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) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m128i          vfitab;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m256d          dummy_mask,cutoff_mask;
 111     __m128           tmpmask0,tmpmask1;
 112     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 113     __m256d          one     = _mm256_set1_pd(1.0);
 114     __m256d          two     = _mm256_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 138     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 139     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 140     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm256_setzero_pd();
 176         fiy0             = _mm256_setzero_pd();
 177         fiz0             = _mm256_setzero_pd();
 178         fix1             = _mm256_setzero_pd();
 179         fiy1             = _mm256_setzero_pd();
 180         fiz1             = _mm256_setzero_pd();
 181         fix2             = _mm256_setzero_pd();
 182         fiy2             = _mm256_setzero_pd();
 183         fiz2             = _mm256_setzero_pd();
 184         fix3             = _mm256_setzero_pd();
 185         fiy3             = _mm256_setzero_pd();
 186         fiz3             = _mm256_setzero_pd();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_pd();
 190         vvdwsum          = _mm256_setzero_pd();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm256_sub_pd(ix0,jx0);
 213             dy00             = _mm256_sub_pd(iy0,jy0);
 214             dz00             = _mm256_sub_pd(iz0,jz0);
 215             dx10             = _mm256_sub_pd(ix1,jx0);
 216             dy10             = _mm256_sub_pd(iy1,jy0);
 217             dz10             = _mm256_sub_pd(iz1,jz0);
 218             dx20             = _mm256_sub_pd(ix2,jx0);
 219             dy20             = _mm256_sub_pd(iy2,jy0);
 220             dz20             = _mm256_sub_pd(iz2,jz0);
 221             dx30             = _mm256_sub_pd(ix3,jx0);
 222             dy30             = _mm256_sub_pd(iy3,jy0);
 223             dz30             = _mm256_sub_pd(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 227             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 228             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 229             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 230
 231             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 232             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 233             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 234             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 235
 236             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 237             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 238             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 242                                                                  charge+jnrC+0,charge+jnrD+0);
 243             vdwjidx0A        = 2*vdwtype[jnrA+0];
 244             vdwjidx0B        = 2*vdwtype[jnrB+0];
 245             vdwjidx0C        = 2*vdwtype[jnrC+0];
 246             vdwjidx0D        = 2*vdwtype[jnrD+0];
 247
 248             fjx0             = _mm256_setzero_pd();
 249             fjy0             = _mm256_setzero_pd();
 250             fjz0             = _mm256_setzero_pd();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             r00              = _mm256_mul_pd(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 260                                             vdwioffsetptr0+vdwjidx0B,
 261                                             vdwioffsetptr0+vdwjidx0C,
 262                                             vdwioffsetptr0+vdwjidx0D,
 263                                             &c6_00,&c12_00);
 264
 265             /* Calculate table index by multiplying r with table scale and truncate to integer */
 266             rt               = _mm256_mul_pd(r00,vftabscale);
 267             vfitab           = _mm256_cvttpd_epi32(rt);
 268             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 269             vfitab           = _mm_slli_epi32(vfitab,3);
 270
 271             /* CUBIC SPLINE TABLE DISPERSION */
 272             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 273             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 274             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 275             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 276             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 277             Heps             = _mm256_mul_pd(vfeps,H);
 278             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 279             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 280             vvdw6            = _mm256_mul_pd(c6_00,VV);
 281             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 282             fvdw6            = _mm256_mul_pd(c6_00,FF);
 283
 284             /* CUBIC SPLINE TABLE REPULSION */
 285             vfitab           = _mm_add_epi32(vfitab,ifour);
 286             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 287             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 288             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 289             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 290             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 291             Heps             = _mm256_mul_pd(vfeps,H);
 292             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 293             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 294             vvdw12           = _mm256_mul_pd(c12_00,VV);
 295             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 296             fvdw12           = _mm256_mul_pd(c12_00,FF);
 297             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 298             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 302
 303             fscal            = fvdw;
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm256_mul_pd(fscal,dx00);
 307             ty               = _mm256_mul_pd(fscal,dy00);
 308             tz               = _mm256_mul_pd(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm256_add_pd(fix0,tx);
