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) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 101     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m256d          dummy_mask,cutoff_mask;
 107     __m128           tmpmask0,tmpmask1;
 108     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 109     __m256d          one     = _mm256_set1_pd(1.0);
 110     __m256d          two     = _mm256_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_pd(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     vftab            = kernel_data->table_elec->data;
 129     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 134     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 135     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _mm256_setzero_pd();
 172         fiy0             = _mm256_setzero_pd();
 173         fiz0             = _mm256_setzero_pd();
 174         fix1             = _mm256_setzero_pd();
 175         fiy1             = _mm256_setzero_pd();
 176         fiz1             = _mm256_setzero_pd();
 177         fix2             = _mm256_setzero_pd();
 178         fiy2             = _mm256_setzero_pd();
 179         fiz2             = _mm256_setzero_pd();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm256_setzero_pd();
 183         vvdwsum          = _mm256_setzero_pd();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 202                                                  &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm256_sub_pd(ix0,jx0);
 206             dy00             = _mm256_sub_pd(iy0,jy0);
 207             dz00             = _mm256_sub_pd(iz0,jz0);
 208             dx10             = _mm256_sub_pd(ix1,jx0);
 209             dy10             = _mm256_sub_pd(iy1,jy0);
 210             dz10             = _mm256_sub_pd(iz1,jz0);
 211             dx20             = _mm256_sub_pd(ix2,jx0);
 212             dy20             = _mm256_sub_pd(iy2,jy0);
 213             dz20             = _mm256_sub_pd(iz2,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 217             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 218             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 219
 220             rinv00           = avx256_invsqrt_d(rsq00);
 221             rinv10           = avx256_invsqrt_d(rsq10);
 222             rinv20           = avx256_invsqrt_d(rsq20);
 223
 224             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 228                                                                  charge+jnrC+0,charge+jnrD+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231             vdwjidx0C        = 2*vdwtype[jnrC+0];
 232             vdwjidx0D        = 2*vdwtype[jnrD+0];
 233
 234             fjx0             = _mm256_setzero_pd();
 235             fjy0             = _mm256_setzero_pd();
 236             fjz0             = _mm256_setzero_pd();
 237
 238             /**************************
 239              * CALCULATE INTERACTIONS *
 240              **************************/
 241
 242             r00              = _mm256_mul_pd(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             qq00             = _mm256_mul_pd(iq0,jq0);
 246             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 247                                             vdwioffsetptr0+vdwjidx0B,
 248                                             vdwioffsetptr0+vdwjidx0C,
 249                                             vdwioffsetptr0+vdwjidx0D,
 250                                             &c6_00,&c12_00);
 251
 252             /* Calculate table index by multiplying r with table scale and truncate to integer */
 253             rt               = _mm256_mul_pd(r00,vftabscale);
 254             vfitab           = _mm256_cvttpd_epi32(rt);
 255             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 256             vfitab           = _mm_slli_epi32(vfitab,2);
 257
 258             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 259             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 260             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 261             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 262             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 263             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 264             Heps             = _mm256_mul_pd(vfeps,H);
 265             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 266             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 267             velec            = _mm256_mul_pd(qq00,VV);
 268             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 269             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 270
 271             /* LENNARD-JONES DISPERSION/REPULSION */
 272
 273             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 274             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 275             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 276             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 277             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velecsum         = _mm256_add_pd(velecsum,velec);
 281             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 282
 283             fscal            = _mm256_add_pd(felec,fvdw);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm256_mul_pd(fscal,dx00);
 287             ty               = _mm256_mul_pd(fscal,dy00);
 288             tz               = _mm256_mul_pd(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm256_add_pd(fix0,tx);
 292             fiy0             = _mm256_add_pd(fiy0,ty);
 293             fiz0             = _mm256_add_pd(fiz0,tz);
 294
 295             fjx0             = _mm256_add_pd(fjx0,tx);
 296             fjy0             = _mm256_add_pd(fjy0,ty);
 297             fjz0             = _mm256_add_pd(fjz0,tz);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r10              = _mm256_mul_pd(rsq10,rinv10);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm256_mul_pd(iq1,jq0);
