src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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_ElecRF_VdwCSTab_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m128i          vfitab;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256d          dummy_mask,cutoff_mask;
 113     __m128           tmpmask0,tmpmask1;
 114     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 115     __m256d          one     = _mm256_set1_pd(1.0);
 116     __m256d          two     = _mm256_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     krf              = _mm256_set1_pd(fr->ic->k_rf);
 131     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 132     crf              = _mm256_set1_pd(fr->ic->c_rf);
 133     nvdwtype         = fr->ntype;
 134     vdwparam         = fr->nbfp;
 135     vdwtype          = mdatoms->typeA;
 136
 137     vftab            = kernel_data->table_vdw->data;
 138     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 143     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 144     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 145     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 179
 180         fix0             = _mm256_setzero_pd();
 181         fiy0             = _mm256_setzero_pd();
 182         fiz0             = _mm256_setzero_pd();
 183         fix1             = _mm256_setzero_pd();
 184         fiy1             = _mm256_setzero_pd();
 185         fiz1             = _mm256_setzero_pd();
 186         fix2             = _mm256_setzero_pd();
 187         fiy2             = _mm256_setzero_pd();
 188         fiz2             = _mm256_setzero_pd();
 189         fix3             = _mm256_setzero_pd();
 190         fiy3             = _mm256_setzero_pd();
 191         fiz3             = _mm256_setzero_pd();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_pd();
 195         vvdwsum          = _mm256_setzero_pd();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             j_coord_offsetA  = DIM*jnrA;
 207             j_coord_offsetB  = DIM*jnrB;
 208             j_coord_offsetC  = DIM*jnrC;
 209             j_coord_offsetD  = DIM*jnrD;
 210
 211             /* load j atom coordinates */
 212             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 214                                                  &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm256_sub_pd(ix0,jx0);
 218             dy00             = _mm256_sub_pd(iy0,jy0);
 219             dz00             = _mm256_sub_pd(iz0,jz0);
 220             dx10             = _mm256_sub_pd(ix1,jx0);
 221             dy10             = _mm256_sub_pd(iy1,jy0);
 222             dz10             = _mm256_sub_pd(iz1,jz0);
 223             dx20             = _mm256_sub_pd(ix2,jx0);
 224             dy20             = _mm256_sub_pd(iy2,jy0);
 225             dz20             = _mm256_sub_pd(iz2,jz0);
 226             dx30             = _mm256_sub_pd(ix3,jx0);
 227             dy30             = _mm256_sub_pd(iy3,jy0);
 228             dz30             = _mm256_sub_pd(iz3,jz0);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 232             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 233             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 234             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 235
 236             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 237             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 238             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 239             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 240
 241             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 242             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 243             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 247                                                                  charge+jnrC+0,charge+jnrD+0);
 248             vdwjidx0A        = 2*vdwtype[jnrA+0];
 249             vdwjidx0B        = 2*vdwtype[jnrB+0];
 250             vdwjidx0C        = 2*vdwtype[jnrC+0];
 251             vdwjidx0D        = 2*vdwtype[jnrD+0];
 252
 253             fjx0             = _mm256_setzero_pd();
 254             fjy0             = _mm256_setzero_pd();
 255             fjz0             = _mm256_setzero_pd();
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             r00              = _mm256_mul_pd(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 265                                             vdwioffsetptr0+vdwjidx0B,
 266                                             vdwioffsetptr0+vdwjidx0C,
 267                                             vdwioffsetptr0+vdwjidx0D,
 268                                             &c6_00,&c12_00);
 269
 270             /* Calculate table index by multiplying r with table scale and truncate to integer */
 271             rt               = _mm256_mul_pd(r00,vftabscale);
 272             vfitab           = _mm256_cvttpd_epi32(rt);
 273             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 274             vfitab           = _mm_slli_epi32(vfitab,3);
 275
 276             /* CUBIC SPLINE TABLE DISPERSION */
 277             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 278             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 279             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 280             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 281             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 282             Heps             = _mm256_mul_pd(vfeps,H);
