src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwLJ_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_VdwLJ_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_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->data;
 130     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->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             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 229                                                                  charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm256_setzero_pd();
 236             fjy0             = _mm256_setzero_pd();
 237             fjz0             = _mm256_setzero_pd();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm256_mul_pd(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm256_mul_pd(iq0,jq0);
 247             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 248                                             vdwioffsetptr0+vdwjidx0B,
 249                                             vdwioffsetptr0+vdwjidx0C,
 250                                             vdwioffsetptr0+vdwjidx0D,
 251                                             &c6_00,&c12_00);
 252
 253             /* Calculate table index by multiplying r with table scale and truncate to integer */
 254             rt               = _mm256_mul_pd(r00,vftabscale);
 255             vfitab           = _mm256_cvttpd_epi32(rt);
 256             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 257             vfitab           = _mm_slli_epi32(vfitab,2);
 258
 259             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 260             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 261             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 262             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 263             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 264             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 265             Heps             = _mm256_mul_pd(vfeps,H);
 266             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 267             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 268             velec            = _mm256_mul_pd(qq00,VV);
 269             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 270             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 271
 272             /* LENNARD-JONES DISPERSION/REPULSION */
 273
 274             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 275             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 276             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 277             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 278             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm256_add_pd(velecsum,velec);
 282             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 283
 284             fscal            = _mm256_add_pd(felec,fvdw);
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm256_mul_pd(fscal,dx00);
 288             ty               = _mm256_mul_pd(fscal,dy00);
 289             tz               = _mm256_mul_pd(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm256_add_pd(fix0,tx);
 293             fiy0             = _mm256_add_pd(fiy0,ty);
 294             fiz0             = _mm256_add_pd(fiz0,tz);
 295
 296             fjx0             = _mm256_add_pd(fjx0,tx);
 297             fjy0             = _mm256_add_pd(fjy0,ty);
 298             fjz0             = _mm256_add_pd(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r10              = _mm256_mul_pd(rsq10,rinv10);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _mm256_mul_pd(iq1,jq0);
 308
 309             /* Calculate table index by multiplying r with table scale and truncate to integer */
 310             rt               = _mm256_mul_pd(r10,vftabscale);
 311             vfitab           = _mm256_cvttpd_epi32(rt);
 312             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 313             vfitab           = _mm_slli_epi32(vfitab,2);
 314
 315             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 316             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 317             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 318             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 319             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 320             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 321             Heps             = _mm256_mul_pd(vfeps,H);
 322             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 323             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 324             velec            = _mm256_mul_pd(qq10,VV);
 325             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 326             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm256_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm256_mul_pd(fscal,dx10);
 335             ty               = _mm256_mul_pd(fscal,dy10);
 336             tz               = _mm256_mul_pd(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm256_add_pd(fix1,tx);
 340             fiy1             = _mm256_add_pd(fiy1,ty);
 341             fiz1             = _mm256_add_pd(fiz1,tz);
 342
 343             fjx0             = _mm256_add_pd(fjx0,tx);
 344             fjy0             = _mm256_add_pd(fjy0,ty);
 345             fjz0             = _mm256_add_pd(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r20              = _mm256_mul_pd(rsq20,rinv20);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm256_mul_pd(iq2,jq0);
 355
 356             /* Calculate table index by multiplying r with table scale and truncate to integer */
 357             rt               = _mm256_mul_pd(r20,vftabscale);
 358             vfitab           = _mm256_cvttpd_epi32(rt);
 359             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 360             vfitab           = _mm_slli_epi32(vfitab,2);
 361
 362             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 363             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 364             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 365             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 366             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 367             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 368             Heps             = _mm256_mul_pd(vfeps,H);
 369             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 370             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 371             velec            = _mm256_mul_pd(qq20,VV);
 372             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 373             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velecsum         = _mm256_add_pd(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm256_mul_pd(fscal,dx20);
 382             ty               = _mm256_mul_pd(fscal,dy20);
 383             tz               = _mm256_mul_pd(fscal,dz20);
 384
 385             /* Update vectorial force */
 386             fix2             = _mm256_add_pd(fix2,tx);
 387             fiy2             = _mm256_add_pd(fiy2,ty);
 388             fiz2             = _mm256_add_pd(fiz2,tz);
 389
 390             fjx0             = _mm256_add_pd(fjx0,tx);
 391             fjy0             = _mm256_add_pd(fjy0,ty);
 392             fjz0             = _mm256_add_pd(fjz0,tz);
 393
 394             fjptrA             = f+j_coord_offsetA;
 395             fjptrB             = f+j_coord_offsetB;
 396             fjptrC             = f+j_coord_offsetC;
 397             fjptrD             = f+j_coord_offsetD;
 398
 399             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 400
 401             /* Inner loop uses 145 flops */
 402         }
 403
 404         if(jidx<j_index_end)
 405         {
 406
 407             /* Get j neighbor index, and coordinate index */
 408             jnrlistA         = jjnr[jidx];
 409             jnrlistB         = jjnr[jidx+1];
 410             jnrlistC         = jjnr[jidx+2];
 411             jnrlistD         = jjnr[jidx+3];
 412             /* Sign of each element will be negative for non-real atoms.
