src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_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     krf              = _mm256_set1_pd(fr->ic->k_rf);
 128     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 129     crf              = _mm256_set1_pd(fr->ic->c_rf);
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_vdw->data;
 135     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 140     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 141     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 145     rcutoff_scalar   = fr->rcoulomb;
 146     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 147     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = jnrC = jnrD = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153     j_coord_offsetC = 0;
 154     j_coord_offsetD = 0;
 155
 156     outeriter        = 0;
 157     inneriter        = 0;
 158
 159     for(iidx=0;iidx<4*DIM;iidx++)
 160     {
 161         scratch[iidx] = 0.0;
 162     }
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 181
 182         fix0             = _mm256_setzero_pd();
 183         fiy0             = _mm256_setzero_pd();
 184         fiz0             = _mm256_setzero_pd();
 185         fix1             = _mm256_setzero_pd();
 186         fiy1             = _mm256_setzero_pd();
 187         fiz1             = _mm256_setzero_pd();
 188         fix2             = _mm256_setzero_pd();
 189         fiy2             = _mm256_setzero_pd();
 190         fiz2             = _mm256_setzero_pd();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm256_setzero_pd();
 194         vvdwsum          = _mm256_setzero_pd();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209
 210             /* load j atom coordinates */
 211             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 212                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 213                                                  &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm256_sub_pd(ix0,jx0);
 217             dy00             = _mm256_sub_pd(iy0,jy0);
 218             dz00             = _mm256_sub_pd(iz0,jz0);
 219             dx10             = _mm256_sub_pd(ix1,jx0);
 220             dy10             = _mm256_sub_pd(iy1,jy0);
 221             dz10             = _mm256_sub_pd(iz1,jz0);
 222             dx20             = _mm256_sub_pd(ix2,jx0);
 223             dy20             = _mm256_sub_pd(iy2,jy0);
 224             dz20             = _mm256_sub_pd(iz2,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 228             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 229             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 230
 231             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 232             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 233             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 234
 235             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 236             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 237             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 241                                                                  charge+jnrC+0,charge+jnrD+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246
 247             fjx0             = _mm256_setzero_pd();
 248             fjy0             = _mm256_setzero_pd();
 249             fjz0             = _mm256_setzero_pd();
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 256             {
 257
 258             r00              = _mm256_mul_pd(rsq00,rinv00);
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq00             = _mm256_mul_pd(iq0,jq0);
 262             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 263                                             vdwioffsetptr0+vdwjidx0B,
 264                                             vdwioffsetptr0+vdwjidx0C,
 265                                             vdwioffsetptr0+vdwjidx0D,
 266                                             &c6_00,&c12_00);
 267
 268             /* Calculate table index by multiplying r with table scale and truncate to integer */
 269             rt               = _mm256_mul_pd(r00,vftabscale);
 270             vfitab           = _mm256_cvttpd_epi32(rt);
 271             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 272             vfitab           = _mm_slli_epi32(vfitab,3);
 273
 274             /* REACTION-FIELD ELECTROSTATICS */
 275             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 276             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 277
 278             /* CUBIC SPLINE TABLE DISPERSION */
 279             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 280             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 281             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 282             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 283             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 284             Heps             = _mm256_mul_pd(vfeps,H);
 285             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 286             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 287             vvdw6            = _mm256_mul_pd(c6_00,VV);
