src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJSh_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_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_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          ewitab;
  93     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  95     real             *ewtab;
  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     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 119     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 120     beta2            = _mm256_mul_pd(beta,beta);
 121     beta3            = _mm256_mul_pd(beta,beta2);
 122
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 130     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 131     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 132     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->rcoulomb;
 136     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 137     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 138
 139     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 140     rvdw             = _mm256_set1_pd(fr->rvdw);
 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             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 232             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 233             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 234
 235             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 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             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 257             {
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 261                                             vdwioffsetptr0+vdwjidx0B,
 262                                             vdwioffsetptr0+vdwjidx0C,
 263                                             vdwioffsetptr0+vdwjidx0D,
 264                                             &c6_00,&c12_00);
 265
 266             /* LENNARD-JONES DISPERSION/REPULSION */
 267
 268             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 269             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 270             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 271             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 272                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 273             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 274
 275             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 279             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 280
 281             fscal            = fvdw;
 282
 283             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm256_mul_pd(fscal,dx00);
 287             ty               = _mm256_mul_pd(fscal,dy00);
 288             tz               = _mm256_mul_pd(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm256_add_pd(fix0,tx);
 292             fiy0             = _mm256_add_pd(fiy0,ty);
 293             fiz0             = _mm256_add_pd(fiz0,tz);
 294
 295             fjx0             = _mm256_add_pd(fjx0,tx);
 296             fjy0             = _mm256_add_pd(fjy0,ty);
 297             fjz0             = _mm256_add_pd(fjz0,tz);
 298
 299             }
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 306             {
 307
 308             r10              = _mm256_mul_pd(rsq10,rinv10);
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm256_mul_pd(iq1,jq0);
 312
 313             /* EWALD ELECTROSTATICS */
 314
 315             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 316             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 317             ewitab           = _mm256_cvttpd_epi32(ewrt);
 318             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 319             ewitab           = _mm_slli_epi32(ewitab,2);
 320             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 321             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 322             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 323             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 324             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 325             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 326             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 327             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 328             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 329
 330             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm256_and_pd(velec,cutoff_mask);
 334             velecsum         = _mm256_add_pd(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm256_mul_pd(fscal,dx10);
 342             ty               = _mm256_mul_pd(fscal,dy10);
 343             tz               = _mm256_mul_pd(fscal,dz10);
 344
 345             /* Update vectorial force */
 346             fix1             = _mm256_add_pd(fix1,tx);
 347             fiy1             = _mm256_add_pd(fiy1,ty);
 348             fiz1             = _mm256_add_pd(fiz1,tz);
 349
 350             fjx0             = _mm256_add_pd(fjx0,tx);
 351             fjy0             = _mm256_add_pd(fjy0,ty);
 352             fjz0             = _mm256_add_pd(fjz0,tz);
 353
 354             }
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 361             {
 362
 363             r20              = _mm256_mul_pd(rsq20,rinv20);
 364
 365             /* Compute parameters for interactions between i and j atoms */
 366             qq20             = _mm256_mul_pd(iq2,jq0);
 367
 368             /* EWALD ELECTROSTATICS */
 369
 370             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 371             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 372             ewitab           = _mm256_cvttpd_epi32(ewrt);
 373             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 374             ewitab           = _mm_slli_epi32(ewitab,2);
 375             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 376             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 377             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 378             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 379             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 380             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 381             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 382             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 383             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 384
 385             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm256_and_pd(velec,cutoff_mask);
 389             velecsum         = _mm256_add_pd(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm256_mul_pd(fscal,dx20);
 397             ty               = _mm256_mul_pd(fscal,dy20);
 398             tz               = _mm256_mul_pd(fscal,dz20);
 399
 400             /* Update vectorial force */
 401             fix2             = _mm256_add_pd(fix2,tx);