 413              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 414              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 415              */
 416             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 417
 418             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 419             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 420             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 421
 422             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 423             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 424             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 425             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 426             j_coord_offsetA  = DIM*jnrA;
 427             j_coord_offsetB  = DIM*jnrB;
 428             j_coord_offsetC  = DIM*jnrC;
 429             j_coord_offsetD  = DIM*jnrD;
 430
 431             /* load j atom coordinates */
 432             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 433                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 434                                                  &jx0,&jy0,&jz0);
 435
 436             /* Calculate displacement vector */
 437             dx00             = _mm256_sub_pd(ix0,jx0);
 438             dy00             = _mm256_sub_pd(iy0,jy0);
 439             dz00             = _mm256_sub_pd(iz0,jz0);
 440             dx10             = _mm256_sub_pd(ix1,jx0);
 441             dy10             = _mm256_sub_pd(iy1,jy0);
 442             dz10             = _mm256_sub_pd(iz1,jz0);
 443             dx20             = _mm256_sub_pd(ix2,jx0);
 444             dy20             = _mm256_sub_pd(iy2,jy0);
 445             dz20             = _mm256_sub_pd(iz2,jz0);
 446
 447             /* Calculate squared distance and things based on it */
 448             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 449             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 450             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 451
 452             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 453             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 454             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 455
 456             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 457
 458             /* Load parameters for j particles */
 459             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 460                                                                  charge+jnrC+0,charge+jnrD+0);
 461             vdwjidx0A        = 2*vdwtype[jnrA+0];
 462             vdwjidx0B        = 2*vdwtype[jnrB+0];
 463             vdwjidx0C        = 2*vdwtype[jnrC+0];
 464             vdwjidx0D        = 2*vdwtype[jnrD+0];
 465
 466             fjx0             = _mm256_setzero_pd();
 467             fjy0             = _mm256_setzero_pd();
 468             fjz0             = _mm256_setzero_pd();
 469
 470             /**************************
 471              * CALCULATE INTERACTIONS *
 472              **************************/
 473
 474             r00              = _mm256_mul_pd(rsq00,rinv00);
 475             r00              = _mm256_andnot_pd(dummy_mask,r00);
 476
 477             /* Compute parameters for interactions between i and j atoms */
 478             qq00             = _mm256_mul_pd(iq0,jq0);
 479             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 480                                             vdwioffsetptr0+vdwjidx0B,
 481                                             vdwioffsetptr0+vdwjidx0C,
 482                                             vdwioffsetptr0+vdwjidx0D,
 483                                             &c6_00,&c12_00);
 484
 485             /* Calculate table index by multiplying r with table scale and truncate to integer */
 486             rt               = _mm256_mul_pd(r00,vftabscale);
 487             vfitab           = _mm256_cvttpd_epi32(rt);
 488             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 489             vfitab           = _mm_slli_epi32(vfitab,2);
 490
 491             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 492             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 493             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 494             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 495             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 496             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 497             Heps             = _mm256_mul_pd(vfeps,H);
 498             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 499             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 500             velec            = _mm256_mul_pd(qq00,VV);
 501             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 502             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 503
 504             /* LENNARD-JONES DISPERSION/REPULSION */
 505
 506             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 507             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 508             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 509             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 510             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velec            = _mm256_andnot_pd(dummy_mask,velec);
 514             velecsum         = _mm256_add_pd(velecsum,velec);
 515             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 516             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 517
