src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m128i          vfitab;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m256d          dummy_mask,cutoff_mask;
 111     __m128           tmpmask0,tmpmask1;
 112     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 113     __m256d          one     = _mm256_set1_pd(1.0);
 114     __m256d          two     = _mm256_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_elec_vdw->data;
 133     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 138     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 139     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 140     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm256_setzero_pd();
 176         fiy0             = _mm256_setzero_pd();
 177         fiz0             = _mm256_setzero_pd();
 178         fix1             = _mm256_setzero_pd();
 179         fiy1             = _mm256_setzero_pd();
 180         fiz1             = _mm256_setzero_pd();
 181         fix2             = _mm256_setzero_pd();
 182         fiy2             = _mm256_setzero_pd();
 183         fiz2             = _mm256_setzero_pd();
 184         fix3             = _mm256_setzero_pd();
 185         fiy3             = _mm256_setzero_pd();
 186         fiz3             = _mm256_setzero_pd();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_pd();
 190         vvdwsum          = _mm256_setzero_pd();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm256_sub_pd(ix0,jx0);
 213             dy00             = _mm256_sub_pd(iy0,jy0);
 214             dz00             = _mm256_sub_pd(iz0,jz0);
 215             dx10             = _mm256_sub_pd(ix1,jx0);
 216             dy10             = _mm256_sub_pd(iy1,jy0);
 217             dz10             = _mm256_sub_pd(iz1,jz0);
 218             dx20             = _mm256_sub_pd(ix2,jx0);
 219             dy20             = _mm256_sub_pd(iy2,jy0);
 220             dz20             = _mm256_sub_pd(iz2,jz0);
 221             dx30             = _mm256_sub_pd(ix3,jx0);
 222             dy30             = _mm256_sub_pd(iy3,jy0);
 223             dz30             = _mm256_sub_pd(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 227             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 228             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 229             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 230
 231             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 232             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 233             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 234             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 238                                                                  charge+jnrC+0,charge+jnrD+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243
 244             fjx0             = _mm256_setzero_pd();
 245             fjy0             = _mm256_setzero_pd();
 246             fjz0             = _mm256_setzero_pd();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm256_mul_pd(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 256                                             vdwioffsetptr0+vdwjidx0B,
 257                                             vdwioffsetptr0+vdwjidx0C,
 258                                             vdwioffsetptr0+vdwjidx0D,
 259                                             &c6_00,&c12_00);
 260
 261             /* Calculate table index by multiplying r with table scale and truncate to integer */
 262             rt               = _mm256_mul_pd(r00,vftabscale);
 263             vfitab           = _mm256_cvttpd_epi32(rt);
 264             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 265             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 266
 267             /* CUBIC SPLINE TABLE DISPERSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 270             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 271             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 272             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 273             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 274             Heps             = _mm256_mul_pd(vfeps,H);
 275             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 276             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 277             vvdw6            = _mm256_mul_pd(c6_00,VV);
 278             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 279             fvdw6            = _mm256_mul_pd(c6_00,FF);
 280
 281             /* CUBIC SPLINE TABLE REPULSION */
 282             vfitab           = _mm_add_epi32(vfitab,ifour);
 283             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 284             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 285             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 286             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 287             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 288             Heps             = _mm256_mul_pd(vfeps,H);
 289             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 290             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 291             vvdw12           = _mm256_mul_pd(c12_00,VV);
 292             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 293             fvdw12           = _mm256_mul_pd(c12_00,FF);
 294             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 295             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 299
 300             fscal            = fvdw;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm256_mul_pd(fscal,dx00);
 304             ty               = _mm256_mul_pd(fscal,dy00);
 305             tz               = _mm256_mul_pd(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm256_add_pd(fix0,tx);
 309             fiy0             = _mm256_add_pd(fiy0,ty);
 310             fiz0             = _mm256_add_pd(fiz0,tz);
 311
 312             fjx0             = _mm256_add_pd(fjx0,tx);
 313             fjy0             = _mm256_add_pd(fjy0,ty);
 314             fjz0             = _mm256_add_pd(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r10              = _mm256_mul_pd(rsq10,rinv10);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm256_mul_pd(iq1,jq0);
 324
 325             /* Calculate table index by multiplying r with table scale and truncate to integer */
 326             rt               = _mm256_mul_pd(r10,vftabscale);
 327             vfitab           = _mm256_cvttpd_epi32(rt);
 328             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 329             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 330
 331             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 332             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 333             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 334             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 335             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 336             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 337             Heps             = _mm256_mul_pd(vfeps,H);
 338             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 339             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 340             velec            = _mm256_mul_pd(qq10,VV);