 312             fiy0             = _mm256_add_pd(fiy0,ty);
 313             fiz0             = _mm256_add_pd(fiz0,tz);
 314
 315             fjx0             = _mm256_add_pd(fjx0,tx);
 316             fjy0             = _mm256_add_pd(fjy0,ty);
 317             fjz0             = _mm256_add_pd(fjz0,tz);
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm256_mul_pd(iq1,jq0);
 325
 326             /* COULOMB ELECTROSTATICS */
 327             velec            = _mm256_mul_pd(qq10,rinv10);
 328             felec            = _mm256_mul_pd(velec,rinvsq10);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm256_add_pd(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_pd(fscal,dx10);
 337             ty               = _mm256_mul_pd(fscal,dy10);
 338             tz               = _mm256_mul_pd(fscal,dz10);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm256_add_pd(fix1,tx);
 342             fiy1             = _mm256_add_pd(fiy1,ty);
 343             fiz1             = _mm256_add_pd(fiz1,tz);
 344
 345             fjx0             = _mm256_add_pd(fjx0,tx);
 346             fjy0             = _mm256_add_pd(fjy0,ty);
 347             fjz0             = _mm256_add_pd(fjz0,tz);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm256_mul_pd(iq2,jq0);
 355
 356             /* COULOMB ELECTROSTATICS */
 357             velec            = _mm256_mul_pd(qq20,rinv20);
 358             felec            = _mm256_mul_pd(velec,rinvsq20);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm256_add_pd(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm256_mul_pd(fscal,dx20);
 367             ty               = _mm256_mul_pd(fscal,dy20);
 368             tz               = _mm256_mul_pd(fscal,dz20);
 369
 370             /* Update vectorial force */
 371             fix2             = _mm256_add_pd(fix2,tx);
 372             fiy2             = _mm256_add_pd(fiy2,ty);
 373             fiz2             = _mm256_add_pd(fiz2,tz);
 374
 375             fjx0             = _mm256_add_pd(fjx0,tx);
 376             fjy0             = _mm256_add_pd(fjy0,ty);
 377             fjz0             = _mm256_add_pd(fjz0,tz);
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq30             = _mm256_mul_pd(iq3,jq0);
 385
 386             /* COULOMB ELECTROSTATICS */
 387             velec            = _mm256_mul_pd(qq30,rinv30);
 388             felec            = _mm256_mul_pd(velec,rinvsq30);
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm256_add_pd(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm256_mul_pd(fscal,dx30);
 397             ty               = _mm256_mul_pd(fscal,dy30);
 398             tz               = _mm256_mul_pd(fscal,dz30);
 399
 400             /* Update vectorial force */
 401             fix3             = _mm256_add_pd(fix3,tx);
 402             fiy3             = _mm256_add_pd(fiy3,ty);
 403             fiz3             = _mm256_add_pd(fiz3,tz);
 404
 405             fjx0             = _mm256_add_pd(fjx0,tx);
 406             fjy0             = _mm256_add_pd(fjy0,ty);
 407             fjz0             = _mm256_add_pd(fjz0,tz);
 408
 409             fjptrA             = f+j_coord_offsetA;
 410             fjptrB             = f+j_coord_offsetB;
 411             fjptrC             = f+j_coord_offsetC;
 412             fjptrD             = f+j_coord_offsetD;
 413
 414             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 415
 416             /* Inner loop uses 140 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             /* Sign of each element will be negative for non-real atoms.
 428              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 429              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 430              */
 431             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 432
 433             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 434             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 435             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 436
 437             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 438             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 439             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 440             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 441             j_coord_offsetA  = DIM*jnrA;
 442             j_coord_offsetB  = DIM*jnrB;
 443             j_coord_offsetC  = DIM*jnrC;
 444             j_coord_offsetD  = DIM*jnrD;
 445
 446             /* load j atom coordinates */
 447             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 448                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 449                                                  &jx0,&jy0,&jz0);
 450
 451             /* Calculate displacement vector */
 452             dx00             = _mm256_sub_pd(ix0,jx0);
 453             dy00             = _mm256_sub_pd(iy0,jy0);
 454             dz00             = _mm256_sub_pd(iz0,jz0);
 455             dx10             = _mm256_sub_pd(ix1,jx0);
 456             dy10             = _mm256_sub_pd(iy1,jy0);
 457             dz10             = _mm256_sub_pd(iz1,jz0);
 458             dx20             = _mm256_sub_pd(ix2,jx0);
 459             dy20             = _mm256_sub_pd(iy2,jy0);
 460             dz20             = _mm256_sub_pd(iz2,jz0);
 461             dx30             = _mm256_sub_pd(ix3,jx0);
 462             dy30             = _mm256_sub_pd(iy3,jy0);
 463             dz30             = _mm256_sub_pd(iz3,jz0);
 464
 465             /* Calculate squared distance and things based on it */
 466             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 467             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 468             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 469             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 470
 471             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 472             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 473             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 474             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 475
