src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m128i          vfitab;
 104     __m128i          ifour       = _mm_set1_epi32(4);
 105     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 106     real             *vftab;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec_vdw->data;
 130     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 135     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 136     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 137     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm256_setzero_pd();
 173         fiy0             = _mm256_setzero_pd();
 174         fiz0             = _mm256_setzero_pd();
 175         fix1             = _mm256_setzero_pd();
 176         fiy1             = _mm256_setzero_pd();
 177         fiz1             = _mm256_setzero_pd();
 178         fix2             = _mm256_setzero_pd();
 179         fiy2             = _mm256_setzero_pd();
 180         fiz2             = _mm256_setzero_pd();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm256_setzero_pd();
 184         vvdwsum          = _mm256_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm256_sub_pd(ix0,jx0);
 207             dy00             = _mm256_sub_pd(iy0,jy0);
 208             dz00             = _mm256_sub_pd(iz0,jz0);
 209             dx10             = _mm256_sub_pd(ix1,jx0);
 210             dy10             = _mm256_sub_pd(iy1,jy0);
 211             dz10             = _mm256_sub_pd(iz1,jz0);
 212             dx20             = _mm256_sub_pd(ix2,jx0);
 213             dy20             = _mm256_sub_pd(iy2,jy0);
 214             dz20             = _mm256_sub_pd(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 218             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 219             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 222             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 223             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 227                                                                  charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             fjx0             = _mm256_setzero_pd();
 234             fjy0             = _mm256_setzero_pd();
 235             fjz0             = _mm256_setzero_pd();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm256_mul_pd(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm256_mul_pd(iq0,jq0);
 245             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 246                                             vdwioffsetptr0+vdwjidx0B,
 247                                             vdwioffsetptr0+vdwjidx0C,
 248                                             vdwioffsetptr0+vdwjidx0D,
 249                                             &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm256_mul_pd(r00,vftabscale);
 253             vfitab           = _mm256_cvttpd_epi32(rt);
 254             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 255             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 256
 257             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 258             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 259             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 260             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 261             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 262             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 263             Heps             = _mm256_mul_pd(vfeps,H);
 264             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 265             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 266             velec            = _mm256_mul_pd(qq00,VV);
 267             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 268             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 269
 270             /* CUBIC SPLINE TABLE DISPERSION */
 271             vfitab           = _mm_add_epi32(vfitab,ifour);
 272             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 273             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 274             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 275             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 276             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 277             Heps             = _mm256_mul_pd(vfeps,H);
 278             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 279             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 280             vvdw6            = _mm256_mul_pd(c6_00,VV);
 281             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 282             fvdw6            = _mm256_mul_pd(c6_00,FF);
 283
 284             /* CUBIC SPLINE TABLE REPULSION */
 285             vfitab           = _mm_add_epi32(vfitab,ifour);
 286             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 287             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 288             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 289             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 290             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 291             Heps             = _mm256_mul_pd(vfeps,H);
 292             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 293             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 294             vvdw12           = _mm256_mul_pd(c12_00,VV);
 295             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 296             fvdw12           = _mm256_mul_pd(c12_00,FF);
 297             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 298             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm256_add_pd(velecsum,velec);
 302             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 303
 304             fscal            = _mm256_add_pd(felec,fvdw);
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm256_mul_pd(fscal,dx00);
 308             ty               = _mm256_mul_pd(fscal,dy00);
 309             tz               = _mm256_mul_pd(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm256_add_pd(fix0,tx);
 313             fiy0             = _mm256_add_pd(fiy0,ty);
 314             fiz0             = _mm256_add_pd(fiz0,tz);
 315
 316             fjx0             = _mm256_add_pd(fjx0,tx);
 317             fjy0             = _mm256_add_pd(fjy0,ty);
 318             fjz0             = _mm256_add_pd(fjz0,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r10              = _mm256_mul_pd(rsq10,rinv10);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm256_mul_pd(iq1,jq0);
 328
 329             /* Calculate table index by multiplying r with table scale and truncate to integer */
 330             rt               = _mm256_mul_pd(r10,vftabscale);
 331             vfitab           = _mm256_cvttpd_epi32(rt);
 332             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 333             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 334