 341             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 342             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_pd(fscal,dx10);
 351             ty               = _mm256_mul_pd(fscal,dy10);
 352             tz               = _mm256_mul_pd(fscal,dz10);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm256_add_pd(fix1,tx);
 356             fiy1             = _mm256_add_pd(fiy1,ty);
 357             fiz1             = _mm256_add_pd(fiz1,tz);
 358
 359             fjx0             = _mm256_add_pd(fjx0,tx);
 360             fjy0             = _mm256_add_pd(fjy0,ty);
 361             fjz0             = _mm256_add_pd(fjz0,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r20              = _mm256_mul_pd(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm256_mul_pd(iq2,jq0);
 371
 372             /* Calculate table index by multiplying r with table scale and truncate to integer */
 373             rt               = _mm256_mul_pd(r20,vftabscale);
 374             vfitab           = _mm256_cvttpd_epi32(rt);
 375             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 376             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 377
 378             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 379             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 380             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 381             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 382             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 383             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 384             Heps             = _mm256_mul_pd(vfeps,H);
 385             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 386             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 387             velec            = _mm256_mul_pd(qq20,VV);
 388             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 389             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm256_add_pd(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm256_mul_pd(fscal,dx20);
 398             ty               = _mm256_mul_pd(fscal,dy20);
 399             tz               = _mm256_mul_pd(fscal,dz20);
 400
 401             /* Update vectorial force */
 402             fix2             = _mm256_add_pd(fix2,tx);
 403             fiy2             = _mm256_add_pd(fiy2,ty);
 404             fiz2             = _mm256_add_pd(fiz2,tz);
 405
 406             fjx0             = _mm256_add_pd(fjx0,tx);
 407             fjy0             = _mm256_add_pd(fjy0,ty);
 408             fjz0             = _mm256_add_pd(fjz0,tz);
 409
 410             /**************************
 411              * CALCULATE INTERACTIONS *
 412              **************************/
 413
 414             r30              = _mm256_mul_pd(rsq30,rinv30);
 415
 416             /* Compute parameters for interactions between i and j atoms */
 417             qq30             = _mm256_mul_pd(iq3,jq0);
 418
 419             /* Calculate table index by multiplying r with table scale and truncate to integer */
 420             rt               = _mm256_mul_pd(r30,vftabscale);
 421             vfitab           = _mm256_cvttpd_epi32(rt);
 422             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 423             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 424
 425             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 426             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 427             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 428             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 429             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 430             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 431             Heps             = _mm256_mul_pd(vfeps,H);
 432             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 433             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 434             velec            = _mm256_mul_pd(qq30,VV);
 435             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 436             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velecsum         = _mm256_add_pd(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             /* Calculate temporary vectorial force */
 444             tx               = _mm256_mul_pd(fscal,dx30);
 445             ty               = _mm256_mul_pd(fscal,dy30);
 446             tz               = _mm256_mul_pd(fscal,dz30);
 447
 448             /* Update vectorial force */
 449             fix3             = _mm256_add_pd(fix3,tx);
 450             fiy3             = _mm256_add_pd(fiy3,ty);
 451             fiz3             = _mm256_add_pd(fiz3,tz);
 452
 453             fjx0             = _mm256_add_pd(fjx0,tx);
 454             fjy0             = _mm256_add_pd(fjy0,ty);
 455             fjz0             = _mm256_add_pd(fjz0,tz);
 456
 457             fjptrA             = f+j_coord_offsetA;
 458             fjptrB             = f+j_coord_offsetB;
 459             fjptrC             = f+j_coord_offsetC;
 460             fjptrD             = f+j_coord_offsetD;
 461
 462             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 463
 464             /* Inner loop uses 188 flops */
 465         }
 466
 467         if(jidx<j_index_end)
 468         {
 469
 470             /* Get j neighbor index, and coordinate index */
 471             jnrlistA         = jjnr[jidx];
 472             jnrlistB         = jjnr[jidx+1];
 473             jnrlistC         = jjnr[jidx+2];
 474             jnrlistD         = jjnr[jidx+3];
 475             /* Sign of each element will be negative for non-real atoms.
 476              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 477              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 478              */
 479             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 480
 481             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 482             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 483             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 484
 485             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 486             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 487             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 488             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 489             j_coord_offsetA  = DIM*jnrA;
 490             j_coord_offsetB  = DIM*jnrB;
 491             j_coord_offsetC  = DIM*jnrC;
 492             j_coord_offsetD  = DIM*jnrD;
 493
 494             /* load j atom coordinates */
 495             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 496                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 497                                                  &jx0,&jy0,&jz0);
 498
 499             /* Calculate displacement vector */
 500             dx00             = _mm256_sub_pd(ix0,jx0);
 501             dy00             = _mm256_sub_pd(iy0,jy0);
 502             dz00             = _mm256_sub_pd(iz0,jz0);
 503             dx10             = _mm256_sub_pd(ix1,jx0);
 504             dy10             = _mm256_sub_pd(iy1,jy0);
 505             dz10             = _mm256_sub_pd(iz1,jz0);
 506             dx20             = _mm256_sub_pd(ix2,jx0);
 507             dy20             = _mm256_sub_pd(iy2,jy0);