 476             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 477             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 478             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 479
 480             /* Load parameters for j particles */
 481             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 482                                                                  charge+jnrC+0,charge+jnrD+0);
 483             vdwjidx0A        = 2*vdwtype[jnrA+0];
 484             vdwjidx0B        = 2*vdwtype[jnrB+0];
 485             vdwjidx0C        = 2*vdwtype[jnrC+0];
 486             vdwjidx0D        = 2*vdwtype[jnrD+0];
 487
 488             fjx0             = _mm256_setzero_pd();
 489             fjy0             = _mm256_setzero_pd();
 490             fjz0             = _mm256_setzero_pd();
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             r00              = _mm256_mul_pd(rsq00,rinv00);
 497             r00              = _mm256_andnot_pd(dummy_mask,r00);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 501                                             vdwioffsetptr0+vdwjidx0B,
 502                                             vdwioffsetptr0+vdwjidx0C,
 503                                             vdwioffsetptr0+vdwjidx0D,
 504                                             &c6_00,&c12_00);
 505
 506             /* Calculate table index by multiplying r with table scale and truncate to integer */
 507             rt               = _mm256_mul_pd(r00,vftabscale);
 508             vfitab           = _mm256_cvttpd_epi32(rt);
 509             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 510             vfitab           = _mm_slli_epi32(vfitab,3);
 511
 512             /* CUBIC SPLINE TABLE DISPERSION */
 513             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 514             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 515             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 516             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 517             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 518             Heps             = _mm256_mul_pd(vfeps,H);
 519             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 520             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 521             vvdw6            = _mm256_mul_pd(c6_00,VV);
 522             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 523             fvdw6            = _mm256_mul_pd(c6_00,FF);
 524
 525             /* CUBIC SPLINE TABLE REPULSION */
 526             vfitab           = _mm_add_epi32(vfitab,ifour);
 527             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 528             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 529             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 530             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 531             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 532             Heps             = _mm256_mul_pd(vfeps,H);
 533             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 534             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 535             vvdw12           = _mm256_mul_pd(c12_00,VV);
 536             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 537             fvdw12           = _mm256_mul_pd(c12_00,FF);
 538             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 539             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 543             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 544
 545             fscal            = fvdw;
 546
 547             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 548
 549             /* Calculate temporary vectorial force */
 550             tx               = _mm256_mul_pd(fscal,dx00);
 551             ty               = _mm256_mul_pd(fscal,dy00);
 552             tz               = _mm256_mul_pd(fscal,dz00);
 553
 554             /* Update vectorial force */
 555             fix0             = _mm256_add_pd(fix0,tx);
 556             fiy0             = _mm256_add_pd(fiy0,ty);
 557             fiz0             = _mm256_add_pd(fiz0,tz);
 558
 559             fjx0             = _mm256_add_pd(fjx0,tx);
 560             fjy0             = _mm256_add_pd(fjy0,ty);
 561             fjz0             = _mm256_add_pd(fjz0,tz);
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq10             = _mm256_mul_pd(iq1,jq0);
 569
 570             /* COULOMB ELECTROSTATICS */
 571             velec            = _mm256_mul_pd(qq10,rinv10);
 572             felec            = _mm256_mul_pd(velec,rinvsq10);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm256_andnot_pd(dummy_mask,velec);
 576             velecsum         = _mm256_add_pd(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_pd(fscal,dx10);
 584             ty               = _mm256_mul_pd(fscal,dy10);
 585             tz               = _mm256_mul_pd(fscal,dz10);
 586
 587             /* Update vectorial force */
 588             fix1             = _mm256_add_pd(fix1,tx);
 589             fiy1             = _mm256_add_pd(fiy1,ty);
 590             fiz1             = _mm256_add_pd(fiz1,tz);
 591
 592             fjx0             = _mm256_add_pd(fjx0,tx);
 593             fjy0             = _mm256_add_pd(fjy0,ty);
 594             fjz0             = _mm256_add_pd(fjz0,tz);
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq20             = _mm256_mul_pd(iq2,jq0);
 602
 603             /* COULOMB ELECTROSTATICS */
 604             velec            = _mm256_mul_pd(qq20,rinv20);
 605             felec            = _mm256_mul_pd(velec,rinvsq20);
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm256_andnot_pd(dummy_mask,velec);
 609             velecsum         = _mm256_add_pd(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm256_mul_pd(fscal,dx20);
 617             ty               = _mm256_mul_pd(fscal,dy20);
 618             tz               = _mm256_mul_pd(fscal,dz20);
 619
 620             /* Update vectorial force */
 621             fix2             = _mm256_add_pd(fix2,tx);
 622             fiy2             = _mm256_add_pd(fiy2,ty);
 623             fiz2             = _mm256_add_pd(fiz2,tz);
 624