 307
 308             /* Calculate table index by multiplying r with table scale and truncate to integer */
 309             rt               = _mm256_mul_pd(r10,vftabscale);
 310             vfitab           = _mm256_cvttpd_epi32(rt);
 311             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 312             vfitab           = _mm_slli_epi32(vfitab,2);
 313
 314             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 315             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 316             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 317             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 318             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 319             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 320             Heps             = _mm256_mul_pd(vfeps,H);
 321             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 322             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 323             velec            = _mm256_mul_pd(qq10,VV);
 324             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 325             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm256_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm256_mul_pd(fscal,dx10);
 334             ty               = _mm256_mul_pd(fscal,dy10);
 335             tz               = _mm256_mul_pd(fscal,dz10);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm256_add_pd(fix1,tx);
 339             fiy1             = _mm256_add_pd(fiy1,ty);
 340             fiz1             = _mm256_add_pd(fiz1,tz);
 341
 342             fjx0             = _mm256_add_pd(fjx0,tx);
 343             fjy0             = _mm256_add_pd(fjy0,ty);
 344             fjz0             = _mm256_add_pd(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             r20              = _mm256_mul_pd(rsq20,rinv20);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm256_mul_pd(iq2,jq0);
 354
 355             /* Calculate table index by multiplying r with table scale and truncate to integer */
 356             rt               = _mm256_mul_pd(r20,vftabscale);
 357             vfitab           = _mm256_cvttpd_epi32(rt);
 358             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 359             vfitab           = _mm_slli_epi32(vfitab,2);
 360
 361             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 362             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 363             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 364             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 365             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 366             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 367             Heps             = _mm256_mul_pd(vfeps,H);
 368             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 369             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 370             velec            = _mm256_mul_pd(qq20,VV);
 371             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 372             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velecsum         = _mm256_add_pd(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm256_mul_pd(fscal,dx20);
 381             ty               = _mm256_mul_pd(fscal,dy20);
 382             tz               = _mm256_mul_pd(fscal,dz20);
 383
 384             /* Update vectorial force */
 385             fix2             = _mm256_add_pd(fix2,tx);
 386             fiy2             = _mm256_add_pd(fiy2,ty);
 387             fiz2             = _mm256_add_pd(fiz2,tz);
 388
 389             fjx0             = _mm256_add_pd(fjx0,tx);
 390             fjy0             = _mm256_add_pd(fjy0,ty);
 391             fjz0             = _mm256_add_pd(fjz0,tz);
 392
 393             fjptrA             = f+j_coord_offsetA;
 394             fjptrB             = f+j_coord_offsetB;
 395             fjptrC             = f+j_coord_offsetC;
 396             fjptrD             = f+j_coord_offsetD;
 397
 398             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 399
 400             /* Inner loop uses 145 flops */
 401         }
 402
 403         if(jidx<j_index_end)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrlistA         = jjnr[jidx];
 408             jnrlistB         = jjnr[jidx+1];
 409             jnrlistC         = jjnr[jidx+2];
 410             jnrlistD         = jjnr[jidx+3];
 411             /* Sign of each element will be negative for non-real atoms.
 412              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 413              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 414              */
 415             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 416
 417             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 418             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 419             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 420
 421             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 422             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 423             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 424             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 425             j_coord_offsetA  = DIM*jnrA;
 426             j_coord_offsetB  = DIM*jnrB;
 427             j_coord_offsetC  = DIM*jnrC;
 428             j_coord_offsetD  = DIM*jnrD;
 429
 430             /* load j atom coordinates */
 431             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 432                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 433                                                  &jx0,&jy0,&jz0);
 434
 435             /* Calculate displacement vector */
 436             dx00             = _mm256_sub_pd(ix0,jx0);
 437             dy00             = _mm256_sub_pd(iy0,jy0);