 508             dz20             = _mm256_sub_pd(iz2,jz0);
 509             dx30             = _mm256_sub_pd(ix3,jx0);
 510             dy30             = _mm256_sub_pd(iy3,jy0);
 511             dz30             = _mm256_sub_pd(iz3,jz0);
 512
 513             /* Calculate squared distance and things based on it */
 514             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 515             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 516             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 517             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 518
 519             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 520             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 521             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 522             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 523
 524             /* Load parameters for j particles */
 525             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 526                                                                  charge+jnrC+0,charge+jnrD+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529             vdwjidx0C        = 2*vdwtype[jnrC+0];
 530             vdwjidx0D        = 2*vdwtype[jnrD+0];
 531
 532             fjx0             = _mm256_setzero_pd();
 533             fjy0             = _mm256_setzero_pd();
 534             fjz0             = _mm256_setzero_pd();
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             r00              = _mm256_mul_pd(rsq00,rinv00);
 541             r00              = _mm256_andnot_pd(dummy_mask,r00);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 545                                             vdwioffsetptr0+vdwjidx0B,
 546                                             vdwioffsetptr0+vdwjidx0C,
 547                                             vdwioffsetptr0+vdwjidx0D,
 548                                             &c6_00,&c12_00);
 549
 550             /* Calculate table index by multiplying r with table scale and truncate to integer */
 551             rt               = _mm256_mul_pd(r00,vftabscale);
 552             vfitab           = _mm256_cvttpd_epi32(rt);
 553             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 554             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 555
 556             /* CUBIC SPLINE TABLE DISPERSION */
 557             vfitab           = _mm_add_epi32(vfitab,ifour);
 558             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 559             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 560             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 561             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 562             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 563             Heps             = _mm256_mul_pd(vfeps,H);
 564             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 565             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 566             vvdw6            = _mm256_mul_pd(c6_00,VV);
 567             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 568             fvdw6            = _mm256_mul_pd(c6_00,FF);
 569
 570             /* CUBIC SPLINE TABLE REPULSION */
 571             vfitab           = _mm_add_epi32(vfitab,ifour);
 572             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 573             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 574             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 575             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 576             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 577             Heps             = _mm256_mul_pd(vfeps,H);
 578             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 579             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 580             vvdw12           = _mm256_mul_pd(c12_00,VV);
 581             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 582             fvdw12           = _mm256_mul_pd(c12_00,FF);
 583             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 584             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 588             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 589
 590             fscal            = fvdw;
 591
 592             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm256_mul_pd(fscal,dx00);
 596             ty               = _mm256_mul_pd(fscal,dy00);
 597             tz               = _mm256_mul_pd(fscal,dz00);
 598
 599             /* Update vectorial force */
 600             fix0             = _mm256_add_pd(fix0,tx);
 601             fiy0             = _mm256_add_pd(fiy0,ty);
 602             fiz0             = _mm256_add_pd(fiz0,tz);
 603
 604             fjx0             = _mm256_add_pd(fjx0,tx);
 605             fjy0             = _mm256_add_pd(fjy0,ty);
 606             fjz0             = _mm256_add_pd(fjz0,tz);
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             r10              = _mm256_mul_pd(rsq10,rinv10);
 613             r10              = _mm256_andnot_pd(dummy_mask,r10);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq10             = _mm256_mul_pd(iq1,jq0);
 617
 618             /* Calculate table index by multiplying r with table scale and truncate to integer */
 619             rt               = _mm256_mul_pd(r10,vftabscale);
 620             vfitab           = _mm256_cvttpd_epi32(rt);
 621             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 622             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 623
 624             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 625             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 626             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 627             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 628             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 629             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 630             Heps             = _mm256_mul_pd(vfeps,H);
 631             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 632             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 633             velec            = _mm256_mul_pd(qq10,VV);
 634             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 635             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm256_andnot_pd(dummy_mask,velec);
 639             velecsum         = _mm256_add_pd(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm256_mul_pd(fscal,dx10);
 647             ty               = _mm256_mul_pd(fscal,dy10);
 648             tz               = _mm256_mul_pd(fscal,dz10);
 649
 650             /* Update vectorial force */
 651             fix1             = _mm256_add_pd(fix1,tx);
 652             fiy1             = _mm256_add_pd(fiy1,ty);
 653             fiz1             = _mm256_add_pd(fiz1,tz);
 654
 655             fjx0             = _mm256_add_pd(fjx0,tx);
 656             fjy0             = _mm256_add_pd(fjy0,ty);
 657             fjz0             = _mm256_add_pd(fjz0,tz);
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             r20              = _mm256_mul_pd(rsq20,rinv20);
 664             r20              = _mm256_andnot_pd(dummy_mask,r20);