 283             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 284             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 285             vvdw6            = _mm256_mul_pd(c6_00,VV);
 286             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 287             fvdw6            = _mm256_mul_pd(c6_00,FF);
 288
 289             /* CUBIC SPLINE TABLE REPULSION */
 290             vfitab           = _mm_add_epi32(vfitab,ifour);
 291             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 292             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 293             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 294             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 295             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 296             Heps             = _mm256_mul_pd(vfeps,H);
 297             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 298             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 299             vvdw12           = _mm256_mul_pd(c12_00,VV);
 300             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 301             fvdw12           = _mm256_mul_pd(c12_00,FF);
 302             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 303             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 307
 308             fscal            = fvdw;
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_pd(fscal,dx00);
 312             ty               = _mm256_mul_pd(fscal,dy00);
 313             tz               = _mm256_mul_pd(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm256_add_pd(fix0,tx);
 317             fiy0             = _mm256_add_pd(fiy0,ty);
 318             fiz0             = _mm256_add_pd(fiz0,tz);
 319
 320             fjx0             = _mm256_add_pd(fjx0,tx);
 321             fjy0             = _mm256_add_pd(fjy0,ty);
 322             fjz0             = _mm256_add_pd(fjz0,tz);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq10             = _mm256_mul_pd(iq1,jq0);
 330
 331             /* REACTION-FIELD ELECTROSTATICS */
 332             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 333             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm256_add_pd(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm256_mul_pd(fscal,dx10);
 342             ty               = _mm256_mul_pd(fscal,dy10);
 343             tz               = _mm256_mul_pd(fscal,dz10);
 344
 345             /* Update vectorial force */
 346             fix1             = _mm256_add_pd(fix1,tx);
 347             fiy1             = _mm256_add_pd(fiy1,ty);
 348             fiz1             = _mm256_add_pd(fiz1,tz);
 349
 350             fjx0             = _mm256_add_pd(fjx0,tx);
 351             fjy0             = _mm256_add_pd(fjy0,ty);
 352             fjz0             = _mm256_add_pd(fjz0,tz);
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _mm256_mul_pd(iq2,jq0);
 360
 361             /* REACTION-FIELD ELECTROSTATICS */
 362             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 363             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm256_add_pd(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm256_mul_pd(fscal,dx20);
 372             ty               = _mm256_mul_pd(fscal,dy20);
 373             tz               = _mm256_mul_pd(fscal,dz20);
 374
 375             /* Update vectorial force */
 376             fix2             = _mm256_add_pd(fix2,tx);
 377             fiy2             = _mm256_add_pd(fiy2,ty);
 378             fiz2             = _mm256_add_pd(fiz2,tz);
 379
 380             fjx0             = _mm256_add_pd(fjx0,tx);
 381             fjy0             = _mm256_add_pd(fjy0,ty);
 382             fjz0             = _mm256_add_pd(fjz0,tz);
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             /* Compute parameters for interactions between i and j atoms */
 389             qq30             = _mm256_mul_pd(iq3,jq0);
 390
 391             /* REACTION-FIELD ELECTROSTATICS */
 392             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 393             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm256_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm256_mul_pd(fscal,dx30);
 402             ty               = _mm256_mul_pd(fscal,dy30);
 403             tz               = _mm256_mul_pd(fscal,dz30);
 404
 405             /* Update vectorial force */
 406             fix3             = _mm256_add_pd(fix3,tx);
 407             fiy3             = _mm256_add_pd(fiy3,ty);
 408             fiz3             = _mm256_add_pd(fiz3,tz);
 409
 410             fjx0             = _mm256_add_pd(fjx0,tx);
 411             fjy0             = _mm256_add_pd(fjy0,ty);
 412             fjz0             = _mm256_add_pd(fjz0,tz);
 413
 414             fjptrA             = f+j_coord_offsetA;
 415             fjptrB             = f+j_coord_offsetB;
 416             fjptrC             = f+j_coord_offsetC;
 417             fjptrD             = f+j_coord_offsetD;
 418
 419             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 420
 421             /* Inner loop uses 155 flops */
 422         }
 423
 424         if(jidx<j_index_end)
 425         {
 426
 427             /* Get j neighbor index, and coordinate index */
 428             jnrlistA         = jjnr[jidx];
 429             jnrlistB         = jjnr[jidx+1];
 430             jnrlistC         = jjnr[jidx+2];
 431             jnrlistD         = jjnr[jidx+3];
 432             /* Sign of each element will be negative for non-real atoms.