 288             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 289             fvdw6            = _mm256_mul_pd(c6_00,FF);
 290
 291             /* CUBIC SPLINE TABLE REPULSION */
 292             vfitab           = _mm_add_epi32(vfitab,ifour);
 293             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 294             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 295             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 296             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 297             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 298             Heps             = _mm256_mul_pd(vfeps,H);
 299             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 300             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 301             vvdw12           = _mm256_mul_pd(c12_00,VV);
 302             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 303             fvdw12           = _mm256_mul_pd(c12_00,FF);
 304             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 305             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 306
 307             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velec            = _mm256_and_pd(velec,cutoff_mask);
 311             velecsum         = _mm256_add_pd(velecsum,velec);
 312             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 313             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 314
 315             fscal            = _mm256_add_pd(felec,fvdw);
 316
 317             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm256_mul_pd(fscal,dx00);
 321             ty               = _mm256_mul_pd(fscal,dy00);
 322             tz               = _mm256_mul_pd(fscal,dz00);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm256_add_pd(fix0,tx);
 326             fiy0             = _mm256_add_pd(fiy0,ty);
 327             fiz0             = _mm256_add_pd(fiz0,tz);
 328
 329             fjx0             = _mm256_add_pd(fjx0,tx);
 330             fjy0             = _mm256_add_pd(fjy0,ty);
 331             fjz0             = _mm256_add_pd(fjz0,tz);
 332
 333             }
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 340             {
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq10             = _mm256_mul_pd(iq1,jq0);
 344
 345             /* REACTION-FIELD ELECTROSTATICS */
 346             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 347             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 348
 349             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm256_and_pd(velec,cutoff_mask);
 353             velecsum         = _mm256_add_pd(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm256_mul_pd(fscal,dx10);
 361             ty               = _mm256_mul_pd(fscal,dy10);
 362             tz               = _mm256_mul_pd(fscal,dz10);
 363
 364             /* Update vectorial force */
 365             fix1             = _mm256_add_pd(fix1,tx);
 366             fiy1             = _mm256_add_pd(fiy1,ty);
 367             fiz1             = _mm256_add_pd(fiz1,tz);
 368
 369             fjx0             = _mm256_add_pd(fjx0,tx);
 370             fjy0             = _mm256_add_pd(fjy0,ty);
 371             fjz0             = _mm256_add_pd(fjz0,tz);
 372
 373             }
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 380             {
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq20             = _mm256_mul_pd(iq2,jq0);
 384
 385             /* REACTION-FIELD ELECTROSTATICS */
 386             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 387             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 388
 389             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velec            = _mm256_and_pd(velec,cutoff_mask);
 393             velecsum         = _mm256_add_pd(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm256_mul_pd(fscal,dx20);
 401             ty               = _mm256_mul_pd(fscal,dy20);
 402             tz               = _mm256_mul_pd(fscal,dz20);
 403
 404             /* Update vectorial force */
 405             fix2             = _mm256_add_pd(fix2,tx);
 406             fiy2             = _mm256_add_pd(fiy2,ty);
 407             fiz2             = _mm256_add_pd(fiz2,tz);
 408
 409             fjx0             = _mm256_add_pd(fjx0,tx);
 410             fjy0             = _mm256_add_pd(fjy0,ty);
 411             fjz0             = _mm256_add_pd(fjz0,tz);
 412
 413             }
 414
 415             fjptrA             = f+j_coord_offsetA;
 416             fjptrB             = f+j_coord_offsetB;
 417             fjptrC             = f+j_coord_offsetC;
 418             fjptrD             = f+j_coord_offsetD;
 419
 420             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 421
 422             /* Inner loop uses 147 flops */
 423         }
 424
 425         if(jidx<j_index_end)
 426         {
 427
 428             /* Get j neighbor index, and coordinate index */
 429             jnrlistA         = jjnr[jidx];
 430             jnrlistB         = jjnr[jidx+1];
 431             jnrlistC         = jjnr[jidx+2];
 432             jnrlistD         = jjnr[jidx+3];
 433             /* Sign of each element will be negative for non-real atoms.