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq20             = _mm256_mul_pd(iq2,jq0);
 668
 669             /* Calculate table index by multiplying r with table scale and truncate to integer */
 670             rt               = _mm256_mul_pd(r20,vftabscale);
 671             vfitab           = _mm256_cvttpd_epi32(rt);
 672             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 673             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 674
 675             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 676             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 677             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 678             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 679             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 680             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 681             Heps             = _mm256_mul_pd(vfeps,H);
 682             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 683             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 684             velec            = _mm256_mul_pd(qq20,VV);
 685             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 686             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 687
 688             /* Update potential sum for this i atom from the interaction with this j atom. */
 689             velec            = _mm256_andnot_pd(dummy_mask,velec);
 690             velecsum         = _mm256_add_pd(velecsum,velec);
 691
 692             fscal            = felec;
 693
 694             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 695
 696             /* Calculate temporary vectorial force */
 697             tx               = _mm256_mul_pd(fscal,dx20);
 698             ty               = _mm256_mul_pd(fscal,dy20);
 699             tz               = _mm256_mul_pd(fscal,dz20);
 700
 701             /* Update vectorial force */
 702             fix2             = _mm256_add_pd(fix2,tx);
 703             fiy2             = _mm256_add_pd(fiy2,ty);
 704             fiz2             = _mm256_add_pd(fiz2,tz);
 705
 706             fjx0             = _mm256_add_pd(fjx0,tx);
 707             fjy0             = _mm256_add_pd(fjy0,ty);
 708             fjz0             = _mm256_add_pd(fjz0,tz);
 709
 710             /**************************
 711              * CALCULATE INTERACTIONS *
 712              **************************/
 713
 714             r30              = _mm256_mul_pd(rsq30,rinv30);
 715             r30              = _mm256_andnot_pd(dummy_mask,r30);
 716
 717             /* Compute parameters for interactions between i and j atoms */
 718             qq30             = _mm256_mul_pd(iq3,jq0);
 719
 720             /* Calculate table index by multiplying r with table scale and truncate to integer */
 721             rt               = _mm256_mul_pd(r30,vftabscale);
 722             vfitab           = _mm256_cvttpd_epi32(rt);
 723             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 724             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 725
 726             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 727             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 728             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 729             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 730             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 731             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 732             Heps             = _mm256_mul_pd(vfeps,H);
 733             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 734             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 735             velec            = _mm256_mul_pd(qq30,VV);
 736             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 737             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 738
 739             /* Update potential sum for this i atom from the interaction with this j atom. */
 740             velec            = _mm256_andnot_pd(dummy_mask,velec);
 741             velecsum         = _mm256_add_pd(velecsum,velec);
 742
 743             fscal            = felec;
 744
 745             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 746
 747             /* Calculate temporary vectorial force */
 748             tx               = _mm256_mul_pd(fscal,dx30);
 749             ty               = _mm256_mul_pd(fscal,dy30);
 750             tz               = _mm256_mul_pd(fscal,dz30);
 751
 752             /* Update vectorial force */
 753             fix3             = _mm256_add_pd(fix3,tx);
 754             fiy3             = _mm256_add_pd(fiy3,ty);
 755             fiz3             = _mm256_add_pd(fiz3,tz);
 756
 757             fjx0             = _mm256_add_pd(fjx0,tx);
 758             fjy0             = _mm256_add_pd(fjy0,ty);
 759             fjz0             = _mm256_add_pd(fjz0,tz);
 760
 761             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 762             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 763             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 764             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 765
 766             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 767
 768             /* Inner loop uses 192 flops */
 769         }
 770
 771         /* End of innermost loop */
 772
 773         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 774                                                  f+i_coord_offset,fshift+i_shift_offset);
 775
 776         ggid                        = gid[iidx];
 777         /* Update potential energies */
 778         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 779         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 780
 781         /* Increment number of inner iterations */
 782         inneriter                  += j_index_end - j_index_start;
 783
 784         /* Outer loop uses 26 flops */
 785     }
 786
 787     /* Increment number of outer iterations */
 788     outeriter        += nri;
 789
 790     /* Update outer/inner flops */
 791
 792     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*192);
 793 }
 794 /*
 795  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_double
 796  * Electrostatics interaction: CubicSplineTable
 797  * VdW interaction:            CubicSplineTable
 798  * Geometry:                   Water4-Particle
 799  * Calculate force/pot:        Force
 800  */
 801 void
 802 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_double
 803                     (t_nblist                    * gmx_restrict       nlist,
 804                      rvec                        * gmx_restrict          xx,
 805                      rvec                        * gmx_restrict          ff,
 806                      t_forcerec                  * gmx_restrict          fr,
 807                      t_mdatoms                   * gmx_restrict     mdatoms,
 808                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 809                      t_nrnb                      * gmx_restrict        nrnb)
 810 {
 811     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 812      * just 0 for non-waters.
 813      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 814      * jnr indices corresponding to data put in the four positions in the SIMD register.
