src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_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 "gmx_math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_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     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     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 140     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 141     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 176
 177         fix0             = _mm256_setzero_pd();
 178         fiy0             = _mm256_setzero_pd();
 179         fiz0             = _mm256_setzero_pd();
 180         fix1             = _mm256_setzero_pd();
 181         fiy1             = _mm256_setzero_pd();
 182         fiz1             = _mm256_setzero_pd();
 183         fix2             = _mm256_setzero_pd();
 184         fiy2             = _mm256_setzero_pd();
 185         fiz2             = _mm256_setzero_pd();
 186         fix3             = _mm256_setzero_pd();
 187         fiy3             = _mm256_setzero_pd();
 188         fiz3             = _mm256_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_pd();
 192         vvdwsum          = _mm256_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_pd(ix0,jx0);
 215             dy00             = _mm256_sub_pd(iy0,jy0);
 216             dz00             = _mm256_sub_pd(iz0,jz0);
 217             dx10             = _mm256_sub_pd(ix1,jx0);
 218             dy10             = _mm256_sub_pd(iy1,jy0);
 219             dz10             = _mm256_sub_pd(iz1,jz0);
 220             dx20             = _mm256_sub_pd(ix2,jx0);
 221             dy20             = _mm256_sub_pd(iy2,jy0);
 222             dz20             = _mm256_sub_pd(iz2,jz0);
 223             dx30             = _mm256_sub_pd(ix3,jx0);
 224             dy30             = _mm256_sub_pd(iy3,jy0);
 225             dz30             = _mm256_sub_pd(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 229             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 230             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 231             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 232
 233             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 234             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 235             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 236             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 237
 238             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 239             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 240             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 244                                                                  charge+jnrC+0,charge+jnrD+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247             vdwjidx0C        = 2*vdwtype[jnrC+0];
 248             vdwjidx0D        = 2*vdwtype[jnrD+0];
 249
 250             fjx0             = _mm256_setzero_pd();
 251             fjy0             = _mm256_setzero_pd();
 252             fjz0             = _mm256_setzero_pd();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             r00              = _mm256_mul_pd(rsq00,rinv00);
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 262                                             vdwioffsetptr0+vdwjidx0B,
 263                                             vdwioffsetptr0+vdwjidx0C,
 264                                             vdwioffsetptr0+vdwjidx0D,
 265                                             &c6_00,&c12_00);
 266
 267             /* Calculate table index by multiplying r with table scale and truncate to integer */
 268             rt               = _mm256_mul_pd(r00,vftabscale);
 269             vfitab           = _mm256_cvttpd_epi32(rt);
 270             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 271             vfitab           = _mm_slli_epi32(vfitab,3);
 272
 273             /* CUBIC SPLINE TABLE DISPERSION */
 274             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 275             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 276             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 277             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 278             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 279             Heps             = _mm256_mul_pd(vfeps,H);
 280             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 281             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 282             vvdw6            = _mm256_mul_pd(c6_00,VV);
 283             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 284             fvdw6            = _mm256_mul_pd(c6_00,FF);
 285
 286             /* CUBIC SPLINE TABLE REPULSION */
 287             vfitab           = _mm_add_epi32(vfitab,ifour);
 288             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 289             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 290             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 291             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 292             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 293             Heps             = _mm256_mul_pd(vfeps,H);