 438             dz00             = _mm256_sub_pd(iz0,jz0);
 439             dx10             = _mm256_sub_pd(ix1,jx0);
 440             dy10             = _mm256_sub_pd(iy1,jy0);
 441             dz10             = _mm256_sub_pd(iz1,jz0);
 442             dx20             = _mm256_sub_pd(ix2,jx0);
 443             dy20             = _mm256_sub_pd(iy2,jy0);
 444             dz20             = _mm256_sub_pd(iz2,jz0);
 445
 446             /* Calculate squared distance and things based on it */
 447             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 448             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 449             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 450
 451             rinv00           = avx256_invsqrt_d(rsq00);
 452             rinv10           = avx256_invsqrt_d(rsq10);
 453             rinv20           = avx256_invsqrt_d(rsq20);
 454
 455             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 456
 457             /* Load parameters for j particles */
 458             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 459                                                                  charge+jnrC+0,charge+jnrD+0);
 460             vdwjidx0A        = 2*vdwtype[jnrA+0];
 461             vdwjidx0B        = 2*vdwtype[jnrB+0];
 462             vdwjidx0C        = 2*vdwtype[jnrC+0];
 463             vdwjidx0D        = 2*vdwtype[jnrD+0];
 464
 465             fjx0             = _mm256_setzero_pd();
 466             fjy0             = _mm256_setzero_pd();
 467             fjz0             = _mm256_setzero_pd();
 468
 469             /**************************
 470              * CALCULATE INTERACTIONS *
 471              **************************/
 472
 473             r00              = _mm256_mul_pd(rsq00,rinv00);
 474             r00              = _mm256_andnot_pd(dummy_mask,r00);
 475
 476             /* Compute parameters for interactions between i and j atoms */
 477             qq00             = _mm256_mul_pd(iq0,jq0);
 478             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 479                                             vdwioffsetptr0+vdwjidx0B,
 480                                             vdwioffsetptr0+vdwjidx0C,
 481                                             vdwioffsetptr0+vdwjidx0D,
 482                                             &c6_00,&c12_00);
 483
 484             /* Calculate table index by multiplying r with table scale and truncate to integer */
 485             rt               = _mm256_mul_pd(r00,vftabscale);
 486             vfitab           = _mm256_cvttpd_epi32(rt);
 487             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 488             vfitab           = _mm_slli_epi32(vfitab,2);
 489
 490             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 491             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 492             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 493             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 494             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 495             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 496             Heps             = _mm256_mul_pd(vfeps,H);
 497             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 498             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 499             velec            = _mm256_mul_pd(qq00,VV);
 500             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 501             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 502
 503             /* LENNARD-JONES DISPERSION/REPULSION */
 504
 505             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 506             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 507             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 508             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 509             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm256_andnot_pd(dummy_mask,velec);
 513             velecsum         = _mm256_add_pd(velecsum,velec);
 514             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 515             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 516
 517             fscal            = _mm256_add_pd(felec,fvdw);
 518
 519             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_pd(fscal,dx00);
 523             ty               = _mm256_mul_pd(fscal,dy00);
 524             tz               = _mm256_mul_pd(fscal,dz00);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm256_add_pd(fix0,tx);
 528             fiy0             = _mm256_add_pd(fiy0,ty);
 529             fiz0             = _mm256_add_pd(fiz0,tz);
 530
 531             fjx0             = _mm256_add_pd(fjx0,tx);
 532             fjy0             = _mm256_add_pd(fjy0,ty);
 533             fjz0             = _mm256_add_pd(fjz0,tz);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r10              = _mm256_mul_pd(rsq10,rinv10);
 540             r10              = _mm256_andnot_pd(dummy_mask,r10);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq10             = _mm256_mul_pd(iq1,jq0);
 544
 545             /* Calculate table index by multiplying r with table scale and truncate to integer */
 546             rt               = _mm256_mul_pd(r10,vftabscale);
 547             vfitab           = _mm256_cvttpd_epi32(rt);
 548             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 549             vfitab           = _mm_slli_epi32(vfitab,2);
 550
 551             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 552             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 553             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 554             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 555             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 556             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 557             Heps             = _mm256_mul_pd(vfeps,H);