 335             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 336             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 337             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 338             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 339             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 340             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 341             Heps             = _mm256_mul_pd(vfeps,H);
 342             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 343             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 344             velec            = _mm256_mul_pd(qq10,VV);
 345             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 346             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm256_add_pd(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_pd(fscal,dx10);
 355             ty               = _mm256_mul_pd(fscal,dy10);
 356             tz               = _mm256_mul_pd(fscal,dz10);
 357
 358             /* Update vectorial force */
 359             fix1             = _mm256_add_pd(fix1,tx);
 360             fiy1             = _mm256_add_pd(fiy1,ty);
 361             fiz1             = _mm256_add_pd(fiz1,tz);
 362
 363             fjx0             = _mm256_add_pd(fjx0,tx);
 364             fjy0             = _mm256_add_pd(fjy0,ty);
 365             fjz0             = _mm256_add_pd(fjz0,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r20              = _mm256_mul_pd(rsq20,rinv20);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq20             = _mm256_mul_pd(iq2,jq0);
 375
 376             /* Calculate table index by multiplying r with table scale and truncate to integer */
 377             rt               = _mm256_mul_pd(r20,vftabscale);
 378             vfitab           = _mm256_cvttpd_epi32(rt);
 379             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 380             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 381
 382             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 383             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 384             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 385             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 386             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 387             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 388             Heps             = _mm256_mul_pd(vfeps,H);
 389             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 390             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 391             velec            = _mm256_mul_pd(qq20,VV);
 392             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 393             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm256_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm256_mul_pd(fscal,dx20);
 402             ty               = _mm256_mul_pd(fscal,dy20);
 403             tz               = _mm256_mul_pd(fscal,dz20);
 404
 405             /* Update vectorial force */
 406             fix2             = _mm256_add_pd(fix2,tx);
 407             fiy2             = _mm256_add_pd(fiy2,ty);
 408             fiz2             = _mm256_add_pd(fiz2,tz);
 409
 410             fjx0             = _mm256_add_pd(fjx0,tx);
 411             fjy0             = _mm256_add_pd(fjy0,ty);
 412             fjz0             = _mm256_add_pd(fjz0,tz);
 413
 414             fjptrA             = f+j_coord_offsetA;
 415             fjptrB             = f+j_coord_offsetB;
 416             fjptrC             = f+j_coord_offsetC;
 417             fjptrD             = f+j_coord_offsetD;
 418
 419             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 420
 421             /* Inner loop uses 162 flops */
 422         }
 423
 424         if(jidx<j_index_end)
 425         {
 426
 427             /* Get j neighbor index, and coordinate index */
 428             jnrlistA         = jjnr[jidx];
 429             jnrlistB         = jjnr[jidx+1];
 430             jnrlistC         = jjnr[jidx+2];
 431             jnrlistD         = jjnr[jidx+3];
 432             /* Sign of each element will be negative for non-real atoms.
 433              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 434              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 435              */
 436             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 437
 438             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 439             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 440             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 441
 442             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 443             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 444             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 445             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 446             j_coord_offsetA  = DIM*jnrA;
 447             j_coord_offsetB  = DIM*jnrB;
 448             j_coord_offsetC  = DIM*jnrC;
 449             j_coord_offsetD  = DIM*jnrD;
 450
 451             /* load j atom coordinates */
 452             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 453                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 454                                                  &jx0,&jy0,&jz0);
 455
 456             /* Calculate displacement vector */
 457             dx00             = _mm256_sub_pd(ix0,jx0);
 458             dy00             = _mm256_sub_pd(iy0,jy0);
 459             dz00             = _mm256_sub_pd(iz0,jz0);
 460             dx10             = _mm256_sub_pd(ix1,jx0);
 461             dy10             = _mm256_sub_pd(iy1,jy0);
 462             dz10             = _mm256_sub_pd(iz1,jz0);
 463             dx20             = _mm256_sub_pd(ix2,jx0);
 464             dy20             = _mm256_sub_pd(iy2,jy0);
 465             dz20             = _mm256_sub_pd(iz2,jz0);
 466
 467             /* Calculate squared distance and things based on it */
 468             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 469             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 470             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 471
 472             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 473             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 474             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 475
 476             /* Load parameters for j particles */
 477             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 478                                                                  charge+jnrC+0,charge+jnrD+0);
 479             vdwjidx0A        = 2*vdwtype[jnrA+0];
 480             vdwjidx0B        = 2*vdwtype[jnrB+0];
 481             vdwjidx0C        = 2*vdwtype[jnrC+0];
