src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  84     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  91     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  92     __m128i          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m256d          dummy_mask,cutoff_mask;
  97     __m128           tmpmask0,tmpmask1;
  98     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  99     __m256d          one     = _mm256_set1_pd(1.0);
 100     __m256d          two     = _mm256_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm256_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     vftab            = kernel_data->table_vdw->data;
 119     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 124     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 125     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 126     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     for(iidx=0;iidx<4*DIM;iidx++)
 139     {
 140         scratch[iidx] = 0.0;
 141     }
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 160
 161         fix0             = _mm256_setzero_pd();
 162         fiy0             = _mm256_setzero_pd();
 163         fiz0             = _mm256_setzero_pd();
 164         fix1             = _mm256_setzero_pd();
 165         fiy1             = _mm256_setzero_pd();
 166         fiz1             = _mm256_setzero_pd();
 167         fix2             = _mm256_setzero_pd();
 168         fiy2             = _mm256_setzero_pd();
 169         fiz2             = _mm256_setzero_pd();
 170         fix3             = _mm256_setzero_pd();
 171         fiy3             = _mm256_setzero_pd();
 172         fiz3             = _mm256_setzero_pd();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_pd();
 176         vvdwsum          = _mm256_setzero_pd();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 195                                                  &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm256_sub_pd(ix0,jx0);
 199             dy00             = _mm256_sub_pd(iy0,jy0);
 200             dz00             = _mm256_sub_pd(iz0,jz0);
 201             dx10             = _mm256_sub_pd(ix1,jx0);
 202             dy10             = _mm256_sub_pd(iy1,jy0);
 203             dz10             = _mm256_sub_pd(iz1,jz0);
 204             dx20             = _mm256_sub_pd(ix2,jx0);
 205             dy20             = _mm256_sub_pd(iy2,jy0);
 206             dz20             = _mm256_sub_pd(iz2,jz0);
 207             dx30             = _mm256_sub_pd(ix3,jx0);
 208             dy30             = _mm256_sub_pd(iy3,jy0);
 209             dz30             = _mm256_sub_pd(iz3,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 215             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 216
 217             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 218             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 219             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 220             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 221
 222             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 223             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 224             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 228                                                                  charge+jnrC+0,charge+jnrD+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231             vdwjidx0C        = 2*vdwtype[jnrC+0];
 232             vdwjidx0D        = 2*vdwtype[jnrD+0];
 233
 234             fjx0             = _mm256_setzero_pd();
 235             fjy0             = _mm256_setzero_pd();
 236             fjz0             = _mm256_setzero_pd();
 237
 238             /**************************
 239              * CALCULATE INTERACTIONS *
 240              **************************/
 241
 242             r00              = _mm256_mul_pd(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 246                                             vdwioffsetptr0+vdwjidx0B,
 247                                             vdwioffsetptr0+vdwjidx0C,
 248                                             vdwioffsetptr0+vdwjidx0D,
 249                                             &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm256_mul_pd(r00,vftabscale);
 253             vfitab           = _mm256_cvttpd_epi32(rt);
 254             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 255             vfitab           = _mm_slli_epi32(vfitab,3);
 256
 257             /* CUBIC SPLINE TABLE DISPERSION */
 258             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 259             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 260             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 261             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 262             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 263             Heps             = _mm256_mul_pd(vfeps,H);
 264             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 265             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 266             vvdw6            = _mm256_mul_pd(c6_00,VV);
 267             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 268             fvdw6            = _mm256_mul_pd(c6_00,FF);
 269
 270             /* CUBIC SPLINE TABLE REPULSION */
 271             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm256_mul_pd(c12_00,VV);
 281             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 282             fvdw12           = _mm256_mul_pd(c12_00,FF);
 283             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 284             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 288
 289             fscal            = fvdw;
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm256_mul_pd(fscal,dx00);
 293             ty               = _mm256_mul_pd(fscal,dy00);
 294             tz               = _mm256_mul_pd(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm256_add_pd(fix0,tx);
 298             fiy0             = _mm256_add_pd(fiy0,ty);
 299             fiz0             = _mm256_add_pd(fiz0,tz);
 300