 434              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 435              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 436              */
 437             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 438
 439             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 440             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 441             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 442
 443             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 444             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 445             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 446             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 447             j_coord_offsetA  = DIM*jnrA;
 448             j_coord_offsetB  = DIM*jnrB;
 449             j_coord_offsetC  = DIM*jnrC;
 450             j_coord_offsetD  = DIM*jnrD;
 451
 452             /* load j atom coordinates */
 453             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 454                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 455                                                  &jx0,&jy0,&jz0);
 456
 457             /* Calculate displacement vector */
 458             dx00             = _mm256_sub_pd(ix0,jx0);
 459             dy00             = _mm256_sub_pd(iy0,jy0);
 460             dz00             = _mm256_sub_pd(iz0,jz0);
 461             dx10             = _mm256_sub_pd(ix1,jx0);
 462             dy10             = _mm256_sub_pd(iy1,jy0);
 463             dz10             = _mm256_sub_pd(iz1,jz0);
 464             dx20             = _mm256_sub_pd(ix2,jx0);
 465             dy20             = _mm256_sub_pd(iy2,jy0);
 466             dz20             = _mm256_sub_pd(iz2,jz0);
 467
 468             /* Calculate squared distance and things based on it */
 469             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 470             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 471             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 472
 473             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 474             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 475             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 476
 477             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 478             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 479             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 480
 481             /* Load parameters for j particles */
 482             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 483                                                                  charge+jnrC+0,charge+jnrD+0);
 484             vdwjidx0A        = 2*vdwtype[jnrA+0];
 485             vdwjidx0B        = 2*vdwtype[jnrB+0];
 486             vdwjidx0C        = 2*vdwtype[jnrC+0];
 487             vdwjidx0D        = 2*vdwtype[jnrD+0];
 488
 489             fjx0             = _mm256_setzero_pd();
 490             fjy0             = _mm256_setzero_pd();
 491             fjz0             = _mm256_setzero_pd();
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 498             {
 499
 500             r00              = _mm256_mul_pd(rsq00,rinv00);
 501             r00              = _mm256_andnot_pd(dummy_mask,r00);
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             qq00             = _mm256_mul_pd(iq0,jq0);
 505             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 506                                             vdwioffsetptr0+vdwjidx0B,
 507                                             vdwioffsetptr0+vdwjidx0C,
 508                                             vdwioffsetptr0+vdwjidx0D,
 509                                             &c6_00,&c12_00);
 510
 511             /* Calculate table index by multiplying r with table scale and truncate to integer */
 512             rt               = _mm256_mul_pd(r00,vftabscale);
 513             vfitab           = _mm256_cvttpd_epi32(rt);
 514             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 515             vfitab           = _mm_slli_epi32(vfitab,3);
 516
 517             /* REACTION-FIELD ELECTROSTATICS */
 518             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 519             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 520
 521             /* CUBIC SPLINE TABLE DISPERSION */
 522             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 523             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 524             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 525             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 526             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 527             Heps             = _mm256_mul_pd(vfeps,H);
 528             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 529             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 530             vvdw6            = _mm256_mul_pd(c6_00,VV);
 531             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 532             fvdw6            = _mm256_mul_pd(c6_00,FF);
 533
 534             /* CUBIC SPLINE TABLE REPULSION */
 535             vfitab           = _mm_add_epi32(vfitab,ifour);
 536             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 537             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 538             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 539             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 540             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 541             Heps             = _mm256_mul_pd(vfeps,H);
 542             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 543             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 544             vvdw12           = _mm256_mul_pd(c12_00,VV);
 545             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 546             fvdw12           = _mm256_mul_pd(c12_00,FF);
 547             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 548             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 549
 550             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm256_and_pd(velec,cutoff_mask);
 554             velec            = _mm256_andnot_pd(dummy_mask,velec);