 558             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 559             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 560             velec            = _mm256_mul_pd(qq10,VV);
 561             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 562             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velec            = _mm256_andnot_pd(dummy_mask,velec);
 566             velecsum         = _mm256_add_pd(velecsum,velec);
 567
 568             fscal            = felec;
 569
 570             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm256_mul_pd(fscal,dx10);
 574             ty               = _mm256_mul_pd(fscal,dy10);
 575             tz               = _mm256_mul_pd(fscal,dz10);
 576
 577             /* Update vectorial force */
 578             fix1             = _mm256_add_pd(fix1,tx);
 579             fiy1             = _mm256_add_pd(fiy1,ty);
 580             fiz1             = _mm256_add_pd(fiz1,tz);
 581
 582             fjx0             = _mm256_add_pd(fjx0,tx);
 583             fjy0             = _mm256_add_pd(fjy0,ty);
 584             fjz0             = _mm256_add_pd(fjz0,tz);
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             r20              = _mm256_mul_pd(rsq20,rinv20);
 591             r20              = _mm256_andnot_pd(dummy_mask,r20);
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq20             = _mm256_mul_pd(iq2,jq0);
 595
 596             /* Calculate table index by multiplying r with table scale and truncate to integer */
 597             rt               = _mm256_mul_pd(r20,vftabscale);
 598             vfitab           = _mm256_cvttpd_epi32(rt);
 599             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 600             vfitab           = _mm_slli_epi32(vfitab,2);
 601
 602             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 603             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 604             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 605             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 606             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 607             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 608             Heps             = _mm256_mul_pd(vfeps,H);
 609             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 610             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 611             velec            = _mm256_mul_pd(qq20,VV);
 612             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 613             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 614
 615             /* Update potential sum for this i atom from the interaction with this j atom. */
 616             velec            = _mm256_andnot_pd(dummy_mask,velec);
 617             velecsum         = _mm256_add_pd(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm256_mul_pd(fscal,dx20);
 625             ty               = _mm256_mul_pd(fscal,dy20);
 626             tz               = _mm256_mul_pd(fscal,dz20);
 627
 628             /* Update vectorial force */
 629             fix2             = _mm256_add_pd(fix2,tx);
 630             fiy2             = _mm256_add_pd(fiy2,ty);
 631             fiz2             = _mm256_add_pd(fiz2,tz);
 632
 633             fjx0             = _mm256_add_pd(fjx0,tx);
 634             fjy0             = _mm256_add_pd(fjy0,ty);
 635             fjz0             = _mm256_add_pd(fjz0,tz);
 636
 637             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 638             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 639             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 640             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 641
 642             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 643
 644             /* Inner loop uses 148 flops */
 645         }
 646
 647         /* End of innermost loop */
 648
 649         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 650                                                  f+i_coord_offset,fshift+i_shift_offset);
 651
 652         ggid                        = gid[iidx];
 653         /* Update potential energies */
 654         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 655         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 656
 657         /* Increment number of inner iterations */
 658         inneriter                  += j_index_end - j_index_start;
 659
 660         /* Outer loop uses 20 flops */
 661     }
 662
 663     /* Increment number of outer iterations */
 664     outeriter        += nri;
 665
 666     /* Update outer/inner flops */
 667
 668     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 669 }
 670 /*
 671  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_double
 672  * Electrostatics interaction: CubicSplineTable
 673  * VdW interaction:            LennardJones
 674  * Geometry:                   Water3-Particle
 675  * Calculate force/pot:        Force
 676  */
 677 void
 678 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_256_double
 679                     (t_nblist                    * gmx_restrict       nlist,
 680                      rvec                        * gmx_restrict          xx,
 681                      rvec                        * gmx_restrict          ff,
 682                      struct t_forcerec           * gmx_restrict          fr,
 683                      t_mdatoms                   * gmx_restrict     mdatoms,
 684                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 685                      t_nrnb                      * gmx_restrict        nrnb)
 686 {
 687     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 688      * just 0 for non-waters.
 689      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 690      * jnr indices corresponding to data put in the four positions in the SIMD register.