 625             fjx0             = _mm256_add_pd(fjx0,tx);
 626             fjy0             = _mm256_add_pd(fjy0,ty);
 627             fjz0             = _mm256_add_pd(fjz0,tz);
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq30             = _mm256_mul_pd(iq3,jq0);
 635
 636             /* COULOMB ELECTROSTATICS */
 637             velec            = _mm256_mul_pd(qq30,rinv30);
 638             felec            = _mm256_mul_pd(velec,rinvsq30);
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velec            = _mm256_andnot_pd(dummy_mask,velec);
 642             velecsum         = _mm256_add_pd(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm256_mul_pd(fscal,dx30);
 650             ty               = _mm256_mul_pd(fscal,dy30);
 651             tz               = _mm256_mul_pd(fscal,dz30);
 652
 653             /* Update vectorial force */
 654             fix3             = _mm256_add_pd(fix3,tx);
 655             fiy3             = _mm256_add_pd(fiy3,ty);
 656             fiz3             = _mm256_add_pd(fiz3,tz);
 657
 658             fjx0             = _mm256_add_pd(fjx0,tx);
 659             fjy0             = _mm256_add_pd(fjy0,ty);
 660             fjz0             = _mm256_add_pd(fjz0,tz);
 661
 662             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 663             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 664             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 665             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 666
 667             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 668
 669             /* Inner loop uses 141 flops */
 670         }
 671
 672         /* End of innermost loop */
 673
 674         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 675                                                  f+i_coord_offset,fshift+i_shift_offset);
 676
 677         ggid                        = gid[iidx];
 678         /* Update potential energies */
 679         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 680         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 681
 682         /* Increment number of inner iterations */
 683         inneriter                  += j_index_end - j_index_start;
 684
 685         /* Outer loop uses 26 flops */
 686     }
 687
 688     /* Increment number of outer iterations */
 689     outeriter        += nri;
 690
 691     /* Update outer/inner flops */
 692
 693     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 694 }
 695 /*
 696  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_double
 697  * Electrostatics interaction: Coulomb
 698  * VdW interaction:            CubicSplineTable
 699  * Geometry:                   Water4-Particle
 700  * Calculate force/pot:        Force
 701  */
 702 void
 703 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_double
 704                     (t_nblist                    * gmx_restrict       nlist,
 705                      rvec                        * gmx_restrict          xx,
 706                      rvec                        * gmx_restrict          ff,
 707                      t_forcerec                  * gmx_restrict          fr,
 708                      t_mdatoms                   * gmx_restrict     mdatoms,
 709                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 710                      t_nrnb                      * gmx_restrict        nrnb)
 711 {
 712     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 713      * just 0 for non-waters.
 714      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 715      * jnr indices corresponding to data put in the four positions in the SIMD register.
 716      */
 717     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 718     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 719     int              jnrA,jnrB,jnrC,jnrD;
 720     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 721     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 722     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 723     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 724     real             rcutoff_scalar;
 725     real             *shiftvec,*fshift,*x,*f;
 726     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 727     real             scratch[4*DIM];
 728     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 729     real *           vdwioffsetptr0;
 730     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 731     real *           vdwioffsetptr1;
 732     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 733     real *           vdwioffsetptr2;
 734     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 735     real *           vdwioffsetptr3;
 736     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 737     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 738     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 739     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 740     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 741     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 742     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 743     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 744     real             *charge;
 745     int              nvdwtype;
 746     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 747     int              *vdwtype;
 748     real             *vdwparam;
 749     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 750     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 751     __m128i          vfitab;
 752     __m128i          ifour       = _mm_set1_epi32(4);
 753     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 754     real             *vftab;
 755     __m256d          dummy_mask,cutoff_mask;
 756     __m128           tmpmask0,tmpmask1;
 757     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 758     __m256d          one     = _mm256_set1_pd(1.0);
 759     __m256d          two     = _mm256_set1_pd(2.0);
 760     x                = xx[0];
 761     f                = ff[0];
 762
 763     nri              = nlist->nri;
 764     iinr             = nlist->iinr;
 765     jindex           = nlist->jindex;
 766     jjnr             = nlist->jjnr;