 402             fiy2             = _mm256_add_pd(fiy2,ty);
 403             fiz2             = _mm256_add_pd(fiz2,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             }
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 416             {
 417
 418             r30              = _mm256_mul_pd(rsq30,rinv30);
 419
 420             /* Compute parameters for interactions between i and j atoms */
 421             qq30             = _mm256_mul_pd(iq3,jq0);
 422
 423             /* EWALD ELECTROSTATICS */
 424
 425             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 426             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 427             ewitab           = _mm256_cvttpd_epi32(ewrt);
 428             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 429             ewitab           = _mm_slli_epi32(ewitab,2);
 430             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 431             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 432             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 433             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 434             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 435             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 436             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 437             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 438             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 439
 440             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velec            = _mm256_and_pd(velec,cutoff_mask);
 444             velecsum         = _mm256_add_pd(velecsum,velec);
 445
 446             fscal            = felec;
 447
 448             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 449
 450             /* Calculate temporary vectorial force */
 451             tx               = _mm256_mul_pd(fscal,dx30);
 452             ty               = _mm256_mul_pd(fscal,dy30);
 453             tz               = _mm256_mul_pd(fscal,dz30);
 454
 455             /* Update vectorial force */
 456             fix3             = _mm256_add_pd(fix3,tx);
 457             fiy3             = _mm256_add_pd(fiy3,ty);
 458             fiz3             = _mm256_add_pd(fiz3,tz);
 459
 460             fjx0             = _mm256_add_pd(fjx0,tx);
 461             fjy0             = _mm256_add_pd(fjy0,ty);
 462             fjz0             = _mm256_add_pd(fjz0,tz);
 463
 464             }
 465
 466             fjptrA             = f+j_coord_offsetA;
 467             fjptrB             = f+j_coord_offsetB;
 468             fjptrC             = f+j_coord_offsetC;
 469             fjptrD             = f+j_coord_offsetD;
 470
 471             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 472
 473             /* Inner loop uses 182 flops */
 474         }
 475
 476         if(jidx<j_index_end)
 477         {
 478
 479             /* Get j neighbor index, and coordinate index */
 480             jnrlistA         = jjnr[jidx];
 481             jnrlistB         = jjnr[jidx+1];
 482             jnrlistC         = jjnr[jidx+2];
 483             jnrlistD         = jjnr[jidx+3];
 484             /* Sign of each element will be negative for non-real atoms.
 485              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 486              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 487              */
 488             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 489
 490             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 491             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 492             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 493
 494             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 495             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 496             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 497             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 498             j_coord_offsetA  = DIM*jnrA;
 499             j_coord_offsetB  = DIM*jnrB;
 500             j_coord_offsetC  = DIM*jnrC;
 501             j_coord_offsetD  = DIM*jnrD;
 502
 503             /* load j atom coordinates */
 504             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 505                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 506                                                  &jx0,&jy0,&jz0);
 507
 508             /* Calculate displacement vector */
 509             dx00             = _mm256_sub_pd(ix0,jx0);
 510             dy00             = _mm256_sub_pd(iy0,jy0);
 511             dz00             = _mm256_sub_pd(iz0,jz0);
 512             dx10             = _mm256_sub_pd(ix1,jx0);
 513             dy10             = _mm256_sub_pd(iy1,jy0);
 514             dz10             = _mm256_sub_pd(iz1,jz0);
 515             dx20             = _mm256_sub_pd(ix2,jx0);
 516             dy20             = _mm256_sub_pd(iy2,jy0);
 517             dz20             = _mm256_sub_pd(iz2,jz0);
 518             dx30             = _mm256_sub_pd(ix3,jx0);
 519             dy30             = _mm256_sub_pd(iy3,jy0);
 520             dz30             = _mm256_sub_pd(iz3,jz0);
 521
 522             /* Calculate squared distance and things based on it */
 523             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 524             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 525             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 526             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 527
 528             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 529             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 530             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 531
 532             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 533             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 534             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 535             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 536
 537             /* Load parameters for j particles */
 538             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 539                                                                  charge+jnrC+0,charge+jnrD+0);
 540             vdwjidx0A        = 2*vdwtype[jnrA+0];
 541             vdwjidx0B        = 2*vdwtype[jnrB+0];
 542             vdwjidx0C        = 2*vdwtype[jnrC+0];
 543             vdwjidx0D        = 2*vdwtype[jnrD+0];
 544
 545             fjx0             = _mm256_setzero_pd();
 546             fjy0             = _mm256_setzero_pd();
 547             fjz0             = _mm256_setzero_pd();
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 554             {
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 558                                             vdwioffsetptr0+vdwjidx0B,
 559                                             vdwioffsetptr0+vdwjidx0C,
 560                                             vdwioffsetptr0+vdwjidx0D,