 691      */
 692     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 693     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 694     int              jnrA,jnrB,jnrC,jnrD;
 695     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 696     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 697     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 698     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 699     real             rcutoff_scalar;
 700     real             *shiftvec,*fshift,*x,*f;
 701     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 702     real             scratch[4*DIM];
 703     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 704     real *           vdwioffsetptr0;
 705     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 706     real *           vdwioffsetptr1;
 707     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 708     real *           vdwioffsetptr2;
 709     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 710     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 711     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 712     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 713     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 714     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 715     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 716     real             *charge;
 717     int              nvdwtype;
 718     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 719     int              *vdwtype;
 720     real             *vdwparam;
 721     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 722     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 723     __m128i          vfitab;
 724     __m128i          ifour       = _mm_set1_epi32(4);
 725     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 726     real             *vftab;
 727     __m256d          dummy_mask,cutoff_mask;
 728     __m128           tmpmask0,tmpmask1;
 729     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 730     __m256d          one     = _mm256_set1_pd(1.0);
 731     __m256d          two     = _mm256_set1_pd(2.0);
 732     x                = xx[0];
 733     f                = ff[0];
 734
 735     nri              = nlist->nri;
 736     iinr             = nlist->iinr;
 737     jindex           = nlist->jindex;
 738     jjnr             = nlist->jjnr;
 739     shiftidx         = nlist->shift;
 740     gid              = nlist->gid;
 741     shiftvec         = fr->shift_vec[0];
 742     fshift           = fr->fshift[0];
 743     facel            = _mm256_set1_pd(fr->ic->epsfac);
 744     charge           = mdatoms->chargeA;
 745     nvdwtype         = fr->ntype;
 746     vdwparam         = fr->nbfp;
 747     vdwtype          = mdatoms->typeA;
 748
 749     vftab            = kernel_data->table_elec->data;
 750     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 751
 752     /* Setup water-specific parameters */
 753     inr              = nlist->iinr[0];
 754     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 755     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 756     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 757     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 758
 759     /* Avoid stupid compiler warnings */
 760     jnrA = jnrB = jnrC = jnrD = 0;
 761     j_coord_offsetA = 0;
 762     j_coord_offsetB = 0;
 763     j_coord_offsetC = 0;
 764     j_coord_offsetD = 0;
 765
 766     outeriter        = 0;
 767     inneriter        = 0;
 768
 769     for(iidx=0;iidx<4*DIM;iidx++)
 770     {
 771         scratch[iidx] = 0.0;
 772     }
 773
 774     /* Start outer loop over neighborlists */
 775     for(iidx=0; iidx<nri; iidx++)
 776     {
 777         /* Load shift vector for this list */
 778         i_shift_offset   = DIM*shiftidx[iidx];
 779
 780         /* Load limits for loop over neighbors */
 781         j_index_start    = jindex[iidx];
 782         j_index_end      = jindex[iidx+1];
 783
 784         /* Get outer coordinate index */
 785         inr              = iinr[iidx];
 786         i_coord_offset   = DIM*inr;
 787
 788         /* Load i particle coords and add shift vector */
 789         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 790                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 791
 792         fix0             = _mm256_setzero_pd();
 793         fiy0             = _mm256_setzero_pd();
 794         fiz0             = _mm256_setzero_pd();
 795         fix1             = _mm256_setzero_pd();
 796         fiy1             = _mm256_setzero_pd();
 797         fiz1             = _mm256_setzero_pd();
 798         fix2             = _mm256_setzero_pd();
 799         fiy2             = _mm256_setzero_pd();
 800         fiz2             = _mm256_setzero_pd();
 801
 802         /* Start inner kernel loop */
 803         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 804         {
 805
 806             /* Get j neighbor index, and coordinate index */
 807             jnrA             = jjnr[jidx];
 808             jnrB             = jjnr[jidx+1];
 809             jnrC             = jjnr[jidx+2];
 810             jnrD             = jjnr[jidx+3];
 811             j_coord_offsetA  = DIM*jnrA;
 812             j_coord_offsetB  = DIM*jnrB;
 813             j_coord_offsetC  = DIM*jnrC;
 814             j_coord_offsetD  = DIM*jnrD;
 815
 816             /* load j atom coordinates */
 817             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 818                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 819                                                  &jx0,&jy0,&jz0);
 820
 821             /* Calculate displacement vector */
 822             dx00             = _mm256_sub_pd(ix0,jx0);
 823             dy00             = _mm256_sub_pd(iy0,jy0);
 824             dz00             = _mm256_sub_pd(iz0,jz0);
 825             dx10             = _mm256_sub_pd(ix1,jx0);
 826             dy10             = _mm256_sub_pd(iy1,jy0);
 827             dz10             = _mm256_sub_pd(iz1,jz0);
 828             dx20             = _mm256_sub_pd(ix2,jx0);
 829             dy20             = _mm256_sub_pd(iy2,jy0);
 830             dz20             = _mm256_sub_pd(iz2,jz0);
 831
 832             /* Calculate squared distance and things based on it */
 833             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 834             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 835             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 836
 837             rinv00           = avx256_invsqrt_d(rsq00);
 838             rinv10           = avx256_invsqrt_d(rsq10);
 839             rinv20           = avx256_invsqrt_d(rsq20);
 840
 841             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 842
 843             /* Load parameters for j particles */
 844             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 845                                                                  charge+jnrC+0,charge+jnrD+0);
 846             vdwjidx0A        = 2*vdwtype[jnrA+0];
 847             vdwjidx0B        = 2*vdwtype[jnrB+0];
 848             vdwjidx0C        = 2*vdwtype[jnrC+0];
 849             vdwjidx0D        = 2*vdwtype[jnrD+0];
 850
 851             fjx0             = _mm256_setzero_pd();
 852             fjy0             = _mm256_setzero_pd();
