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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     vftab            = kernel_data->table_elec->data;
 132     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 137     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 138     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm256_setzero_pd();
 175         fiy0             = _mm256_setzero_pd();
 176         fiz0             = _mm256_setzero_pd();
 177         fix1             = _mm256_setzero_pd();
 178         fiy1             = _mm256_setzero_pd();
 179         fiz1             = _mm256_setzero_pd();
 180         fix2             = _mm256_setzero_pd();
 181         fiy2             = _mm256_setzero_pd();
 182         fiz2             = _mm256_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm256_setzero_pd();
 186         vvdwsum          = _mm256_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_pd(ix0,jx0);
 209             dy00             = _mm256_sub_pd(iy0,jy0);
 210             dz00             = _mm256_sub_pd(iz0,jz0);
 211             dx10             = _mm256_sub_pd(ix1,jx0);
 212             dy10             = _mm256_sub_pd(iy1,jy0);
 213             dz10             = _mm256_sub_pd(iz1,jz0);
 214             dx20             = _mm256_sub_pd(ix2,jx0);
 215             dy20             = _mm256_sub_pd(iy2,jy0);
 216             dz20             = _mm256_sub_pd(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 224             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 225             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 226
 227             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 228
 229             /* Load parameters for j particles */
 230             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 231                                                                  charge+jnrC+0,charge+jnrD+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234             vdwjidx0C        = 2*vdwtype[jnrC+0];
 235             vdwjidx0D        = 2*vdwtype[jnrD+0];
 236
 237             fjx0             = _mm256_setzero_pd();
 238             fjy0             = _mm256_setzero_pd();
 239             fjz0             = _mm256_setzero_pd();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             r00              = _mm256_mul_pd(rsq00,rinv00);
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             qq00             = _mm256_mul_pd(iq0,jq0);
 249             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 250                                             vdwioffsetptr0+vdwjidx0B,
 251                                             vdwioffsetptr0+vdwjidx0C,
 252                                             vdwioffsetptr0+vdwjidx0D,
 253                                             &c6_00,&c12_00);
 254
 255             /* Calculate table index by multiplying r with table scale and truncate to integer */
 256             rt               = _mm256_mul_pd(r00,vftabscale);
 257             vfitab           = _mm256_cvttpd_epi32(rt);
 258             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 259             vfitab           = _mm_slli_epi32(vfitab,2);
 260
 261             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 262             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 263             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 264             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 265             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 266             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 267             Heps             = _mm256_mul_pd(vfeps,H);
 268             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 269             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 270             velec            = _mm256_mul_pd(qq00,VV);
 271             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 272             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 273
 274             /* LENNARD-JONES DISPERSION/REPULSION */
 275
 276             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 277             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 278             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 279             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 280             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _mm256_add_pd(velecsum,velec);
 284             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 285
 286             fscal            = _mm256_add_pd(felec,fvdw);
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm256_mul_pd(fscal,dx00);
 290             ty               = _mm256_mul_pd(fscal,dy00);
 291             tz               = _mm256_mul_pd(fscal,dz00);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm256_add_pd(fix0,tx);
 295             fiy0             = _mm256_add_pd(fiy0,ty);
 296             fiz0             = _mm256_add_pd(fiz0,tz);
 297
 298             fjx0             = _mm256_add_pd(fjx0,tx);
 299             fjy0             = _mm256_add_pd(fjy0,ty);
 300             fjz0             = _mm256_add_pd(fjz0,tz);
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             r10              = _mm256_mul_pd(rsq10,rinv10);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm256_mul_pd(iq1,jq0);
 310
 311             /* Calculate table index by multiplying r with table scale and truncate to integer */
 312             rt               = _mm256_mul_pd(r10,vftabscale);
 313             vfitab           = _mm256_cvttpd_epi32(rt);
 314             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 315             vfitab           = _mm_slli_epi32(vfitab,2);
 316
 317             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 318             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 319             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 320             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 321             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 322             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 323             Heps             = _mm256_mul_pd(vfeps,H);