 294             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 295             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 296             vvdw12           = _mm256_mul_pd(c12_00,VV);
 297             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 298             fvdw12           = _mm256_mul_pd(c12_00,FF);
 299             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 300             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 304
 305             fscal            = fvdw;
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm256_mul_pd(fscal,dx00);
 309             ty               = _mm256_mul_pd(fscal,dy00);
 310             tz               = _mm256_mul_pd(fscal,dz00);
 311
 312             /* Update vectorial force */
 313             fix0             = _mm256_add_pd(fix0,tx);
 314             fiy0             = _mm256_add_pd(fiy0,ty);
 315             fiz0             = _mm256_add_pd(fiz0,tz);
 316
 317             fjx0             = _mm256_add_pd(fjx0,tx);
 318             fjy0             = _mm256_add_pd(fjy0,ty);
 319             fjz0             = _mm256_add_pd(fjz0,tz);
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             /* Compute parameters for interactions between i and j atoms */
 326             qq10             = _mm256_mul_pd(iq1,jq0);
 327
 328             /* COULOMB ELECTROSTATICS */
 329             velec            = _mm256_mul_pd(qq10,rinv10);
 330             felec            = _mm256_mul_pd(velec,rinvsq10);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm256_add_pd(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             /* Calculate temporary vectorial force */
 338             tx               = _mm256_mul_pd(fscal,dx10);
 339             ty               = _mm256_mul_pd(fscal,dy10);
 340             tz               = _mm256_mul_pd(fscal,dz10);
 341
 342             /* Update vectorial force */
 343             fix1             = _mm256_add_pd(fix1,tx);
 344             fiy1             = _mm256_add_pd(fiy1,ty);
 345             fiz1             = _mm256_add_pd(fiz1,tz);
 346
 347             fjx0             = _mm256_add_pd(fjx0,tx);
 348             fjy0             = _mm256_add_pd(fjy0,ty);
 349             fjz0             = _mm256_add_pd(fjz0,tz);
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq20             = _mm256_mul_pd(iq2,jq0);
 357
 358             /* COULOMB ELECTROSTATICS */
 359             velec            = _mm256_mul_pd(qq20,rinv20);
 360             felec            = _mm256_mul_pd(velec,rinvsq20);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velecsum         = _mm256_add_pd(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             /* Calculate temporary vectorial force */
 368             tx               = _mm256_mul_pd(fscal,dx20);
 369             ty               = _mm256_mul_pd(fscal,dy20);
 370             tz               = _mm256_mul_pd(fscal,dz20);
 371
 372             /* Update vectorial force */
 373             fix2             = _mm256_add_pd(fix2,tx);
 374             fiy2             = _mm256_add_pd(fiy2,ty);
 375             fiz2             = _mm256_add_pd(fiz2,tz);
 376
 377             fjx0             = _mm256_add_pd(fjx0,tx);
 378             fjy0             = _mm256_add_pd(fjy0,ty);
 379             fjz0             = _mm256_add_pd(fjz0,tz);
 380
 381             /**************************
 382              * CALCULATE INTERACTIONS *
 383              **************************/
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq30             = _mm256_mul_pd(iq3,jq0);
 387
 388             /* COULOMB ELECTROSTATICS */
 389             velec            = _mm256_mul_pd(qq30,rinv30);
 390             felec            = _mm256_mul_pd(velec,rinvsq30);
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm256_add_pd(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm256_mul_pd(fscal,dx30);
 399             ty               = _mm256_mul_pd(fscal,dy30);
 400             tz               = _mm256_mul_pd(fscal,dz30);
 401
 402             /* Update vectorial force */
 403             fix3             = _mm256_add_pd(fix3,tx);
 404             fiy3             = _mm256_add_pd(fiy3,ty);
 405             fiz3             = _mm256_add_pd(fiz3,tz);
 406
 407             fjx0             = _mm256_add_pd(fjx0,tx);
 408             fjy0             = _mm256_add_pd(fjy0,ty);
 409             fjz0             = _mm256_add_pd(fjz0,tz);
 410
 411             fjptrA             = f+j_coord_offsetA;
 412             fjptrB             = f+j_coord_offsetB;
 413             fjptrC             = f+j_coord_offsetC;
 414             fjptrD             = f+j_coord_offsetD;
 415
 416             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 417
 418             /* Inner loop uses 140 flops */
 419         }
 420
 421         if(jidx<j_index_end)
 422         {
 423
 424             /* Get j neighbor index, and coordinate index */
 425             jnrlistA         = jjnr[jidx];
 426             jnrlistB         = jjnr[jidx+1];
 427             jnrlistC         = jjnr[jidx+2];
 428             jnrlistD         = jjnr[jidx+3];
 429             /* Sign of each element will be negative for non-real atoms.