 815      */
 816     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 817     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 818     int              jnrA,jnrB,jnrC,jnrD;
 819     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 820     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 821     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 822     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 823     real             rcutoff_scalar;
 824     real             *shiftvec,*fshift,*x,*f;
 825     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 826     real             scratch[4*DIM];
 827     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 828     real *           vdwioffsetptr0;
 829     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 830     real *           vdwioffsetptr1;
 831     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 832     real *           vdwioffsetptr2;
 833     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 834     real *           vdwioffsetptr3;
 835     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 836     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 837     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 838     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 839     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 840     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 841     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 842     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 843     real             *charge;
 844     int              nvdwtype;
 845     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 846     int              *vdwtype;
 847     real             *vdwparam;
 848     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 849     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 850     __m128i          vfitab;
 851     __m128i          ifour       = _mm_set1_epi32(4);
 852     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 853     real             *vftab;
 854     __m256d          dummy_mask,cutoff_mask;
 855     __m128           tmpmask0,tmpmask1;
 856     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 857     __m256d          one     = _mm256_set1_pd(1.0);
 858     __m256d          two     = _mm256_set1_pd(2.0);
 859     x                = xx[0];
 860     f                = ff[0];
 861
 862     nri              = nlist->nri;
 863     iinr             = nlist->iinr;
 864     jindex           = nlist->jindex;
 865     jjnr             = nlist->jjnr;
 866     shiftidx         = nlist->shift;
 867     gid              = nlist->gid;
 868     shiftvec         = fr->shift_vec[0];
 869     fshift           = fr->fshift[0];
 870     facel            = _mm256_set1_pd(fr->epsfac);
 871     charge           = mdatoms->chargeA;
 872     nvdwtype         = fr->ntype;
 873     vdwparam         = fr->nbfp;
 874     vdwtype          = mdatoms->typeA;
 875
 876     vftab            = kernel_data->table_elec_vdw->data;
 877     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 878
 879     /* Setup water-specific parameters */
 880     inr              = nlist->iinr[0];
 881     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 882     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 883     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 884     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 885
 886     /* Avoid stupid compiler warnings */
 887     jnrA = jnrB = jnrC = jnrD = 0;
 888     j_coord_offsetA = 0;
 889     j_coord_offsetB = 0;
 890     j_coord_offsetC = 0;
 891     j_coord_offsetD = 0;
 892
 893     outeriter        = 0;
 894     inneriter        = 0;
 895
 896     for(iidx=0;iidx<4*DIM;iidx++)
 897     {
 898         scratch[iidx] = 0.0;
 899     }
 900
 901     /* Start outer loop over neighborlists */
 902     for(iidx=0; iidx<nri; iidx++)
 903     {
 904         /* Load shift vector for this list */
 905         i_shift_offset   = DIM*shiftidx[iidx];
 906
 907         /* Load limits for loop over neighbors */
 908         j_index_start    = jindex[iidx];
 909         j_index_end      = jindex[iidx+1];
 910
 911         /* Get outer coordinate index */
 912         inr              = iinr[iidx];
 913         i_coord_offset   = DIM*inr;
 914
 915         /* Load i particle coords and add shift vector */
 916         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 917                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 918
 919         fix0             = _mm256_setzero_pd();
 920         fiy0             = _mm256_setzero_pd();
 921         fiz0             = _mm256_setzero_pd();
 922         fix1             = _mm256_setzero_pd();
 923         fiy1             = _mm256_setzero_pd();
 924         fiz1             = _mm256_setzero_pd();
 925         fix2             = _mm256_setzero_pd();
 926         fiy2             = _mm256_setzero_pd();
 927         fiz2             = _mm256_setzero_pd();
 928         fix3             = _mm256_setzero_pd();
 929         fiy3             = _mm256_setzero_pd();
 930         fiz3             = _mm256_setzero_pd();
 931
 932         /* Start inner kernel loop */
 933         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 934         {
 935
 936             /* Get j neighbor index, and coordinate index */
 937             jnrA             = jjnr[jidx];
 938             jnrB             = jjnr[jidx+1];
 939             jnrC             = jjnr[jidx+2];
 940             jnrD             = jjnr[jidx+3];
 941             j_coord_offsetA  = DIM*jnrA;
 942             j_coord_offsetB  = DIM*jnrB;
 943             j_coord_offsetC  = DIM*jnrC;
 944             j_coord_offsetD  = DIM*jnrD;
 945
 946             /* load j atom coordinates */
 947             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 948                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 949                                                  &jx0,&jy0,&jz0);
 950
 951             /* Calculate displacement vector */
 952             dx00             = _mm256_sub_pd(ix0,jx0);
 953             dy00             = _mm256_sub_pd(iy0,jy0);
 954             dz00             = _mm256_sub_pd(iz0,jz0);
 955             dx10             = _mm256_sub_pd(ix1,jx0);
 956             dy10             = _mm256_sub_pd(iy1,jy0);
 957             dz10             = _mm256_sub_pd(iz1,jz0);
 958             dx20             = _mm256_sub_pd(ix2,jx0);
 959             dy20             = _mm256_sub_pd(iy2,jy0);
 960             dz20             = _mm256_sub_pd(iz2,jz0);
 961             dx30             = _mm256_sub_pd(ix3,jx0);
 962             dy30             = _mm256_sub_pd(iy3,jy0);
 963             dz30             = _mm256_sub_pd(iz3,jz0);
 964
 965             /* Calculate squared distance and things based on it */
 966             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 967             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 968             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 969             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 970
 971             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 972             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 973             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 974             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 975
 976             /* Load parameters for j particles */
 977             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 978                                                                  charge+jnrC+0,charge+jnrD+0);
 979             vdwjidx0A        = 2*vdwtype[jnrA+0];
 980             vdwjidx0B        = 2*vdwtype[jnrB+0];
 981             vdwjidx0C        = 2*vdwtype[jnrC+0];
 982             vdwjidx0D        = 2*vdwtype[jnrD+0];
 983
 984             fjx0             = _mm256_setzero_pd();
 985             fjy0             = _mm256_setzero_pd();
 986             fjz0             = _mm256_setzero_pd();
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             r00              = _mm256_mul_pd(rsq00,rinv00);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 996                                             vdwioffsetptr0+vdwjidx0B,
 997                                             vdwioffsetptr0+vdwjidx0C,
 998                                             vdwioffsetptr0+vdwjidx0D,
 999                                             &c6_00,&c12_00);
1000
1001             /* Calculate table index by multiplying r with table scale and truncate to integer */
1002             rt               = _mm256_mul_pd(r00,vftabscale);
1003             vfitab           = _mm256_cvttpd_epi32(rt);
1004             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1005             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1006
1007             /* CUBIC SPLINE TABLE DISPERSION */