 518             fscal            = _mm256_add_pd(felec,fvdw);
 519
 520             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 521
 522             /* Calculate temporary vectorial force */
 523             tx               = _mm256_mul_pd(fscal,dx00);
 524             ty               = _mm256_mul_pd(fscal,dy00);
 525             tz               = _mm256_mul_pd(fscal,dz00);
 526
 527             /* Update vectorial force */
 528             fix0             = _mm256_add_pd(fix0,tx);
 529             fiy0             = _mm256_add_pd(fiy0,ty);
 530             fiz0             = _mm256_add_pd(fiz0,tz);
 531
 532             fjx0             = _mm256_add_pd(fjx0,tx);
 533             fjy0             = _mm256_add_pd(fjy0,ty);
 534             fjz0             = _mm256_add_pd(fjz0,tz);
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             r10              = _mm256_mul_pd(rsq10,rinv10);
 541             r10              = _mm256_andnot_pd(dummy_mask,r10);
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq10             = _mm256_mul_pd(iq1,jq0);
 545
 546             /* Calculate table index by multiplying r with table scale and truncate to integer */
 547             rt               = _mm256_mul_pd(r10,vftabscale);
 548             vfitab           = _mm256_cvttpd_epi32(rt);
 549             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 550             vfitab           = _mm_slli_epi32(vfitab,2);
 551
 552             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 553             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 554             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 555             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 556             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 557             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 558             Heps             = _mm256_mul_pd(vfeps,H);
 559             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 560             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 561             velec            = _mm256_mul_pd(qq10,VV);
 562             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 563             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _mm256_andnot_pd(dummy_mask,velec);
 567             velecsum         = _mm256_add_pd(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm256_mul_pd(fscal,dx10);
 575             ty               = _mm256_mul_pd(fscal,dy10);
 576             tz               = _mm256_mul_pd(fscal,dz10);
 577
 578             /* Update vectorial force */
 579             fix1             = _mm256_add_pd(fix1,tx);
 580             fiy1             = _mm256_add_pd(fiy1,ty);
 581             fiz1             = _mm256_add_pd(fiz1,tz);
 582
 583             fjx0             = _mm256_add_pd(fjx0,tx);
 584             fjy0             = _mm256_add_pd(fjy0,ty);
 585             fjz0             = _mm256_add_pd(fjz0,tz);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r20              = _mm256_mul_pd(rsq20,rinv20);
 592             r20              = _mm256_andnot_pd(dummy_mask,r20);
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq20             = _mm256_mul_pd(iq2,jq0);
 596
 597             /* Calculate table index by multiplying r with table scale and truncate to integer */
 598             rt               = _mm256_mul_pd(r20,vftabscale);
 599             vfitab           = _mm256_cvttpd_epi32(rt);
 600             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 601             vfitab           = _mm_slli_epi32(vfitab,2);
 602
 603             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 604             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 605             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 606             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 607             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 608             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 609             Heps             = _mm256_mul_pd(vfeps,H);
 610             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 611             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 612             velec            = _mm256_mul_pd(qq20,VV);
 613             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 614             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm256_andnot_pd(dummy_mask,velec);
 618             velecsum         = _mm256_add_pd(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm256_mul_pd(fscal,dx20);
 626             ty               = _mm256_mul_pd(fscal,dy20);
 627             tz               = _mm256_mul_pd(fscal,dz20);
 628
 629             /* Update vectorial force */
 630             fix2             = _mm256_add_pd(fix2,tx);
 631             fiy2             = _mm256_add_pd(fiy2,ty);
 632             fiz2             = _mm256_add_pd(fiz2,tz);
 633
 634             fjx0             = _mm256_add_pd(fjx0,tx);
 635             fjy0             = _mm256_add_pd(fjy0,ty);
 636             fjz0             = _mm256_add_pd(fjz0,tz);
 637
 638             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 639             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 640             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 641             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 642
 643             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 644
 645             /* Inner loop uses 148 flops */