 433              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 434              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 435              */
 436             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 437
 438             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 439             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 440             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 441
 442             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 443             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 444             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 445             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 446             j_coord_offsetA  = DIM*jnrA;
 447             j_coord_offsetB  = DIM*jnrB;
 448             j_coord_offsetC  = DIM*jnrC;
 449             j_coord_offsetD  = DIM*jnrD;
 450
 451             /* load j atom coordinates */
 452             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 453                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 454                                                  &jx0,&jy0,&jz0);
 455
 456             /* Calculate displacement vector */
 457             dx00             = _mm256_sub_pd(ix0,jx0);
 458             dy00             = _mm256_sub_pd(iy0,jy0);
 459             dz00             = _mm256_sub_pd(iz0,jz0);
 460             dx10             = _mm256_sub_pd(ix1,jx0);
 461             dy10             = _mm256_sub_pd(iy1,jy0);
 462             dz10             = _mm256_sub_pd(iz1,jz0);
 463             dx20             = _mm256_sub_pd(ix2,jx0);
 464             dy20             = _mm256_sub_pd(iy2,jy0);
 465             dz20             = _mm256_sub_pd(iz2,jz0);
 466             dx30             = _mm256_sub_pd(ix3,jx0);
 467             dy30             = _mm256_sub_pd(iy3,jy0);
 468             dz30             = _mm256_sub_pd(iz3,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 472             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 473             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 474             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 475
 476             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 477             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 478             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 479             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 480
 481             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 482             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 483             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 484
 485             /* Load parameters for j particles */
 486             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 487                                                                  charge+jnrC+0,charge+jnrD+0);
 488             vdwjidx0A        = 2*vdwtype[jnrA+0];
 489             vdwjidx0B        = 2*vdwtype[jnrB+0];
 490             vdwjidx0C        = 2*vdwtype[jnrC+0];
 491             vdwjidx0D        = 2*vdwtype[jnrD+0];
 492
 493             fjx0             = _mm256_setzero_pd();
 494             fjy0             = _mm256_setzero_pd();
 495             fjz0             = _mm256_setzero_pd();
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             r00              = _mm256_mul_pd(rsq00,rinv00);
 502             r00              = _mm256_andnot_pd(dummy_mask,r00);
 503
 504             /* Compute parameters for interactions between i and j atoms */
 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             /* CUBIC SPLINE TABLE DISPERSION */
 518             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 519             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 520             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 521             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 522             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 523             Heps             = _mm256_mul_pd(vfeps,H);
 524             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 525             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 526             vvdw6            = _mm256_mul_pd(c6_00,VV);
 527             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 528             fvdw6            = _mm256_mul_pd(c6_00,FF);
 529
 530             /* CUBIC SPLINE TABLE REPULSION */
 531             vfitab           = _mm_add_epi32(vfitab,ifour);
 532             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 533             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 534             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 535             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 536             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 537             Heps             = _mm256_mul_pd(vfeps,H);
 538             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 539             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 540             vvdw12           = _mm256_mul_pd(c12_00,VV);
 541             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 542             fvdw12           = _mm256_mul_pd(c12_00,FF);
 543             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 544             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 548             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 549
 550             fscal            = fvdw;
 551
 552             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm256_mul_pd(fscal,dx00);
 556             ty               = _mm256_mul_pd(fscal,dy00);
 557             tz               = _mm256_mul_pd(fscal,dz00);
 558
 559             /* Update vectorial force */
 560             fix0             = _mm256_add_pd(fix0,tx);
 561             fiy0             = _mm256_add_pd(fiy0,ty);
 562             fiz0             = _mm256_add_pd(fiz0,tz);
 563
 564             fjx0             = _mm256_add_pd(fjx0,tx);