 430              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 431              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 432              */
 433             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 434
 435             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 436             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 437             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 438
 439             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 440             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 441             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 442             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 443             j_coord_offsetA  = DIM*jnrA;
 444             j_coord_offsetB  = DIM*jnrB;
 445             j_coord_offsetC  = DIM*jnrC;
 446             j_coord_offsetD  = DIM*jnrD;
 447
 448             /* load j atom coordinates */
 449             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 450                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 451                                                  &jx0,&jy0,&jz0);
 452
 453             /* Calculate displacement vector */
 454             dx00             = _mm256_sub_pd(ix0,jx0);
 455             dy00             = _mm256_sub_pd(iy0,jy0);
 456             dz00             = _mm256_sub_pd(iz0,jz0);
 457             dx10             = _mm256_sub_pd(ix1,jx0);
 458             dy10             = _mm256_sub_pd(iy1,jy0);
 459             dz10             = _mm256_sub_pd(iz1,jz0);
 460             dx20             = _mm256_sub_pd(ix2,jx0);
 461             dy20             = _mm256_sub_pd(iy2,jy0);
 462             dz20             = _mm256_sub_pd(iz2,jz0);
 463             dx30             = _mm256_sub_pd(ix3,jx0);
 464             dy30             = _mm256_sub_pd(iy3,jy0);
 465             dz30             = _mm256_sub_pd(iz3,jz0);
 466
 467             /* Calculate squared distance and things based on it */
 468             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 469             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 470             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 471             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 472
 473             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 474             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 475             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 476             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 477
 478             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 479             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 480             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 481
 482             /* Load parameters for j particles */
 483             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 484                                                                  charge+jnrC+0,charge+jnrD+0);
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487             vdwjidx0C        = 2*vdwtype[jnrC+0];
 488             vdwjidx0D        = 2*vdwtype[jnrD+0];
 489
 490             fjx0             = _mm256_setzero_pd();
 491             fjy0             = _mm256_setzero_pd();
 492             fjz0             = _mm256_setzero_pd();
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r00              = _mm256_mul_pd(rsq00,rinv00);
 499             r00              = _mm256_andnot_pd(dummy_mask,r00);
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 503                                             vdwioffsetptr0+vdwjidx0B,
 504                                             vdwioffsetptr0+vdwjidx0C,
 505                                             vdwioffsetptr0+vdwjidx0D,
 506                                             &c6_00,&c12_00);
 507
 508             /* Calculate table index by multiplying r with table scale and truncate to integer */
 509             rt               = _mm256_mul_pd(r00,vftabscale);
 510             vfitab           = _mm256_cvttpd_epi32(rt);
 511             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 512             vfitab           = _mm_slli_epi32(vfitab,3);
 513
 514             /* CUBIC SPLINE TABLE DISPERSION */
 515             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 516             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 517             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 518             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 519             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 520             Heps             = _mm256_mul_pd(vfeps,H);
 521             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 522             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 523             vvdw6            = _mm256_mul_pd(c6_00,VV);
 524             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 525             fvdw6            = _mm256_mul_pd(c6_00,FF);
 526
 527             /* CUBIC SPLINE TABLE REPULSION */
 528             vfitab           = _mm_add_epi32(vfitab,ifour);
 529             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 530             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 531             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 532             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 533             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 534             Heps             = _mm256_mul_pd(vfeps,H);
 535             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 536             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 537             vvdw12           = _mm256_mul_pd(c12_00,VV);
 538             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 539             fvdw12           = _mm256_mul_pd(c12_00,FF);
 540             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 541             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 545             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 546
 547             fscal            = fvdw;
 548
 549             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 550
 551             /* Calculate temporary vectorial force */