 482             vdwjidx0D        = 2*vdwtype[jnrD+0];
 483
 484             fjx0             = _mm256_setzero_pd();
 485             fjy0             = _mm256_setzero_pd();
 486             fjz0             = _mm256_setzero_pd();
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             r00              = _mm256_mul_pd(rsq00,rinv00);
 493             r00              = _mm256_andnot_pd(dummy_mask,r00);
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq00             = _mm256_mul_pd(iq0,jq0);
 497             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 498                                             vdwioffsetptr0+vdwjidx0B,
 499                                             vdwioffsetptr0+vdwjidx0C,
 500                                             vdwioffsetptr0+vdwjidx0D,
 501                                             &c6_00,&c12_00);
 502
 503             /* Calculate table index by multiplying r with table scale and truncate to integer */
 504             rt               = _mm256_mul_pd(r00,vftabscale);
 505             vfitab           = _mm256_cvttpd_epi32(rt);
 506             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 507             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 508
 509             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 510             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 511             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 512             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 513             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 514             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 515             Heps             = _mm256_mul_pd(vfeps,H);
 516             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 517             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 518             velec            = _mm256_mul_pd(qq00,VV);
 519             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 520             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 521
 522             /* CUBIC SPLINE TABLE DISPERSION */
 523             vfitab           = _mm_add_epi32(vfitab,ifour);
 524             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 525             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 526             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 527             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 528             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 529             Heps             = _mm256_mul_pd(vfeps,H);
 530             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 531             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 532             vvdw6            = _mm256_mul_pd(c6_00,VV);
 533             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 534             fvdw6            = _mm256_mul_pd(c6_00,FF);
 535
 536             /* CUBIC SPLINE TABLE REPULSION */
 537             vfitab           = _mm_add_epi32(vfitab,ifour);
 538             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 539             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 540             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 541             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 542             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 543             Heps             = _mm256_mul_pd(vfeps,H);
 544             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 545             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 546             vvdw12           = _mm256_mul_pd(c12_00,VV);
 547             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 548             fvdw12           = _mm256_mul_pd(c12_00,FF);
 549             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 550             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm256_andnot_pd(dummy_mask,velec);
 554             velecsum         = _mm256_add_pd(velecsum,velec);
 555             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 556             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 557
 558             fscal            = _mm256_add_pd(felec,fvdw);
 559
 560             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 561
 562             /* Calculate temporary vectorial force */
 563             tx               = _mm256_mul_pd(fscal,dx00);
 564             ty               = _mm256_mul_pd(fscal,dy00);
 565             tz               = _mm256_mul_pd(fscal,dz00);
 566
 567             /* Update vectorial force */
 568             fix0             = _mm256_add_pd(fix0,tx);
 569             fiy0             = _mm256_add_pd(fiy0,ty);
 570             fiz0             = _mm256_add_pd(fiz0,tz);
 571
 572             fjx0             = _mm256_add_pd(fjx0,tx);
 573             fjy0             = _mm256_add_pd(fjy0,ty);
 574             fjz0             = _mm256_add_pd(fjz0,tz);
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             r10              = _mm256_mul_pd(rsq10,rinv10);
 581             r10              = _mm256_andnot_pd(dummy_mask,r10);
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq10             = _mm256_mul_pd(iq1,jq0);
 585
 586             /* Calculate table index by multiplying r with table scale and truncate to integer */
 587             rt               = _mm256_mul_pd(r10,vftabscale);
 588             vfitab           = _mm256_cvttpd_epi32(rt);
 589             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 590             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 591
 592             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 593             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 594             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 595             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 596             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 597             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 598             Heps             = _mm256_mul_pd(vfeps,H);
 599             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 600             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 601             velec            = _mm256_mul_pd(qq10,VV);
 602             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 603             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm256_andnot_pd(dummy_mask,velec);
 607             velecsum         = _mm256_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm256_mul_pd(fscal,dx10);
 615             ty               = _mm256_mul_pd(fscal,dy10);
 616             tz               = _mm256_mul_pd(fscal,dz10);
 617
 618             /* Update vectorial force */
 619             fix1             = _mm256_add_pd(fix1,tx);
 620             fiy1             = _mm256_add_pd(fiy1,ty);