 853             fjz0             = _mm256_setzero_pd();
 854
 855             /**************************
 856              * CALCULATE INTERACTIONS *
 857              **************************/
 858
 859             r00              = _mm256_mul_pd(rsq00,rinv00);
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             qq00             = _mm256_mul_pd(iq0,jq0);
 863             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 864                                             vdwioffsetptr0+vdwjidx0B,
 865                                             vdwioffsetptr0+vdwjidx0C,
 866                                             vdwioffsetptr0+vdwjidx0D,
 867                                             &c6_00,&c12_00);
 868
 869             /* Calculate table index by multiplying r with table scale and truncate to integer */
 870             rt               = _mm256_mul_pd(r00,vftabscale);
 871             vfitab           = _mm256_cvttpd_epi32(rt);
 872             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 873             vfitab           = _mm_slli_epi32(vfitab,2);
 874
 875             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 876             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 877             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 878             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 879             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 880             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 881             Heps             = _mm256_mul_pd(vfeps,H);
 882             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 883             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 884             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 885
 886             /* LENNARD-JONES DISPERSION/REPULSION */
 887
 888             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 889             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 890
 891             fscal            = _mm256_add_pd(felec,fvdw);
 892
 893             /* Calculate temporary vectorial force */
 894             tx               = _mm256_mul_pd(fscal,dx00);
 895             ty               = _mm256_mul_pd(fscal,dy00);
 896             tz               = _mm256_mul_pd(fscal,dz00);
 897
 898             /* Update vectorial force */
 899             fix0             = _mm256_add_pd(fix0,tx);
 900             fiy0             = _mm256_add_pd(fiy0,ty);
 901             fiz0             = _mm256_add_pd(fiz0,tz);
 902
 903             fjx0             = _mm256_add_pd(fjx0,tx);
 904             fjy0             = _mm256_add_pd(fjy0,ty);
 905             fjz0             = _mm256_add_pd(fjz0,tz);
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             r10              = _mm256_mul_pd(rsq10,rinv10);
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             qq10             = _mm256_mul_pd(iq1,jq0);
 915
 916             /* Calculate table index by multiplying r with table scale and truncate to integer */
 917             rt               = _mm256_mul_pd(r10,vftabscale);
 918             vfitab           = _mm256_cvttpd_epi32(rt);
 919             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 920             vfitab           = _mm_slli_epi32(vfitab,2);
 921
 922             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 923             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 924             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 925             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 926             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 927             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 928             Heps             = _mm256_mul_pd(vfeps,H);
 929             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 930             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 931             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 932
 933             fscal            = felec;
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm256_mul_pd(fscal,dx10);
 937             ty               = _mm256_mul_pd(fscal,dy10);
 938             tz               = _mm256_mul_pd(fscal,dz10);
 939
 940             /* Update vectorial force */
 941             fix1             = _mm256_add_pd(fix1,tx);
 942             fiy1             = _mm256_add_pd(fiy1,ty);
 943             fiz1             = _mm256_add_pd(fiz1,tz);
 944
 945             fjx0             = _mm256_add_pd(fjx0,tx);
 946             fjy0             = _mm256_add_pd(fjy0,ty);
 947             fjz0             = _mm256_add_pd(fjz0,tz);
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             r20              = _mm256_mul_pd(rsq20,rinv20);
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq20             = _mm256_mul_pd(iq2,jq0);
 957
 958             /* Calculate table index by multiplying r with table scale and truncate to integer */
 959             rt               = _mm256_mul_pd(r20,vftabscale);
 960             vfitab           = _mm256_cvttpd_epi32(rt);
 961             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 962             vfitab           = _mm_slli_epi32(vfitab,2);
 963
 964             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 965             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 966             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 967             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 968             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 969             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 970             Heps             = _mm256_mul_pd(vfeps,H);
 971             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 972             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 973             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 974
 975             fscal            = felec;
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_pd(fscal,dx20);
 979             ty               = _mm256_mul_pd(fscal,dy20);
 980             tz               = _mm256_mul_pd(fscal,dz20);
 981
 982             /* Update vectorial force */
 983             fix2             = _mm256_add_pd(fix2,tx);
 984             fiy2             = _mm256_add_pd(fiy2,ty);
 985             fiz2             = _mm256_add_pd(fiz2,tz);
 986
 987             fjx0             = _mm256_add_pd(fjx0,tx);
 988             fjy0             = _mm256_add_pd(fjy0,ty);
 989             fjz0             = _mm256_add_pd(fjz0,tz);
 990
 991             fjptrA             = f+j_coord_offsetA;
 992             fjptrB             = f+j_coord_offsetB;
 993             fjptrC             = f+j_coord_offsetC;
 994             fjptrD             = f+j_coord_offsetD;
 995
 996             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 997
 998             /* Inner loop uses 128 flops */
 999         }
1000
1001         if(jidx<j_index_end)
1002         {
1003
1004             /* Get j neighbor index, and coordinate index */
1005             jnrlistA         = jjnr[jidx];
1006             jnrlistB         = jjnr[jidx+1];
1007             jnrlistC         = jjnr[jidx+2];
1008             jnrlistD         = jjnr[jidx+3];
1009             /* Sign of each element will be negative for non-real atoms.