 324             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 325             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 326             velec            = _mm256_mul_pd(qq10,VV);
 327             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 328             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm256_add_pd(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_pd(fscal,dx10);
 337             ty               = _mm256_mul_pd(fscal,dy10);
 338             tz               = _mm256_mul_pd(fscal,dz10);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm256_add_pd(fix1,tx);
 342             fiy1             = _mm256_add_pd(fiy1,ty);
 343             fiz1             = _mm256_add_pd(fiz1,tz);
 344
 345             fjx0             = _mm256_add_pd(fjx0,tx);
 346             fjy0             = _mm256_add_pd(fjy0,ty);
 347             fjz0             = _mm256_add_pd(fjz0,tz);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             r20              = _mm256_mul_pd(rsq20,rinv20);
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq20             = _mm256_mul_pd(iq2,jq0);
 357
 358             /* Calculate table index by multiplying r with table scale and truncate to integer */
 359             rt               = _mm256_mul_pd(r20,vftabscale);
 360             vfitab           = _mm256_cvttpd_epi32(rt);
 361             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 362             vfitab           = _mm_slli_epi32(vfitab,2);
 363
 364             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 365             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 366             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 367             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 368             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 369             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 370             Heps             = _mm256_mul_pd(vfeps,H);
 371             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 372             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 373             velec            = _mm256_mul_pd(qq20,VV);
 374             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 375             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm256_add_pd(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm256_mul_pd(fscal,dx20);
 384             ty               = _mm256_mul_pd(fscal,dy20);
 385             tz               = _mm256_mul_pd(fscal,dz20);
 386
 387             /* Update vectorial force */
 388             fix2             = _mm256_add_pd(fix2,tx);
 389             fiy2             = _mm256_add_pd(fiy2,ty);
 390             fiz2             = _mm256_add_pd(fiz2,tz);
 391
 392             fjx0             = _mm256_add_pd(fjx0,tx);
 393             fjy0             = _mm256_add_pd(fjy0,ty);
 394             fjz0             = _mm256_add_pd(fjz0,tz);
 395
 396             fjptrA             = f+j_coord_offsetA;
 397             fjptrB             = f+j_coord_offsetB;
 398             fjptrC             = f+j_coord_offsetC;
 399             fjptrD             = f+j_coord_offsetD;
 400
 401             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 402
 403             /* Inner loop uses 145 flops */
 404         }
 405
 406         if(jidx<j_index_end)
 407         {
 408
 409             /* Get j neighbor index, and coordinate index */
 410             jnrlistA         = jjnr[jidx];
 411             jnrlistB         = jjnr[jidx+1];
 412             jnrlistC         = jjnr[jidx+2];
 413             jnrlistD         = jjnr[jidx+3];
 414             /* Sign of each element will be negative for non-real atoms.
 415              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 416              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 417              */
 418             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 419
 420             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 421             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 422             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 423
 424             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 425             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 426             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 427             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 428             j_coord_offsetA  = DIM*jnrA;
 429             j_coord_offsetB  = DIM*jnrB;
 430             j_coord_offsetC  = DIM*jnrC;
 431             j_coord_offsetD  = DIM*jnrD;
 432
 433             /* load j atom coordinates */
 434             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 435                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 436                                                  &jx0,&jy0,&jz0);
 437
 438             /* Calculate displacement vector */
 439             dx00             = _mm256_sub_pd(ix0,jx0);
 440             dy00             = _mm256_sub_pd(iy0,jy0);
 441             dz00             = _mm256_sub_pd(iz0,jz0);
 442             dx10             = _mm256_sub_pd(ix1,jx0);
 443             dy10             = _mm256_sub_pd(iy1,jy0);
 444             dz10             = _mm256_sub_pd(iz1,jz0);
 445             dx20             = _mm256_sub_pd(ix2,jx0);
 446             dy20             = _mm256_sub_pd(iy2,jy0);
 447             dz20             = _mm256_sub_pd(iz2,jz0);
 448
 449             /* Calculate squared distance and things based on it */
 450             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 451             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 452             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 453
 454             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 455             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 456             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 457
 458             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 459
 460             /* Load parameters for j particles */
 461             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 462                                                                  charge+jnrC+0,charge+jnrD+0);
 463             vdwjidx0A        = 2*vdwtype[jnrA+0];