 621             fiz1             = _mm256_add_pd(fiz1,tz);
 622
 623             fjx0             = _mm256_add_pd(fjx0,tx);
 624             fjy0             = _mm256_add_pd(fjy0,ty);
 625             fjz0             = _mm256_add_pd(fjz0,tz);
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             r20              = _mm256_mul_pd(rsq20,rinv20);
 632             r20              = _mm256_andnot_pd(dummy_mask,r20);
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq20             = _mm256_mul_pd(iq2,jq0);
 636
 637             /* Calculate table index by multiplying r with table scale and truncate to integer */
 638             rt               = _mm256_mul_pd(r20,vftabscale);
 639             vfitab           = _mm256_cvttpd_epi32(rt);
 640             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 641             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 642
 643             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 644             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 645             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 646             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 647             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 648             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 649             Heps             = _mm256_mul_pd(vfeps,H);
 650             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 651             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 652             velec            = _mm256_mul_pd(qq20,VV);
 653             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 654             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             velec            = _mm256_andnot_pd(dummy_mask,velec);
 658             velecsum         = _mm256_add_pd(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm256_mul_pd(fscal,dx20);
 666             ty               = _mm256_mul_pd(fscal,dy20);
 667             tz               = _mm256_mul_pd(fscal,dz20);
 668
 669             /* Update vectorial force */
 670             fix2             = _mm256_add_pd(fix2,tx);
 671             fiy2             = _mm256_add_pd(fiy2,ty);
 672             fiz2             = _mm256_add_pd(fiz2,tz);
 673
 674             fjx0             = _mm256_add_pd(fjx0,tx);
 675             fjy0             = _mm256_add_pd(fjy0,ty);
 676             fjz0             = _mm256_add_pd(fjz0,tz);
 677
 678             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 679             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 680             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 681             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 682
 683             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 684
 685             /* Inner loop uses 165 flops */
 686         }
 687
 688         /* End of innermost loop */
 689
 690         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 691                                                  f+i_coord_offset,fshift+i_shift_offset);
 692
 693         ggid                        = gid[iidx];
 694         /* Update potential energies */
 695         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 696         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 697
 698         /* Increment number of inner iterations */
 699         inneriter                  += j_index_end - j_index_start;
 700
 701         /* Outer loop uses 20 flops */
 702     }
 703
 704     /* Increment number of outer iterations */
 705     outeriter        += nri;
 706
 707     /* Update outer/inner flops */
 708
 709     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*165);
 710 }
 711 /*
 712  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_double
 713  * Electrostatics interaction: CubicSplineTable
 714  * VdW interaction:            CubicSplineTable
 715  * Geometry:                   Water3-Particle
 716  * Calculate force/pot:        Force
 717  */
 718 void
 719 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_double
 720                     (t_nblist                    * gmx_restrict       nlist,
 721                      rvec                        * gmx_restrict          xx,
 722                      rvec                        * gmx_restrict          ff,
 723                      t_forcerec                  * gmx_restrict          fr,
 724                      t_mdatoms                   * gmx_restrict     mdatoms,
 725                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 726                      t_nrnb                      * gmx_restrict        nrnb)
 727 {
 728     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 729      * just 0 for non-waters.
 730      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 731      * jnr indices corresponding to data put in the four positions in the SIMD register.
 732      */
 733     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 734     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 735     int              jnrA,jnrB,jnrC,jnrD;
 736     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 737     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 738     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 739     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 740     real             rcutoff_scalar;
 741     real             *shiftvec,*fshift,*x,*f;
 742     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 743     real             scratch[4*DIM];
 744     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 745     real *           vdwioffsetptr0;
 746     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 747     real *           vdwioffsetptr1;
 748     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 749     real *           vdwioffsetptr2;
 750     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 751     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 752     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 753     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 754     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 755     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 756     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 757     real             *charge;
 758     int              nvdwtype;
 759     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 760     int              *vdwtype;
 761     real             *vdwparam;
 762     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 763     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 764     __m128i          vfitab;
 765     __m128i          ifour       = _mm_set1_epi32(4);
 766     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 767     real             *vftab;
 768     __m256d          dummy_mask,cutoff_mask;
 769     __m128           tmpmask0,tmpmask1;