 555             velecsum         = _mm256_add_pd(velecsum,velec);
 556             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 557             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 558             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 559
 560             fscal            = _mm256_add_pd(felec,fvdw);
 561
 562             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 563
 564             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm256_mul_pd(fscal,dx00);
 568             ty               = _mm256_mul_pd(fscal,dy00);
 569             tz               = _mm256_mul_pd(fscal,dz00);
 570
 571             /* Update vectorial force */
 572             fix0             = _mm256_add_pd(fix0,tx);
 573             fiy0             = _mm256_add_pd(fiy0,ty);
 574             fiz0             = _mm256_add_pd(fiz0,tz);
 575
 576             fjx0             = _mm256_add_pd(fjx0,tx);
 577             fjy0             = _mm256_add_pd(fjy0,ty);
 578             fjz0             = _mm256_add_pd(fjz0,tz);
 579
 580             }
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 587             {
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq10             = _mm256_mul_pd(iq1,jq0);
 591
 592             /* REACTION-FIELD ELECTROSTATICS */
 593             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 594             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 595
 596             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velec            = _mm256_and_pd(velec,cutoff_mask);
 600             velec            = _mm256_andnot_pd(dummy_mask,velec);
 601             velecsum         = _mm256_add_pd(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 606
 607             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm256_mul_pd(fscal,dx10);
 611             ty               = _mm256_mul_pd(fscal,dy10);
 612             tz               = _mm256_mul_pd(fscal,dz10);
 613
 614             /* Update vectorial force */
 615             fix1             = _mm256_add_pd(fix1,tx);
 616             fiy1             = _mm256_add_pd(fiy1,ty);
 617             fiz1             = _mm256_add_pd(fiz1,tz);
 618
 619             fjx0             = _mm256_add_pd(fjx0,tx);
 620             fjy0             = _mm256_add_pd(fjy0,ty);
 621             fjz0             = _mm256_add_pd(fjz0,tz);
 622
 623             }
 624
 625             /**************************
 626              * CALCULATE INTERACTIONS *
 627              **************************/
 628
 629             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 630             {
 631
 632             /* Compute parameters for interactions between i and j atoms */
 633             qq20             = _mm256_mul_pd(iq2,jq0);
 634
 635             /* REACTION-FIELD ELECTROSTATICS */
 636             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 637             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 638
 639             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm256_and_pd(velec,cutoff_mask);
 643             velec            = _mm256_andnot_pd(dummy_mask,velec);
 644             velecsum         = _mm256_add_pd(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 649
 650             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm256_mul_pd(fscal,dx20);
 654             ty               = _mm256_mul_pd(fscal,dy20);
 655             tz               = _mm256_mul_pd(fscal,dz20);
 656
 657             /* Update vectorial force */
 658             fix2             = _mm256_add_pd(fix2,tx);
 659             fiy2             = _mm256_add_pd(fiy2,ty);
 660             fiz2             = _mm256_add_pd(fiz2,tz);
 661
 662             fjx0             = _mm256_add_pd(fjx0,tx);
 663             fjy0             = _mm256_add_pd(fjy0,ty);
 664             fjz0             = _mm256_add_pd(fjz0,tz);
 665
 666             }
 667
 668             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 669             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 670             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 671             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 672
 673             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 674
 675             /* Inner loop uses 148 flops */
 676         }
 677
 678         /* End of innermost loop */
 679
 680         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 681                                                  f+i_coord_offset,fshift+i_shift_offset);
 682
 683         ggid                        = gid[iidx];
 684         /* Update potential energies */
 685         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 686         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 687
 688         /* Increment number of inner iterations */
 689         inneriter                  += j_index_end - j_index_start;
 690
 691         /* Outer loop uses 20 flops */
 692     }
 693
 694     /* Increment number of outer iterations */
 695     outeriter        += nri;
 696
 697     /* Update outer/inner flops */
 698
 699     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 700 }
 701 /*
 702  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_double
 703  * Electrostatics interaction: ReactionField
 704  * VdW interaction:            CubicSplineTable
 705  * Geometry:                   Water3-Particle
 706  * Calculate force/pot:        Force
 707  */
 708 void
 709 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_double
 710                     (t_nblist                    * gmx_restrict       nlist,
 711                      rvec                        * gmx_restrict          xx,
 712                      rvec                        * gmx_restrict          ff,
 713                      t_forcerec                  * gmx_restrict          fr,
 714                      t_mdatoms                   * gmx_restrict     mdatoms,
 715                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 716                      t_nrnb                      * gmx_restrict        nrnb)
 717 {
 718     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 719      * just 0 for non-waters.
 720      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 721      * jnr indices corresponding to data put in the four positions in the SIMD register.