 552             tx               = _mm256_mul_pd(fscal,dx00);
 553             ty               = _mm256_mul_pd(fscal,dy00);
 554             tz               = _mm256_mul_pd(fscal,dz00);
 555
 556             /* Update vectorial force */
 557             fix0             = _mm256_add_pd(fix0,tx);
 558             fiy0             = _mm256_add_pd(fiy0,ty);
 559             fiz0             = _mm256_add_pd(fiz0,tz);
 560
 561             fjx0             = _mm256_add_pd(fjx0,tx);
 562             fjy0             = _mm256_add_pd(fjy0,ty);
 563             fjz0             = _mm256_add_pd(fjz0,tz);
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq10             = _mm256_mul_pd(iq1,jq0);
 571
 572             /* COULOMB ELECTROSTATICS */
 573             velec            = _mm256_mul_pd(qq10,rinv10);
 574             felec            = _mm256_mul_pd(velec,rinvsq10);
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velec            = _mm256_andnot_pd(dummy_mask,velec);
 578             velecsum         = _mm256_add_pd(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_pd(fscal,dx10);
 586             ty               = _mm256_mul_pd(fscal,dy10);
 587             tz               = _mm256_mul_pd(fscal,dz10);
 588
 589             /* Update vectorial force */
 590             fix1             = _mm256_add_pd(fix1,tx);
 591             fiy1             = _mm256_add_pd(fiy1,ty);
 592             fiz1             = _mm256_add_pd(fiz1,tz);
 593
 594             fjx0             = _mm256_add_pd(fjx0,tx);
 595             fjy0             = _mm256_add_pd(fjy0,ty);
 596             fjz0             = _mm256_add_pd(fjz0,tz);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq20             = _mm256_mul_pd(iq2,jq0);
 604
 605             /* COULOMB ELECTROSTATICS */
 606             velec            = _mm256_mul_pd(qq20,rinv20);
 607             felec            = _mm256_mul_pd(velec,rinvsq20);
 608
 609             /* Update potential sum for this i atom from the interaction with this j atom. */
 610             velec            = _mm256_andnot_pd(dummy_mask,velec);
 611             velecsum         = _mm256_add_pd(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm256_mul_pd(fscal,dx20);
 619             ty               = _mm256_mul_pd(fscal,dy20);
 620             tz               = _mm256_mul_pd(fscal,dz20);
 621
 622             /* Update vectorial force */
 623             fix2             = _mm256_add_pd(fix2,tx);
 624             fiy2             = _mm256_add_pd(fiy2,ty);
 625             fiz2             = _mm256_add_pd(fiz2,tz);
 626
 627             fjx0             = _mm256_add_pd(fjx0,tx);
 628             fjy0             = _mm256_add_pd(fjy0,ty);
 629             fjz0             = _mm256_add_pd(fjz0,tz);
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             /* Compute parameters for interactions between i and j atoms */
 636             qq30             = _mm256_mul_pd(iq3,jq0);
 637
 638             /* COULOMB ELECTROSTATICS */
 639             velec            = _mm256_mul_pd(qq30,rinv30);
 640             felec            = _mm256_mul_pd(velec,rinvsq30);
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velec            = _mm256_andnot_pd(dummy_mask,velec);
 644             velecsum         = _mm256_add_pd(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm256_mul_pd(fscal,dx30);
 652             ty               = _mm256_mul_pd(fscal,dy30);
 653             tz               = _mm256_mul_pd(fscal,dz30);
 654
 655             /* Update vectorial force */
 656             fix3             = _mm256_add_pd(fix3,tx);
 657             fiy3             = _mm256_add_pd(fiy3,ty);
 658             fiz3             = _mm256_add_pd(fiz3,tz);
 659
 660             fjx0             = _mm256_add_pd(fjx0,tx);
 661             fjy0             = _mm256_add_pd(fjy0,ty);
 662             fjz0             = _mm256_add_pd(fjz0,tz);
 663
 664             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 665             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 666             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 667             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 668
 669             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 670
 671             /* Inner loop uses 141 flops */
 672         }
 673
 674         /* End of innermost loop */
 675
 676         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 677                                                  f+i_coord_offset,fshift+i_shift_offset);
 678
 679         ggid                        = gid[iidx];
 680         /* Update potential energies */
 681         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 682         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 683
 684         /* Increment number of inner iterations */
 685         inneriter                  += j_index_end - j_index_start;
 686
 687         /* Outer loop uses 26 flops */
 688     }
 689
 690     /* Increment number of outer iterations */
 691     outeriter        += nri;
 692
 693     /* Update outer/inner flops */
 694
 695     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 696 }
 697 /*
 698  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_double
 699  * Electrostatics interaction: Coulomb
 700  * VdW interaction:            CubicSplineTable
 701  * Geometry:                   Water4-Particle
 702  * Calculate force/pot:        Force
 703  */
 704 void
 705 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_double
 706                     (t_nblist                    * gmx_restrict       nlist,
 707                      rvec                        * gmx_restrict          xx,
 708                      rvec                        * gmx_restrict          ff,
 709                      t_forcerec                  * gmx_restrict          fr,
 710                      t_mdatoms                   * gmx_restrict     mdatoms,
 711                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 712                      t_nrnb                      * gmx_restrict        nrnb)
 713 {
 714     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 715      * just 0 for non-waters.
 716      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 717      * jnr indices corresponding to data put in the four positions in the SIMD register.