 646         }
 647
 648         /* End of innermost loop */
 649
 650         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 651                                                  f+i_coord_offset,fshift+i_shift_offset);
 652
 653         ggid                        = gid[iidx];
 654         /* Update potential energies */
 655         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 656         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 657
 658         /* Increment number of inner iterations */
 659         inneriter                  += j_index_end - j_index_start;
 660
 661         /* Outer loop uses 20 flops */
 662     }
 663
 664     /* Increment number of outer iterations */
 665     outeriter        += nri;
 666
 667     /* Update outer/inner flops */
 668
 669     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 670 }
 671 /*
 672  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_double
 673  * Electrostatics interaction: CubicSplineTable
 674  * VdW interaction:            LennardJones
 675  * Geometry:                   Water3-Particle
 676  * Calculate force/pot:        Force
 677  */
 678 void
 679 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_double
 680                     (t_nblist                    * gmx_restrict       nlist,
 681                      rvec                        * gmx_restrict          xx,
 682                      rvec                        * gmx_restrict          ff,
 683                      t_forcerec                  * gmx_restrict          fr,
 684                      t_mdatoms                   * gmx_restrict     mdatoms,
 685                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 686                      t_nrnb                      * gmx_restrict        nrnb)
 687 {
 688     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 689      * just 0 for non-waters.
 690      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 691      * jnr indices corresponding to data put in the four positions in the SIMD register.
 692      */
 693     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 694     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 695     int              jnrA,jnrB,jnrC,jnrD;
 696     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 697     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 698     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 699     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 700     real             rcutoff_scalar;
 701     real             *shiftvec,*fshift,*x,*f;
 702     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 703     real             scratch[4*DIM];
 704     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 705     real *           vdwioffsetptr0;
 706     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 707     real *           vdwioffsetptr1;
 708     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 709     real *           vdwioffsetptr2;
 710     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 711     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 712     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 713     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 714     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 715     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 716     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 717     real             *charge;
 718     int              nvdwtype;
 719     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 720     int              *vdwtype;
 721     real             *vdwparam;
 722     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 723     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 724     __m128i          vfitab;
 725     __m128i          ifour       = _mm_set1_epi32(4);
 726     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 727     real             *vftab;
 728     __m256d          dummy_mask,cutoff_mask;
 729     __m128           tmpmask0,tmpmask1;
 730     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 731     __m256d          one     = _mm256_set1_pd(1.0);
 732     __m256d          two     = _mm256_set1_pd(2.0);
 733     x                = xx[0];
 734     f                = ff[0];
 735
 736     nri              = nlist->nri;
 737     iinr             = nlist->iinr;
 738     jindex           = nlist->jindex;
 739     jjnr             = nlist->jjnr;
 740     shiftidx         = nlist->shift;
 741     gid              = nlist->gid;
 742     shiftvec         = fr->shift_vec[0];
 743     fshift           = fr->fshift[0];
 744     facel            = _mm256_set1_pd(fr->epsfac);
 745     charge           = mdatoms->chargeA;
 746     nvdwtype         = fr->ntype;
 747     vdwparam         = fr->nbfp;
 748     vdwtype          = mdatoms->typeA;
 749
 750     vftab            = kernel_data->table_elec->data;
 751     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 752
 753     /* Setup water-specific parameters */
 754     inr              = nlist->iinr[0];
 755     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 756     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 757     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 758     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 759
 760     /* Avoid stupid compiler warnings */
 761     jnrA = jnrB = jnrC = jnrD = 0;
 762     j_coord_offsetA = 0;
 763     j_coord_offsetB = 0;
 764     j_coord_offsetC = 0;
 765     j_coord_offsetD = 0;
 766
 767     outeriter        = 0;
 768     inneriter        = 0;
 769