 561                                             &c6_00,&c12_00);
 562
 563             /* LENNARD-JONES DISPERSION/REPULSION */
 564
 565             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 566             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 567             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 568             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 569                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 570             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 571
 572             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 576             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 577             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 578
 579             fscal            = fvdw;
 580
 581             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 582
 583             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_pd(fscal,dx00);
 587             ty               = _mm256_mul_pd(fscal,dy00);
 588             tz               = _mm256_mul_pd(fscal,dz00);
 589
 590             /* Update vectorial force */
 591             fix0             = _mm256_add_pd(fix0,tx);
 592             fiy0             = _mm256_add_pd(fiy0,ty);
 593             fiz0             = _mm256_add_pd(fiz0,tz);
 594
 595             fjx0             = _mm256_add_pd(fjx0,tx);
 596             fjy0             = _mm256_add_pd(fjy0,ty);
 597             fjz0             = _mm256_add_pd(fjz0,tz);
 598
 599             }
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 606             {
 607
 608             r10              = _mm256_mul_pd(rsq10,rinv10);
 609             r10              = _mm256_andnot_pd(dummy_mask,r10);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq10             = _mm256_mul_pd(iq1,jq0);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 617             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 618             ewitab           = _mm256_cvttpd_epi32(ewrt);
 619             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 620             ewitab           = _mm_slli_epi32(ewitab,2);
 621             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 622             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 623             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 624             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 625             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 626             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 627             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 628             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 629             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 630
 631             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 632
 633             /* Update potential sum for this i atom from the interaction with this j atom. */
 634             velec            = _mm256_and_pd(velec,cutoff_mask);
 635             velec            = _mm256_andnot_pd(dummy_mask,velec);
 636             velecsum         = _mm256_add_pd(velecsum,velec);
 637
 638             fscal            = felec;
 639
 640             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 641
 642             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 643
 644             /* Calculate temporary vectorial force */
 645             tx               = _mm256_mul_pd(fscal,dx10);
 646             ty               = _mm256_mul_pd(fscal,dy10);
 647             tz               = _mm256_mul_pd(fscal,dz10);
 648
 649             /* Update vectorial force */
 650             fix1             = _mm256_add_pd(fix1,tx);
 651             fiy1             = _mm256_add_pd(fiy1,ty);
 652             fiz1             = _mm256_add_pd(fiz1,tz);
 653
 654             fjx0             = _mm256_add_pd(fjx0,tx);
 655             fjy0             = _mm256_add_pd(fjy0,ty);
 656             fjz0             = _mm256_add_pd(fjz0,tz);
 657
 658             }
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 665             {
 666
 667             r20              = _mm256_mul_pd(rsq20,rinv20);
 668             r20              = _mm256_andnot_pd(dummy_mask,r20);
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq20             = _mm256_mul_pd(iq2,jq0);
 672
 673             /* EWALD ELECTROSTATICS */
 674
 675             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 676             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 677             ewitab           = _mm256_cvttpd_epi32(ewrt);
 678             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 679             ewitab           = _mm_slli_epi32(ewitab,2);
 680             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 681             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 682             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 683             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 684             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 685             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 686             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 687             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 688             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 689
 690             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 691
 692             /* Update potential sum for this i atom from the interaction with this j atom. */
 693             velec            = _mm256_and_pd(velec,cutoff_mask);
 694             velec            = _mm256_andnot_pd(dummy_mask,velec);
 695             velecsum         = _mm256_add_pd(velecsum,velec);
 696
 697             fscal            = felec;
 698
 699             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 700
 701             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 702
 703             /* Calculate temporary vectorial force */
 704             tx               = _mm256_mul_pd(fscal,dx20);
 705             ty               = _mm256_mul_pd(fscal,dy20);
 706             tz               = _mm256_mul_pd(fscal,dz20);
 707
 708             /* Update vectorial force */
 709             fix2             = _mm256_add_pd(fix2,tx);
 710             fiy2             = _mm256_add_pd(fiy2,ty);
 711             fiz2             = _mm256_add_pd(fiz2,tz);
 712
 713             fjx0             = _mm256_add_pd(fjx0,tx);
 714             fjy0             = _mm256_add_pd(fjy0,ty);
 715             fjz0             = _mm256_add_pd(fjz0,tz);
 716
 717             }
 718
 719             /**************************
 720              * CALCULATE INTERACTIONS *
 721              **************************/
 722
 723             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 724             {
 725
 726             r30              = _mm256_mul_pd(rsq30,rinv30);
 727             r30              = _mm256_andnot_pd(dummy_mask,r30);