 301             fjx0             = _mm256_add_pd(fjx0,tx);
 302             fjy0             = _mm256_add_pd(fjy0,ty);
 303             fjz0             = _mm256_add_pd(fjz0,tz);
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq10             = _mm256_mul_pd(iq1,jq0);
 311
 312             /* COULOMB ELECTROSTATICS */
 313             velec            = _mm256_mul_pd(qq10,rinv10);
 314             felec            = _mm256_mul_pd(velec,rinvsq10);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm256_add_pd(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             /* Calculate temporary vectorial force */
 322             tx               = _mm256_mul_pd(fscal,dx10);
 323             ty               = _mm256_mul_pd(fscal,dy10);
 324             tz               = _mm256_mul_pd(fscal,dz10);
 325
 326             /* Update vectorial force */
 327             fix1             = _mm256_add_pd(fix1,tx);
 328             fiy1             = _mm256_add_pd(fiy1,ty);
 329             fiz1             = _mm256_add_pd(fiz1,tz);
 330
 331             fjx0             = _mm256_add_pd(fjx0,tx);
 332             fjy0             = _mm256_add_pd(fjy0,ty);
 333             fjz0             = _mm256_add_pd(fjz0,tz);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq20             = _mm256_mul_pd(iq2,jq0);
 341
 342             /* COULOMB ELECTROSTATICS */
 343             velec            = _mm256_mul_pd(qq20,rinv20);
 344             felec            = _mm256_mul_pd(velec,rinvsq20);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm256_add_pd(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_pd(fscal,dx20);
 353             ty               = _mm256_mul_pd(fscal,dy20);
 354             tz               = _mm256_mul_pd(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm256_add_pd(fix2,tx);
 358             fiy2             = _mm256_add_pd(fiy2,ty);
 359             fiz2             = _mm256_add_pd(fiz2,tz);
 360
 361             fjx0             = _mm256_add_pd(fjx0,tx);
 362             fjy0             = _mm256_add_pd(fjy0,ty);
 363             fjz0             = _mm256_add_pd(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq30             = _mm256_mul_pd(iq3,jq0);
 371
 372             /* COULOMB ELECTROSTATICS */
 373             velec            = _mm256_mul_pd(qq30,rinv30);
 374             felec            = _mm256_mul_pd(velec,rinvsq30);
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm256_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm256_mul_pd(fscal,dx30);
 383             ty               = _mm256_mul_pd(fscal,dy30);
 384             tz               = _mm256_mul_pd(fscal,dz30);
 385
 386             /* Update vectorial force */
 387             fix3             = _mm256_add_pd(fix3,tx);
 388             fiy3             = _mm256_add_pd(fiy3,ty);
 389             fiz3             = _mm256_add_pd(fiz3,tz);
 390
 391             fjx0             = _mm256_add_pd(fjx0,tx);
 392             fjy0             = _mm256_add_pd(fjy0,ty);
 393             fjz0             = _mm256_add_pd(fjz0,tz);
 394
 395             fjptrA             = f+j_coord_offsetA;
 396             fjptrB             = f+j_coord_offsetB;
 397             fjptrC             = f+j_coord_offsetC;
 398             fjptrD             = f+j_coord_offsetD;
 399
 400             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 401
 402             /* Inner loop uses 140 flops */
 403         }
 404
 405         if(jidx<j_index_end)
 406         {
 407
 408             /* Get j neighbor index, and coordinate index */
 409             jnrlistA         = jjnr[jidx];
 410             jnrlistB         = jjnr[jidx+1];
 411             jnrlistC         = jjnr[jidx+2];
 412             jnrlistD         = jjnr[jidx+3];
 413             /* Sign of each element will be negative for non-real atoms.
 414              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 415              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 416              */
 417             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 418
 419             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 420             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 421             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 422
 423             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 424             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 425             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 426             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 427             j_coord_offsetA  = DIM*jnrA;
 428             j_coord_offsetB  = DIM*jnrB;
 429             j_coord_offsetC  = DIM*jnrC;
 430             j_coord_offsetD  = DIM*jnrD;
 431
 432             /* load j atom coordinates */
 433             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 434                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 435                                                  &jx0,&jy0,&jz0);
 436
 437             /* Calculate displacement vector */
 438             dx00             = _mm256_sub_pd(ix0,jx0);
 439             dy00             = _mm256_sub_pd(iy0,jy0);
 440             dz00             = _mm256_sub_pd(iz0,jz0);
 441             dx10             = _mm256_sub_pd(ix1,jx0);
 442             dy10             = _mm256_sub_pd(iy1,jy0);
 443             dz10             = _mm256_sub_pd(iz1,jz0);
 444             dx20             = _mm256_sub_pd(ix2,jx0);
 445             dy20             = _mm256_sub_pd(iy2,jy0);
 446             dz20             = _mm256_sub_pd(iz2,jz0);
 447             dx30             = _mm256_sub_pd(ix3,jx0);
 448             dy30             = _mm256_sub_pd(iy3,jy0);
 449             dz30             = _mm256_sub_pd(iz3,jz0);
 450
 451             /* Calculate squared distance and things based on it */
 452             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 453             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 454             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 455             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 456