 722      */
 723     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 724     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 725     int              jnrA,jnrB,jnrC,jnrD;
 726     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 727     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 728     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 729     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 730     real             rcutoff_scalar;
 731     real             *shiftvec,*fshift,*x,*f;
 732     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 733     real             scratch[4*DIM];
 734     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 735     real *           vdwioffsetptr0;
 736     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 737     real *           vdwioffsetptr1;
 738     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 739     real *           vdwioffsetptr2;
 740     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 741     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 742     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 743     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 744     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 745     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 746     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 747     real             *charge;
 748     int              nvdwtype;
 749     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 750     int              *vdwtype;
 751     real             *vdwparam;
 752     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 753     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 754     __m128i          vfitab;
 755     __m128i          ifour       = _mm_set1_epi32(4);
 756     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 757     real             *vftab;
 758     __m256d          dummy_mask,cutoff_mask;
 759     __m128           tmpmask0,tmpmask1;
 760     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 761     __m256d          one     = _mm256_set1_pd(1.0);
 762     __m256d          two     = _mm256_set1_pd(2.0);
 763     x                = xx[0];
 764     f                = ff[0];
 765
 766     nri              = nlist->nri;
 767     iinr             = nlist->iinr;
 768     jindex           = nlist->jindex;
 769     jjnr             = nlist->jjnr;
 770     shiftidx         = nlist->shift;
 771     gid              = nlist->gid;
 772     shiftvec         = fr->shift_vec[0];
 773     fshift           = fr->fshift[0];
 774     facel            = _mm256_set1_pd(fr->epsfac);
 775     charge           = mdatoms->chargeA;
 776     krf              = _mm256_set1_pd(fr->ic->k_rf);
 777     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 778     crf              = _mm256_set1_pd(fr->ic->c_rf);
 779     nvdwtype         = fr->ntype;
 780     vdwparam         = fr->nbfp;
 781     vdwtype          = mdatoms->typeA;
 782
 783     vftab            = kernel_data->table_vdw->data;
 784     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 785
 786     /* Setup water-specific parameters */
 787     inr              = nlist->iinr[0];
 788     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 789     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 790     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 791     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 792
 793     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 794     rcutoff_scalar   = fr->rcoulomb;
 795     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 796     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 797
 798     /* Avoid stupid compiler warnings */
 799     jnrA = jnrB = jnrC = jnrD = 0;
 800     j_coord_offsetA = 0;
 801     j_coord_offsetB = 0;
 802     j_coord_offsetC = 0;
 803     j_coord_offsetD = 0;
 804
 805     outeriter        = 0;
 806     inneriter        = 0;
 807
 808     for(iidx=0;iidx<4*DIM;iidx++)
 809     {
 810         scratch[iidx] = 0.0;
 811     }
 812
 813     /* Start outer loop over neighborlists */
 814     for(iidx=0; iidx<nri; iidx++)
 815     {
 816         /* Load shift vector for this list */
 817         i_shift_offset   = DIM*shiftidx[iidx];
 818
 819         /* Load limits for loop over neighbors */
 820         j_index_start    = jindex[iidx];
 821         j_index_end      = jindex[iidx+1];
 822
 823         /* Get outer coordinate index */
 824         inr              = iinr[iidx];
 825         i_coord_offset   = DIM*inr;
 826
 827         /* Load i particle coords and add shift vector */
 828         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 829                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 830
 831         fix0             = _mm256_setzero_pd();
 832         fiy0             = _mm256_setzero_pd();
 833         fiz0             = _mm256_setzero_pd();
 834         fix1             = _mm256_setzero_pd();
 835         fiy1             = _mm256_setzero_pd();
 836         fiz1             = _mm256_setzero_pd();
 837         fix2             = _mm256_setzero_pd();
 838         fiy2             = _mm256_setzero_pd();
 839         fiz2             = _mm256_setzero_pd();
 840
 841         /* Start inner kernel loop */
 842         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 843         {
 844
 845             /* Get j neighbor index, and coordinate index */
 846             jnrA             = jjnr[jidx];
 847             jnrB             = jjnr[jidx+1];
 848             jnrC             = jjnr[jidx+2];
 849             jnrD             = jjnr[jidx+3];
 850             j_coord_offsetA  = DIM*jnrA;
 851             j_coord_offsetB  = DIM*jnrB;
 852             j_coord_offsetC  = DIM*jnrC;
 853             j_coord_offsetD  = DIM*jnrD;
 854
 855             /* load j atom coordinates */
 856             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 857                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 858                                                  &jx0,&jy0,&jz0);
 859
 860             /* Calculate displacement vector */
 861             dx00             = _mm256_sub_pd(ix0,jx0);
 862             dy00             = _mm256_sub_pd(iy0,jy0);
 863             dz00             = _mm256_sub_pd(iz0,jz0);
 864             dx10             = _mm256_sub_pd(ix1,jx0);
 865             dy10             = _mm256_sub_pd(iy1,jy0);
 866             dz10             = _mm256_sub_pd(iz1,jz0);
 867             dx20             = _mm256_sub_pd(ix2,jx0);
 868             dy20             = _mm256_sub_pd(iy2,jy0);
 869             dz20             = _mm256_sub_pd(iz2,jz0);
 870
 871             /* Calculate squared distance and things based on it */
 872             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 873             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 874             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 875