 728
 729             /* Compute parameters for interactions between i and j atoms */
 730             qq30             = _mm256_mul_pd(iq3,jq0);
 731
 732             /* EWALD ELECTROSTATICS */
 733
 734             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 735             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 736             ewitab           = _mm256_cvttpd_epi32(ewrt);
 737             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 738             ewitab           = _mm_slli_epi32(ewitab,2);
 739             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 740             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 741             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 742             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 743             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 744             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 745             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 746             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 747             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 748
 749             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 750
 751             /* Update potential sum for this i atom from the interaction with this j atom. */
 752             velec            = _mm256_and_pd(velec,cutoff_mask);
 753             velec            = _mm256_andnot_pd(dummy_mask,velec);
 754             velecsum         = _mm256_add_pd(velecsum,velec);
 755
 756             fscal            = felec;
 757
 758             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 759
 760             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 761
 762             /* Calculate temporary vectorial force */
 763             tx               = _mm256_mul_pd(fscal,dx30);
 764             ty               = _mm256_mul_pd(fscal,dy30);
 765             tz               = _mm256_mul_pd(fscal,dz30);
 766
 767             /* Update vectorial force */
 768             fix3             = _mm256_add_pd(fix3,tx);
 769             fiy3             = _mm256_add_pd(fiy3,ty);
 770             fiz3             = _mm256_add_pd(fiz3,tz);
 771
 772             fjx0             = _mm256_add_pd(fjx0,tx);
 773             fjy0             = _mm256_add_pd(fjy0,ty);
 774             fjz0             = _mm256_add_pd(fjz0,tz);
 775
 776             }
 777
 778             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 779             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 780             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 781             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 782
 783             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 784
 785             /* Inner loop uses 185 flops */
 786         }
 787
 788         /* End of innermost loop */
 789
 790         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 791                                                  f+i_coord_offset,fshift+i_shift_offset);
 792
 793         ggid                        = gid[iidx];
 794         /* Update potential energies */
 795         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 796         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 797
 798         /* Increment number of inner iterations */
 799         inneriter                  += j_index_end - j_index_start;
 800
 801         /* Outer loop uses 26 flops */
 802     }
 803
 804     /* Increment number of outer iterations */
 805     outeriter        += nri;
 806
 807     /* Update outer/inner flops */
 808
 809     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*185);
 810 }
 811 /*
 812  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_double
 813  * Electrostatics interaction: Ewald
 814  * VdW interaction:            LennardJones
 815  * Geometry:                   Water4-Particle
 816  * Calculate force/pot:        Force
 817  */
 818 void
 819 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_double
 820                     (t_nblist * gmx_restrict                nlist,
 821                      rvec * gmx_restrict                    xx,
 822                      rvec * gmx_restrict                    ff,
 823                      t_forcerec * gmx_restrict              fr,
 824                      t_mdatoms * gmx_restrict               mdatoms,
 825                      nb_kernel_data_t * gmx_restrict        kernel_data,
 826                      t_nrnb * gmx_restrict                  nrnb)
 827 {
 828     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 829      * just 0 for non-waters.
 830      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 831      * jnr indices corresponding to data put in the four positions in the SIMD register.
 832      */
 833     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 834     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 835     int              jnrA,jnrB,jnrC,jnrD;
 836     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 837     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 838     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 839     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 840     real             rcutoff_scalar;
 841     real             *shiftvec,*fshift,*x,*f;
 842     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 843     real             scratch[4*DIM];
 844     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 845     real *           vdwioffsetptr0;
 846     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 847     real *           vdwioffsetptr1;
 848     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 849     real *           vdwioffsetptr2;
 850     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 851     real *           vdwioffsetptr3;
 852     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 853     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 854     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 855     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 856     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 857     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 858     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 859     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 860     real             *charge;
 861     int              nvdwtype;
 862     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 863     int              *vdwtype;
 864     real             *vdwparam;
 865     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 866     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 867     __m128i          ewitab;
 868     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 869     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 870     real             *ewtab;
 871     __m256d          dummy_mask,cutoff_mask;
 872     __m128           tmpmask0,tmpmask1;