 464             vdwjidx0B        = 2*vdwtype[jnrB+0];
 465             vdwjidx0C        = 2*vdwtype[jnrC+0];
 466             vdwjidx0D        = 2*vdwtype[jnrD+0];
 467
 468             fjx0             = _mm256_setzero_pd();
 469             fjy0             = _mm256_setzero_pd();
 470             fjz0             = _mm256_setzero_pd();
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             r00              = _mm256_mul_pd(rsq00,rinv00);
 477             r00              = _mm256_andnot_pd(dummy_mask,r00);
 478
 479             /* Compute parameters for interactions between i and j atoms */
 480             qq00             = _mm256_mul_pd(iq0,jq0);
 481             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 482                                             vdwioffsetptr0+vdwjidx0B,
 483                                             vdwioffsetptr0+vdwjidx0C,
 484                                             vdwioffsetptr0+vdwjidx0D,
 485                                             &c6_00,&c12_00);
 486
 487             /* Calculate table index by multiplying r with table scale and truncate to integer */
 488             rt               = _mm256_mul_pd(r00,vftabscale);
 489             vfitab           = _mm256_cvttpd_epi32(rt);
 490             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 491             vfitab           = _mm_slli_epi32(vfitab,2);
 492
 493             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 494             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 495             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 496             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 497             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 498             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 499             Heps             = _mm256_mul_pd(vfeps,H);
 500             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 501             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 502             velec            = _mm256_mul_pd(qq00,VV);
 503             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 504             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 505
 506             /* LENNARD-JONES DISPERSION/REPULSION */
 507
 508             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 509             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 510             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 511             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 512             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velec            = _mm256_andnot_pd(dummy_mask,velec);
 516             velecsum         = _mm256_add_pd(velecsum,velec);
 517             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 518             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 519
 520             fscal            = _mm256_add_pd(felec,fvdw);
 521
 522             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm256_mul_pd(fscal,dx00);
 526             ty               = _mm256_mul_pd(fscal,dy00);
 527             tz               = _mm256_mul_pd(fscal,dz00);
 528
 529             /* Update vectorial force */
 530             fix0             = _mm256_add_pd(fix0,tx);
 531             fiy0             = _mm256_add_pd(fiy0,ty);
 532             fiz0             = _mm256_add_pd(fiz0,tz);
 533
 534             fjx0             = _mm256_add_pd(fjx0,tx);
 535             fjy0             = _mm256_add_pd(fjy0,ty);
 536             fjz0             = _mm256_add_pd(fjz0,tz);
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             r10              = _mm256_mul_pd(rsq10,rinv10);
 543             r10              = _mm256_andnot_pd(dummy_mask,r10);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq10             = _mm256_mul_pd(iq1,jq0);
 547
 548             /* Calculate table index by multiplying r with table scale and truncate to integer */
 549             rt               = _mm256_mul_pd(r10,vftabscale);
 550             vfitab           = _mm256_cvttpd_epi32(rt);
 551             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 552             vfitab           = _mm_slli_epi32(vfitab,2);
 553
 554             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 555             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 556             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 557             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 558             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 559             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 560             Heps             = _mm256_mul_pd(vfeps,H);
 561             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 562             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 563             velec            = _mm256_mul_pd(qq10,VV);
 564             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 565             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm256_andnot_pd(dummy_mask,velec);
 569             velecsum         = _mm256_add_pd(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm256_mul_pd(fscal,dx10);
 577             ty               = _mm256_mul_pd(fscal,dy10);
 578             tz               = _mm256_mul_pd(fscal,dz10);
 579
 580             /* Update vectorial force */
 581             fix1             = _mm256_add_pd(fix1,tx);
 582             fiy1             = _mm256_add_pd(fiy1,ty);
 583             fiz1             = _mm256_add_pd(fiz1,tz);
 584
 585             fjx0             = _mm256_add_pd(fjx0,tx);
 586             fjy0             = _mm256_add_pd(fjy0,ty);
 587             fjz0             = _mm256_add_pd(fjz0,tz);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             r20              = _mm256_mul_pd(rsq20,rinv20);
 594             r20              = _mm256_andnot_pd(dummy_mask,r20);
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq20             = _mm256_mul_pd(iq2,jq0);
 598
 599             /* Calculate table index by multiplying r with table scale and truncate to integer */
 600             rt               = _mm256_mul_pd(r20,vftabscale);
 601             vfitab           = _mm256_cvttpd_epi32(rt);