1010              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1011              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1012              */
1013             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1014
1015             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1016             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1017             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1018
1019             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1020             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1021             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1022             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1023             j_coord_offsetA  = DIM*jnrA;
1024             j_coord_offsetB  = DIM*jnrB;
1025             j_coord_offsetC  = DIM*jnrC;
1026             j_coord_offsetD  = DIM*jnrD;
1027
1028             /* load j atom coordinates */
1029             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1030                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1031                                                  &jx0,&jy0,&jz0);
1032
1033             /* Calculate displacement vector */
1034             dx00             = _mm256_sub_pd(ix0,jx0);
1035             dy00             = _mm256_sub_pd(iy0,jy0);
1036             dz00             = _mm256_sub_pd(iz0,jz0);
1037             dx10             = _mm256_sub_pd(ix1,jx0);
1038             dy10             = _mm256_sub_pd(iy1,jy0);
1039             dz10             = _mm256_sub_pd(iz1,jz0);
1040             dx20             = _mm256_sub_pd(ix2,jx0);
1041             dy20             = _mm256_sub_pd(iy2,jy0);
1042             dz20             = _mm256_sub_pd(iz2,jz0);
1043
1044             /* Calculate squared distance and things based on it */
1045             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1046             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1047             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1048
1049             rinv00           = avx256_invsqrt_d(rsq00);
1050             rinv10           = avx256_invsqrt_d(rsq10);
1051             rinv20           = avx256_invsqrt_d(rsq20);
1052
1053             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1054
1055             /* Load parameters for j particles */
1056             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1057                                                                  charge+jnrC+0,charge+jnrD+0);
1058             vdwjidx0A        = 2*vdwtype[jnrA+0];
1059             vdwjidx0B        = 2*vdwtype[jnrB+0];
1060             vdwjidx0C        = 2*vdwtype[jnrC+0];
1061             vdwjidx0D        = 2*vdwtype[jnrD+0];
1062
1063             fjx0             = _mm256_setzero_pd();
1064             fjy0             = _mm256_setzero_pd();
1065             fjz0             = _mm256_setzero_pd();
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             r00              = _mm256_mul_pd(rsq00,rinv00);
1072             r00              = _mm256_andnot_pd(dummy_mask,r00);
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             qq00             = _mm256_mul_pd(iq0,jq0);
1076             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1077                                             vdwioffsetptr0+vdwjidx0B,
1078                                             vdwioffsetptr0+vdwjidx0C,
1079                                             vdwioffsetptr0+vdwjidx0D,
1080                                             &c6_00,&c12_00);
1081
1082             /* Calculate table index by multiplying r with table scale and truncate to integer */
1083             rt               = _mm256_mul_pd(r00,vftabscale);
1084             vfitab           = _mm256_cvttpd_epi32(rt);
1085             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1086             vfitab           = _mm_slli_epi32(vfitab,2);
1087
1088             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1089             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1090             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1091             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1092             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1093             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1094             Heps             = _mm256_mul_pd(vfeps,H);
1095             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1096             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1097             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1098
1099             /* LENNARD-JONES DISPERSION/REPULSION */
1100
1101             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1102             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1103
1104             fscal            = _mm256_add_pd(felec,fvdw);
1105
1106             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1107
1108             /* Calculate temporary vectorial force */
1109             tx               = _mm256_mul_pd(fscal,dx00);
1110             ty               = _mm256_mul_pd(fscal,dy00);
1111             tz               = _mm256_mul_pd(fscal,dz00);
1112
1113             /* Update vectorial force */
1114             fix0             = _mm256_add_pd(fix0,tx);
1115             fiy0             = _mm256_add_pd(fiy0,ty);
1116             fiz0             = _mm256_add_pd(fiz0,tz);
1117
1118             fjx0             = _mm256_add_pd(fjx0,tx);