 873     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 874     __m256d          one     = _mm256_set1_pd(1.0);
 875     __m256d          two     = _mm256_set1_pd(2.0);
 876     x                = xx[0];
 877     f                = ff[0];
 878
 879     nri              = nlist->nri;
 880     iinr             = nlist->iinr;
 881     jindex           = nlist->jindex;
 882     jjnr             = nlist->jjnr;
 883     shiftidx         = nlist->shift;
 884     gid              = nlist->gid;
 885     shiftvec         = fr->shift_vec[0];
 886     fshift           = fr->fshift[0];
 887     facel            = _mm256_set1_pd(fr->epsfac);
 888     charge           = mdatoms->chargeA;
 889     nvdwtype         = fr->ntype;
 890     vdwparam         = fr->nbfp;
 891     vdwtype          = mdatoms->typeA;
 892
 893     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 894     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 895     beta2            = _mm256_mul_pd(beta,beta);
 896     beta3            = _mm256_mul_pd(beta,beta2);
 897
 898     ewtab            = fr->ic->tabq_coul_F;
 899     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 900     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 901
 902     /* Setup water-specific parameters */
 903     inr              = nlist->iinr[0];
 904     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 905     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 906     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 907     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 908
 909     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 910     rcutoff_scalar   = fr->rcoulomb;
 911     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 912     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 913
 914     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 915     rvdw             = _mm256_set1_pd(fr->rvdw);
 916
 917     /* Avoid stupid compiler warnings */
 918     jnrA = jnrB = jnrC = jnrD = 0;
 919     j_coord_offsetA = 0;
 920     j_coord_offsetB = 0;
 921     j_coord_offsetC = 0;
 922     j_coord_offsetD = 0;
 923
 924     outeriter        = 0;
 925     inneriter        = 0;
 926
 927     for(iidx=0;iidx<4*DIM;iidx++)
 928     {
 929         scratch[iidx] = 0.0;
 930     }
 931
 932     /* Start outer loop over neighborlists */
 933     for(iidx=0; iidx<nri; iidx++)
 934     {
 935         /* Load shift vector for this list */
 936         i_shift_offset   = DIM*shiftidx[iidx];
 937
 938         /* Load limits for loop over neighbors */
 939         j_index_start    = jindex[iidx];
 940         j_index_end      = jindex[iidx+1];
 941
 942         /* Get outer coordinate index */
 943         inr              = iinr[iidx];
 944         i_coord_offset   = DIM*inr;
 945
 946         /* Load i particle coords and add shift vector */
 947         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 948                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 949
 950         fix0             = _mm256_setzero_pd();
 951         fiy0             = _mm256_setzero_pd();
 952         fiz0             = _mm256_setzero_pd();
 953         fix1             = _mm256_setzero_pd();
 954         fiy1             = _mm256_setzero_pd();
 955         fiz1             = _mm256_setzero_pd();
 956         fix2             = _mm256_setzero_pd();
 957         fiy2             = _mm256_setzero_pd();
 958         fiz2             = _mm256_setzero_pd();
 959         fix3             = _mm256_setzero_pd();
 960         fiy3             = _mm256_setzero_pd();
 961         fiz3             = _mm256_setzero_pd();
 962
 963         /* Start inner kernel loop */
 964         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 965         {
 966
 967             /* Get j neighbor index, and coordinate index */
 968             jnrA             = jjnr[jidx];
 969             jnrB             = jjnr[jidx+1];
 970             jnrC             = jjnr[jidx+2];
 971             jnrD             = jjnr[jidx+3];
 972             j_coord_offsetA  = DIM*jnrA;
 973             j_coord_offsetB  = DIM*jnrB;
 974             j_coord_offsetC  = DIM*jnrC;
 975             j_coord_offsetD  = DIM*jnrD;
 976
 977             /* load j atom coordinates */
 978             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 979                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 980                                                  &jx0,&jy0,&jz0);
 981
 982             /* Calculate displacement vector */
 983             dx00             = _mm256_sub_pd(ix0,jx0);
 984             dy00             = _mm256_sub_pd(iy0,jy0);
 985             dz00             = _mm256_sub_pd(iz0,jz0);
 986             dx10             = _mm256_sub_pd(ix1,jx0);
 987             dy10             = _mm256_sub_pd(iy1,jy0);
 988             dz10             = _mm256_sub_pd(iz1,jz0);
 989             dx20             = _mm256_sub_pd(ix2,jx0);
 990             dy20             = _mm256_sub_pd(iy2,jy0);
 991             dz20             = _mm256_sub_pd(iz2,jz0);
 992             dx30             = _mm256_sub_pd(ix3,jx0);
 993             dy30             = _mm256_sub_pd(iy3,jy0);
 994             dz30             = _mm256_sub_pd(iz3,jz0);
 995
 996             /* Calculate squared distance and things based on it */
 997             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 998             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 999             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1000             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1001
1002             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1003             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1004             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1005
1006             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1007             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1008             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1009             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1010
1011             /* Load parameters for j particles */
1012             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1013                                                                  charge+jnrC+0,charge+jnrD+0);
1014             vdwjidx0A        = 2*vdwtype[jnrA+0];
1015             vdwjidx0B        = 2*vdwtype[jnrB+0];
1016             vdwjidx0C        = 2*vdwtype[jnrC+0];
1017             vdwjidx0D        = 2*vdwtype[jnrD+0];
1018
1019             fjx0             = _mm256_setzero_pd();
1020             fjy0             = _mm256_setzero_pd();
1021             fjz0             = _mm256_setzero_pd();
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1028             {
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1032                                             vdwioffsetptr0+vdwjidx0B,
1033                                             vdwioffsetptr0+vdwjidx0C,
1034                                             vdwioffsetptr0+vdwjidx0D,