 602             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 603             vfitab           = _mm_slli_epi32(vfitab,2);
 604
 605             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 606             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 607             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 608             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 609             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 610             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 611             Heps             = _mm256_mul_pd(vfeps,H);
 612             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 613             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 614             velec            = _mm256_mul_pd(qq20,VV);
 615             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 616             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm256_andnot_pd(dummy_mask,velec);
 620             velecsum         = _mm256_add_pd(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm256_mul_pd(fscal,dx20);
 628             ty               = _mm256_mul_pd(fscal,dy20);
 629             tz               = _mm256_mul_pd(fscal,dz20);
 630
 631             /* Update vectorial force */
 632             fix2             = _mm256_add_pd(fix2,tx);
 633             fiy2             = _mm256_add_pd(fiy2,ty);
 634             fiz2             = _mm256_add_pd(fiz2,tz);
 635
 636             fjx0             = _mm256_add_pd(fjx0,tx);
 637             fjy0             = _mm256_add_pd(fjy0,ty);
 638             fjz0             = _mm256_add_pd(fjz0,tz);
 639
 640             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 641             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 642             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 643             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 644
 645             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 646
 647             /* Inner loop uses 148 flops */
 648         }
 649
 650         /* End of innermost loop */
 651
 652         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 653                                                  f+i_coord_offset,fshift+i_shift_offset);
 654
 655         ggid                        = gid[iidx];
 656         /* Update potential energies */
 657         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 658         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 659
 660         /* Increment number of inner iterations */
 661         inneriter                  += j_index_end - j_index_start;
 662
 663         /* Outer loop uses 20 flops */
 664     }
 665
 666     /* Increment number of outer iterations */
 667     outeriter        += nri;
 668
 669     /* Update outer/inner flops */
 670
 671     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 672 }
 673 /*
 674  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_double
 675  * Electrostatics interaction: CubicSplineTable
 676  * VdW interaction:            LennardJones
 677  * Geometry:                   Water3-Particle
 678  * Calculate force/pot:        Force
 679  */
 680 void
 681 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_double
 682                     (t_nblist                    * gmx_restrict       nlist,
 683                      rvec                        * gmx_restrict          xx,
 684                      rvec                        * gmx_restrict          ff,
 685                      t_forcerec                  * gmx_restrict          fr,
 686                      t_mdatoms                   * gmx_restrict     mdatoms,
 687                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 688                      t_nrnb                      * gmx_restrict        nrnb)
 689 {
 690     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 691      * just 0 for non-waters.
 692      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 693      * jnr indices corresponding to data put in the four positions in the SIMD register.
 694      */
 695     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 696     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 697     int              jnrA,jnrB,jnrC,jnrD;
 698     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 699     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 700     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 701     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 702     real             rcutoff_scalar;
 703     real             *shiftvec,*fshift,*x,*f;
 704     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 705     real             scratch[4*DIM];
 706     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 707     real *           vdwioffsetptr0;
 708     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 709     real *           vdwioffsetptr1;
 710     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 711     real *           vdwioffsetptr2;
 712     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 713     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 714     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 715     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 716     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 717     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 718     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 719     real             *charge;
 720     int              nvdwtype;
 721     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 722     int              *vdwtype;
 723     real             *vdwparam;
 724     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 725     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 726     __m128i          vfitab;
 727     __m128i          ifour       = _mm_set1_epi32(4);
 728     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 729     real             *vftab;
 730     __m256d          dummy_mask,cutoff_mask;
 731     __m128           tmpmask0,tmpmask1;
 732     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 733     __m256d          one     = _mm256_set1_pd(1.0);
 734     __m256d          two     = _mm256_set1_pd(2.0);
 735     x                = xx[0];