 770     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 771     __m256d          one     = _mm256_set1_pd(1.0);
 772     __m256d          two     = _mm256_set1_pd(2.0);
 773     x                = xx[0];
 774     f                = ff[0];
 775
 776     nri              = nlist->nri;
 777     iinr             = nlist->iinr;
 778     jindex           = nlist->jindex;
 779     jjnr             = nlist->jjnr;
 780     shiftidx         = nlist->shift;
 781     gid              = nlist->gid;
 782     shiftvec         = fr->shift_vec[0];
 783     fshift           = fr->fshift[0];
 784     facel            = _mm256_set1_pd(fr->epsfac);
 785     charge           = mdatoms->chargeA;
 786     nvdwtype         = fr->ntype;
 787     vdwparam         = fr->nbfp;
 788     vdwtype          = mdatoms->typeA;
 789
 790     vftab            = kernel_data->table_elec_vdw->data;
 791     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 792
 793     /* Setup water-specific parameters */
 794     inr              = nlist->iinr[0];
 795     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 796     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 797     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 798     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 799
 800     /* Avoid stupid compiler warnings */
 801     jnrA = jnrB = jnrC = jnrD = 0;
 802     j_coord_offsetA = 0;
 803     j_coord_offsetB = 0;
 804     j_coord_offsetC = 0;
 805     j_coord_offsetD = 0;
 806
 807     outeriter        = 0;
 808     inneriter        = 0;
 809
 810     for(iidx=0;iidx<4*DIM;iidx++)
 811     {
 812         scratch[iidx] = 0.0;
 813     }
 814
 815     /* Start outer loop over neighborlists */
 816     for(iidx=0; iidx<nri; iidx++)
 817     {
 818         /* Load shift vector for this list */
 819         i_shift_offset   = DIM*shiftidx[iidx];
 820
 821         /* Load limits for loop over neighbors */
 822         j_index_start    = jindex[iidx];
 823         j_index_end      = jindex[iidx+1];
 824
 825         /* Get outer coordinate index */
 826         inr              = iinr[iidx];
 827         i_coord_offset   = DIM*inr;
 828
 829         /* Load i particle coords and add shift vector */
 830         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 831                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 832
 833         fix0             = _mm256_setzero_pd();
 834         fiy0             = _mm256_setzero_pd();
 835         fiz0             = _mm256_setzero_pd();
 836         fix1             = _mm256_setzero_pd();
 837         fiy1             = _mm256_setzero_pd();
 838         fiz1             = _mm256_setzero_pd();
 839         fix2             = _mm256_setzero_pd();
 840         fiy2             = _mm256_setzero_pd();
 841         fiz2             = _mm256_setzero_pd();
 842
 843         /* Start inner kernel loop */
 844         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 845         {
 846
 847             /* Get j neighbor index, and coordinate index */
 848             jnrA             = jjnr[jidx];
 849             jnrB             = jjnr[jidx+1];
 850             jnrC             = jjnr[jidx+2];
 851             jnrD             = jjnr[jidx+3];
 852             j_coord_offsetA  = DIM*jnrA;
 853             j_coord_offsetB  = DIM*jnrB;
 854             j_coord_offsetC  = DIM*jnrC;
 855             j_coord_offsetD  = DIM*jnrD;
 856
 857             /* load j atom coordinates */
 858             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 859                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 860                                                  &jx0,&jy0,&jz0);
 861
 862             /* Calculate displacement vector */
 863             dx00             = _mm256_sub_pd(ix0,jx0);
 864             dy00             = _mm256_sub_pd(iy0,jy0);
 865             dz00             = _mm256_sub_pd(iz0,jz0);
 866             dx10             = _mm256_sub_pd(ix1,jx0);
 867             dy10             = _mm256_sub_pd(iy1,jy0);
 868             dz10             = _mm256_sub_pd(iz1,jz0);
 869             dx20             = _mm256_sub_pd(ix2,jx0);
 870             dy20             = _mm256_sub_pd(iy2,jy0);
 871             dz20             = _mm256_sub_pd(iz2,jz0);
 872
 873             /* Calculate squared distance and things based on it */
 874             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 875             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 876             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 877
 878             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 879             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 880             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 881
 882             /* Load parameters for j particles */
 883             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 884                                                                  charge+jnrC+0,charge+jnrD+0);
 885             vdwjidx0A        = 2*vdwtype[jnrA+0];
 886             vdwjidx0B        = 2*vdwtype[jnrB+0];
 887             vdwjidx0C        = 2*vdwtype[jnrC+0];
 888             vdwjidx0D        = 2*vdwtype[jnrD+0];
 889
 890             fjx0             = _mm256_setzero_pd();
 891             fjy0             = _mm256_setzero_pd();
 892             fjz0             = _mm256_setzero_pd();
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             r00              = _mm256_mul_pd(rsq00,rinv00);
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq00             = _mm256_mul_pd(iq0,jq0);
 902             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 903                                             vdwioffsetptr0+vdwjidx0B,
 904                                             vdwioffsetptr0+vdwjidx0C,
 905                                             vdwioffsetptr0+vdwjidx0D,
 906                                             &c6_00,&c12_00);
 907
 908             /* Calculate table index by multiplying r with table scale and truncate to integer */
 909             rt               = _mm256_mul_pd(r00,vftabscale);
 910             vfitab           = _mm256_cvttpd_epi32(rt);
 911             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 912             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 913
 914             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 915             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 916             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 917             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 918             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 919             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 920             Heps             = _mm256_mul_pd(vfeps,H);
 921             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 922             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 923             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 924