1035                                             &c6_00,&c12_00);
1036
1037             /* LENNARD-JONES DISPERSION/REPULSION */
1038
1039             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1040             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1041
1042             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1043
1044             fscal            = fvdw;
1045
1046             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm256_mul_pd(fscal,dx00);
1050             ty               = _mm256_mul_pd(fscal,dy00);
1051             tz               = _mm256_mul_pd(fscal,dz00);
1052
1053             /* Update vectorial force */
1054             fix0             = _mm256_add_pd(fix0,tx);
1055             fiy0             = _mm256_add_pd(fiy0,ty);
1056             fiz0             = _mm256_add_pd(fiz0,tz);
1057
1058             fjx0             = _mm256_add_pd(fjx0,tx);
1059             fjy0             = _mm256_add_pd(fjy0,ty);
1060             fjz0             = _mm256_add_pd(fjz0,tz);
1061
1062             }
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1069             {
1070
1071             r10              = _mm256_mul_pd(rsq10,rinv10);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq10             = _mm256_mul_pd(iq1,jq0);
1075
1076             /* EWALD ELECTROSTATICS */
1077
1078             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1079             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1080             ewitab           = _mm256_cvttpd_epi32(ewrt);
1081             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1082             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1083                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1084                                             &ewtabF,&ewtabFn);
1085             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1086             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1087
1088             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm256_mul_pd(fscal,dx10);
1096             ty               = _mm256_mul_pd(fscal,dy10);
1097             tz               = _mm256_mul_pd(fscal,dz10);
1098
1099             /* Update vectorial force */
1100             fix1             = _mm256_add_pd(fix1,tx);
1101             fiy1             = _mm256_add_pd(fiy1,ty);
1102             fiz1             = _mm256_add_pd(fiz1,tz);
1103
1104             fjx0             = _mm256_add_pd(fjx0,tx);
1105             fjy0             = _mm256_add_pd(fjy0,ty);
1106             fjz0             = _mm256_add_pd(fjz0,tz);
1107
1108             }
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1115             {
1116
1117             r20              = _mm256_mul_pd(rsq20,rinv20);
1118
1119             /* Compute parameters for interactions between i and j atoms */
1120             qq20             = _mm256_mul_pd(iq2,jq0);
1121
1122             /* EWALD ELECTROSTATICS */
1123
1124             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1125             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1126             ewitab           = _mm256_cvttpd_epi32(ewrt);
1127             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1128             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1129                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1130                                             &ewtabF,&ewtabFn);
1131             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1132             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1133
1134             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1135
1136             fscal            = felec;
1137
1138             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm256_mul_pd(fscal,dx20);
1142             ty               = _mm256_mul_pd(fscal,dy20);
1143             tz               = _mm256_mul_pd(fscal,dz20);
1144
1145             /* Update vectorial force */
1146             fix2             = _mm256_add_pd(fix2,tx);
1147             fiy2             = _mm256_add_pd(fiy2,ty);
1148             fiz2             = _mm256_add_pd(fiz2,tz);
1149
1150             fjx0             = _mm256_add_pd(fjx0,tx);
1151             fjy0             = _mm256_add_pd(fjy0,ty);
1152             fjz0             = _mm256_add_pd(fjz0,tz);
1153
1154             }
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1161             {
1162
1163             r30              = _mm256_mul_pd(rsq30,rinv30);
1164
1165             /* Compute parameters for interactions between i and j atoms */
1166             qq30             = _mm256_mul_pd(iq3,jq0);
1167
1168             /* EWALD ELECTROSTATICS */
1169
1170             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1171             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1172             ewitab           = _mm256_cvttpd_epi32(ewrt);
1173             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1174             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1175                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1176                                             &ewtabF,&ewtabFn);
1177             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1178             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1179
1180             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1181
1182             fscal            = felec;
1183
1184             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1185
1186             /* Calculate temporary vectorial force */
1187             tx               = _mm256_mul_pd(fscal,dx30);
1188             ty               = _mm256_mul_pd(fscal,dy30);
1189             tz               = _mm256_mul_pd(fscal,dz30);
1190
1191             /* Update vectorial force */
1192             fix3             = _mm256_add_pd(fix3,tx);
1193             fiy3             = _mm256_add_pd(fiy3,ty);
1194             fiz3             = _mm256_add_pd(fiz3,tz);
1195
1196             fjx0             = _mm256_add_pd(fjx0,tx);
1197             fjy0             = _mm256_add_pd(fjy0,ty);
1198             fjz0             = _mm256_add_pd(fjz0,tz);
1199
1200             }
1201
1202             fjptrA             = f+j_coord_offsetA;
1203             fjptrB             = f+j_coord_offsetB;
1204             fjptrC             = f+j_coord_offsetC;
1205             fjptrD             = f+j_coord_offsetD;
1206
1207             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1208
1209             /* Inner loop uses 150 flops */
1210         }
1211
1212         if(jidx<j_index_end)
1213         {
1214
1215             /* Get j neighbor index, and coordinate index */
1216             jnrlistA         = jjnr[jidx];
1217             jnrlistB         = jjnr[jidx+1];
1218             jnrlistC         = jjnr[jidx+2];
1219             jnrlistD         = jjnr[jidx+3];
1220             /* Sign of each element will be negative for non-real atoms.