 736     f                = ff[0];
 737
 738     nri              = nlist->nri;
 739     iinr             = nlist->iinr;
 740     jindex           = nlist->jindex;
 741     jjnr             = nlist->jjnr;
 742     shiftidx         = nlist->shift;
 743     gid              = nlist->gid;
 744     shiftvec         = fr->shift_vec[0];
 745     fshift           = fr->fshift[0];
 746     facel            = _mm256_set1_pd(fr->epsfac);
 747     charge           = mdatoms->chargeA;
 748     nvdwtype         = fr->ntype;
 749     vdwparam         = fr->nbfp;
 750     vdwtype          = mdatoms->typeA;
 751
 752     vftab            = kernel_data->table_elec->data;
 753     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 754
 755     /* Setup water-specific parameters */
 756     inr              = nlist->iinr[0];
 757     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 758     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 759     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 760     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 761
 762     /* Avoid stupid compiler warnings */
 763     jnrA = jnrB = jnrC = jnrD = 0;
 764     j_coord_offsetA = 0;
 765     j_coord_offsetB = 0;
 766     j_coord_offsetC = 0;
 767     j_coord_offsetD = 0;
 768
 769     outeriter        = 0;
 770     inneriter        = 0;
 771
 772     for(iidx=0;iidx<4*DIM;iidx++)
 773     {
 774         scratch[iidx] = 0.0;
 775     }
 776
 777     /* Start outer loop over neighborlists */
 778     for(iidx=0; iidx<nri; iidx++)
 779     {
 780         /* Load shift vector for this list */
 781         i_shift_offset   = DIM*shiftidx[iidx];
 782
 783         /* Load limits for loop over neighbors */
 784         j_index_start    = jindex[iidx];
 785         j_index_end      = jindex[iidx+1];
 786
 787         /* Get outer coordinate index */
 788         inr              = iinr[iidx];
 789         i_coord_offset   = DIM*inr;
 790
 791         /* Load i particle coords and add shift vector */
 792         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 793                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 794
 795         fix0             = _mm256_setzero_pd();
 796         fiy0             = _mm256_setzero_pd();
 797         fiz0             = _mm256_setzero_pd();
 798         fix1             = _mm256_setzero_pd();
 799         fiy1             = _mm256_setzero_pd();
 800         fiz1             = _mm256_setzero_pd();
 801         fix2             = _mm256_setzero_pd();
 802         fiy2             = _mm256_setzero_pd();
 803         fiz2             = _mm256_setzero_pd();
 804
 805         /* Start inner kernel loop */
 806         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 807         {
 808
 809             /* Get j neighbor index, and coordinate index */
 810             jnrA             = jjnr[jidx];
 811             jnrB             = jjnr[jidx+1];
 812             jnrC             = jjnr[jidx+2];
 813             jnrD             = jjnr[jidx+3];
 814             j_coord_offsetA  = DIM*jnrA;
 815             j_coord_offsetB  = DIM*jnrB;
 816             j_coord_offsetC  = DIM*jnrC;
 817             j_coord_offsetD  = DIM*jnrD;
 818
 819             /* load j atom coordinates */
 820             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 821                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 822                                                  &jx0,&jy0,&jz0);
 823
 824             /* Calculate displacement vector */
 825             dx00             = _mm256_sub_pd(ix0,jx0);
 826             dy00             = _mm256_sub_pd(iy0,jy0);
 827             dz00             = _mm256_sub_pd(iz0,jz0);
 828             dx10             = _mm256_sub_pd(ix1,jx0);
 829             dy10             = _mm256_sub_pd(iy1,jy0);
 830             dz10             = _mm256_sub_pd(iz1,jz0);
 831             dx20             = _mm256_sub_pd(ix2,jx0);
 832             dy20             = _mm256_sub_pd(iy2,jy0);
 833             dz20             = _mm256_sub_pd(iz2,jz0);
 834
 835             /* Calculate squared distance and things based on it */
 836             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 837             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 838             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 839
 840             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 841             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 842             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 843
 844             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 845
 846             /* Load parameters for j particles */
 847             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 848                                                                  charge+jnrC+0,charge+jnrD+0);
 849             vdwjidx0A        = 2*vdwtype[jnrA+0];
 850             vdwjidx0B        = 2*vdwtype[jnrB+0];
 851             vdwjidx0C        = 2*vdwtype[jnrC+0];
 852             vdwjidx0D        = 2*vdwtype[jnrD+0];
 853
 854             fjx0             = _mm256_setzero_pd();
 855             fjy0             = _mm256_setzero_pd();
 856             fjz0             = _mm256_setzero_pd();
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             r00              = _mm256_mul_pd(rsq00,rinv00);
 863
 864             /* Compute parameters for interactions between i and j atoms */
 865             qq00             = _mm256_mul_pd(iq0,jq0);
 866             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 867                                             vdwioffsetptr0+vdwjidx0B,
 868                                             vdwioffsetptr0+vdwjidx0C,
 869                                             vdwioffsetptr0+vdwjidx0D,
 870                                             &c6_00,&c12_00);
 871
 872             /* Calculate table index by multiplying r with table scale and truncate to integer */
 873             rt               = _mm256_mul_pd(r00,vftabscale);
 874             vfitab           = _mm256_cvttpd_epi32(rt);
 875             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 876             vfitab           = _mm_slli_epi32(vfitab,2);
 877