 457             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 458             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 459             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 460             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 461
 462             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 463             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 464             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 465
 466             /* Load parameters for j particles */
 467             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 468                                                                  charge+jnrC+0,charge+jnrD+0);
 469             vdwjidx0A        = 2*vdwtype[jnrA+0];
 470             vdwjidx0B        = 2*vdwtype[jnrB+0];
 471             vdwjidx0C        = 2*vdwtype[jnrC+0];
 472             vdwjidx0D        = 2*vdwtype[jnrD+0];
 473
 474             fjx0             = _mm256_setzero_pd();
 475             fjy0             = _mm256_setzero_pd();
 476             fjz0             = _mm256_setzero_pd();
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r00              = _mm256_mul_pd(rsq00,rinv00);
 483             r00              = _mm256_andnot_pd(dummy_mask,r00);
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 487                                             vdwioffsetptr0+vdwjidx0B,
 488                                             vdwioffsetptr0+vdwjidx0C,
 489                                             vdwioffsetptr0+vdwjidx0D,
 490                                             &c6_00,&c12_00);
 491
 492             /* Calculate table index by multiplying r with table scale and truncate to integer */
 493             rt               = _mm256_mul_pd(r00,vftabscale);
 494             vfitab           = _mm256_cvttpd_epi32(rt);
 495             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 496             vfitab           = _mm_slli_epi32(vfitab,3);
 497
 498             /* CUBIC SPLINE TABLE DISPERSION */
 499             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 500             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 501             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 502             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 503             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 504             Heps             = _mm256_mul_pd(vfeps,H);
 505             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 506             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 507             vvdw6            = _mm256_mul_pd(c6_00,VV);
 508             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 509             fvdw6            = _mm256_mul_pd(c6_00,FF);
 510
 511             /* CUBIC SPLINE TABLE REPULSION */
 512             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm256_mul_pd(c12_00,VV);
 522             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 523             fvdw12           = _mm256_mul_pd(c12_00,FF);
 524             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 525             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 529             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 530
 531             fscal            = fvdw;
 532
 533             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 534
 535             /* Calculate temporary vectorial force */
 536             tx               = _mm256_mul_pd(fscal,dx00);
 537             ty               = _mm256_mul_pd(fscal,dy00);
 538             tz               = _mm256_mul_pd(fscal,dz00);
 539
 540             /* Update vectorial force */
 541             fix0             = _mm256_add_pd(fix0,tx);
 542             fiy0             = _mm256_add_pd(fiy0,ty);
 543             fiz0             = _mm256_add_pd(fiz0,tz);
 544
 545             fjx0             = _mm256_add_pd(fjx0,tx);
 546             fjy0             = _mm256_add_pd(fjy0,ty);
 547             fjz0             = _mm256_add_pd(fjz0,tz);
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq10             = _mm256_mul_pd(iq1,jq0);
 555
 556             /* COULOMB ELECTROSTATICS */
 557             velec            = _mm256_mul_pd(qq10,rinv10);
 558             felec            = _mm256_mul_pd(velec,rinvsq10);
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm256_andnot_pd(dummy_mask,velec);
 562             velecsum         = _mm256_add_pd(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm256_mul_pd(fscal,dx10);
 570             ty               = _mm256_mul_pd(fscal,dy10);
 571             tz               = _mm256_mul_pd(fscal,dz10);
 572
 573             /* Update vectorial force */
 574             fix1             = _mm256_add_pd(fix1,tx);
 575             fiy1             = _mm256_add_pd(fiy1,ty);
 576             fiz1             = _mm256_add_pd(fiz1,tz);
 577
 578             fjx0             = _mm256_add_pd(fjx0,tx);
 579             fjy0             = _mm256_add_pd(fjy0,ty);
 580             fjz0             = _mm256_add_pd(fjz0,tz);
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq20             = _mm256_mul_pd(iq2,jq0);
 588
 589             /* COULOMB ELECTROSTATICS */
 590             velec            = _mm256_mul_pd(qq20,rinv20);
 591             felec            = _mm256_mul_pd(velec,rinvsq20);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm256_andnot_pd(dummy_mask,velec);
 595             velecsum         = _mm256_add_pd(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm256_mul_pd(fscal,dx20);
 603             ty               = _mm256_mul_pd(fscal,dy20);
 604             tz               = _mm256_mul_pd(fscal,dz20);
 605
 606             /* Update vectorial force */
 607             fix2             = _mm256_add_pd(fix2,tx);
 608             fiy2             = _mm256_add_pd(fiy2,ty);
 609             fiz2             = _mm256_add_pd(fiz2,tz);
 610
 611             fjx0             = _mm256_add_pd(fjx0,tx);
 612             fjy0             = _mm256_add_pd(fjy0,ty);
 613             fjz0             = _mm256_add_pd(fjz0,tz);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq30             = _mm256_mul_pd(iq3,jq0);