 565             fjy0             = _mm256_add_pd(fjy0,ty);
 566             fjz0             = _mm256_add_pd(fjz0,tz);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq10             = _mm256_mul_pd(iq1,jq0);
 574
 575             /* REACTION-FIELD ELECTROSTATICS */
 576             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 577             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velec            = _mm256_andnot_pd(dummy_mask,velec);
 581             velecsum         = _mm256_add_pd(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm256_mul_pd(fscal,dx10);
 589             ty               = _mm256_mul_pd(fscal,dy10);
 590             tz               = _mm256_mul_pd(fscal,dz10);
 591
 592             /* Update vectorial force */
 593             fix1             = _mm256_add_pd(fix1,tx);
 594             fiy1             = _mm256_add_pd(fiy1,ty);
 595             fiz1             = _mm256_add_pd(fiz1,tz);
 596
 597             fjx0             = _mm256_add_pd(fjx0,tx);
 598             fjy0             = _mm256_add_pd(fjy0,ty);
 599             fjz0             = _mm256_add_pd(fjz0,tz);
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq20             = _mm256_mul_pd(iq2,jq0);
 607
 608             /* REACTION-FIELD ELECTROSTATICS */
 609             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 610             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm256_andnot_pd(dummy_mask,velec);
 614             velecsum         = _mm256_add_pd(velecsum,velec);
 615
 616             fscal            = felec;
 617
 618             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 619
 620             /* Calculate temporary vectorial force */
 621             tx               = _mm256_mul_pd(fscal,dx20);
 622             ty               = _mm256_mul_pd(fscal,dy20);
 623             tz               = _mm256_mul_pd(fscal,dz20);
 624
 625             /* Update vectorial force */
 626             fix2             = _mm256_add_pd(fix2,tx);
 627             fiy2             = _mm256_add_pd(fiy2,ty);
 628             fiz2             = _mm256_add_pd(fiz2,tz);
 629
 630             fjx0             = _mm256_add_pd(fjx0,tx);
 631             fjy0             = _mm256_add_pd(fjy0,ty);
 632             fjz0             = _mm256_add_pd(fjz0,tz);
 633
 634             /**************************
 635              * CALCULATE INTERACTIONS *
 636              **************************/
 637
 638             /* Compute parameters for interactions between i and j atoms */
 639             qq30             = _mm256_mul_pd(iq3,jq0);
 640
 641             /* REACTION-FIELD ELECTROSTATICS */
 642             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 643             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 644
 645             /* Update potential sum for this i atom from the interaction with this j atom. */
 646             velec            = _mm256_andnot_pd(dummy_mask,velec);
 647             velecsum         = _mm256_add_pd(velecsum,velec);
 648
 649             fscal            = felec;
 650
 651             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 652
 653             /* Calculate temporary vectorial force */
 654             tx               = _mm256_mul_pd(fscal,dx30);
 655             ty               = _mm256_mul_pd(fscal,dy30);
 656             tz               = _mm256_mul_pd(fscal,dz30);
 657
 658             /* Update vectorial force */
 659             fix3             = _mm256_add_pd(fix3,tx);
 660             fiy3             = _mm256_add_pd(fiy3,ty);
 661             fiz3             = _mm256_add_pd(fiz3,tz);
 662
 663             fjx0             = _mm256_add_pd(fjx0,tx);
 664             fjy0             = _mm256_add_pd(fjy0,ty);
 665             fjz0             = _mm256_add_pd(fjz0,tz);
 666
 667             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 668             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 669             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 670             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 671
 672             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 673
 674             /* Inner loop uses 156 flops */
 675         }
 676
 677         /* End of innermost loop */
 678
 679         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 680                                                  f+i_coord_offset,fshift+i_shift_offset);
 681
 682         ggid                        = gid[iidx];
 683         /* Update potential energies */
 684         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 685         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 686
 687         /* Increment number of inner iterations */
 688         inneriter                  += j_index_end - j_index_start;
 689
 690         /* Outer loop uses 26 flops */
 691     }
 692
 693     /* Increment number of outer iterations */
 694     outeriter        += nri;
 695
 696     /* Update outer/inner flops */
 697
 698     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*156);
 699 }
 700 /*
 701  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_double
 702  * Electrostatics interaction: ReactionField
 703  * VdW interaction:            CubicSplineTable
 704  * Geometry:                   Water4-Particle
 705  * Calculate force/pot:        Force
 706  */
 707 void
 708 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_double
 709                     (t_nblist                    * gmx_restrict       nlist,
 710                      rvec                        * gmx_restrict          xx,
 711                      rvec                        * gmx_restrict          ff,
 712                      t_forcerec                  * gmx_restrict          fr,
 713                      t_mdatoms                   * gmx_restrict     mdatoms,
 714                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 715                      t_nrnb                      * gmx_restrict        nrnb)
 716 {
 717     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 718      * just 0 for non-waters.