 767     shiftidx         = nlist->shift;
 768     gid              = nlist->gid;
 769     shiftvec         = fr->shift_vec[0];
 770     fshift           = fr->fshift[0];
 771     facel            = _mm256_set1_pd(fr->epsfac);
 772     charge           = mdatoms->chargeA;
 773     nvdwtype         = fr->ntype;
 774     vdwparam         = fr->nbfp;
 775     vdwtype          = mdatoms->typeA;
 776
 777     vftab            = kernel_data->table_vdw->data;
 778     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 779
 780     /* Setup water-specific parameters */
 781     inr              = nlist->iinr[0];
 782     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 783     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 784     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 785     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 786
 787     /* Avoid stupid compiler warnings */
 788     jnrA = jnrB = jnrC = jnrD = 0;
 789     j_coord_offsetA = 0;
 790     j_coord_offsetB = 0;
 791     j_coord_offsetC = 0;
 792     j_coord_offsetD = 0;
 793
 794     outeriter        = 0;
 795     inneriter        = 0;
 796
 797     for(iidx=0;iidx<4*DIM;iidx++)
 798     {
 799         scratch[iidx] = 0.0;
 800     }
 801
 802     /* Start outer loop over neighborlists */
 803     for(iidx=0; iidx<nri; iidx++)
 804     {
 805         /* Load shift vector for this list */
 806         i_shift_offset   = DIM*shiftidx[iidx];
 807
 808         /* Load limits for loop over neighbors */
 809         j_index_start    = jindex[iidx];
 810         j_index_end      = jindex[iidx+1];
 811
 812         /* Get outer coordinate index */
 813         inr              = iinr[iidx];
 814         i_coord_offset   = DIM*inr;
 815
 816         /* Load i particle coords and add shift vector */
 817         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 818                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 819
 820         fix0             = _mm256_setzero_pd();
 821         fiy0             = _mm256_setzero_pd();
 822         fiz0             = _mm256_setzero_pd();
 823         fix1             = _mm256_setzero_pd();
 824         fiy1             = _mm256_setzero_pd();
 825         fiz1             = _mm256_setzero_pd();
 826         fix2             = _mm256_setzero_pd();
 827         fiy2             = _mm256_setzero_pd();
 828         fiz2             = _mm256_setzero_pd();
 829         fix3             = _mm256_setzero_pd();
 830         fiy3             = _mm256_setzero_pd();
 831         fiz3             = _mm256_setzero_pd();
 832
 833         /* Start inner kernel loop */
 834         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 835         {
 836
 837             /* Get j neighbor index, and coordinate index */
 838             jnrA             = jjnr[jidx];
 839             jnrB             = jjnr[jidx+1];
 840             jnrC             = jjnr[jidx+2];
 841             jnrD             = jjnr[jidx+3];
 842             j_coord_offsetA  = DIM*jnrA;
 843             j_coord_offsetB  = DIM*jnrB;
 844             j_coord_offsetC  = DIM*jnrC;
 845             j_coord_offsetD  = DIM*jnrD;
 846
 847             /* load j atom coordinates */
 848             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 849                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 850                                                  &jx0,&jy0,&jz0);
 851
 852             /* Calculate displacement vector */
 853             dx00             = _mm256_sub_pd(ix0,jx0);
 854             dy00             = _mm256_sub_pd(iy0,jy0);
 855             dz00             = _mm256_sub_pd(iz0,jz0);
 856             dx10             = _mm256_sub_pd(ix1,jx0);
 857             dy10             = _mm256_sub_pd(iy1,jy0);
 858             dz10             = _mm256_sub_pd(iz1,jz0);
 859             dx20             = _mm256_sub_pd(ix2,jx0);
 860             dy20             = _mm256_sub_pd(iy2,jy0);
 861             dz20             = _mm256_sub_pd(iz2,jz0);
 862             dx30             = _mm256_sub_pd(ix3,jx0);
 863             dy30             = _mm256_sub_pd(iy3,jy0);
 864             dz30             = _mm256_sub_pd(iz3,jz0);
 865
 866             /* Calculate squared distance and things based on it */
 867             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 868             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 869             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 870             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 871
 872             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 873             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 874             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 875             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 876
 877             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 878             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 879             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 880
 881             /* Load parameters for j particles */
 882             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 883                                                                  charge+jnrC+0,charge+jnrD+0);
 884             vdwjidx0A        = 2*vdwtype[jnrA+0];
 885             vdwjidx0B        = 2*vdwtype[jnrB+0];
 886             vdwjidx0C        = 2*vdwtype[jnrC+0];
 887             vdwjidx0D        = 2*vdwtype[jnrD+0];
 888
 889             fjx0             = _mm256_setzero_pd();
 890             fjy0             = _mm256_setzero_pd();
 891             fjz0             = _mm256_setzero_pd();
 892
 893             /**************************
 894              * CALCULATE INTERACTIONS *
 895              **************************/
 896
 897             r00              = _mm256_mul_pd(rsq00,rinv00);
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 901                                             vdwioffsetptr0+vdwjidx0B,
 902                                             vdwioffsetptr0+vdwjidx0C,
 903                                             vdwioffsetptr0+vdwjidx0D,
 904                                             &c6_00,&c12_00);
 905