 621
 622             /* COULOMB ELECTROSTATICS */
 623             velec            = _mm256_mul_pd(qq30,rinv30);
 624             felec            = _mm256_mul_pd(velec,rinvsq30);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm256_andnot_pd(dummy_mask,velec);
 628             velecsum         = _mm256_add_pd(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_pd(fscal,dx30);
 636             ty               = _mm256_mul_pd(fscal,dy30);
 637             tz               = _mm256_mul_pd(fscal,dz30);
 638
 639             /* Update vectorial force */
 640             fix3             = _mm256_add_pd(fix3,tx);
 641             fiy3             = _mm256_add_pd(fiy3,ty);
 642             fiz3             = _mm256_add_pd(fiz3,tz);
 643
 644             fjx0             = _mm256_add_pd(fjx0,tx);
 645             fjy0             = _mm256_add_pd(fjy0,ty);
 646             fjz0             = _mm256_add_pd(fjz0,tz);
 647
 648             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 649             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 650             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 651             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 652
 653             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 654
 655             /* Inner loop uses 141 flops */
 656         }
 657
 658         /* End of innermost loop */
 659
 660         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 661                                                  f+i_coord_offset,fshift+i_shift_offset);
 662
 663         ggid                        = gid[iidx];
 664         /* Update potential energies */
 665         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 666         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 667
 668         /* Increment number of inner iterations */
 669         inneriter                  += j_index_end - j_index_start;
 670
 671         /* Outer loop uses 26 flops */
 672     }
 673
 674     /* Increment number of outer iterations */
 675     outeriter        += nri;
 676
 677     /* Update outer/inner flops */
 678
 679     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 680 }
 681 /*
 682  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_double
 683  * Electrostatics interaction: Coulomb
 684  * VdW interaction:            CubicSplineTable
 685  * Geometry:                   Water4-Particle
 686  * Calculate force/pot:        Force
 687  */
 688 void
 689 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_double
 690                     (t_nblist * gmx_restrict                nlist,
 691                      rvec * gmx_restrict                    xx,
 692                      rvec * gmx_restrict                    ff,
 693                      t_forcerec * gmx_restrict              fr,
 694                      t_mdatoms * gmx_restrict               mdatoms,
 695                      nb_kernel_data_t * gmx_restrict        kernel_data,
 696                      t_nrnb * gmx_restrict                  nrnb)
 697 {
 698     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 699      * just 0 for non-waters.
 700      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 701      * jnr indices corresponding to data put in the four positions in the SIMD register.
 702      */
 703     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 704     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 705     int              jnrA,jnrB,jnrC,jnrD;
 706     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 707     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 708     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 709     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 710     real             rcutoff_scalar;
 711     real             *shiftvec,*fshift,*x,*f;
 712     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 713     real             scratch[4*DIM];
 714     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 715     real *           vdwioffsetptr0;
 716     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 717     real *           vdwioffsetptr1;
 718     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 719     real *           vdwioffsetptr2;
 720     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 721     real *           vdwioffsetptr3;
 722     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 723     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 724     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 725     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 726     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 727     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 728     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 729     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 730     real             *charge;
 731     int              nvdwtype;
 732     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 733     int              *vdwtype;
 734     real             *vdwparam;
 735     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 736     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 737     __m128i          vfitab;
 738     __m128i          ifour       = _mm_set1_epi32(4);
 739     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 740     real             *vftab;
 741     __m256d          dummy_mask,cutoff_mask;
 742     __m128           tmpmask0,tmpmask1;
 743     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 744     __m256d          one     = _mm256_set1_pd(1.0);
 745     __m256d          two     = _mm256_set1_pd(2.0);
 746     x                = xx[0];
 747     f                = ff[0];
 748
 749     nri              = nlist->nri;
 750     iinr             = nlist->iinr;
 751     jindex           = nlist->jindex;
 752     jjnr             = nlist->jjnr;
 753     shiftidx         = nlist->shift;
 754     gid              = nlist->gid;
 755     shiftvec         = fr->shift_vec[0];
 756     fshift           = fr->fshift[0];
 757     facel            = _mm256_set1_pd(fr->epsfac);