1221              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1222              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1223              */
1224             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1225
1226             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1227             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1228             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1229
1230             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1231             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1232             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1233             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1234             j_coord_offsetA  = DIM*jnrA;
1235             j_coord_offsetB  = DIM*jnrB;
1236             j_coord_offsetC  = DIM*jnrC;
1237             j_coord_offsetD  = DIM*jnrD;
1238
1239             /* load j atom coordinates */
1240             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1241                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1242                                                  &jx0,&jy0,&jz0);
1243
1244             /* Calculate displacement vector */
1245             dx00             = _mm256_sub_pd(ix0,jx0);
1246             dy00             = _mm256_sub_pd(iy0,jy0);
1247             dz00             = _mm256_sub_pd(iz0,jz0);
1248             dx10             = _mm256_sub_pd(ix1,jx0);
1249             dy10             = _mm256_sub_pd(iy1,jy0);
1250             dz10             = _mm256_sub_pd(iz1,jz0);
1251             dx20             = _mm256_sub_pd(ix2,jx0);
1252             dy20             = _mm256_sub_pd(iy2,jy0);
1253             dz20             = _mm256_sub_pd(iz2,jz0);
1254             dx30             = _mm256_sub_pd(ix3,jx0);
1255             dy30             = _mm256_sub_pd(iy3,jy0);
1256             dz30             = _mm256_sub_pd(iz3,jz0);
1257
1258             /* Calculate squared distance and things based on it */
1259             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1260             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1261             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1262             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1263
1264             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1265             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1266             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1267
1268             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1269             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1270             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1271             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1272
1273             /* Load parameters for j particles */
1274             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1275                                                                  charge+jnrC+0,charge+jnrD+0);
1276             vdwjidx0A        = 2*vdwtype[jnrA+0];
1277             vdwjidx0B        = 2*vdwtype[jnrB+0];
1278             vdwjidx0C        = 2*vdwtype[jnrC+0];
1279             vdwjidx0D        = 2*vdwtype[jnrD+0];
1280
1281             fjx0             = _mm256_setzero_pd();
1282             fjy0             = _mm256_setzero_pd();
1283             fjz0             = _mm256_setzero_pd();
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1290             {
1291
1292             /* Compute parameters for interactions between i and j atoms */
1293             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1294                                             vdwioffsetptr0+vdwjidx0B,
1295                                             vdwioffsetptr0+vdwjidx0C,
1296                                             vdwioffsetptr0+vdwjidx0D,
1297                                             &c6_00,&c12_00);
1298
1299             /* LENNARD-JONES DISPERSION/REPULSION */
1300
1301             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1302             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1303
1304             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1305
1306             fscal            = fvdw;
1307
1308             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1309
1310             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1311
1312             /* Calculate temporary vectorial force */
1313             tx               = _mm256_mul_pd(fscal,dx00);
1314             ty               = _mm256_mul_pd(fscal,dy00);
1315             tz               = _mm256_mul_pd(fscal,dz00);
1316
1317             /* Update vectorial force */
1318             fix0             = _mm256_add_pd(fix0,tx);
1319             fiy0             = _mm256_add_pd(fiy0,ty);
1320             fiz0             = _mm256_add_pd(fiz0,tz);
1321
1322             fjx0             = _mm256_add_pd(fjx0,tx);
1323             fjy0             = _mm256_add_pd(fjy0,ty);
1324             fjz0             = _mm256_add_pd(fjz0,tz);
1325
1326             }
1327
1328             /**************************
1329              * CALCULATE INTERACTIONS *
1330              **************************/
1331
1332             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1333             {
1334
1335             r10              = _mm256_mul_pd(rsq10,rinv10);
1336             r10              = _mm256_andnot_pd(dummy_mask,r10);
1337
1338             /* Compute parameters for interactions between i and j atoms */
1339             qq10             = _mm256_mul_pd(iq1,jq0);
1340
1341             /* EWALD ELECTROSTATICS */
1342
1343             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1344             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1345             ewitab           = _mm256_cvttpd_epi32(ewrt);
1346             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1347             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1348                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1349                                             &ewtabF,&ewtabFn);
1350             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1351             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1352