 718      */
 719     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 720     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 721     int              jnrA,jnrB,jnrC,jnrD;
 722     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 723     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 724     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 725     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 726     real             rcutoff_scalar;
 727     real             *shiftvec,*fshift,*x,*f;
 728     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 729     real             scratch[4*DIM];
 730     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 731     real *           vdwioffsetptr0;
 732     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 733     real *           vdwioffsetptr1;
 734     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 735     real *           vdwioffsetptr2;
 736     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 737     real *           vdwioffsetptr3;
 738     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 739     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 740     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 741     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 742     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 743     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 744     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 745     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 746     real             *charge;
 747     int              nvdwtype;
 748     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 749     int              *vdwtype;
 750     real             *vdwparam;
 751     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 752     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 753     __m128i          vfitab;
 754     __m128i          ifour       = _mm_set1_epi32(4);
 755     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 756     real             *vftab;
 757     __m256d          dummy_mask,cutoff_mask;
 758     __m128           tmpmask0,tmpmask1;
 759     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 760     __m256d          one     = _mm256_set1_pd(1.0);
 761     __m256d          two     = _mm256_set1_pd(2.0);
 762     x                = xx[0];
 763     f                = ff[0];
 764
 765     nri              = nlist->nri;
 766     iinr             = nlist->iinr;
 767     jindex           = nlist->jindex;
 768     jjnr             = nlist->jjnr;
 769     shiftidx         = nlist->shift;
 770     gid              = nlist->gid;
 771     shiftvec         = fr->shift_vec[0];
 772     fshift           = fr->fshift[0];
 773     facel            = _mm256_set1_pd(fr->epsfac);
 774     charge           = mdatoms->chargeA;
 775     nvdwtype         = fr->ntype;
 776     vdwparam         = fr->nbfp;
 777     vdwtype          = mdatoms->typeA;
 778
 779     vftab            = kernel_data->table_vdw->data;
 780     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 781
 782     /* Setup water-specific parameters */
 783     inr              = nlist->iinr[0];
 784     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 785     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 786     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 787     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 788
 789     /* Avoid stupid compiler warnings */
 790     jnrA = jnrB = jnrC = jnrD = 0;
 791     j_coord_offsetA = 0;
 792     j_coord_offsetB = 0;
 793     j_coord_offsetC = 0;
 794     j_coord_offsetD = 0;
 795
 796     outeriter        = 0;
 797     inneriter        = 0;
 798
 799     for(iidx=0;iidx<4*DIM;iidx++)
 800     {
 801         scratch[iidx] = 0.0;
 802     }
 803
 804     /* Start outer loop over neighborlists */
 805     for(iidx=0; iidx<nri; iidx++)
 806     {
 807         /* Load shift vector for this list */
 808         i_shift_offset   = DIM*shiftidx[iidx];
 809
 810         /* Load limits for loop over neighbors */
 811         j_index_start    = jindex[iidx];
 812         j_index_end      = jindex[iidx+1];
 813
 814         /* Get outer coordinate index */
 815         inr              = iinr[iidx];
 816         i_coord_offset   = DIM*inr;
 817
 818         /* Load i particle coords and add shift vector */
 819         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 820                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 821
 822         fix0             = _mm256_setzero_pd();
 823         fiy0             = _mm256_setzero_pd();
 824         fiz0             = _mm256_setzero_pd();
 825         fix1             = _mm256_setzero_pd();
 826         fiy1             = _mm256_setzero_pd();
 827         fiz1             = _mm256_setzero_pd();
 828         fix2             = _mm256_setzero_pd();
 829         fiy2             = _mm256_setzero_pd();
 830         fiz2             = _mm256_setzero_pd();
 831         fix3             = _mm256_setzero_pd();
 832         fiy3             = _mm256_setzero_pd();
 833         fiz3             = _mm256_setzero_pd();
 834
 835         /* Start inner kernel loop */
 836         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 837         {
 838
 839             /* Get j neighbor index, and coordinate index */
 840             jnrA             = jjnr[jidx];
 841             jnrB             = jjnr[jidx+1];
 842             jnrC             = jjnr[jidx+2];
 843             jnrD             = jjnr[jidx+3];
 844             j_coord_offsetA  = DIM*jnrA;
 845             j_coord_offsetB  = DIM*jnrB;
 846             j_coord_offsetC  = DIM*jnrC;
 847             j_coord_offsetD  = DIM*jnrD;
 848
 849             /* load j atom coordinates */
 850             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 851                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 852                                                  &jx0,&jy0,&jz0);
 853
 854             /* Calculate displacement vector */
 855             dx00             = _mm256_sub_pd(ix0,jx0);