 758     charge           = mdatoms->chargeA;
 759     nvdwtype         = fr->ntype;
 760     vdwparam         = fr->nbfp;
 761     vdwtype          = mdatoms->typeA;
 762
 763     vftab            = kernel_data->table_vdw->data;
 764     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 765
 766     /* Setup water-specific parameters */
 767     inr              = nlist->iinr[0];
 768     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 769     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 770     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 771     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 772
 773     /* Avoid stupid compiler warnings */
 774     jnrA = jnrB = jnrC = jnrD = 0;
 775     j_coord_offsetA = 0;
 776     j_coord_offsetB = 0;
 777     j_coord_offsetC = 0;
 778     j_coord_offsetD = 0;
 779
 780     outeriter        = 0;
 781     inneriter        = 0;
 782
 783     for(iidx=0;iidx<4*DIM;iidx++)
 784     {
 785         scratch[iidx] = 0.0;
 786     }
 787
 788     /* Start outer loop over neighborlists */
 789     for(iidx=0; iidx<nri; iidx++)
 790     {
 791         /* Load shift vector for this list */
 792         i_shift_offset   = DIM*shiftidx[iidx];
 793
 794         /* Load limits for loop over neighbors */
 795         j_index_start    = jindex[iidx];
 796         j_index_end      = jindex[iidx+1];
 797
 798         /* Get outer coordinate index */
 799         inr              = iinr[iidx];
 800         i_coord_offset   = DIM*inr;
 801
 802         /* Load i particle coords and add shift vector */
 803         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 804                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 805
 806         fix0             = _mm256_setzero_pd();
 807         fiy0             = _mm256_setzero_pd();
 808         fiz0             = _mm256_setzero_pd();
 809         fix1             = _mm256_setzero_pd();
 810         fiy1             = _mm256_setzero_pd();
 811         fiz1             = _mm256_setzero_pd();
 812         fix2             = _mm256_setzero_pd();
 813         fiy2             = _mm256_setzero_pd();
 814         fiz2             = _mm256_setzero_pd();
 815         fix3             = _mm256_setzero_pd();
 816         fiy3             = _mm256_setzero_pd();
 817         fiz3             = _mm256_setzero_pd();
 818
 819         /* Start inner kernel loop */
 820         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 821         {
 822
 823             /* Get j neighbor index, and coordinate index */
 824             jnrA             = jjnr[jidx];
 825             jnrB             = jjnr[jidx+1];
 826             jnrC             = jjnr[jidx+2];
 827             jnrD             = jjnr[jidx+3];
 828             j_coord_offsetA  = DIM*jnrA;
 829             j_coord_offsetB  = DIM*jnrB;
 830             j_coord_offsetC  = DIM*jnrC;
 831             j_coord_offsetD  = DIM*jnrD;
 832
 833             /* load j atom coordinates */
 834             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 835                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 836                                                  &jx0,&jy0,&jz0);
 837
 838             /* Calculate displacement vector */
 839             dx00             = _mm256_sub_pd(ix0,jx0);
 840             dy00             = _mm256_sub_pd(iy0,jy0);
 841             dz00             = _mm256_sub_pd(iz0,jz0);
 842             dx10             = _mm256_sub_pd(ix1,jx0);
 843             dy10             = _mm256_sub_pd(iy1,jy0);
 844             dz10             = _mm256_sub_pd(iz1,jz0);
 845             dx20             = _mm256_sub_pd(ix2,jx0);
 846             dy20             = _mm256_sub_pd(iy2,jy0);
 847             dz20             = _mm256_sub_pd(iz2,jz0);
 848             dx30             = _mm256_sub_pd(ix3,jx0);
 849             dy30             = _mm256_sub_pd(iy3,jy0);
 850             dz30             = _mm256_sub_pd(iz3,jz0);
 851
 852             /* Calculate squared distance and things based on it */
 853             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 854             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 855             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 856             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 857
 858             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 859             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 860             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 861             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 862
 863             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 864             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 865             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 866
 867             /* Load parameters for j particles */
 868             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 869                                                                  charge+jnrC+0,charge+jnrD+0);
 870             vdwjidx0A        = 2*vdwtype[jnrA+0];
 871             vdwjidx0B        = 2*vdwtype[jnrB+0];
 872             vdwjidx0C        = 2*vdwtype[jnrC+0];
 873             vdwjidx0D        = 2*vdwtype[jnrD+0];
 874
 875             fjx0             = _mm256_setzero_pd();
 876             fjy0             = _mm256_setzero_pd();
 877             fjz0             = _mm256_setzero_pd();
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             r00              = _mm256_mul_pd(rsq00,rinv00);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 887                                             vdwioffsetptr0+vdwjidx0B,
 888                                             vdwioffsetptr0+vdwjidx0C,
 889                                             vdwioffsetptr0+vdwjidx0D,
 890                                             &c6_00,&c12_00);
 891
 892             /* Calculate table index by multiplying r with table scale and truncate to integer */
 893             rt               = _mm256_mul_pd(r00,vftabscale);
 894             vfitab           = _mm256_cvttpd_epi32(rt);