 925             /* CUBIC SPLINE TABLE DISPERSION */
 926             vfitab           = _mm_add_epi32(vfitab,ifour);
 927             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 928             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 929             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 930             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 931             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 932             Heps             = _mm256_mul_pd(vfeps,H);
 933             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 934             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 935             fvdw6            = _mm256_mul_pd(c6_00,FF);
 936
 937             /* CUBIC SPLINE TABLE REPULSION */
 938             vfitab           = _mm_add_epi32(vfitab,ifour);
 939             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 940             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 941             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 942             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 943             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 944             Heps             = _mm256_mul_pd(vfeps,H);
 945             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 946             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 947             fvdw12           = _mm256_mul_pd(c12_00,FF);
 948             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 949
 950             fscal            = _mm256_add_pd(felec,fvdw);
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm256_mul_pd(fscal,dx00);
 954             ty               = _mm256_mul_pd(fscal,dy00);
 955             tz               = _mm256_mul_pd(fscal,dz00);
 956
 957             /* Update vectorial force */
 958             fix0             = _mm256_add_pd(fix0,tx);
 959             fiy0             = _mm256_add_pd(fiy0,ty);
 960             fiz0             = _mm256_add_pd(fiz0,tz);
 961
 962             fjx0             = _mm256_add_pd(fjx0,tx);
 963             fjy0             = _mm256_add_pd(fjy0,ty);
 964             fjz0             = _mm256_add_pd(fjz0,tz);
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             r10              = _mm256_mul_pd(rsq10,rinv10);
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq10             = _mm256_mul_pd(iq1,jq0);
 974
 975             /* Calculate table index by multiplying r with table scale and truncate to integer */
 976             rt               = _mm256_mul_pd(r10,vftabscale);
 977             vfitab           = _mm256_cvttpd_epi32(rt);
 978             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 979             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 980
 981             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 982             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 983             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 984             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 985             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 986             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 987             Heps             = _mm256_mul_pd(vfeps,H);
 988             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 989             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 990             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 991
 992             fscal            = felec;
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm256_mul_pd(fscal,dx10);
 996             ty               = _mm256_mul_pd(fscal,dy10);
 997             tz               = _mm256_mul_pd(fscal,dz10);
 998
 999             /* Update vectorial force */
1000             fix1             = _mm256_add_pd(fix1,tx);
1001             fiy1             = _mm256_add_pd(fiy1,ty);
1002             fiz1             = _mm256_add_pd(fiz1,tz);
1003
1004             fjx0             = _mm256_add_pd(fjx0,tx);
1005             fjy0             = _mm256_add_pd(fjy0,ty);
1006             fjz0             = _mm256_add_pd(fjz0,tz);
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             r20              = _mm256_mul_pd(rsq20,rinv20);
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq20             = _mm256_mul_pd(iq2,jq0);
1016
1017             /* Calculate table index by multiplying r with table scale and truncate to integer */
1018             rt               = _mm256_mul_pd(r20,vftabscale);
1019             vfitab           = _mm256_cvttpd_epi32(rt);
1020             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1021             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1022
1023             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1024             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1025             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1026             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1027             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1028             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1029             Heps             = _mm256_mul_pd(vfeps,H);
1030             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1031             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1032             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1033
1034             fscal            = felec;
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_pd(fscal,dx20);
1038             ty               = _mm256_mul_pd(fscal,dy20);
1039             tz               = _mm256_mul_pd(fscal,dz20);
1040
1041             /* Update vectorial force */
1042             fix2             = _mm256_add_pd(fix2,tx);
1043             fiy2             = _mm256_add_pd(fiy2,ty);
1044             fiz2             = _mm256_add_pd(fiz2,tz);
1045
1046             fjx0             = _mm256_add_pd(fjx0,tx);
1047             fjy0             = _mm256_add_pd(fjy0,ty);
1048             fjz0             = _mm256_add_pd(fjz0,tz);
1049
1050             fjptrA             = f+j_coord_offsetA;
1051             fjptrB             = f+j_coord_offsetB;
1052             fjptrC             = f+j_coord_offsetC;
1053             fjptrD             = f+j_coord_offsetD;
1054
1055             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1056
1057             /* Inner loop uses 142 flops */
1058         }
1059
1060         if(jidx<j_index_end)
1061         {
1062
1063             /* Get j neighbor index, and coordinate index */
1064             jnrlistA         = jjnr[jidx];
1065             jnrlistB         = jjnr[jidx+1];
1066             jnrlistC         = jjnr[jidx+2];
1067             jnrlistD         = jjnr[jidx+3];
1068             /* Sign of each element will be negative for non-real atoms.