 770     for(iidx=0;iidx<4*DIM;iidx++)
 771     {
 772         scratch[iidx] = 0.0;
 773     }
 774
 775     /* Start outer loop over neighborlists */
 776     for(iidx=0; iidx<nri; iidx++)
 777     {
 778         /* Load shift vector for this list */
 779         i_shift_offset   = DIM*shiftidx[iidx];
 780
 781         /* Load limits for loop over neighbors */
 782         j_index_start    = jindex[iidx];
 783         j_index_end      = jindex[iidx+1];
 784
 785         /* Get outer coordinate index */
 786         inr              = iinr[iidx];
 787         i_coord_offset   = DIM*inr;
 788
 789         /* Load i particle coords and add shift vector */
 790         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 791                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 792
 793         fix0             = _mm256_setzero_pd();
 794         fiy0             = _mm256_setzero_pd();
 795         fiz0             = _mm256_setzero_pd();
 796         fix1             = _mm256_setzero_pd();
 797         fiy1             = _mm256_setzero_pd();
 798         fiz1             = _mm256_setzero_pd();
 799         fix2             = _mm256_setzero_pd();
 800         fiy2             = _mm256_setzero_pd();
 801         fiz2             = _mm256_setzero_pd();
 802
 803         /* Start inner kernel loop */
 804         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 805         {
 806
 807             /* Get j neighbor index, and coordinate index */
 808             jnrA             = jjnr[jidx];
 809             jnrB             = jjnr[jidx+1];
 810             jnrC             = jjnr[jidx+2];
 811             jnrD             = jjnr[jidx+3];
 812             j_coord_offsetA  = DIM*jnrA;
 813             j_coord_offsetB  = DIM*jnrB;
 814             j_coord_offsetC  = DIM*jnrC;
 815             j_coord_offsetD  = DIM*jnrD;
 816
 817             /* load j atom coordinates */
 818             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 819                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 820                                                  &jx0,&jy0,&jz0);
 821
 822             /* Calculate displacement vector */
 823             dx00             = _mm256_sub_pd(ix0,jx0);
 824             dy00             = _mm256_sub_pd(iy0,jy0);
 825             dz00             = _mm256_sub_pd(iz0,jz0);
 826             dx10             = _mm256_sub_pd(ix1,jx0);
 827             dy10             = _mm256_sub_pd(iy1,jy0);
 828             dz10             = _mm256_sub_pd(iz1,jz0);
 829             dx20             = _mm256_sub_pd(ix2,jx0);
 830             dy20             = _mm256_sub_pd(iy2,jy0);
 831             dz20             = _mm256_sub_pd(iz2,jz0);
 832
 833             /* Calculate squared distance and things based on it */
 834             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 835             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 836             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 837
 838             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 839             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 840             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 841
 842             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 843
 844             /* Load parameters for j particles */
 845             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 846                                                                  charge+jnrC+0,charge+jnrD+0);
 847             vdwjidx0A        = 2*vdwtype[jnrA+0];
 848             vdwjidx0B        = 2*vdwtype[jnrB+0];
 849             vdwjidx0C        = 2*vdwtype[jnrC+0];
 850             vdwjidx0D        = 2*vdwtype[jnrD+0];
 851
 852             fjx0             = _mm256_setzero_pd();
 853             fjy0             = _mm256_setzero_pd();
 854             fjz0             = _mm256_setzero_pd();
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             r00              = _mm256_mul_pd(rsq00,rinv00);
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             qq00             = _mm256_mul_pd(iq0,jq0);
 864             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 865                                             vdwioffsetptr0+vdwjidx0B,
 866                                             vdwioffsetptr0+vdwjidx0C,
 867                                             vdwioffsetptr0+vdwjidx0D,
 868                                             &c6_00,&c12_00);
 869
 870             /* Calculate table index by multiplying r with table scale and truncate to integer */
 871             rt               = _mm256_mul_pd(r00,vftabscale);
 872             vfitab           = _mm256_cvttpd_epi32(rt);
 873             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 874             vfitab           = _mm_slli_epi32(vfitab,2);
 875
 876             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 877             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 878             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 879             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 880             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 881             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 882             Heps             = _mm256_mul_pd(vfeps,H);
 883             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 884             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 885             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 886
 887             /* LENNARD-JONES DISPERSION/REPULSION */