 895             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 896             vfitab           = _mm_slli_epi32(vfitab,3);
 897
 898             /* CUBIC SPLINE TABLE DISPERSION */
 899             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 900             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 901             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 902             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 903             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 904             Heps             = _mm256_mul_pd(vfeps,H);
 905             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 906             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 907             fvdw6            = _mm256_mul_pd(c6_00,FF);
 908
 909             /* CUBIC SPLINE TABLE REPULSION */
 910             vfitab           = _mm_add_epi32(vfitab,ifour);
 911             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 912             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 913             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 914             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 915             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 916             Heps             = _mm256_mul_pd(vfeps,H);
 917             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 918             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 919             fvdw12           = _mm256_mul_pd(c12_00,FF);
 920             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 921
 922             fscal            = fvdw;
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm256_mul_pd(fscal,dx00);
 926             ty               = _mm256_mul_pd(fscal,dy00);
 927             tz               = _mm256_mul_pd(fscal,dz00);
 928
 929             /* Update vectorial force */
 930             fix0             = _mm256_add_pd(fix0,tx);
 931             fiy0             = _mm256_add_pd(fiy0,ty);
 932             fiz0             = _mm256_add_pd(fiz0,tz);
 933
 934             fjx0             = _mm256_add_pd(fjx0,tx);
 935             fjy0             = _mm256_add_pd(fjy0,ty);
 936             fjz0             = _mm256_add_pd(fjz0,tz);
 937
 938             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             /* Compute parameters for interactions between i and j atoms */
 943             qq10             = _mm256_mul_pd(iq1,jq0);
 944
 945             /* COULOMB ELECTROSTATICS */
 946             velec            = _mm256_mul_pd(qq10,rinv10);
 947             felec            = _mm256_mul_pd(velec,rinvsq10);
 948
 949             fscal            = felec;
 950
 951             /* Calculate temporary vectorial force */
 952             tx               = _mm256_mul_pd(fscal,dx10);
 953             ty               = _mm256_mul_pd(fscal,dy10);
 954             tz               = _mm256_mul_pd(fscal,dz10);
 955
 956             /* Update vectorial force */
 957             fix1             = _mm256_add_pd(fix1,tx);
 958             fiy1             = _mm256_add_pd(fiy1,ty);
 959             fiz1             = _mm256_add_pd(fiz1,tz);
 960
 961             fjx0             = _mm256_add_pd(fjx0,tx);
 962             fjy0             = _mm256_add_pd(fjy0,ty);
 963             fjz0             = _mm256_add_pd(fjz0,tz);
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq20             = _mm256_mul_pd(iq2,jq0);
 971
 972             /* COULOMB ELECTROSTATICS */
 973             velec            = _mm256_mul_pd(qq20,rinv20);
 974             felec            = _mm256_mul_pd(velec,rinvsq20);
 975
 976             fscal            = felec;
 977
 978             /* Calculate temporary vectorial force */
 979             tx               = _mm256_mul_pd(fscal,dx20);
 980             ty               = _mm256_mul_pd(fscal,dy20);
 981             tz               = _mm256_mul_pd(fscal,dz20);
 982
 983             /* Update vectorial force */
 984             fix2             = _mm256_add_pd(fix2,tx);
 985             fiy2             = _mm256_add_pd(fiy2,ty);
 986             fiz2             = _mm256_add_pd(fiz2,tz);
 987
 988             fjx0             = _mm256_add_pd(fjx0,tx);
 989             fjy0             = _mm256_add_pd(fjy0,ty);
 990             fjz0             = _mm256_add_pd(fjz0,tz);
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             qq30             = _mm256_mul_pd(iq3,jq0);
 998
 999             /* COULOMB ELECTROSTATICS */
1000             velec            = _mm256_mul_pd(qq30,rinv30);
1001             felec            = _mm256_mul_pd(velec,rinvsq30);
1002
1003             fscal            = felec;
1004
1005             /* Calculate temporary vectorial force */
1006             tx               = _mm256_mul_pd(fscal,dx30);
1007             ty               = _mm256_mul_pd(fscal,dy30);
1008             tz               = _mm256_mul_pd(fscal,dz30);
1009
1010             /* Update vectorial force */
1011             fix3             = _mm256_add_pd(fix3,tx);
1012             fiy3             = _mm256_add_pd(fiy3,ty);
1013             fiz3             = _mm256_add_pd(fiz3,tz);
1014
1015             fjx0             = _mm256_add_pd(fjx0,tx);
1016             fjy0             = _mm256_add_pd(fjy0,ty);
1017             fjz0             = _mm256_add_pd(fjz0,tz);
1018
1019             fjptrA             = f+j_coord_offsetA;
1020             fjptrB             = f+j_coord_offsetB;
1021             fjptrC             = f+j_coord_offsetC;
1022             fjptrD             = f+j_coord_offsetD;
1023
1024             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1025
1026             /* Inner loop uses 129 flops */
1027         }
1028
1029         if(jidx<j_index_end)
1030         {
1031
1032             /* Get j neighbor index, and coordinate index */
1033             jnrlistA         = jjnr[jidx];
1034             jnrlistB         = jjnr[jidx+1];
1035             jnrlistC         = jjnr[jidx+2];
1036             jnrlistD         = jjnr[jidx+3];
1037             /* Sign of each element will be negative for non-real atoms.