1069              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1070              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1071              */
1072             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1073
1074             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1075             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1076             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1077
1078             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1079             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1080             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1081             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1082             j_coord_offsetA  = DIM*jnrA;
1083             j_coord_offsetB  = DIM*jnrB;
1084             j_coord_offsetC  = DIM*jnrC;
1085             j_coord_offsetD  = DIM*jnrD;
1086
1087             /* load j atom coordinates */
1088             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1089                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1090                                                  &jx0,&jy0,&jz0);
1091
1092             /* Calculate displacement vector */
1093             dx00             = _mm256_sub_pd(ix0,jx0);
1094             dy00             = _mm256_sub_pd(iy0,jy0);
1095             dz00             = _mm256_sub_pd(iz0,jz0);
1096             dx10             = _mm256_sub_pd(ix1,jx0);
1097             dy10             = _mm256_sub_pd(iy1,jy0);
1098             dz10             = _mm256_sub_pd(iz1,jz0);
1099             dx20             = _mm256_sub_pd(ix2,jx0);
1100             dy20             = _mm256_sub_pd(iy2,jy0);
1101             dz20             = _mm256_sub_pd(iz2,jz0);
1102
1103             /* Calculate squared distance and things based on it */
1104             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1105             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1106             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1107
1108             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1109             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1110             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1111
1112             /* Load parameters for j particles */
1113             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1114                                                                  charge+jnrC+0,charge+jnrD+0);
1115             vdwjidx0A        = 2*vdwtype[jnrA+0];
1116             vdwjidx0B        = 2*vdwtype[jnrB+0];
1117             vdwjidx0C        = 2*vdwtype[jnrC+0];
1118             vdwjidx0D        = 2*vdwtype[jnrD+0];
1119
1120             fjx0             = _mm256_setzero_pd();
1121             fjy0             = _mm256_setzero_pd();
1122             fjz0             = _mm256_setzero_pd();
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             r00              = _mm256_mul_pd(rsq00,rinv00);
1129             r00              = _mm256_andnot_pd(dummy_mask,r00);
1130
1131             /* Compute parameters for interactions between i and j atoms */
1132             qq00             = _mm256_mul_pd(iq0,jq0);
1133             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1134                                             vdwioffsetptr0+vdwjidx0B,
1135                                             vdwioffsetptr0+vdwjidx0C,
1136                                             vdwioffsetptr0+vdwjidx0D,
1137                                             &c6_00,&c12_00);
1138
1139             /* Calculate table index by multiplying r with table scale and truncate to integer */
1140             rt               = _mm256_mul_pd(r00,vftabscale);
1141             vfitab           = _mm256_cvttpd_epi32(rt);
1142             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1143             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1144
1145             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1146             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1147             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1148             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1149             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1150             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1151             Heps             = _mm256_mul_pd(vfeps,H);
1152             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1153             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1154             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1155
1156             /* CUBIC SPLINE TABLE DISPERSION */
1157             vfitab           = _mm_add_epi32(vfitab,ifour);
1158             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1159             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1160             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1161             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1162             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1163             Heps             = _mm256_mul_pd(vfeps,H);
1164             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1165             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1166             fvdw6            = _mm256_mul_pd(c6_00,FF);
1167
1168             /* CUBIC SPLINE TABLE REPULSION */
1169             vfitab           = _mm_add_epi32(vfitab,ifour);
1170             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1171             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1172             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1173             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1174             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1175             Heps             = _mm256_mul_pd(vfeps,H);
1176             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1177             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1178             fvdw12           = _mm256_mul_pd(c12_00,FF);