 906             /* Calculate table index by multiplying r with table scale and truncate to integer */
 907             rt               = _mm256_mul_pd(r00,vftabscale);
 908             vfitab           = _mm256_cvttpd_epi32(rt);
 909             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 910             vfitab           = _mm_slli_epi32(vfitab,3);
 911
 912             /* CUBIC SPLINE TABLE DISPERSION */
 913             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 914             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 915             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 916             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 917             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 918             Heps             = _mm256_mul_pd(vfeps,H);
 919             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 920             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 921             fvdw6            = _mm256_mul_pd(c6_00,FF);
 922
 923             /* CUBIC SPLINE TABLE REPULSION */
 924             vfitab           = _mm_add_epi32(vfitab,ifour);
 925             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 926             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 927             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 928             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 929             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 930             Heps             = _mm256_mul_pd(vfeps,H);
 931             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 932             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 933             fvdw12           = _mm256_mul_pd(c12_00,FF);
 934             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 935
 936             fscal            = fvdw;
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm256_mul_pd(fscal,dx00);
 940             ty               = _mm256_mul_pd(fscal,dy00);
 941             tz               = _mm256_mul_pd(fscal,dz00);
 942
 943             /* Update vectorial force */
 944             fix0             = _mm256_add_pd(fix0,tx);
 945             fiy0             = _mm256_add_pd(fiy0,ty);
 946             fiz0             = _mm256_add_pd(fiz0,tz);
 947
 948             fjx0             = _mm256_add_pd(fjx0,tx);
 949             fjy0             = _mm256_add_pd(fjy0,ty);
 950             fjz0             = _mm256_add_pd(fjz0,tz);
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq10             = _mm256_mul_pd(iq1,jq0);
 958
 959             /* COULOMB ELECTROSTATICS */
 960             velec            = _mm256_mul_pd(qq10,rinv10);
 961             felec            = _mm256_mul_pd(velec,rinvsq10);
 962
 963             fscal            = felec;
 964
 965             /* Calculate temporary vectorial force */
 966             tx               = _mm256_mul_pd(fscal,dx10);
 967             ty               = _mm256_mul_pd(fscal,dy10);
 968             tz               = _mm256_mul_pd(fscal,dz10);
 969
 970             /* Update vectorial force */
 971             fix1             = _mm256_add_pd(fix1,tx);
 972             fiy1             = _mm256_add_pd(fiy1,ty);
 973             fiz1             = _mm256_add_pd(fiz1,tz);
 974
 975             fjx0             = _mm256_add_pd(fjx0,tx);
 976             fjy0             = _mm256_add_pd(fjy0,ty);
 977             fjz0             = _mm256_add_pd(fjz0,tz);
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq20             = _mm256_mul_pd(iq2,jq0);
 985
 986             /* COULOMB ELECTROSTATICS */
 987             velec            = _mm256_mul_pd(qq20,rinv20);
 988             felec            = _mm256_mul_pd(velec,rinvsq20);
 989
 990             fscal            = felec;
 991
 992             /* Calculate temporary vectorial force */
 993             tx               = _mm256_mul_pd(fscal,dx20);
 994             ty               = _mm256_mul_pd(fscal,dy20);
 995             tz               = _mm256_mul_pd(fscal,dz20);
 996
 997             /* Update vectorial force */
 998             fix2             = _mm256_add_pd(fix2,tx);
 999             fiy2             = _mm256_add_pd(fiy2,ty);
1000             fiz2             = _mm256_add_pd(fiz2,tz);
1001
1002             fjx0             = _mm256_add_pd(fjx0,tx);
1003             fjy0             = _mm256_add_pd(fjy0,ty);
1004             fjz0             = _mm256_add_pd(fjz0,tz);
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq30             = _mm256_mul_pd(iq3,jq0);
1012
1013             /* COULOMB ELECTROSTATICS */
1014             velec            = _mm256_mul_pd(qq30,rinv30);
1015             felec            = _mm256_mul_pd(velec,rinvsq30);
1016
1017             fscal            = felec;
1018
1019             /* Calculate temporary vectorial force */
1020             tx               = _mm256_mul_pd(fscal,dx30);
1021             ty               = _mm256_mul_pd(fscal,dy30);
1022             tz               = _mm256_mul_pd(fscal,dz30);
1023
1024             /* Update vectorial force */
1025             fix3             = _mm256_add_pd(fix3,tx);
1026             fiy3             = _mm256_add_pd(fiy3,ty);
1027             fiz3             = _mm256_add_pd(fiz3,tz);
1028
1029             fjx0             = _mm256_add_pd(fjx0,tx);
1030             fjy0             = _mm256_add_pd(fjy0,ty);
1031             fjz0             = _mm256_add_pd(fjz0,tz);
1032
1033             fjptrA             = f+j_coord_offsetA;
1034             fjptrB             = f+j_coord_offsetB;
1035             fjptrC             = f+j_coord_offsetC;
1036             fjptrD             = f+j_coord_offsetD;
1037
1038             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1039
1040             /* Inner loop uses 129 flops */
1041         }
1042
1043         if(jidx<j_index_end)
1044         {
1045
1046             /* Get j neighbor index, and coordinate index */
1047             jnrlistA         = jjnr[jidx];
1048             jnrlistB         = jjnr[jidx+1];
1049             jnrlistC         = jjnr[jidx+2];
1050             jnrlistD         = jjnr[jidx+3];
1051             /* Sign of each element will be negative for non-real atoms.