1008             vfitab           = _mm_add_epi32(vfitab,ifour);
1009             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1010             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1011             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1012             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1013             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1014             Heps             = _mm256_mul_pd(vfeps,H);
1015             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1016             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1017             fvdw6            = _mm256_mul_pd(c6_00,FF);
1018
1019             /* CUBIC SPLINE TABLE REPULSION */
1020             vfitab           = _mm_add_epi32(vfitab,ifour);
1021             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1022             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1023             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1024             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1025             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1026             Heps             = _mm256_mul_pd(vfeps,H);
1027             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1028             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1029             fvdw12           = _mm256_mul_pd(c12_00,FF);
1030             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1031
1032             fscal            = fvdw;
1033
1034             /* Calculate temporary vectorial force */
1035             tx               = _mm256_mul_pd(fscal,dx00);
1036             ty               = _mm256_mul_pd(fscal,dy00);
1037             tz               = _mm256_mul_pd(fscal,dz00);
1038
1039             /* Update vectorial force */
1040             fix0             = _mm256_add_pd(fix0,tx);
1041             fiy0             = _mm256_add_pd(fiy0,ty);
1042             fiz0             = _mm256_add_pd(fiz0,tz);
1043
1044             fjx0             = _mm256_add_pd(fjx0,tx);
1045             fjy0             = _mm256_add_pd(fjy0,ty);
1046             fjz0             = _mm256_add_pd(fjz0,tz);
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             r10              = _mm256_mul_pd(rsq10,rinv10);
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq10             = _mm256_mul_pd(iq1,jq0);
1056
1057             /* Calculate table index by multiplying r with table scale and truncate to integer */
1058             rt               = _mm256_mul_pd(r10,vftabscale);
1059             vfitab           = _mm256_cvttpd_epi32(rt);
1060             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1061             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1062
1063             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1064             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1065             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1066             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1067             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1068             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1069             Heps             = _mm256_mul_pd(vfeps,H);
1070             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1071             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1072             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1073
1074             fscal            = felec;
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm256_mul_pd(fscal,dx10);
1078             ty               = _mm256_mul_pd(fscal,dy10);
1079             tz               = _mm256_mul_pd(fscal,dz10);
1080
1081             /* Update vectorial force */
1082             fix1             = _mm256_add_pd(fix1,tx);
1083             fiy1             = _mm256_add_pd(fiy1,ty);
1084             fiz1             = _mm256_add_pd(fiz1,tz);
1085
1086             fjx0             = _mm256_add_pd(fjx0,tx);
1087             fjy0             = _mm256_add_pd(fjy0,ty);
1088             fjz0             = _mm256_add_pd(fjz0,tz);
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
1092              **************************/
1093
1094             r20              = _mm256_mul_pd(rsq20,rinv20);
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             qq20             = _mm256_mul_pd(iq2,jq0);
1098
1099             /* Calculate table index by multiplying r with table scale and truncate to integer */
1100             rt               = _mm256_mul_pd(r20,vftabscale);
1101             vfitab           = _mm256_cvttpd_epi32(rt);
1102             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1103             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1104
1105             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1106             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1107             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1108             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1109             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1110             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1111             Heps             = _mm256_mul_pd(vfeps,H);
1112             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1113             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1114             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1115
1116             fscal            = felec;
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm256_mul_pd(fscal,dx20);
1120             ty               = _mm256_mul_pd(fscal,dy20);
1121             tz               = _mm256_mul_pd(fscal,dz20);
1122
1123             /* Update vectorial force */
1124             fix2             = _mm256_add_pd(fix2,tx);
1125             fiy2             = _mm256_add_pd(fiy2,ty);
1126             fiz2             = _mm256_add_pd(fiz2,tz);
1127
1128             fjx0             = _mm256_add_pd(fjx0,tx);
1129             fjy0             = _mm256_add_pd(fjy0,ty);
1130             fjz0             = _mm256_add_pd(fjz0,tz);
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             r30              = _mm256_mul_pd(rsq30,rinv30);
1137
1138             /* Compute parameters for interactions between i and j atoms */
1139             qq30             = _mm256_mul_pd(iq3,jq0);
1140
1141             /* Calculate table index by multiplying r with table scale and truncate to integer */
1142             rt               = _mm256_mul_pd(r30,vftabscale);
1143             vfitab           = _mm256_cvttpd_epi32(rt);
1144             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1145             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1146
1147             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1148             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1149             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1150             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1151             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1152             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1153             Heps             = _mm256_mul_pd(vfeps,H);
1154             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1155             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1156             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1157
1158             fscal            = felec;
1159
1160             /* Calculate temporary vectorial force */
1161             tx               = _mm256_mul_pd(fscal,dx30);
1162             ty               = _mm256_mul_pd(fscal,dy30);
1163             tz               = _mm256_mul_pd(fscal,dz30);
1164
1165             /* Update vectorial force */
1166             fix3             = _mm256_add_pd(fix3,tx);
1167             fiy3             = _mm256_add_pd(fiy3,ty);
1168             fiz3             = _mm256_add_pd(fiz3,tz);
1169
1170             fjx0             = _mm256_add_pd(fjx0,tx);
1171             fjy0             = _mm256_add_pd(fjy0,ty);
1172             fjz0             = _mm256_add_pd(fjz0,tz);
1173
1174             fjptrA             = f+j_coord_offsetA;
1175             fjptrB             = f+j_coord_offsetB;
1176             fjptrC             = f+j_coord_offsetC;
1177             fjptrD             = f+j_coord_offsetD;
1178
1179             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1180
1181             /* Inner loop uses 168 flops */
1182         }
1183
1184         if(jidx<j_index_end)
1185         {
1186
1187             /* Get j neighbor index, and coordinate index */
1188             jnrlistA         = jjnr[jidx];
1189             jnrlistB         = jjnr[jidx+1];
1190             jnrlistC         = jjnr[jidx+2];
1191             jnrlistD         = jjnr[jidx+3];
1192             /* Sign of each element will be negative for non-real atoms.
