
1297
1298             /* CUBIC SPLINE TABLE REPULSION */
1299             vfitab           = _mm_add_epi32(vfitab,ifour);
1300             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1301             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1302             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1303             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1304             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1305             Heps             = _mm256_mul_pd(vfeps,H);
1306             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1307             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1308             fvdw12           = _mm256_mul_pd(c12_00,FF);
1309             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1310
1311             fscal            = fvdw;
1312
1313             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1314
1315             /* Calculate temporary vectorial force */
1316             tx               = _mm256_mul_pd(fscal,dx00);
1317             ty               = _mm256_mul_pd(fscal,dy00);
1318             tz               = _mm256_mul_pd(fscal,dz00);
1319
1320             /* Update vectorial force */
1321             fix0             = _mm256_add_pd(fix0,tx);
1322             fiy0             = _mm256_add_pd(fiy0,ty);
1323             fiz0             = _mm256_add_pd(fiz0,tz);
1324
1325             fjx0             = _mm256_add_pd(fjx0,tx);
1326             fjy0             = _mm256_add_pd(fjy0,ty);
1327             fjz0             = _mm256_add_pd(fjz0,tz);
1328
1329             /**************************
1330              * CALCULATE INTERACTIONS *
1331              **************************/
1332
1333             r10              = _mm256_mul_pd(rsq10,rinv10);
1334             r10              = _mm256_andnot_pd(dummy_mask,r10);
1335
1336             /* Compute parameters for interactions between i and j atoms */
1337             qq10             = _mm256_mul_pd(iq1,jq0);
1338
1339             /* EWALD ELECTROSTATICS */
1340
1341             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1342             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1343             ewitab           = _mm256_cvttpd_epi32(ewrt);
1344             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1345             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1346                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1347                                             &ewtabF,&ewtabFn);
1348             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1349             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1350
1351             fscal            = felec;
1352
1353             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1354
1355             /* Calculate temporary vectorial force */
1356             tx               = _mm256_mul_pd(fscal,dx10);
1357             ty               = _mm256_mul_pd(fscal,dy10);
1358             tz               = _mm256_mul_pd(fscal,dz10);
1359
1360             /* Update vectorial force */
1361             fix1             = _mm256_add_pd(fix1,tx);
1362             fiy1             = _mm256_add_pd(fiy1,ty);
1363             fiz1             = _mm256_add_pd(fiz1,tz);
1364
1365             fjx0             = _mm256_add_pd(fjx0,tx);
1366             fjy0             = _mm256_add_pd(fjy0,ty);
1367             fjz0             = _mm256_add_pd(fjz0,tz);
1368
1369             /**************************
1370              * CALCULATE INTERACTIONS *
1371              **************************/
1372
1373             r20              = _mm256_mul_pd(rsq20,rinv20);
1374             r20              = _mm256_andnot_pd(dummy_mask,r20);
1375
1376             /* Compute parameters for interactions between i and j atoms */
1377             qq20             = _mm256_mul_pd(iq2,jq0);
1378
1379             /* EWALD ELECTROSTATICS */
1380
1381             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1382             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1383             ewitab           = _mm256_cvttpd_epi32(ewrt);
1384             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1385             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1386                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1387                                             &ewtabF,&ewtabFn);
1388             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1389             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1390
1391             fscal            = felec;
1392
1393             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1394
1395             /* Calculate temporary vectorial force */
1396             tx               = _mm256_mul_pd(fscal,dx20);
1397             ty               = _mm256_mul_pd(fscal,dy20);
1398             tz               = _mm256_mul_pd(fscal,dz20);
1399
1400             /* Update vectorial force */
1401             fix2             = _mm256_add_pd(fix2,tx);
1402             fiy2             = _mm256_add_pd(fiy2,ty);
1403             fiz2             = _mm256_add_pd(fiz2,tz);
1404
1405             fjx0             = _mm256_add_pd(fjx0,tx);
1406             fjy0             = _mm256_add_pd(fjy0,ty);
1407             fjz0             = _mm256_add_pd(fjz0,tz);
1408
1409             /**************************
1410              * CALCULATE INTERACTIONS *
1411              **************************/
1412
1413             r30              = _mm256_mul_pd(rsq30,rinv30);
1414             r30              = _mm256_andnot_pd(dummy_mask,r30);
1415
1416             /* Compute parameters for interactions between i and j atoms */
1417             qq30             = _mm256_mul_pd(iq3,jq0);
1418
1419             /* EWALD ELECTROSTATICS */
1420
1421             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1422             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1423             ewitab           = _mm256_cvttpd_epi32(ewrt);
1424             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1425             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1426                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1427                                             &ewtabF,&ewtabFn);
1428             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1429             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1430
1431             fscal            = felec;
1432
1433             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1434
1435             /* Calculate temporary vectorial force */
1436             tx               = _mm256_mul_pd(fscal,dx30);
1437             ty               = _mm256_mul_pd(fscal,dy30);
1438             tz               = _mm256_mul_pd(fscal,dz30);
1439
1440             /* Update vectorial force */
1441             fix3             = _mm256_add_pd(fix3,tx);
1442             fiy3             = _mm256_add_pd(fiy3,ty);
1443             fiz3             = _mm256_add_pd(fiz3,tz);
1444
1445             fjx0             = _mm256_add_pd(fjx0,tx);
1446             fjy0             = _mm256_add_pd(fjy0,ty);
1447             fjz0             = _mm256_add_pd(fjz0,tz);
1448
1449             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1450             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1451             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1452             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1453
1454             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1455
1456             /* Inner loop uses 163 flops */
1457         }
1458
1459         /* End of innermost loop */
1460
1461         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1462                                                  f+i_coord_offset,fshift+i_shift_offset);
1463
1464         /* Increment number of inner iterations */
1465         inneriter                  += j_index_end - j_index_start;
1466
1467         /* Outer loop uses 24 flops */
1468     }
1469
1470     /* Increment number of outer iterations */
1471     outeriter        += nri;
1472
1473     /* Update outer/inner flops */
1474
1475     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*163);
1476 }