 878             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 879             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 880             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 881             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 882             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 883             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 884             Heps             = _mm256_mul_pd(vfeps,H);
 885             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 886             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 887             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 888
 889             /* LENNARD-JONES DISPERSION/REPULSION */
 890
 891             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 892             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 893
 894             fscal            = _mm256_add_pd(felec,fvdw);
 895
 896             /* Calculate temporary vectorial force */
 897             tx               = _mm256_mul_pd(fscal,dx00);
 898             ty               = _mm256_mul_pd(fscal,dy00);
 899             tz               = _mm256_mul_pd(fscal,dz00);
 900
 901             /* Update vectorial force */
 902             fix0             = _mm256_add_pd(fix0,tx);
 903             fiy0             = _mm256_add_pd(fiy0,ty);
 904             fiz0             = _mm256_add_pd(fiz0,tz);
 905
 906             fjx0             = _mm256_add_pd(fjx0,tx);
 907             fjy0             = _mm256_add_pd(fjy0,ty);
 908             fjz0             = _mm256_add_pd(fjz0,tz);
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             r10              = _mm256_mul_pd(rsq10,rinv10);
 915
 916             /* Compute parameters for interactions between i and j atoms */
 917             qq10             = _mm256_mul_pd(iq1,jq0);
 918
 919             /* Calculate table index by multiplying r with table scale and truncate to integer */
 920             rt               = _mm256_mul_pd(r10,vftabscale);
 921             vfitab           = _mm256_cvttpd_epi32(rt);
 922             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 923             vfitab           = _mm_slli_epi32(vfitab,2);
 924
 925             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 926             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 927             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 928             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 929             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 930             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 931             Heps             = _mm256_mul_pd(vfeps,H);
 932             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 933             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 934             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 935
 936             fscal            = felec;
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm256_mul_pd(fscal,dx10);
 940             ty               = _mm256_mul_pd(fscal,dy10);
 941             tz               = _mm256_mul_pd(fscal,dz10);
 942
 943             /* Update vectorial force */
 944             fix1             = _mm256_add_pd(fix1,tx);
 945             fiy1             = _mm256_add_pd(fiy1,ty);
 946             fiz1             = _mm256_add_pd(fiz1,tz);
 947
 948             fjx0             = _mm256_add_pd(fjx0,tx);
 949             fjy0             = _mm256_add_pd(fjy0,ty);
 950             fjz0             = _mm256_add_pd(fjz0,tz);
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             r20              = _mm256_mul_pd(rsq20,rinv20);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq20             = _mm256_mul_pd(iq2,jq0);
 960
 961             /* Calculate table index by multiplying r with table scale and truncate to integer */
 962             rt               = _mm256_mul_pd(r20,vftabscale);
 963             vfitab           = _mm256_cvttpd_epi32(rt);
 964             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 965             vfitab           = _mm_slli_epi32(vfitab,2);
 966
 967             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 968             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 969             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 970             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 971             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 972             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 973             Heps             = _mm256_mul_pd(vfeps,H);
 974             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 975             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 976             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 977
 978             fscal            = felec;
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm256_mul_pd(fscal,dx20);
 982             ty               = _mm256_mul_pd(fscal,dy20);
 983             tz               = _mm256_mul_pd(fscal,dz20);
 984
 985             /* Update vectorial force */
 986             fix2             = _mm256_add_pd(fix2,tx);
 987             fiy2             = _mm256_add_pd(fiy2,ty);
 988             fiz2             = _mm256_add_pd(fiz2,tz);
 989
 990             fjx0             = _mm256_add_pd(fjx0,tx);
 991             fjy0             = _mm256_add_pd(fjy0,ty);
 992             fjz0             = _mm256_add_pd(fjz0,tz);
 993
 994             fjptrA             = f+j_coord_offsetA;
 995             fjptrB             = f+j_coord_offsetB;
 996             fjptrC             = f+j_coord_offsetC;
 997             fjptrD             = f+j_coord_offsetD;
 998
 999             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1000
1001             /* Inner loop uses 128 flops */
1002         }
1003
1004         if(jidx<j_index_end)
1005         {
1006
1007             /* Get j neighbor index, and coordinate index */
1008             jnrlistA         = jjnr[jidx];
1009             jnrlistB         = jjnr[jidx+1];
1010             jnrlistC         = jjnr[jidx+2];
1011             jnrlistD         = jjnr[jidx+3];
1012             /* Sign of each element will be negative for non-real atoms.