1052              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1053              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1054              */
1055             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1056
1057             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1058             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1059             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1060
1061             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1062             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1063             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1064             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1065             j_coord_offsetA  = DIM*jnrA;
1066             j_coord_offsetB  = DIM*jnrB;
1067             j_coord_offsetC  = DIM*jnrC;
1068             j_coord_offsetD  = DIM*jnrD;
1069
1070             /* load j atom coordinates */
1071             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1072                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1073                                                  &jx0,&jy0,&jz0);
1074
1075             /* Calculate displacement vector */
1076             dx00             = _mm256_sub_pd(ix0,jx0);
1077             dy00             = _mm256_sub_pd(iy0,jy0);
1078             dz00             = _mm256_sub_pd(iz0,jz0);
1079             dx10             = _mm256_sub_pd(ix1,jx0);
1080             dy10             = _mm256_sub_pd(iy1,jy0);
1081             dz10             = _mm256_sub_pd(iz1,jz0);
1082             dx20             = _mm256_sub_pd(ix2,jx0);
1083             dy20             = _mm256_sub_pd(iy2,jy0);
1084             dz20             = _mm256_sub_pd(iz2,jz0);
1085             dx30             = _mm256_sub_pd(ix3,jx0);
1086             dy30             = _mm256_sub_pd(iy3,jy0);
1087             dz30             = _mm256_sub_pd(iz3,jz0);
1088
1089             /* Calculate squared distance and things based on it */
1090             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1091             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1092             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1093             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1094
1095             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1096             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1097             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1098             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1099
1100             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1101             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1102             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1103
1104             /* Load parameters for j particles */
1105             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1106                                                                  charge+jnrC+0,charge+jnrD+0);
1107             vdwjidx0A        = 2*vdwtype[jnrA+0];
1108             vdwjidx0B        = 2*vdwtype[jnrB+0];
1109             vdwjidx0C        = 2*vdwtype[jnrC+0];
1110             vdwjidx0D        = 2*vdwtype[jnrD+0];
1111
1112             fjx0             = _mm256_setzero_pd();
1113             fjy0             = _mm256_setzero_pd();
1114             fjz0             = _mm256_setzero_pd();
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             r00              = _mm256_mul_pd(rsq00,rinv00);
1121             r00              = _mm256_andnot_pd(dummy_mask,r00);
1122
1123             /* Compute parameters for interactions between i and j atoms */
1124             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1125                                             vdwioffsetptr0+vdwjidx0B,
1126                                             vdwioffsetptr0+vdwjidx0C,
1127                                             vdwioffsetptr0+vdwjidx0D,
1128                                             &c6_00,&c12_00);
1129
1130             /* Calculate table index by multiplying r with table scale and truncate to integer */
1131             rt               = _mm256_mul_pd(r00,vftabscale);
1132             vfitab           = _mm256_cvttpd_epi32(rt);
1133             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1134             vfitab           = _mm_slli_epi32(vfitab,3);
1135
1136             /* CUBIC SPLINE TABLE DISPERSION */
1137             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1138             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1139             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1140             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1141             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1142             Heps             = _mm256_mul_pd(vfeps,H);
1143             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1144             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1145             fvdw6            = _mm256_mul_pd(c6_00,FF);
1146
1147             /* CUBIC SPLINE TABLE REPULSION */
1148             vfitab           = _mm_add_epi32(vfitab,ifour);
1149             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1150             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1151             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1152             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1153             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1154             Heps             = _mm256_mul_pd(vfeps,H);