1193              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1194              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1195              */
1196             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1197
1198             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1199             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1200             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1201
1202             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1203             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1204             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1205             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1206             j_coord_offsetA  = DIM*jnrA;
1207             j_coord_offsetB  = DIM*jnrB;
1208             j_coord_offsetC  = DIM*jnrC;
1209             j_coord_offsetD  = DIM*jnrD;
1210
1211             /* load j atom coordinates */
1212             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1213                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1214                                                  &jx0,&jy0,&jz0);
1215
1216             /* Calculate displacement vector */
1217             dx00             = _mm256_sub_pd(ix0,jx0);
1218             dy00             = _mm256_sub_pd(iy0,jy0);
1219             dz00             = _mm256_sub_pd(iz0,jz0);
1220             dx10             = _mm256_sub_pd(ix1,jx0);
1221             dy10             = _mm256_sub_pd(iy1,jy0);
1222             dz10             = _mm256_sub_pd(iz1,jz0);
1223             dx20             = _mm256_sub_pd(ix2,jx0);
1224             dy20             = _mm256_sub_pd(iy2,jy0);
1225             dz20             = _mm256_sub_pd(iz2,jz0);
1226             dx30             = _mm256_sub_pd(ix3,jx0);
1227             dy30             = _mm256_sub_pd(iy3,jy0);
1228             dz30             = _mm256_sub_pd(iz3,jz0);
1229
1230             /* Calculate squared distance and things based on it */
1231             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1232             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1233             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1234             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1235
1236             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1237             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1238             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1239             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1240
1241             /* Load parameters for j particles */
1242             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1243                                                                  charge+jnrC+0,charge+jnrD+0);
1244             vdwjidx0A        = 2*vdwtype[jnrA+0];
1245             vdwjidx0B        = 2*vdwtype[jnrB+0];
1246             vdwjidx0C        = 2*vdwtype[jnrC+0];
1247             vdwjidx0D        = 2*vdwtype[jnrD+0];
1248
1249             fjx0             = _mm256_setzero_pd();
1250             fjy0             = _mm256_setzero_pd();
1251             fjz0             = _mm256_setzero_pd();
1252
1253             /**************************
1254              * CALCULATE INTERACTIONS *
1255              **************************/
1256
1257             r00              = _mm256_mul_pd(rsq00,rinv00);
1258             r00              = _mm256_andnot_pd(dummy_mask,r00);
1259
1260             /* Compute parameters for interactions between i and j atoms */
1261             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1262                                             vdwioffsetptr0+vdwjidx0B,
1263                                             vdwioffsetptr0+vdwjidx0C,
1264                                             vdwioffsetptr0+vdwjidx0D,
1265                                             &c6_00,&c12_00);
1266
1267             /* Calculate table index by multiplying r with table scale and truncate to integer */
1268             rt               = _mm256_mul_pd(r00,vftabscale);
1269             vfitab           = _mm256_cvttpd_epi32(rt);
1270             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1271             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1272
1273             /* CUBIC SPLINE TABLE DISPERSION */
1274             vfitab           = _mm_add_epi32(vfitab,ifour);
1275             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1276             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1277             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1278             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1279             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1280             Heps             = _mm256_mul_pd(vfeps,H);
1281             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1282             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1283             fvdw6            = _mm256_mul_pd(c6_00,FF);
1284
1285             /* CUBIC SPLINE TABLE REPULSION */
1286             vfitab           = _mm_add_epi32(vfitab,ifour);
1287             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1288             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1289             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1290             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1291             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1292             Heps             = _mm256_mul_pd(vfeps,H);
1293             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1294             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1295             fvdw12           = _mm256_mul_pd(c12_00,FF);
1296             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1297
1298             fscal            = fvdw;
1299
1300             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1301
1302             /* Calculate temporary vectorial force */
1303             tx               = _mm256_mul_pd(fscal,dx00);
1304             ty               = _mm256_mul_pd(fscal,dy00);
1305             tz               = _mm256_mul_pd(fscal,dz00);
1306
1307             /* Update vectorial force */
1308             fix0             = _mm256_add_pd(fix0,tx);
1309             fiy0             = _mm256_add_pd(fiy0,ty);
1310             fiz0             = _mm256_add_pd(fiz0,tz);
1311
1312             fjx0             = _mm256_add_pd(fjx0,tx);
1313             fjy0             = _mm256_add_pd(fjy0,ty);
1314             fjz0             = _mm256_add_pd(fjz0,tz);
1315
1316             /**************************
1317              * CALCULATE INTERACTIONS *
1318              **************************/
1319
1320             r10              = _mm256_mul_pd(rsq10,rinv10);
1321             r10              = _mm256_andnot_pd(dummy_mask,r10);
1322
1323             /* Compute parameters for interactions between i and j atoms */
1324             qq10             = _mm256_mul_pd(iq1,jq0);
1325
1326             /* Calculate table index by multiplying r with table scale and truncate to integer */
1327             rt               = _mm256_mul_pd(r10,vftabscale);