 719      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 720      * jnr indices corresponding to data put in the four positions in the SIMD register.
 721      */
 722     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 723     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 724     int              jnrA,jnrB,jnrC,jnrD;
 725     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 726     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 727     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 728     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 729     real             rcutoff_scalar;
 730     real             *shiftvec,*fshift,*x,*f;
 731     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 732     real             scratch[4*DIM];
 733     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 734     real *           vdwioffsetptr0;
 735     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 736     real *           vdwioffsetptr1;
 737     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 738     real *           vdwioffsetptr2;
 739     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 740     real *           vdwioffsetptr3;
 741     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 742     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 743     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 744     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 745     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 746     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 747     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 748     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 749     real             *charge;
 750     int              nvdwtype;
 751     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 752     int              *vdwtype;
 753     real             *vdwparam;
 754     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 755     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 756     __m128i          vfitab;
 757     __m128i          ifour       = _mm_set1_epi32(4);
 758     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 759     real             *vftab;
 760     __m256d          dummy_mask,cutoff_mask;
 761     __m128           tmpmask0,tmpmask1;
 762     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 763     __m256d          one     = _mm256_set1_pd(1.0);
 764     __m256d          two     = _mm256_set1_pd(2.0);
 765     x                = xx[0];
 766     f                = ff[0];
 767
 768     nri              = nlist->nri;
 769     iinr             = nlist->iinr;
 770     jindex           = nlist->jindex;
 771     jjnr             = nlist->jjnr;
 772     shiftidx         = nlist->shift;
 773     gid              = nlist->gid;
 774     shiftvec         = fr->shift_vec[0];
 775     fshift           = fr->fshift[0];
 776     facel            = _mm256_set1_pd(fr->epsfac);
 777     charge           = mdatoms->chargeA;
 778     krf              = _mm256_set1_pd(fr->ic->k_rf);
 779     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 780     crf              = _mm256_set1_pd(fr->ic->c_rf);
 781     nvdwtype         = fr->ntype;
 782     vdwparam         = fr->nbfp;
 783     vdwtype          = mdatoms->typeA;
 784
 785     vftab            = kernel_data->table_vdw->data;
 786     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 787
 788     /* Setup water-specific parameters */
 789     inr              = nlist->iinr[0];
 790     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 791     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 792     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 793     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 794
 795     /* Avoid stupid compiler warnings */
 796     jnrA = jnrB = jnrC = jnrD = 0;
 797     j_coord_offsetA = 0;
 798     j_coord_offsetB = 0;
 799     j_coord_offsetC = 0;
 800     j_coord_offsetD = 0;
 801
 802     outeriter        = 0;
 803     inneriter        = 0;
 804
 805     for(iidx=0;iidx<4*DIM;iidx++)
 806     {
 807         scratch[iidx] = 0.0;
 808     }
 809
 810     /* Start outer loop over neighborlists */
 811     for(iidx=0; iidx<nri; iidx++)
 812     {
 813         /* Load shift vector for this list */
 814         i_shift_offset   = DIM*shiftidx[iidx];
 815
 816         /* Load limits for loop over neighbors */
 817         j_index_start    = jindex[iidx];
 818         j_index_end      = jindex[iidx+1];
 819
 820         /* Get outer coordinate index */
 821         inr              = iinr[iidx];
 822         i_coord_offset   = DIM*inr;
 823
 824         /* Load i particle coords and add shift vector */
 825         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 826                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 827
 828         fix0             = _mm256_setzero_pd();
 829         fiy0             = _mm256_setzero_pd();
 830         fiz0             = _mm256_setzero_pd();
 831         fix1             = _mm256_setzero_pd();
 832         fiy1             = _mm256_setzero_pd();
 833         fiz1             = _mm256_setzero_pd();
 834         fix2             = _mm256_setzero_pd();
 835         fiy2             = _mm256_setzero_pd();
 836         fiz2             = _mm256_setzero_pd();
 837         fix3             = _mm256_setzero_pd();
 838         fiy3             = _mm256_setzero_pd();
 839         fiz3             = _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             dx30             = _mm256_sub_pd(ix3,jx0);