1038              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1039              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1040              */
1041             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1042
1043             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1044             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1045             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1046
1047             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1048             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1049             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1050             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1051             j_coord_offsetA  = DIM*jnrA;
1052             j_coord_offsetB  = DIM*jnrB;
1053             j_coord_offsetC  = DIM*jnrC;
1054             j_coord_offsetD  = DIM*jnrD;
1055
1056             /* load j atom coordinates */
1057             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1058                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1059                                                  &jx0,&jy0,&jz0);
1060
1061             /* Calculate displacement vector */
1062             dx00             = _mm256_sub_pd(ix0,jx0);
1063             dy00             = _mm256_sub_pd(iy0,jy0);
1064             dz00             = _mm256_sub_pd(iz0,jz0);
1065             dx10             = _mm256_sub_pd(ix1,jx0);
1066             dy10             = _mm256_sub_pd(iy1,jy0);
1067             dz10             = _mm256_sub_pd(iz1,jz0);
1068             dx20             = _mm256_sub_pd(ix2,jx0);
1069             dy20             = _mm256_sub_pd(iy2,jy0);
1070             dz20             = _mm256_sub_pd(iz2,jz0);
1071             dx30             = _mm256_sub_pd(ix3,jx0);
1072             dy30             = _mm256_sub_pd(iy3,jy0);
1073             dz30             = _mm256_sub_pd(iz3,jz0);
1074
1075             /* Calculate squared distance and things based on it */
1076             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1077             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1078             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1079             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1080
1081             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1082             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1083             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1084             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1085
1086             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1087             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1088             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1089
1090             /* Load parameters for j particles */
1091             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1092                                                                  charge+jnrC+0,charge+jnrD+0);
1093             vdwjidx0A        = 2*vdwtype[jnrA+0];
1094             vdwjidx0B        = 2*vdwtype[jnrB+0];
1095             vdwjidx0C        = 2*vdwtype[jnrC+0];
1096             vdwjidx0D        = 2*vdwtype[jnrD+0];
1097
1098             fjx0             = _mm256_setzero_pd();
1099             fjy0             = _mm256_setzero_pd();
1100             fjz0             = _mm256_setzero_pd();
1101
1102             /**************************
1103              * CALCULATE INTERACTIONS *
1104              **************************/
1105
1106             r00              = _mm256_mul_pd(rsq00,rinv00);
1107             r00              = _mm256_andnot_pd(dummy_mask,r00);
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1111                                             vdwioffsetptr0+vdwjidx0B,
1112                                             vdwioffsetptr0+vdwjidx0C,
1113                                             vdwioffsetptr0+vdwjidx0D,
1114                                             &c6_00,&c12_00);
1115
1116             /* Calculate table index by multiplying r with table scale and truncate to integer */
1117             rt               = _mm256_mul_pd(r00,vftabscale);
1118             vfitab           = _mm256_cvttpd_epi32(rt);
1119             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1120             vfitab           = _mm_slli_epi32(vfitab,3);
1121
1122             /* CUBIC SPLINE TABLE DISPERSION */
1123             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1124             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1125             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1126             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1127             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1128             Heps             = _mm256_mul_pd(vfeps,H);
1129             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1130             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1131             fvdw6            = _mm256_mul_pd(c6_00,FF);
1132
1133             /* CUBIC SPLINE TABLE REPULSION */
1134             vfitab           = _mm_add_epi32(vfitab,ifour);
1135             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1136             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1137             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1138             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1139             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1140             Heps             = _mm256_mul_pd(vfeps,H);