 888
 889             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 890             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 891
 892             fscal            = _mm256_add_pd(felec,fvdw);
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm256_mul_pd(fscal,dx00);
 896             ty               = _mm256_mul_pd(fscal,dy00);
 897             tz               = _mm256_mul_pd(fscal,dz00);
 898
 899             /* Update vectorial force */
 900             fix0             = _mm256_add_pd(fix0,tx);
 901             fiy0             = _mm256_add_pd(fiy0,ty);
 902             fiz0             = _mm256_add_pd(fiz0,tz);
 903
 904             fjx0             = _mm256_add_pd(fjx0,tx);
 905             fjy0             = _mm256_add_pd(fjy0,ty);
 906             fjz0             = _mm256_add_pd(fjz0,tz);
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             r10              = _mm256_mul_pd(rsq10,rinv10);
 913
 914             /* Compute parameters for interactions between i and j atoms */
 915             qq10             = _mm256_mul_pd(iq1,jq0);
 916
 917             /* Calculate table index by multiplying r with table scale and truncate to integer */
 918             rt               = _mm256_mul_pd(r10,vftabscale);
 919             vfitab           = _mm256_cvttpd_epi32(rt);
 920             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 921             vfitab           = _mm_slli_epi32(vfitab,2);
 922
 923             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 924             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 925             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 926             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 927             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 928             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 929             Heps             = _mm256_mul_pd(vfeps,H);
 930             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 931             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 932             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 933
 934             fscal            = felec;
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm256_mul_pd(fscal,dx10);
 938             ty               = _mm256_mul_pd(fscal,dy10);
 939             tz               = _mm256_mul_pd(fscal,dz10);
 940
 941             /* Update vectorial force */
 942             fix1             = _mm256_add_pd(fix1,tx);
 943             fiy1             = _mm256_add_pd(fiy1,ty);
 944             fiz1             = _mm256_add_pd(fiz1,tz);
 945
 946             fjx0             = _mm256_add_pd(fjx0,tx);
 947             fjy0             = _mm256_add_pd(fjy0,ty);
 948             fjz0             = _mm256_add_pd(fjz0,tz);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             r20              = _mm256_mul_pd(rsq20,rinv20);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq20             = _mm256_mul_pd(iq2,jq0);
 958
 959             /* Calculate table index by multiplying r with table scale and truncate to integer */
 960             rt               = _mm256_mul_pd(r20,vftabscale);
 961             vfitab           = _mm256_cvttpd_epi32(rt);
 962             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 963             vfitab           = _mm_slli_epi32(vfitab,2);
 964
 965             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 966             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 967             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 968             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 969             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 970             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 971             Heps             = _mm256_mul_pd(vfeps,H);
 972             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 973             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 974             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 975
 976             fscal            = felec;
 977
 978             /* Calculate temporary vectorial force */
 979             tx               = _mm256_mul_pd(fscal,dx20);
 980             ty               = _mm256_mul_pd(fscal,dy20);
 981             tz               = _mm256_mul_pd(fscal,dz20);
 982
 983             /* Update vectorial force */
 984             fix2             = _mm256_add_pd(fix2,tx);
 985             fiy2             = _mm256_add_pd(fiy2,ty);
 986             fiz2             = _mm256_add_pd(fiz2,tz);
 987
 988             fjx0             = _mm256_add_pd(fjx0,tx);
 989             fjy0             = _mm256_add_pd(fjy0,ty);
 990             fjz0             = _mm256_add_pd(fjz0,tz);
 991
 992             fjptrA             = f+j_coord_offsetA;
 993             fjptrB             = f+j_coord_offsetB;
 994             fjptrC             = f+j_coord_offsetC;
 995             fjptrD             = f+j_coord_offsetD;
 996
 997             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 998
 999             /* Inner loop uses 128 flops */
1000         }
1001
1002         if(jidx<j_index_end)
1003         {
1004
1005             /* Get j neighbor index, and coordinate index */
1006             jnrlistA         = jjnr[jidx];
1007             jnrlistB         = jjnr[jidx+1];
1008             jnrlistC         = jjnr[jidx+2];
1009             jnrlistD         = jjnr[jidx+3];
1010             /* Sign of each element will be negative for non-real atoms.