 871             dy30             = _mm256_sub_pd(iy3,jy0);
 872             dz30             = _mm256_sub_pd(iz3,jz0);
 873
 874             /* Calculate squared distance and things based on it */
 875             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 876             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 877             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 878             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 879
 880             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 881             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 882             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 883             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 884
 885             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 886             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 887             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 888
 889             /* Load parameters for j particles */
 890             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 891                                                                  charge+jnrC+0,charge+jnrD+0);
 892             vdwjidx0A        = 2*vdwtype[jnrA+0];
 893             vdwjidx0B        = 2*vdwtype[jnrB+0];
 894             vdwjidx0C        = 2*vdwtype[jnrC+0];
 895             vdwjidx0D        = 2*vdwtype[jnrD+0];
 896
 897             fjx0             = _mm256_setzero_pd();
 898             fjy0             = _mm256_setzero_pd();
 899             fjz0             = _mm256_setzero_pd();
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             r00              = _mm256_mul_pd(rsq00,rinv00);
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 909                                             vdwioffsetptr0+vdwjidx0B,
 910                                             vdwioffsetptr0+vdwjidx0C,
 911                                             vdwioffsetptr0+vdwjidx0D,
 912                                             &c6_00,&c12_00);
 913
 914             /* Calculate table index by multiplying r with table scale and truncate to integer */
 915             rt               = _mm256_mul_pd(r00,vftabscale);
 916             vfitab           = _mm256_cvttpd_epi32(rt);
 917             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 918             vfitab           = _mm_slli_epi32(vfitab,3);
 919
 920             /* CUBIC SPLINE TABLE DISPERSION */
 921             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 922             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 923             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 924             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 925             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 926             Heps             = _mm256_mul_pd(vfeps,H);
 927             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 928             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 929             fvdw6            = _mm256_mul_pd(c6_00,FF);
 930
 931             /* CUBIC SPLINE TABLE REPULSION */
 932             vfitab           = _mm_add_epi32(vfitab,ifour);
 933             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 934             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 935             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 936             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 937             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 938             Heps             = _mm256_mul_pd(vfeps,H);
 939             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 940             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 941             fvdw12           = _mm256_mul_pd(c12_00,FF);
 942             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 943
 944             fscal            = fvdw;
 945
 946             /* Calculate temporary vectorial force */
 947             tx               = _mm256_mul_pd(fscal,dx00);
 948             ty               = _mm256_mul_pd(fscal,dy00);
 949             tz               = _mm256_mul_pd(fscal,dz00);
 950
 951             /* Update vectorial force */
 952             fix0             = _mm256_add_pd(fix0,tx);
 953             fiy0             = _mm256_add_pd(fiy0,ty);
 954             fiz0             = _mm256_add_pd(fiz0,tz);
 955
 956             fjx0             = _mm256_add_pd(fjx0,tx);
 957             fjy0             = _mm256_add_pd(fjy0,ty);
 958             fjz0             = _mm256_add_pd(fjz0,tz);
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq10             = _mm256_mul_pd(iq1,jq0);
 966
 967             /* REACTION-FIELD ELECTROSTATICS */
 968             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 969
 970             fscal            = felec;
 971
 972             /* Calculate temporary vectorial force */
 973             tx               = _mm256_mul_pd(fscal,dx10);
 974             ty               = _mm256_mul_pd(fscal,dy10);
 975             tz               = _mm256_mul_pd(fscal,dz10);
 976
 977             /* Update vectorial force */
 978             fix1             = _mm256_add_pd(fix1,tx);
 979             fiy1             = _mm256_add_pd(fiy1,ty);
 980             fiz1             = _mm256_add_pd(fiz1,tz);
 981
 982             fjx0             = _mm256_add_pd(fjx0,tx);
 983             fjy0             = _mm256_add_pd(fjy0,ty);
 984             fjz0             = _mm256_add_pd(fjz0,tz);