1179             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1180
1181             fscal            = _mm256_add_pd(felec,fvdw);
1182
1183             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm256_mul_pd(fscal,dx00);
1187             ty               = _mm256_mul_pd(fscal,dy00);
1188             tz               = _mm256_mul_pd(fscal,dz00);
1189
1190             /* Update vectorial force */
1191             fix0             = _mm256_add_pd(fix0,tx);
1192             fiy0             = _mm256_add_pd(fiy0,ty);
1193             fiz0             = _mm256_add_pd(fiz0,tz);
1194
1195             fjx0             = _mm256_add_pd(fjx0,tx);
1196             fjy0             = _mm256_add_pd(fjy0,ty);
1197             fjz0             = _mm256_add_pd(fjz0,tz);
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             r10              = _mm256_mul_pd(rsq10,rinv10);
1204             r10              = _mm256_andnot_pd(dummy_mask,r10);
1205
1206             /* Compute parameters for interactions between i and j atoms */
1207             qq10             = _mm256_mul_pd(iq1,jq0);
1208
1209             /* Calculate table index by multiplying r with table scale and truncate to integer */
1210             rt               = _mm256_mul_pd(r10,vftabscale);
1211             vfitab           = _mm256_cvttpd_epi32(rt);
1212             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1213             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1214
1215             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1216             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1217             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1218             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1219             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1220             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1221             Heps             = _mm256_mul_pd(vfeps,H);
1222             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1223             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1224             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1225
1226             fscal            = felec;
1227
1228             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm256_mul_pd(fscal,dx10);
1232             ty               = _mm256_mul_pd(fscal,dy10);
1233             tz               = _mm256_mul_pd(fscal,dz10);
1234
1235             /* Update vectorial force */
1236             fix1             = _mm256_add_pd(fix1,tx);
1237             fiy1             = _mm256_add_pd(fiy1,ty);
1238             fiz1             = _mm256_add_pd(fiz1,tz);
1239
1240             fjx0             = _mm256_add_pd(fjx0,tx);
1241             fjy0             = _mm256_add_pd(fjy0,ty);
1242             fjz0             = _mm256_add_pd(fjz0,tz);
1243
1244             /**************************
1245              * CALCULATE INTERACTIONS *
1246              **************************/
1247
1248             r20              = _mm256_mul_pd(rsq20,rinv20);
1249             r20              = _mm256_andnot_pd(dummy_mask,r20);
1250
1251             /* Compute parameters for interactions between i and j atoms */
1252             qq20             = _mm256_mul_pd(iq2,jq0);
1253
1254             /* Calculate table index by multiplying r with table scale and truncate to integer */
1255             rt               = _mm256_mul_pd(r20,vftabscale);
1256             vfitab           = _mm256_cvttpd_epi32(rt);
1257             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1258             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1259
1260             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1261             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1262             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1263             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1264             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1265             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1266             Heps             = _mm256_mul_pd(vfeps,H);
1267             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1268             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1269             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1270
1271             fscal            = felec;
1272
1273             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1274
1275             /* Calculate temporary vectorial force */
1276             tx               = _mm256_mul_pd(fscal,dx20);
1277             ty               = _mm256_mul_pd(fscal,dy20);
1278             tz               = _mm256_mul_pd(fscal,dz20);
1279
1280             /* Update vectorial force */
1281             fix2             = _mm256_add_pd(fix2,tx);
1282             fiy2             = _mm256_add_pd(fiy2,ty);
1283             fiz2             = _mm256_add_pd(fiz2,tz);
1284
1285             fjx0             = _mm256_add_pd(fjx0,tx);
1286             fjy0             = _mm256_add_pd(fjy0,ty);
1287             fjz0             = _mm256_add_pd(fjz0,tz);
1288
1289             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1290             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1291             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1292             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1293
1294             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1295
1296             /* Inner loop uses 145 flops */
1297         }
1298
1299         /* End of innermost loop */
1300
1301         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1302                                                  f+i_coord_offset,fshift+i_shift_offset);
1303
1304         /* Increment number of inner iterations */
1305         inneriter                  += j_index_end - j_index_start;
1306
1307         /* Outer loop uses 18 flops */
1308     }
1309
1310     /* Increment number of outer iterations */
1311     outeriter        += nri;
1312
1313     /* Update outer/inner flops */
1314
1315     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*145);
1316 }