 876             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 877             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 878             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 879
 880             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 881             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 882             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 883
 884             /* Load parameters for j particles */
 885             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 886                                                                  charge+jnrC+0,charge+jnrD+0);
 887             vdwjidx0A        = 2*vdwtype[jnrA+0];
 888             vdwjidx0B        = 2*vdwtype[jnrB+0];
 889             vdwjidx0C        = 2*vdwtype[jnrC+0];
 890             vdwjidx0D        = 2*vdwtype[jnrD+0];
 891
 892             fjx0             = _mm256_setzero_pd();
 893             fjy0             = _mm256_setzero_pd();
 894             fjz0             = _mm256_setzero_pd();
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 901             {
 902
 903             r00              = _mm256_mul_pd(rsq00,rinv00);
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             qq00             = _mm256_mul_pd(iq0,jq0);
 907             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 908                                             vdwioffsetptr0+vdwjidx0B,
 909                                             vdwioffsetptr0+vdwjidx0C,
 910                                             vdwioffsetptr0+vdwjidx0D,
 911                                             &c6_00,&c12_00);
 912
 913             /* Calculate table index by multiplying r with table scale and truncate to integer */
 914             rt               = _mm256_mul_pd(r00,vftabscale);
 915             vfitab           = _mm256_cvttpd_epi32(rt);
 916             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 917             vfitab           = _mm_slli_epi32(vfitab,3);
 918
 919             /* REACTION-FIELD ELECTROSTATICS */
 920             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 921
 922             /* CUBIC SPLINE TABLE DISPERSION */
 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             fvdw6            = _mm256_mul_pd(c6_00,FF);
 932
 933             /* CUBIC SPLINE TABLE REPULSION */
 934             vfitab           = _mm_add_epi32(vfitab,ifour);
 935             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 936             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 937             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 938             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 939             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 940             Heps             = _mm256_mul_pd(vfeps,H);
 941             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 942             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 943             fvdw12           = _mm256_mul_pd(c12_00,FF);
 944             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 945
 946             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 947
 948             fscal            = _mm256_add_pd(felec,fvdw);
 949
 950             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm256_mul_pd(fscal,dx00);
 954             ty               = _mm256_mul_pd(fscal,dy00);
 955             tz               = _mm256_mul_pd(fscal,dz00);
 956
 957             /* Update vectorial force */
 958             fix0             = _mm256_add_pd(fix0,tx);
 959             fiy0             = _mm256_add_pd(fiy0,ty);
 960             fiz0             = _mm256_add_pd(fiz0,tz);
 961
 962             fjx0             = _mm256_add_pd(fjx0,tx);
 963             fjy0             = _mm256_add_pd(fjy0,ty);
 964             fjz0             = _mm256_add_pd(fjz0,tz);
 965
 966             }
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 973             {
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq10             = _mm256_mul_pd(iq1,jq0);
 977
 978             /* REACTION-FIELD ELECTROSTATICS */
 979             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 980
 981             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 982
 983             fscal            = felec;
 984
 985             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 986
 987             /* Calculate temporary vectorial force */
 988             tx               = _mm256_mul_pd(fscal,dx10);
 989             ty               = _mm256_mul_pd(fscal,dy10);
 990             tz               = _mm256_mul_pd(fscal,dz10);
 991
 992             /* Update vectorial force */
 993             fix1             = _mm256_add_pd(fix1,tx);
 994             fiy1             = _mm256_add_pd(fiy1,ty);
 995             fiz1             = _mm256_add_pd(fiz1,tz);
 996
 997             fjx0             = _mm256_add_pd(fjx0,tx);
 998             fjy0             = _mm256_add_pd(fjy0,ty);
 999             fjz0             = _mm256_add_pd(fjz0,tz);
1000
1001             }
1002
1003             /**************************
1004              * CALCULATE INTERACTIONS *
1005              **************************/
1006
1007             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1008             {
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq20             = _mm256_mul_pd(iq2,jq0);
1012
1013             /* REACTION-FIELD ELECTROSTATICS */
1014             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1015
1016             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1017
1018             fscal            = felec;
1019
1020             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm256_mul_pd(fscal,dx20);
1024             ty               = _mm256_mul_pd(fscal,dy20);
1025             tz               = _mm256_mul_pd(fscal,dz20);
1026
1027             /* Update vectorial force */
1028             fix2             = _mm256_add_pd(fix2,tx);
1029             fiy2             = _mm256_add_pd(fiy2,ty);
1030             fiz2             = _mm256_add_pd(fiz2,tz);
1031
1032             fjx0             = _mm256_add_pd(fjx0,tx);
1033             fjy0             = _mm256_add_pd(fjy0,ty);
1034             fjz0             = _mm256_add_pd(fjz0,tz);
1035
1036             }
1037
1038             fjptrA             = f+j_coord_offsetA;
1039             fjptrB             = f+j_coord_offsetB;
1040             fjptrC             = f+j_coord_offsetC;
1041             fjptrD             = f+j_coord_offsetD;
1042
1043             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1044
1045             /* Inner loop uses 120 flops */
1046         }
1047
1048         if(jidx<j_index_end)
1049         {
1050
1051             /* Get j neighbor index, and coordinate index */
1052             jnrlistA         = jjnr[jidx];
1053             jnrlistB         = jjnr[jidx+1];
1054             jnrlistC         = jjnr[jidx+2];
1055             jnrlistD         = jjnr[jidx+3];
1056             /* Sign of each element will be negative for non-real atoms.