1013              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1014              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1015              */
1016             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1017
1018             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1019             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1020             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1021
1022             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1023             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1024             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1025             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1026             j_coord_offsetA  = DIM*jnrA;
1027             j_coord_offsetB  = DIM*jnrB;
1028             j_coord_offsetC  = DIM*jnrC;
1029             j_coord_offsetD  = DIM*jnrD;
1030
1031             /* load j atom coordinates */
1032             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1033                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1034                                                  &jx0,&jy0,&jz0);
1035
1036             /* Calculate displacement vector */
1037             dx00             = _mm256_sub_pd(ix0,jx0);
1038             dy00             = _mm256_sub_pd(iy0,jy0);
1039             dz00             = _mm256_sub_pd(iz0,jz0);
1040             dx10             = _mm256_sub_pd(ix1,jx0);
1041             dy10             = _mm256_sub_pd(iy1,jy0);
1042             dz10             = _mm256_sub_pd(iz1,jz0);
1043             dx20             = _mm256_sub_pd(ix2,jx0);
1044             dy20             = _mm256_sub_pd(iy2,jy0);
1045             dz20             = _mm256_sub_pd(iz2,jz0);
1046
1047             /* Calculate squared distance and things based on it */
1048             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1049             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1050             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1051
1052             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1053             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1054             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1055
1056             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1057
1058             /* Load parameters for j particles */
1059             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1060                                                                  charge+jnrC+0,charge+jnrD+0);
1061             vdwjidx0A        = 2*vdwtype[jnrA+0];
1062             vdwjidx0B        = 2*vdwtype[jnrB+0];
1063             vdwjidx0C        = 2*vdwtype[jnrC+0];
1064             vdwjidx0D        = 2*vdwtype[jnrD+0];
1065
1066             fjx0             = _mm256_setzero_pd();
1067             fjy0             = _mm256_setzero_pd();
1068             fjz0             = _mm256_setzero_pd();
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             r00              = _mm256_mul_pd(rsq00,rinv00);
1075             r00              = _mm256_andnot_pd(dummy_mask,r00);
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq00             = _mm256_mul_pd(iq0,jq0);
1079             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1080                                             vdwioffsetptr0+vdwjidx0B,
1081                                             vdwioffsetptr0+vdwjidx0C,
1082                                             vdwioffsetptr0+vdwjidx0D,
1083                                             &c6_00,&c12_00);
1084
1085             /* Calculate table index by multiplying r with table scale and truncate to integer */
1086             rt               = _mm256_mul_pd(r00,vftabscale);
1087             vfitab           = _mm256_cvttpd_epi32(rt);
1088             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1089             vfitab           = _mm_slli_epi32(vfitab,2);
1090
1091             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1092             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1093             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1094             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1095             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1096             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1097             Heps             = _mm256_mul_pd(vfeps,H);
1098             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1099             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1100             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1101
1102             /* LENNARD-JONES DISPERSION/REPULSION */
1103
1104             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1105             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1106
1107             fscal            = _mm256_add_pd(felec,fvdw);
1108
1109             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1110
1111             /* Calculate temporary vectorial force */
1112             tx               = _mm256_mul_pd(fscal,dx00);
1113             ty               = _mm256_mul_pd(fscal,dy00);
1114             tz               = _mm256_mul_pd(fscal,dz00);
1115
1116             /* Update vectorial force */
1117             fix0             = _mm256_add_pd(fix0,tx);
1118             fiy0             = _mm256_add_pd(fiy0,ty);
1119             fiz0             = _mm256_add_pd(fiz0,tz);