1353             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1354
1355             fscal            = felec;
1356
1357             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1358
1359             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1360
1361             /* Calculate temporary vectorial force */
1362             tx               = _mm256_mul_pd(fscal,dx10);
1363             ty               = _mm256_mul_pd(fscal,dy10);
1364             tz               = _mm256_mul_pd(fscal,dz10);
1365
1366             /* Update vectorial force */
1367             fix1             = _mm256_add_pd(fix1,tx);
1368             fiy1             = _mm256_add_pd(fiy1,ty);
1369             fiz1             = _mm256_add_pd(fiz1,tz);
1370
1371             fjx0             = _mm256_add_pd(fjx0,tx);
1372             fjy0             = _mm256_add_pd(fjy0,ty);
1373             fjz0             = _mm256_add_pd(fjz0,tz);
1374
1375             }
1376
1377             /**************************
1378              * CALCULATE INTERACTIONS *
1379              **************************/
1380
1381             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1382             {
1383
1384             r20              = _mm256_mul_pd(rsq20,rinv20);
1385             r20              = _mm256_andnot_pd(dummy_mask,r20);
1386
1387             /* Compute parameters for interactions between i and j atoms */
1388             qq20             = _mm256_mul_pd(iq2,jq0);
1389
1390             /* EWALD ELECTROSTATICS */
1391
1392             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1393             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1394             ewitab           = _mm256_cvttpd_epi32(ewrt);
1395             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1396             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1397                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1398                                             &ewtabF,&ewtabFn);
1399             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1400             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1401
1402             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1403
1404             fscal            = felec;
1405
1406             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1407
1408             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1409
1410             /* Calculate temporary vectorial force */
1411             tx               = _mm256_mul_pd(fscal,dx20);
1412             ty               = _mm256_mul_pd(fscal,dy20);
1413             tz               = _mm256_mul_pd(fscal,dz20);
1414
1415             /* Update vectorial force */
1416             fix2             = _mm256_add_pd(fix2,tx);
1417             fiy2             = _mm256_add_pd(fiy2,ty);
1418             fiz2             = _mm256_add_pd(fiz2,tz);
1419
1420             fjx0             = _mm256_add_pd(fjx0,tx);
1421             fjy0             = _mm256_add_pd(fjy0,ty);
1422             fjz0             = _mm256_add_pd(fjz0,tz);
1423
1424             }
1425
1426             /**************************
1427              * CALCULATE INTERACTIONS *
1428              **************************/
1429
1430             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1431             {
1432
1433             r30              = _mm256_mul_pd(rsq30,rinv30);
1434             r30              = _mm256_andnot_pd(dummy_mask,r30);
1435
1436             /* Compute parameters for interactions between i and j atoms */
1437             qq30             = _mm256_mul_pd(iq3,jq0);
1438
1439             /* EWALD ELECTROSTATICS */
1440
1441             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1442             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1443             ewitab           = _mm256_cvttpd_epi32(ewrt);
1444             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1445             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1446                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1447                                             &ewtabF,&ewtabFn);
1448             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1449             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1450
1451             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1452
1453             fscal            = felec;
1454
1455             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1456
1457             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1458
1459             /* Calculate temporary vectorial force */
1460             tx               = _mm256_mul_pd(fscal,dx30);
1461             ty               = _mm256_mul_pd(fscal,dy30);
1462             tz               = _mm256_mul_pd(fscal,dz30);
1463
1464             /* Update vectorial force */
1465             fix3             = _mm256_add_pd(fix3,tx);
1466             fiy3             = _mm256_add_pd(fiy3,ty);
1467             fiz3             = _mm256_add_pd(fiz3,tz);
1468
1469             fjx0             = _mm256_add_pd(fjx0,tx);
1470             fjy0             = _mm256_add_pd(fjy0,ty);
1471             fjz0             = _mm256_add_pd(fjz0,tz);
1472
1473             }
1474
1475             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1476             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1477             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1478             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1479
1480             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1481
1482             /* Inner loop uses 153 flops */
1483         }
1484
1485         /* End of innermost loop */
1486
1487         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1488                                                  f+i_coord_offset,fshift+i_shift_offset);
1489
1490         /* Increment number of inner iterations */
1491         inneriter                  += j_index_end - j_index_start;
1492
1493         /* Outer loop uses 24 flops */
1494     }
1495
1496     /* Increment number of outer iterations */
1497     outeriter        += nri;
1498
1499     /* Update outer/inner flops */
1500
1501     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*153);
1502 }