1119             fjy0             = _mm256_add_pd(fjy0,ty);
1120             fjz0             = _mm256_add_pd(fjz0,tz);
1121
1122             /**************************
1123              * CALCULATE INTERACTIONS *
1124              **************************/
1125
1126             r10              = _mm256_mul_pd(rsq10,rinv10);
1127             r10              = _mm256_andnot_pd(dummy_mask,r10);
1128
1129             /* Compute parameters for interactions between i and j atoms */
1130             qq10             = _mm256_mul_pd(iq1,jq0);
1131
1132             /* Calculate table index by multiplying r with table scale and truncate to integer */
1133             rt               = _mm256_mul_pd(r10,vftabscale);
1134             vfitab           = _mm256_cvttpd_epi32(rt);
1135             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1136             vfitab           = _mm_slli_epi32(vfitab,2);
1137
1138             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1139             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1140             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1141             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1142             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1143             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1144             Heps             = _mm256_mul_pd(vfeps,H);
1145             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1146             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1147             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1148
1149             fscal            = felec;
1150
1151             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1152
1153             /* Calculate temporary vectorial force */
1154             tx               = _mm256_mul_pd(fscal,dx10);
1155             ty               = _mm256_mul_pd(fscal,dy10);
1156             tz               = _mm256_mul_pd(fscal,dz10);
1157
1158             /* Update vectorial force */
1159             fix1             = _mm256_add_pd(fix1,tx);
1160             fiy1             = _mm256_add_pd(fiy1,ty);
1161             fiz1             = _mm256_add_pd(fiz1,tz);
1162
1163             fjx0             = _mm256_add_pd(fjx0,tx);
1164             fjy0             = _mm256_add_pd(fjy0,ty);
1165             fjz0             = _mm256_add_pd(fjz0,tz);
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             r20              = _mm256_mul_pd(rsq20,rinv20);
1172             r20              = _mm256_andnot_pd(dummy_mask,r20);
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             qq20             = _mm256_mul_pd(iq2,jq0);
1176
1177             /* Calculate table index by multiplying r with table scale and truncate to integer */
1178             rt               = _mm256_mul_pd(r20,vftabscale);
1179             vfitab           = _mm256_cvttpd_epi32(rt);
1180             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1181             vfitab           = _mm_slli_epi32(vfitab,2);
1182
1183             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1184             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1185             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1186             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1187             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1188             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1189             Heps             = _mm256_mul_pd(vfeps,H);
1190             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1191             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1192             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1193
1194             fscal            = felec;
1195
1196             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1197
1198             /* Calculate temporary vectorial force */
1199             tx               = _mm256_mul_pd(fscal,dx20);
1200             ty               = _mm256_mul_pd(fscal,dy20);
1201             tz               = _mm256_mul_pd(fscal,dz20);
1202
1203             /* Update vectorial force */
1204             fix2             = _mm256_add_pd(fix2,tx);
1205             fiy2             = _mm256_add_pd(fiy2,ty);
1206             fiz2             = _mm256_add_pd(fiz2,tz);
1207
1208             fjx0             = _mm256_add_pd(fjx0,tx);
1209             fjy0             = _mm256_add_pd(fjy0,ty);
1210             fjz0             = _mm256_add_pd(fjz0,tz);
1211
1212             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1213             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1214             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1215             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1216
1217             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1218
1219             /* Inner loop uses 131 flops */
1220         }
1221
1222         /* End of innermost loop */
1223
1224         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1225                                                  f+i_coord_offset,fshift+i_shift_offset);
1226
1227         /* Increment number of inner iterations */
1228         inneriter                  += j_index_end - j_index_start;
1229
1230         /* Outer loop uses 18 flops */
1231     }
1232
1233     /* Increment number of outer iterations */
1234     outeriter        += nri;
1235
1236     /* Update outer/inner flops */
1237
1238     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1239 }