 856             dy00             = _mm256_sub_pd(iy0,jy0);
 857             dz00             = _mm256_sub_pd(iz0,jz0);
 858             dx10             = _mm256_sub_pd(ix1,jx0);
 859             dy10             = _mm256_sub_pd(iy1,jy0);
 860             dz10             = _mm256_sub_pd(iz1,jz0);
 861             dx20             = _mm256_sub_pd(ix2,jx0);
 862             dy20             = _mm256_sub_pd(iy2,jy0);
 863             dz20             = _mm256_sub_pd(iz2,jz0);
 864             dx30             = _mm256_sub_pd(ix3,jx0);
 865             dy30             = _mm256_sub_pd(iy3,jy0);
 866             dz30             = _mm256_sub_pd(iz3,jz0);
 867
 868             /* Calculate squared distance and things based on it */
 869             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 870             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 871             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 872             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 873
 874             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 875             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 876             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 877             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 878
 879             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 880             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 881             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 882
 883             /* Load parameters for j particles */
 884             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 885                                                                  charge+jnrC+0,charge+jnrD+0);
 886             vdwjidx0A        = 2*vdwtype[jnrA+0];
 887             vdwjidx0B        = 2*vdwtype[jnrB+0];
 888             vdwjidx0C        = 2*vdwtype[jnrC+0];
 889             vdwjidx0D        = 2*vdwtype[jnrD+0];
 890
 891             fjx0             = _mm256_setzero_pd();
 892             fjy0             = _mm256_setzero_pd();
 893             fjz0             = _mm256_setzero_pd();
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             r00              = _mm256_mul_pd(rsq00,rinv00);
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 903                                             vdwioffsetptr0+vdwjidx0B,
 904                                             vdwioffsetptr0+vdwjidx0C,
 905                                             vdwioffsetptr0+vdwjidx0D,
 906                                             &c6_00,&c12_00);
 907
 908             /* Calculate table index by multiplying r with table scale and truncate to integer */
 909             rt               = _mm256_mul_pd(r00,vftabscale);
 910             vfitab           = _mm256_cvttpd_epi32(rt);
 911             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 912             vfitab           = _mm_slli_epi32(vfitab,3);
 913
 914             /* CUBIC SPLINE TABLE DISPERSION */
 915             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 916             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 917             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 918             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 919             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 920             Heps             = _mm256_mul_pd(vfeps,H);
 921             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 922             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 923             fvdw6            = _mm256_mul_pd(c6_00,FF);
 924
 925             /* CUBIC SPLINE TABLE REPULSION */
 926             vfitab           = _mm_add_epi32(vfitab,ifour);
 927             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 928             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 929             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 930             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 931             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 932             Heps             = _mm256_mul_pd(vfeps,H);
 933             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 934             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 935             fvdw12           = _mm256_mul_pd(c12_00,FF);
 936             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 937
 938             fscal            = fvdw;
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm256_mul_pd(fscal,dx00);
 942             ty               = _mm256_mul_pd(fscal,dy00);
 943             tz               = _mm256_mul_pd(fscal,dz00);
 944
 945             /* Update vectorial force */
 946             fix0             = _mm256_add_pd(fix0,tx);
 947             fiy0             = _mm256_add_pd(fiy0,ty);
 948             fiz0             = _mm256_add_pd(fiz0,tz);
 949
 950             fjx0             = _mm256_add_pd(fjx0,tx);
 951             fjy0             = _mm256_add_pd(fjy0,ty);
 952             fjz0             = _mm256_add_pd(fjz0,tz);
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq10             = _mm256_mul_pd(iq1,jq0);
 960
 961             /* COULOMB ELECTROSTATICS */
 962             velec            = _mm256_mul_pd(qq10,rinv10);
 963             felec            = _mm256_mul_pd(velec,rinvsq10);
 964
 965             fscal            = felec;
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm256_mul_pd(fscal,dx10);
 969             ty               = _mm256_mul_pd(fscal,dy10);
 970             tz               = _mm256_mul_pd(fscal,dz10);
 971
 972             /* Update vectorial force */
 973             fix1             = _mm256_add_pd(fix1,tx);
 974             fiy1             = _mm256_add_pd(fiy1,ty);
 975             fiz1             = _mm256_add_pd(fiz1,tz);
 976
 977             fjx0             = _mm256_add_pd(fjx0,tx);
 978             fjy0             = _mm256_add_pd(fjy0,ty);
 979             fjz0             = _mm256_add_pd(fjz0,tz);
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             qq20             = _mm256_mul_pd(iq2,jq0);