1155             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1156             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1157             fvdw12           = _mm256_mul_pd(c12_00,FF);
1158             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1159
1160             fscal            = fvdw;
1161
1162             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1163
1164             /* Calculate temporary vectorial force */
1165             tx               = _mm256_mul_pd(fscal,dx00);
1166             ty               = _mm256_mul_pd(fscal,dy00);
1167             tz               = _mm256_mul_pd(fscal,dz00);
1168
1169             /* Update vectorial force */
1170             fix0             = _mm256_add_pd(fix0,tx);
1171             fiy0             = _mm256_add_pd(fiy0,ty);
1172             fiz0             = _mm256_add_pd(fiz0,tz);
1173
1174             fjx0             = _mm256_add_pd(fjx0,tx);
1175             fjy0             = _mm256_add_pd(fjy0,ty);
1176             fjz0             = _mm256_add_pd(fjz0,tz);
1177
1178             /**************************
1179              * CALCULATE INTERACTIONS *
1180              **************************/
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             qq10             = _mm256_mul_pd(iq1,jq0);
1184
1185             /* COULOMB ELECTROSTATICS */
1186             velec            = _mm256_mul_pd(qq10,rinv10);
1187             felec            = _mm256_mul_pd(velec,rinvsq10);
1188
1189             fscal            = felec;
1190
1191             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1192
1193             /* Calculate temporary vectorial force */
1194             tx               = _mm256_mul_pd(fscal,dx10);
1195             ty               = _mm256_mul_pd(fscal,dy10);
1196             tz               = _mm256_mul_pd(fscal,dz10);
1197
1198             /* Update vectorial force */
1199             fix1             = _mm256_add_pd(fix1,tx);
1200             fiy1             = _mm256_add_pd(fiy1,ty);
1201             fiz1             = _mm256_add_pd(fiz1,tz);
1202
1203             fjx0             = _mm256_add_pd(fjx0,tx);
1204             fjy0             = _mm256_add_pd(fjy0,ty);
1205             fjz0             = _mm256_add_pd(fjz0,tz);
1206
1207             /**************************
1208              * CALCULATE INTERACTIONS *
1209              **************************/
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq20             = _mm256_mul_pd(iq2,jq0);
1213
1214             /* COULOMB ELECTROSTATICS */
1215             velec            = _mm256_mul_pd(qq20,rinv20);
1216             felec            = _mm256_mul_pd(velec,rinvsq20);
1217
1218             fscal            = felec;
1219
1220             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1221
1222             /* Calculate temporary vectorial force */
1223             tx               = _mm256_mul_pd(fscal,dx20);
1224             ty               = _mm256_mul_pd(fscal,dy20);
1225             tz               = _mm256_mul_pd(fscal,dz20);
1226
1227             /* Update vectorial force */
1228             fix2             = _mm256_add_pd(fix2,tx);
1229             fiy2             = _mm256_add_pd(fiy2,ty);
1230             fiz2             = _mm256_add_pd(fiz2,tz);
1231
1232             fjx0             = _mm256_add_pd(fjx0,tx);
1233             fjy0             = _mm256_add_pd(fjy0,ty);
1234             fjz0             = _mm256_add_pd(fjz0,tz);
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq30             = _mm256_mul_pd(iq3,jq0);
1242
1243             /* COULOMB ELECTROSTATICS */
1244             velec            = _mm256_mul_pd(qq30,rinv30);
1245             felec            = _mm256_mul_pd(velec,rinvsq30);
1246
1247             fscal            = felec;
1248
1249             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1250
1251             /* Calculate temporary vectorial force */
1252             tx               = _mm256_mul_pd(fscal,dx30);
1253             ty               = _mm256_mul_pd(fscal,dy30);
1254             tz               = _mm256_mul_pd(fscal,dz30);
1255
1256             /* Update vectorial force */
1257             fix3             = _mm256_add_pd(fix3,tx);
1258             fiy3             = _mm256_add_pd(fiy3,ty);
1259             fiz3             = _mm256_add_pd(fiz3,tz);
1260
1261             fjx0             = _mm256_add_pd(fjx0,tx);
1262             fjy0             = _mm256_add_pd(fjy0,ty);
1263             fjz0             = _mm256_add_pd(fjz0,tz);
1264
1265             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1266             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1267             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1268             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1269
1270             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1271
1272             /* Inner loop uses 130 flops */
1273         }
1274
1275         /* End of innermost loop */
1276
1277         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1278                                                  f+i_coord_offset,fshift+i_shift_offset);
1279
1280         /* Increment number of inner iterations */
1281         inneriter                  += j_index_end - j_index_start;
1282
1283         /* Outer loop uses 24 flops */
1284     }
1285
1286     /* Increment number of outer iterations */
1287     outeriter        += nri;
1288
1289     /* Update outer/inner flops */
1290
1291     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1292 }