1141             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1142             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1143             fvdw12           = _mm256_mul_pd(c12_00,FF);
1144             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1145
1146             fscal            = fvdw;
1147
1148             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm256_mul_pd(fscal,dx00);
1152             ty               = _mm256_mul_pd(fscal,dy00);
1153             tz               = _mm256_mul_pd(fscal,dz00);
1154
1155             /* Update vectorial force */
1156             fix0             = _mm256_add_pd(fix0,tx);
1157             fiy0             = _mm256_add_pd(fiy0,ty);
1158             fiz0             = _mm256_add_pd(fiz0,tz);
1159
1160             fjx0             = _mm256_add_pd(fjx0,tx);
1161             fjy0             = _mm256_add_pd(fjy0,ty);
1162             fjz0             = _mm256_add_pd(fjz0,tz);
1163
1164             /**************************
1165              * CALCULATE INTERACTIONS *
1166              **************************/
1167
1168             /* Compute parameters for interactions between i and j atoms */
1169             qq10             = _mm256_mul_pd(iq1,jq0);
1170
1171             /* COULOMB ELECTROSTATICS */
1172             velec            = _mm256_mul_pd(qq10,rinv10);
1173             felec            = _mm256_mul_pd(velec,rinvsq10);
1174
1175             fscal            = felec;
1176
1177             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1178
1179             /* Calculate temporary vectorial force */
1180             tx               = _mm256_mul_pd(fscal,dx10);
1181             ty               = _mm256_mul_pd(fscal,dy10);
1182             tz               = _mm256_mul_pd(fscal,dz10);
1183
1184             /* Update vectorial force */
1185             fix1             = _mm256_add_pd(fix1,tx);
1186             fiy1             = _mm256_add_pd(fiy1,ty);
1187             fiz1             = _mm256_add_pd(fiz1,tz);
1188
1189             fjx0             = _mm256_add_pd(fjx0,tx);
1190             fjy0             = _mm256_add_pd(fjy0,ty);
1191             fjz0             = _mm256_add_pd(fjz0,tz);
1192
1193             /**************************
1194              * CALCULATE INTERACTIONS *
1195              **************************/
1196
1197             /* Compute parameters for interactions between i and j atoms */
1198             qq20             = _mm256_mul_pd(iq2,jq0);
1199
1200             /* COULOMB ELECTROSTATICS */
1201             velec            = _mm256_mul_pd(qq20,rinv20);
1202             felec            = _mm256_mul_pd(velec,rinvsq20);
1203
1204             fscal            = felec;
1205
1206             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1207
1208             /* Calculate temporary vectorial force */
1209             tx               = _mm256_mul_pd(fscal,dx20);
1210             ty               = _mm256_mul_pd(fscal,dy20);
1211             tz               = _mm256_mul_pd(fscal,dz20);
1212
1213             /* Update vectorial force */
1214             fix2             = _mm256_add_pd(fix2,tx);
1215             fiy2             = _mm256_add_pd(fiy2,ty);
1216             fiz2             = _mm256_add_pd(fiz2,tz);
1217
1218             fjx0             = _mm256_add_pd(fjx0,tx);
1219             fjy0             = _mm256_add_pd(fjy0,ty);
1220             fjz0             = _mm256_add_pd(fjz0,tz);
1221
1222             /**************************
1223              * CALCULATE INTERACTIONS *
1224              **************************/
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq30             = _mm256_mul_pd(iq3,jq0);
1228
1229             /* COULOMB ELECTROSTATICS */
1230             velec            = _mm256_mul_pd(qq30,rinv30);
1231             felec            = _mm256_mul_pd(velec,rinvsq30);
1232
1233             fscal            = felec;
1234
1235             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1236
1237             /* Calculate temporary vectorial force */
1238             tx               = _mm256_mul_pd(fscal,dx30);
1239             ty               = _mm256_mul_pd(fscal,dy30);
1240             tz               = _mm256_mul_pd(fscal,dz30);
1241
1242             /* Update vectorial force */
1243             fix3             = _mm256_add_pd(fix3,tx);
1244             fiy3             = _mm256_add_pd(fiy3,ty);
1245             fiz3             = _mm256_add_pd(fiz3,tz);
1246
1247             fjx0             = _mm256_add_pd(fjx0,tx);
1248             fjy0             = _mm256_add_pd(fjy0,ty);
1249             fjz0             = _mm256_add_pd(fjz0,tz);
1250
1251             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1252             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1253             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1254             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1255
1256             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1257
1258             /* Inner loop uses 130 flops */
1259         }
1260
1261         /* End of innermost loop */
1262
1263         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1264                                                  f+i_coord_offset,fshift+i_shift_offset);
1265
1266         /* Increment number of inner iterations */
1267         inneriter                  += j_index_end - j_index_start;
1268
1269         /* Outer loop uses 24 flops */
1270     }
1271
1272     /* Increment number of outer iterations */
1273     outeriter        += nri;
1274
1275     /* Update outer/inner flops */
1276
1277     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1278 }