1057              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1058              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1059              */
1060             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1061
1062             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1063             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1064             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1065
1066             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1067             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1068             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1069             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1070             j_coord_offsetA  = DIM*jnrA;
1071             j_coord_offsetB  = DIM*jnrB;
1072             j_coord_offsetC  = DIM*jnrC;
1073             j_coord_offsetD  = DIM*jnrD;
1074
1075             /* load j atom coordinates */
1076             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1077                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1078                                                  &jx0,&jy0,&jz0);
1079
1080             /* Calculate displacement vector */
1081             dx00             = _mm256_sub_pd(ix0,jx0);
1082             dy00             = _mm256_sub_pd(iy0,jy0);
1083             dz00             = _mm256_sub_pd(iz0,jz0);
1084             dx10             = _mm256_sub_pd(ix1,jx0);
1085             dy10             = _mm256_sub_pd(iy1,jy0);
1086             dz10             = _mm256_sub_pd(iz1,jz0);
1087             dx20             = _mm256_sub_pd(ix2,jx0);
1088             dy20             = _mm256_sub_pd(iy2,jy0);
1089             dz20             = _mm256_sub_pd(iz2,jz0);
1090
1091             /* Calculate squared distance and things based on it */
1092             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1093             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1094             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1095
1096             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1097             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1098             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1099
1100             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1101             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1102             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1103
1104             /* Load parameters for j particles */
1105             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1106                                                                  charge+jnrC+0,charge+jnrD+0);
1107             vdwjidx0A        = 2*vdwtype[jnrA+0];
1108             vdwjidx0B        = 2*vdwtype[jnrB+0];
1109             vdwjidx0C        = 2*vdwtype[jnrC+0];
1110             vdwjidx0D        = 2*vdwtype[jnrD+0];
1111
1112             fjx0             = _mm256_setzero_pd();
1113             fjy0             = _mm256_setzero_pd();
1114             fjz0             = _mm256_setzero_pd();
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1121             {
1122
1123             r00              = _mm256_mul_pd(rsq00,rinv00);
1124             r00              = _mm256_andnot_pd(dummy_mask,r00);
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq00             = _mm256_mul_pd(iq0,jq0);
1128             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1129                                             vdwioffsetptr0+vdwjidx0B,
1130                                             vdwioffsetptr0+vdwjidx0C,
1131                                             vdwioffsetptr0+vdwjidx0D,
1132                                             &c6_00,&c12_00);
1133
1134             /* Calculate table index by multiplying r with table scale and truncate to integer */
1135             rt               = _mm256_mul_pd(r00,vftabscale);
1136             vfitab           = _mm256_cvttpd_epi32(rt);
1137             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1138             vfitab           = _mm_slli_epi32(vfitab,3);
1139
1140             /* REACTION-FIELD ELECTROSTATICS */
1141             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1142
1143             /* CUBIC SPLINE TABLE DISPERSION */
1144             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1145             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1146             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1147             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1148             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1149             Heps             = _mm256_mul_pd(vfeps,H);
1150             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1151             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1152             fvdw6            = _mm256_mul_pd(c6_00,FF);
1153
1154             /* CUBIC SPLINE TABLE REPULSION */
1155             vfitab           = _mm_add_epi32(vfitab,ifour);
1156             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1157             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1158             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1159             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1160             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1161             Heps             = _mm256_mul_pd(vfeps,H);