1011              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1012              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1013              */
1014             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1015
1016             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1017             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1018             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1019
1020             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1021             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1022             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1023             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1024             j_coord_offsetA  = DIM*jnrA;
1025             j_coord_offsetB  = DIM*jnrB;
1026             j_coord_offsetC  = DIM*jnrC;
1027             j_coord_offsetD  = DIM*jnrD;
1028
1029             /* load j atom coordinates */
1030             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1031                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1032                                                  &jx0,&jy0,&jz0);
1033
1034             /* Calculate displacement vector */
1035             dx00             = _mm256_sub_pd(ix0,jx0);
1036             dy00             = _mm256_sub_pd(iy0,jy0);
1037             dz00             = _mm256_sub_pd(iz0,jz0);
1038             dx10             = _mm256_sub_pd(ix1,jx0);
1039             dy10             = _mm256_sub_pd(iy1,jy0);
1040             dz10             = _mm256_sub_pd(iz1,jz0);
1041             dx20             = _mm256_sub_pd(ix2,jx0);
1042             dy20             = _mm256_sub_pd(iy2,jy0);
1043             dz20             = _mm256_sub_pd(iz2,jz0);
1044
1045             /* Calculate squared distance and things based on it */
1046             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1047             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1048             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1049
1050             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1051             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1052             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1053
1054             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1055
1056             /* Load parameters for j particles */
1057             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1058                                                                  charge+jnrC+0,charge+jnrD+0);
1059             vdwjidx0A        = 2*vdwtype[jnrA+0];
1060             vdwjidx0B        = 2*vdwtype[jnrB+0];
1061             vdwjidx0C        = 2*vdwtype[jnrC+0];
1062             vdwjidx0D        = 2*vdwtype[jnrD+0];
1063
1064             fjx0             = _mm256_setzero_pd();
1065             fjy0             = _mm256_setzero_pd();
1066             fjz0             = _mm256_setzero_pd();
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             r00              = _mm256_mul_pd(rsq00,rinv00);
1073             r00              = _mm256_andnot_pd(dummy_mask,r00);
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq00             = _mm256_mul_pd(iq0,jq0);
1077             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1078                                             vdwioffsetptr0+vdwjidx0B,
1079                                             vdwioffsetptr0+vdwjidx0C,
1080                                             vdwioffsetptr0+vdwjidx0D,
1081                                             &c6_00,&c12_00);
1082
1083             /* Calculate table index by multiplying r with table scale and truncate to integer */
1084             rt               = _mm256_mul_pd(r00,vftabscale);
1085             vfitab           = _mm256_cvttpd_epi32(rt);
1086             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1087             vfitab           = _mm_slli_epi32(vfitab,2);
1088
1089             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1090             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1091             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1092             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1093             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1094             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1095             Heps             = _mm256_mul_pd(vfeps,H);
1096             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1097             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1098             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1099
1100             /* LENNARD-JONES DISPERSION/REPULSION */
1101
1102             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1103             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1104
1105             fscal            = _mm256_add_pd(felec,fvdw);
1106
1107             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm256_mul_pd(fscal,dx00);
1111             ty               = _mm256_mul_pd(fscal,dy00);
1112             tz               = _mm256_mul_pd(fscal,dz00);
1113
1114             /* Update vectorial force */
1115             fix0             = _mm256_add_pd(fix0,tx);
1116             fiy0             = _mm256_add_pd(fiy0,ty);
1117             fiz0             = _mm256_add_pd(fiz0,tz);
1118