 985
 986             /**************************
 987              * CALCULATE INTERACTIONS *
 988              **************************/
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq20             = _mm256_mul_pd(iq2,jq0);
 992
 993             /* REACTION-FIELD ELECTROSTATICS */
 994             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 995
 996             fscal            = felec;
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm256_mul_pd(fscal,dx20);
1000             ty               = _mm256_mul_pd(fscal,dy20);
1001             tz               = _mm256_mul_pd(fscal,dz20);
1002
1003             /* Update vectorial force */
1004             fix2             = _mm256_add_pd(fix2,tx);
1005             fiy2             = _mm256_add_pd(fiy2,ty);
1006             fiz2             = _mm256_add_pd(fiz2,tz);
1007
1008             fjx0             = _mm256_add_pd(fjx0,tx);
1009             fjy0             = _mm256_add_pd(fjy0,ty);
1010             fjz0             = _mm256_add_pd(fjz0,tz);
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq30             = _mm256_mul_pd(iq3,jq0);
1018
1019             /* REACTION-FIELD ELECTROSTATICS */
1020             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1021
1022             fscal            = felec;
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm256_mul_pd(fscal,dx30);
1026             ty               = _mm256_mul_pd(fscal,dy30);
1027             tz               = _mm256_mul_pd(fscal,dz30);
1028
1029             /* Update vectorial force */
1030             fix3             = _mm256_add_pd(fix3,tx);
1031             fiy3             = _mm256_add_pd(fiy3,ty);
1032             fiz3             = _mm256_add_pd(fiz3,tz);
1033
1034             fjx0             = _mm256_add_pd(fjx0,tx);
1035             fjy0             = _mm256_add_pd(fjy0,ty);
1036             fjz0             = _mm256_add_pd(fjz0,tz);
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 132 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             dx30             = _mm256_sub_pd(ix3,jx0);
1091             dy30             = _mm256_sub_pd(iy3,jy0);
1092             dz30             = _mm256_sub_pd(iz3,jz0);
1093
1094             /* Calculate squared distance and things based on it */
1095             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1096             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1097             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1098             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1099
1100             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1101             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1102             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1103             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1104
1105             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1106             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1107             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1108
1109             /* Load parameters for j particles */
1110             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1111                                                                  charge+jnrC+0,charge+jnrD+0);
1112             vdwjidx0A        = 2*vdwtype[jnrA+0];
1113             vdwjidx0B        = 2*vdwtype[jnrB+0];
1114             vdwjidx0C        = 2*vdwtype[jnrC+0];
1115             vdwjidx0D        = 2*vdwtype[jnrD+0];
1116
1117             fjx0             = _mm256_setzero_pd();
1118             fjy0             = _mm256_setzero_pd();
1119             fjz0             = _mm256_setzero_pd();
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             r00              = _mm256_mul_pd(rsq00,rinv00);
1126             r00              = _mm256_andnot_pd(dummy_mask,r00);
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1130                                             vdwioffsetptr0+vdwjidx0B,
1131                                             vdwioffsetptr0+vdwjidx0C,
1132                                             vdwioffsetptr0+vdwjidx0D,
1133                                             &c6_00,&c12_00);
1134
1135             /* Calculate table index by multiplying r with table scale and truncate to integer */
1136             rt               = _mm256_mul_pd(r00,vftabscale);
1137             vfitab           = _mm256_cvttpd_epi32(rt);
1138             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1139             vfitab           = _mm_slli_epi32(vfitab,3);
1140
1141             /* CUBIC SPLINE TABLE DISPERSION */
1142             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1143             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1144             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1145             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1146             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1147             Heps             = _mm256_mul_pd(vfeps,H);
1148             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1149             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1150             fvdw6            = _mm256_mul_pd(c6_00,FF);
1151
1152             /* CUBIC SPLINE TABLE REPULSION */
1153             vfitab           = _mm_add_epi32(vfitab,ifour);