 987
 988             /* COULOMB ELECTROSTATICS */
 989             velec            = _mm256_mul_pd(qq20,rinv20);
 990             felec            = _mm256_mul_pd(velec,rinvsq20);
 991
 992             fscal            = felec;
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm256_mul_pd(fscal,dx20);
 996             ty               = _mm256_mul_pd(fscal,dy20);
 997             tz               = _mm256_mul_pd(fscal,dz20);
 998
 999             /* Update vectorial force */
1000             fix2             = _mm256_add_pd(fix2,tx);
1001             fiy2             = _mm256_add_pd(fiy2,ty);
1002             fiz2             = _mm256_add_pd(fiz2,tz);
1003
1004             fjx0             = _mm256_add_pd(fjx0,tx);
1005             fjy0             = _mm256_add_pd(fjy0,ty);
1006             fjz0             = _mm256_add_pd(fjz0,tz);
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             qq30             = _mm256_mul_pd(iq3,jq0);
1014
1015             /* COULOMB ELECTROSTATICS */
1016             velec            = _mm256_mul_pd(qq30,rinv30);
1017             felec            = _mm256_mul_pd(velec,rinvsq30);
1018
1019             fscal            = felec;
1020
1021             /* Calculate temporary vectorial force */
1022             tx               = _mm256_mul_pd(fscal,dx30);
1023             ty               = _mm256_mul_pd(fscal,dy30);
1024             tz               = _mm256_mul_pd(fscal,dz30);
1025
1026             /* Update vectorial force */
1027             fix3             = _mm256_add_pd(fix3,tx);
1028             fiy3             = _mm256_add_pd(fiy3,ty);
1029             fiz3             = _mm256_add_pd(fiz3,tz);
1030
1031             fjx0             = _mm256_add_pd(fjx0,tx);
1032             fjy0             = _mm256_add_pd(fjy0,ty);
1033             fjz0             = _mm256_add_pd(fjz0,tz);
1034
1035             fjptrA             = f+j_coord_offsetA;
1036             fjptrB             = f+j_coord_offsetB;
1037             fjptrC             = f+j_coord_offsetC;
1038             fjptrD             = f+j_coord_offsetD;
1039
1040             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1041
1042             /* Inner loop uses 129 flops */
1043         }
1044
1045         if(jidx<j_index_end)
1046         {
1047
1048             /* Get j neighbor index, and coordinate index */
1049             jnrlistA         = jjnr[jidx];
1050             jnrlistB         = jjnr[jidx+1];
1051             jnrlistC         = jjnr[jidx+2];
1052             jnrlistD         = jjnr[jidx+3];
1053             /* Sign of each element will be negative for non-real atoms.
1054              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1055              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1056              */
1057             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1058
1059             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1060             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1061             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1062
1063             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1064             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1065             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1066             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1067             j_coord_offsetA  = DIM*jnrA;
1068             j_coord_offsetB  = DIM*jnrB;
1069             j_coord_offsetC  = DIM*jnrC;
1070             j_coord_offsetD  = DIM*jnrD;
1071
1072             /* load j atom coordinates */
1073             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1074                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1075                                                  &jx0,&jy0,&jz0);
1076
1077             /* Calculate displacement vector */
1078             dx00             = _mm256_sub_pd(ix0,jx0);
1079             dy00             = _mm256_sub_pd(iy0,jy0);
1080             dz00             = _mm256_sub_pd(iz0,jz0);
1081             dx10             = _mm256_sub_pd(ix1,jx0);
1082             dy10             = _mm256_sub_pd(iy1,jy0);
1083             dz10             = _mm256_sub_pd(iz1,jz0);
1084             dx20             = _mm256_sub_pd(ix2,jx0);
1085             dy20             = _mm256_sub_pd(iy2,jy0);
1086             dz20             = _mm256_sub_pd(iz2,jz0);
1087             dx30             = _mm256_sub_pd(ix3,jx0);
1088             dy30             = _mm256_sub_pd(iy3,jy0);
1089             dz30             = _mm256_sub_pd(iz3,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             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1096
1097             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1098             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1099             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1100             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1101
1102             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1103             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1104             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1105
1106             /* Load parameters for j particles */
1107             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1108                                                                  charge+jnrC+0,charge+jnrD+0);
1109             vdwjidx0A        = 2*vdwtype[jnrA+0];
1110             vdwjidx0B        = 2*vdwtype[jnrB+0];
1111             vdwjidx0C        = 2*vdwtype[jnrC+0];
1112             vdwjidx0D        = 2*vdwtype[jnrD+0];
1113
1114             fjx0             = _mm256_setzero_pd();
1115             fjy0             = _mm256_setzero_pd();
1116             fjz0             = _mm256_setzero_pd();
1117
1118             /**************************
1119              * CALCULATE INTERACTIONS *
1120              **************************/
1121
1122             r00              = _mm256_mul_pd(rsq00,rinv00);