1119             fjx0             = _mm256_add_pd(fjx0,tx);
1120             fjy0             = _mm256_add_pd(fjy0,ty);
1121             fjz0             = _mm256_add_pd(fjz0,tz);
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             r10              = _mm256_mul_pd(rsq10,rinv10);
1128             r10              = _mm256_andnot_pd(dummy_mask,r10);
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             qq10             = _mm256_mul_pd(iq1,jq0);
1132
1133             /* Calculate table index by multiplying r with table scale and truncate to integer */
1134             rt               = _mm256_mul_pd(r10,vftabscale);
1135             vfitab           = _mm256_cvttpd_epi32(rt);
1136             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1137             vfitab           = _mm_slli_epi32(vfitab,2);
1138
1139             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1140             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1141             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1142             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1143             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1144             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1145             Heps             = _mm256_mul_pd(vfeps,H);
1146             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1147             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1148             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1149
1150             fscal            = felec;
1151
1152             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm256_mul_pd(fscal,dx10);
1156             ty               = _mm256_mul_pd(fscal,dy10);
1157             tz               = _mm256_mul_pd(fscal,dz10);
1158
1159             /* Update vectorial force */
1160             fix1             = _mm256_add_pd(fix1,tx);
1161             fiy1             = _mm256_add_pd(fiy1,ty);
1162             fiz1             = _mm256_add_pd(fiz1,tz);
1163
1164             fjx0             = _mm256_add_pd(fjx0,tx);
1165             fjy0             = _mm256_add_pd(fjy0,ty);
1166             fjz0             = _mm256_add_pd(fjz0,tz);
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             r20              = _mm256_mul_pd(rsq20,rinv20);
1173             r20              = _mm256_andnot_pd(dummy_mask,r20);
1174
1175             /* Compute parameters for interactions between i and j atoms */
1176             qq20             = _mm256_mul_pd(iq2,jq0);
1177
1178             /* Calculate table index by multiplying r with table scale and truncate to integer */
1179             rt               = _mm256_mul_pd(r20,vftabscale);
1180             vfitab           = _mm256_cvttpd_epi32(rt);
1181             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1182             vfitab           = _mm_slli_epi32(vfitab,2);
1183
1184             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1185             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1186             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1187             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1188             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1189             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1190             Heps             = _mm256_mul_pd(vfeps,H);
1191             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1192             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1193             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1194
1195             fscal            = felec;
1196
1197             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1198
1199             /* Calculate temporary vectorial force */
1200             tx               = _mm256_mul_pd(fscal,dx20);
1201             ty               = _mm256_mul_pd(fscal,dy20);
1202             tz               = _mm256_mul_pd(fscal,dz20);
1203
1204             /* Update vectorial force */
1205             fix2             = _mm256_add_pd(fix2,tx);
1206             fiy2             = _mm256_add_pd(fiy2,ty);
1207             fiz2             = _mm256_add_pd(fiz2,tz);
1208
1209             fjx0             = _mm256_add_pd(fjx0,tx);
1210             fjy0             = _mm256_add_pd(fjy0,ty);
1211             fjz0             = _mm256_add_pd(fjz0,tz);
1212
1213             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1214             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1215             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1216             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1217
1218             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1219
1220             /* Inner loop uses 131 flops */
1221         }
1222
1223         /* End of innermost loop */
1224
1225         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1226                                                  f+i_coord_offset,fshift+i_shift_offset);
1227
1228         /* Increment number of inner iterations */
1229         inneriter                  += j_index_end - j_index_start;
1230
1231         /* Outer loop uses 18 flops */
1232     }
1233
1234     /* Increment number of outer iterations */
1235     outeriter        += nri;
1236
1237     /* Update outer/inner flops */
1238
1239     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1240 }