1154             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1155             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1156             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1157             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1158             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1159             Heps             = _mm256_mul_pd(vfeps,H);
1160             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1161             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1162             fvdw12           = _mm256_mul_pd(c12_00,FF);
1163             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1164
1165             fscal            = fvdw;
1166
1167             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm256_mul_pd(fscal,dx00);
1171             ty               = _mm256_mul_pd(fscal,dy00);
1172             tz               = _mm256_mul_pd(fscal,dz00);
1173
1174             /* Update vectorial force */
1175             fix0             = _mm256_add_pd(fix0,tx);
1176             fiy0             = _mm256_add_pd(fiy0,ty);
1177             fiz0             = _mm256_add_pd(fiz0,tz);
1178
1179             fjx0             = _mm256_add_pd(fjx0,tx);
1180             fjy0             = _mm256_add_pd(fjy0,ty);
1181             fjz0             = _mm256_add_pd(fjz0,tz);
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             /* Compute parameters for interactions between i and j atoms */
1188             qq10             = _mm256_mul_pd(iq1,jq0);
1189
1190             /* REACTION-FIELD ELECTROSTATICS */
1191             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1192
1193             fscal            = felec;
1194
1195             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1196
1197             /* Calculate temporary vectorial force */
1198             tx               = _mm256_mul_pd(fscal,dx10);
1199             ty               = _mm256_mul_pd(fscal,dy10);
1200             tz               = _mm256_mul_pd(fscal,dz10);
1201
1202             /* Update vectorial force */
1203             fix1             = _mm256_add_pd(fix1,tx);
1204             fiy1             = _mm256_add_pd(fiy1,ty);
1205             fiz1             = _mm256_add_pd(fiz1,tz);
1206
1207             fjx0             = _mm256_add_pd(fjx0,tx);
1208             fjy0             = _mm256_add_pd(fjy0,ty);
1209             fjz0             = _mm256_add_pd(fjz0,tz);
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             /* Compute parameters for interactions between i and j atoms */
1216             qq20             = _mm256_mul_pd(iq2,jq0);
1217
1218             /* REACTION-FIELD ELECTROSTATICS */
1219             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1220
1221             fscal            = felec;
1222
1223             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1224
1225             /* Calculate temporary vectorial force */
1226             tx               = _mm256_mul_pd(fscal,dx20);
1227             ty               = _mm256_mul_pd(fscal,dy20);
1228             tz               = _mm256_mul_pd(fscal,dz20);
1229
1230             /* Update vectorial force */
1231             fix2             = _mm256_add_pd(fix2,tx);
1232             fiy2             = _mm256_add_pd(fiy2,ty);
1233             fiz2             = _mm256_add_pd(fiz2,tz);
1234
1235             fjx0             = _mm256_add_pd(fjx0,tx);
1236             fjy0             = _mm256_add_pd(fjy0,ty);
1237             fjz0             = _mm256_add_pd(fjz0,tz);
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             /* Compute parameters for interactions between i and j atoms */
1244             qq30             = _mm256_mul_pd(iq3,jq0);
1245
1246             /* REACTION-FIELD ELECTROSTATICS */
1247             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1248
1249             fscal            = felec;
1250
1251             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1252
1253             /* Calculate temporary vectorial force */
1254             tx               = _mm256_mul_pd(fscal,dx30);
1255             ty               = _mm256_mul_pd(fscal,dy30);
1256             tz               = _mm256_mul_pd(fscal,dz30);
1257
1258             /* Update vectorial force */
1259             fix3             = _mm256_add_pd(fix3,tx);
1260             fiy3             = _mm256_add_pd(fiy3,ty);
1261             fiz3             = _mm256_add_pd(fiz3,tz);
1262
1263             fjx0             = _mm256_add_pd(fjx0,tx);
1264             fjy0             = _mm256_add_pd(fjy0,ty);
1265             fjz0             = _mm256_add_pd(fjz0,tz);
1266
1267             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1268             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1269             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1270             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1271
1272             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1273
1274             /* Inner loop uses 133 flops */
1275         }
1276
1277         /* End of innermost loop */
1278
1279         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1280                                                  f+i_coord_offset,fshift+i_shift_offset);
1281
1282         /* Increment number of inner iterations */
1283         inneriter                  += j_index_end - j_index_start;
1284
1285         /* Outer loop uses 24 flops */
1286     }
1287
1288     /* Increment number of outer iterations */
1289     outeriter        += nri;
1290
1291     /* Update outer/inner flops */
1292
1293     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*133);
1294 }