1123             r00              = _mm256_andnot_pd(dummy_mask,r00);
1124
1125             /* Compute parameters for interactions between i and j atoms */
1126             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1127                                             vdwioffsetptr0+vdwjidx0B,
1128                                             vdwioffsetptr0+vdwjidx0C,
1129                                             vdwioffsetptr0+vdwjidx0D,
1130                                             &c6_00,&c12_00);
1131
1132             /* Calculate table index by multiplying r with table scale and truncate to integer */
1133             rt               = _mm256_mul_pd(r00,vftabscale);
1134             vfitab           = _mm256_cvttpd_epi32(rt);
1135             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1136             vfitab           = _mm_slli_epi32(vfitab,3);
1137
1138             /* CUBIC SPLINE TABLE DISPERSION */
1139             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1140             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1141             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1142             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1143             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1144             Heps             = _mm256_mul_pd(vfeps,H);
1145             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1146             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1147             fvdw6            = _mm256_mul_pd(c6_00,FF);
1148
1149             /* CUBIC SPLINE TABLE REPULSION */
1150             vfitab           = _mm_add_epi32(vfitab,ifour);
1151             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1152             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1153             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1154             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1155             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1156             Heps             = _mm256_mul_pd(vfeps,H);
1157             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1158             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1159             fvdw12           = _mm256_mul_pd(c12_00,FF);
1160             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1161
1162             fscal            = fvdw;
1163
1164             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1165
1166             /* Calculate temporary vectorial force */
1167             tx               = _mm256_mul_pd(fscal,dx00);
1168             ty               = _mm256_mul_pd(fscal,dy00);
1169             tz               = _mm256_mul_pd(fscal,dz00);
1170
1171             /* Update vectorial force */
1172             fix0             = _mm256_add_pd(fix0,tx);
1173             fiy0             = _mm256_add_pd(fiy0,ty);
1174             fiz0             = _mm256_add_pd(fiz0,tz);
1175
1176             fjx0             = _mm256_add_pd(fjx0,tx);
1177             fjy0             = _mm256_add_pd(fjy0,ty);
1178             fjz0             = _mm256_add_pd(fjz0,tz);
1179
1180             /**************************
1181              * CALCULATE INTERACTIONS *
1182              **************************/
1183
1184             /* Compute parameters for interactions between i and j atoms */
1185             qq10             = _mm256_mul_pd(iq1,jq0);
1186
1187             /* COULOMB ELECTROSTATICS */
1188             velec            = _mm256_mul_pd(qq10,rinv10);
1189             felec            = _mm256_mul_pd(velec,rinvsq10);
1190
1191             fscal            = felec;
1192
1193             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1194
1195             /* Calculate temporary vectorial force */
1196             tx               = _mm256_mul_pd(fscal,dx10);
1197             ty               = _mm256_mul_pd(fscal,dy10);
1198             tz               = _mm256_mul_pd(fscal,dz10);
1199
1200             /* Update vectorial force */
1201             fix1             = _mm256_add_pd(fix1,tx);
1202             fiy1             = _mm256_add_pd(fiy1,ty);
1203             fiz1             = _mm256_add_pd(fiz1,tz);
1204
1205             fjx0             = _mm256_add_pd(fjx0,tx);
1206             fjy0             = _mm256_add_pd(fjy0,ty);
1207             fjz0             = _mm256_add_pd(fjz0,tz);
1208
1209             /**************************
1210              * CALCULATE INTERACTIONS *
1211              **************************/
1212
1213             /* Compute parameters for interactions between i and j atoms */
1214             qq20             = _mm256_mul_pd(iq2,jq0);
1215
1216             /* COULOMB ELECTROSTATICS */
1217             velec            = _mm256_mul_pd(qq20,rinv20);
1218             felec            = _mm256_mul_pd(velec,rinvsq20);
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1223
1224             /* Calculate temporary vectorial force */
1225             tx               = _mm256_mul_pd(fscal,dx20);
1226             ty               = _mm256_mul_pd(fscal,dy20);
1227             tz               = _mm256_mul_pd(fscal,dz20);
1228
1229             /* Update vectorial force */
1230             fix2             = _mm256_add_pd(fix2,tx);
1231             fiy2             = _mm256_add_pd(fiy2,ty);
1232             fiz2             = _mm256_add_pd(fiz2,tz);
1233
1234             fjx0             = _mm256_add_pd(fjx0,tx);
1235             fjy0             = _mm256_add_pd(fjy0,ty);
1236             fjz0             = _mm256_add_pd(fjz0,tz);
1237
1238             /**************************
1239              * CALCULATE INTERACTIONS *
1240              **************************/
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             qq30             = _mm256_mul_pd(iq3,jq0);
1244
1245             /* COULOMB ELECTROSTATICS */
1246             velec            = _mm256_mul_pd(qq30,rinv30);
1247             felec            = _mm256_mul_pd(velec,rinvsq30);
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 130 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*130);
1294 }