1120
1121             fjx0             = _mm256_add_pd(fjx0,tx);
1122             fjy0             = _mm256_add_pd(fjy0,ty);
1123             fjz0             = _mm256_add_pd(fjz0,tz);
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             r10              = _mm256_mul_pd(rsq10,rinv10);
1130             r10              = _mm256_andnot_pd(dummy_mask,r10);
1131
1132             /* Compute parameters for interactions between i and j atoms */
1133             qq10             = _mm256_mul_pd(iq1,jq0);
1134
1135             /* Calculate table index by multiplying r with table scale and truncate to integer */
1136             rt               = _mm256_mul_pd(r10,vftabscale);
1137             vfitab           = _mm256_cvttpd_epi32(rt);
1138             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1139             vfitab           = _mm_slli_epi32(vfitab,2);
1140
1141             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1142             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1143             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1144             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1145             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1146             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1147             Heps             = _mm256_mul_pd(vfeps,H);
1148             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1149             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1150             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1151
1152             fscal            = felec;
1153
1154             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1155
1156             /* Calculate temporary vectorial force */
1157             tx               = _mm256_mul_pd(fscal,dx10);
1158             ty               = _mm256_mul_pd(fscal,dy10);
1159             tz               = _mm256_mul_pd(fscal,dz10);
1160
1161             /* Update vectorial force */
1162             fix1             = _mm256_add_pd(fix1,tx);
1163             fiy1             = _mm256_add_pd(fiy1,ty);
1164             fiz1             = _mm256_add_pd(fiz1,tz);
1165
1166             fjx0             = _mm256_add_pd(fjx0,tx);
1167             fjy0             = _mm256_add_pd(fjy0,ty);
1168             fjz0             = _mm256_add_pd(fjz0,tz);
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             r20              = _mm256_mul_pd(rsq20,rinv20);
1175             r20              = _mm256_andnot_pd(dummy_mask,r20);
1176
1177             /* Compute parameters for interactions between i and j atoms */
1178             qq20             = _mm256_mul_pd(iq2,jq0);
1179
1180             /* Calculate table index by multiplying r with table scale and truncate to integer */
1181             rt               = _mm256_mul_pd(r20,vftabscale);
1182             vfitab           = _mm256_cvttpd_epi32(rt);
1183             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1184             vfitab           = _mm_slli_epi32(vfitab,2);
1185
1186             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1187             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1188             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1189             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1190             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1191             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1192             Heps             = _mm256_mul_pd(vfeps,H);
1193             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1194             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1195             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1200
1201             /* Calculate temporary vectorial force */
1202             tx               = _mm256_mul_pd(fscal,dx20);
1203             ty               = _mm256_mul_pd(fscal,dy20);
1204             tz               = _mm256_mul_pd(fscal,dz20);
1205
1206             /* Update vectorial force */
1207             fix2             = _mm256_add_pd(fix2,tx);
1208             fiy2             = _mm256_add_pd(fiy2,ty);
1209             fiz2             = _mm256_add_pd(fiz2,tz);
1210
1211             fjx0             = _mm256_add_pd(fjx0,tx);
1212             fjy0             = _mm256_add_pd(fjy0,ty);
1213             fjz0             = _mm256_add_pd(fjz0,tz);
1214
1215             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1216             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1217             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1218             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1219
1220             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1221
1222             /* Inner loop uses 131 flops */
1223         }
1224
1225         /* End of innermost loop */
1226
1227         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1228                                                  f+i_coord_offset,fshift+i_shift_offset);
1229
1230         /* Increment number of inner iterations */
1231         inneriter                  += j_index_end - j_index_start;
1232
1233         /* Outer loop uses 18 flops */
1234     }
1235
1236     /* Increment number of outer iterations */
1237     outeriter        += nri;
1238
1239     /* Update outer/inner flops */
1240
1241     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1242 }