1162             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1163             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1164             fvdw12           = _mm256_mul_pd(c12_00,FF);
1165             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1166
1167             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1168
1169             fscal            = _mm256_add_pd(felec,fvdw);
1170
1171             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1172
1173             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1174
1175             /* Calculate temporary vectorial force */
1176             tx               = _mm256_mul_pd(fscal,dx00);
1177             ty               = _mm256_mul_pd(fscal,dy00);
1178             tz               = _mm256_mul_pd(fscal,dz00);
1179
1180             /* Update vectorial force */
1181             fix0             = _mm256_add_pd(fix0,tx);
1182             fiy0             = _mm256_add_pd(fiy0,ty);
1183             fiz0             = _mm256_add_pd(fiz0,tz);
1184
1185             fjx0             = _mm256_add_pd(fjx0,tx);
1186             fjy0             = _mm256_add_pd(fjy0,ty);
1187             fjz0             = _mm256_add_pd(fjz0,tz);
1188
1189             }
1190
1191             /**************************
1192              * CALCULATE INTERACTIONS *
1193              **************************/
1194
1195             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1196             {
1197
1198             /* Compute parameters for interactions between i and j atoms */
1199             qq10             = _mm256_mul_pd(iq1,jq0);
1200
1201             /* REACTION-FIELD ELECTROSTATICS */
1202             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1203
1204             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1205
1206             fscal            = felec;
1207
1208             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1209
1210             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1211
1212             /* Calculate temporary vectorial force */
1213             tx               = _mm256_mul_pd(fscal,dx10);
1214             ty               = _mm256_mul_pd(fscal,dy10);
1215             tz               = _mm256_mul_pd(fscal,dz10);
1216
1217             /* Update vectorial force */
1218             fix1             = _mm256_add_pd(fix1,tx);
1219             fiy1             = _mm256_add_pd(fiy1,ty);
1220             fiz1             = _mm256_add_pd(fiz1,tz);
1221
1222             fjx0             = _mm256_add_pd(fjx0,tx);
1223             fjy0             = _mm256_add_pd(fjy0,ty);
1224             fjz0             = _mm256_add_pd(fjz0,tz);
1225
1226             }
1227
1228             /**************************
1229              * CALCULATE INTERACTIONS *
1230              **************************/
1231
1232             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1233             {
1234
1235             /* Compute parameters for interactions between i and j atoms */
1236             qq20             = _mm256_mul_pd(iq2,jq0);
1237
1238             /* REACTION-FIELD ELECTROSTATICS */
1239             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1240
1241             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1242
1243             fscal            = felec;
1244
1245             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1246
1247             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1248
1249             /* Calculate temporary vectorial force */
1250             tx               = _mm256_mul_pd(fscal,dx20);
1251             ty               = _mm256_mul_pd(fscal,dy20);
1252             tz               = _mm256_mul_pd(fscal,dz20);
1253
1254             /* Update vectorial force */
1255             fix2             = _mm256_add_pd(fix2,tx);
1256             fiy2             = _mm256_add_pd(fiy2,ty);
1257             fiz2             = _mm256_add_pd(fiz2,tz);
1258
1259             fjx0             = _mm256_add_pd(fjx0,tx);
1260             fjy0             = _mm256_add_pd(fjy0,ty);
1261             fjz0             = _mm256_add_pd(fjz0,tz);
1262
1263             }
1264
1265             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1266             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1267             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1268             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1269
1270             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1271
1272             /* Inner loop uses 121 flops */
1273         }
1274
1275         /* End of innermost loop */
1276
1277         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1278                                                  f+i_coord_offset,fshift+i_shift_offset);
1279
1280         /* Increment number of inner iterations */
1281         inneriter                  += j_index_end - j_index_start;
1282
1283         /* Outer loop uses 18 flops */
1284     }
1285
1286     /* Increment number of outer iterations */
1287     outeriter        += nri;
1288
1289     /* Update outer/inner flops */
1290
1291     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1292 }