1328             vfitab           = _mm256_cvttpd_epi32(rt);
1329             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1330             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1331
1332             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1333             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1334             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1335             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1336             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1337             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1338             Heps             = _mm256_mul_pd(vfeps,H);
1339             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1340             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1341             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1342
1343             fscal            = felec;
1344
1345             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1346
1347             /* Calculate temporary vectorial force */
1348             tx               = _mm256_mul_pd(fscal,dx10);
1349             ty               = _mm256_mul_pd(fscal,dy10);
1350             tz               = _mm256_mul_pd(fscal,dz10);
1351
1352             /* Update vectorial force */
1353             fix1             = _mm256_add_pd(fix1,tx);
1354             fiy1             = _mm256_add_pd(fiy1,ty);
1355             fiz1             = _mm256_add_pd(fiz1,tz);
1356
1357             fjx0             = _mm256_add_pd(fjx0,tx);
1358             fjy0             = _mm256_add_pd(fjy0,ty);
1359             fjz0             = _mm256_add_pd(fjz0,tz);
1360
1361             /**************************
1362              * CALCULATE INTERACTIONS *
1363              **************************/
1364
1365             r20              = _mm256_mul_pd(rsq20,rinv20);
1366             r20              = _mm256_andnot_pd(dummy_mask,r20);
1367
1368             /* Compute parameters for interactions between i and j atoms */
1369             qq20             = _mm256_mul_pd(iq2,jq0);
1370
1371             /* Calculate table index by multiplying r with table scale and truncate to integer */
1372             rt               = _mm256_mul_pd(r20,vftabscale);
1373             vfitab           = _mm256_cvttpd_epi32(rt);
1374             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1375             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1376
1377             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1378             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1379             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1380             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1381             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1382             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1383             Heps             = _mm256_mul_pd(vfeps,H);
1384             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1385             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1386             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1387
1388             fscal            = felec;
1389
1390             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1391
1392             /* Calculate temporary vectorial force */
1393             tx               = _mm256_mul_pd(fscal,dx20);
1394             ty               = _mm256_mul_pd(fscal,dy20);
1395             tz               = _mm256_mul_pd(fscal,dz20);
1396
1397             /* Update vectorial force */
1398             fix2             = _mm256_add_pd(fix2,tx);
1399             fiy2             = _mm256_add_pd(fiy2,ty);
1400             fiz2             = _mm256_add_pd(fiz2,tz);
1401
1402             fjx0             = _mm256_add_pd(fjx0,tx);
1403             fjy0             = _mm256_add_pd(fjy0,ty);
1404             fjz0             = _mm256_add_pd(fjz0,tz);
1405
1406             /**************************
1407              * CALCULATE INTERACTIONS *
1408              **************************/
1409
1410             r30              = _mm256_mul_pd(rsq30,rinv30);
1411             r30              = _mm256_andnot_pd(dummy_mask,r30);
1412
1413             /* Compute parameters for interactions between i and j atoms */
1414             qq30             = _mm256_mul_pd(iq3,jq0);
1415
1416             /* Calculate table index by multiplying r with table scale and truncate to integer */
1417             rt               = _mm256_mul_pd(r30,vftabscale);
1418             vfitab           = _mm256_cvttpd_epi32(rt);
1419             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1420             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1421
1422             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1423             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1424             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1425             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1426             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1427             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1428             Heps             = _mm256_mul_pd(vfeps,H);
1429             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1430             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1431             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1432
1433             fscal            = felec;
1434
1435             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1436
1437             /* Calculate temporary vectorial force */
1438             tx               = _mm256_mul_pd(fscal,dx30);
1439             ty               = _mm256_mul_pd(fscal,dy30);
1440             tz               = _mm256_mul_pd(fscal,dz30);
1441
1442             /* Update vectorial force */
1443             fix3             = _mm256_add_pd(fix3,tx);
1444             fiy3             = _mm256_add_pd(fiy3,ty);
1445             fiz3             = _mm256_add_pd(fiz3,tz);
1446
1447             fjx0             = _mm256_add_pd(fjx0,tx);
1448             fjy0             = _mm256_add_pd(fjy0,ty);
1449             fjz0             = _mm256_add_pd(fjz0,tz);
1450
1451             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1452             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1453             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1454             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1455
1456             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1457
1458             /* Inner loop uses 172 flops */
1459         }
1460
1461         /* End of innermost loop */
1462
1463         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1464                                                  f+i_coord_offset,fshift+i_shift_offset);
1465
1466         /* Increment number of inner iterations */
1467         inneriter                  += j_index_end - j_index_start;
1468
1469         /* Outer loop uses 24 flops */
1470     }
1471
1472     /* Increment number of outer iterations */
1473     outeriter        += nri;
1474
1475     /* Update outer/inner flops */
1476
1477     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*172);
1478 }