src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  98     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  99     __m128i          ewitab;
 100     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 102     real             *ewtab;
 103     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 104     real             rswitch_scalar,d_scalar;
 105     __m256d          dummy_mask,cutoff_mask;
 106     __m128           tmpmask0,tmpmask1;
 107     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 108     __m256d          one     = _mm256_set1_pd(1.0);
 109     __m256d          two     = _mm256_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 128     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 129     beta2            = _mm256_mul_pd(beta,beta);
 130     beta3            = _mm256_mul_pd(beta,beta2);
 131
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->rcoulomb;
 138     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 139     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 140
 141     rswitch_scalar   = fr->rcoulomb_switch;
 142     rswitch          = _mm256_set1_pd(rswitch_scalar);
 143     /* Setup switch parameters */
 144     d_scalar         = rcutoff_scalar-rswitch_scalar;
 145     d                = _mm256_set1_pd(d_scalar);
 146     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 147     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 148     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 149     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 150     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 151     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 152
 153     /* Avoid stupid compiler warnings */
 154     jnrA = jnrB = jnrC = jnrD = 0;
 155     j_coord_offsetA = 0;
 156     j_coord_offsetB = 0;
 157     j_coord_offsetC = 0;
 158     j_coord_offsetD = 0;
 159
 160     outeriter        = 0;
 161     inneriter        = 0;
 162
 163     for(iidx=0;iidx<4*DIM;iidx++)
 164     {
 165         scratch[iidx] = 0.0;
 166     }
 167
 168     /* Start outer loop over neighborlists */
 169     for(iidx=0; iidx<nri; iidx++)
 170     {
 171         /* Load shift vector for this list */
 172         i_shift_offset   = DIM*shiftidx[iidx];
 173
 174         /* Load limits for loop over neighbors */
 175         j_index_start    = jindex[iidx];
 176         j_index_end      = jindex[iidx+1];
 177
 178         /* Get outer coordinate index */
 179         inr              = iinr[iidx];
 180         i_coord_offset   = DIM*inr;
 181
 182         /* Load i particle coords and add shift vector */
 183         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 184
 185         fix0             = _mm256_setzero_pd();
 186         fiy0             = _mm256_setzero_pd();
 187         fiz0             = _mm256_setzero_pd();
 188
 189         /* Load parameters for i particles */
 190         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 191         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_pd();
 195         vvdwsum          = _mm256_setzero_pd();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             j_coord_offsetA  = DIM*jnrA;
 207             j_coord_offsetB  = DIM*jnrB;
 208             j_coord_offsetC  = DIM*jnrC;
 209             j_coord_offsetD  = DIM*jnrD;
 210
 211             /* load j atom coordinates */
 212             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 214                                                  &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm256_sub_pd(ix0,jx0);
 218             dy00             = _mm256_sub_pd(iy0,jy0);
 219             dz00             = _mm256_sub_pd(iz0,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 223
 224             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 225
 226             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 230                                                                  charge+jnrC+0,charge+jnrD+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             r00              = _mm256_mul_pd(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm256_mul_pd(iq0,jq0);
 247             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 248                                             vdwioffsetptr0+vdwjidx0B,
 249                                             vdwioffsetptr0+vdwjidx0C,
 250                                             vdwioffsetptr0+vdwjidx0D,
 251                                             &c6_00,&c12_00);
 252
 253             /* EWALD ELECTROSTATICS */
 254
 255             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 256             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 257             ewitab           = _mm256_cvttpd_epi32(ewrt);
 258             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 259             ewitab           = _mm_slli_epi32(ewitab,2);
 260             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 261             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 262             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 263             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 264             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 265             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 266             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 267             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 268             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 269
 270             /* LENNARD-JONES DISPERSION/REPULSION */
 271
 272             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 273             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 274             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 275             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 276             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 277
 278             d                = _mm256_sub_pd(r00,rswitch);
 279             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 280             d2               = _mm256_mul_pd(d,d);
 281             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 282
 283             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 284
 285             /* Evaluate switch function */
 286             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 287             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 288             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 289             velec            = _mm256_mul_pd(velec,sw);
 290             vvdw             = _mm256_mul_pd(vvdw,sw);
 291             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velec            = _mm256_and_pd(velec,cutoff_mask);
 295             velecsum         = _mm256_add_pd(velecsum,velec);
 296             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 297             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 298
 299             fscal            = _mm256_add_pd(felec,fvdw);
 300
 301             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm256_mul_pd(fscal,dx00);
 305             ty               = _mm256_mul_pd(fscal,dy00);
 306             tz               = _mm256_mul_pd(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm256_add_pd(fix0,tx);
 310             fiy0             = _mm256_add_pd(fiy0,ty);
 311             fiz0             = _mm256_add_pd(fiz0,tz);
 312
 313             fjptrA             = f+j_coord_offsetA;
 314             fjptrB             = f+j_coord_offsetB;
 315             fjptrC             = f+j_coord_offsetC;
 316             fjptrD             = f+j_coord_offsetD;
 317             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 318
 319             }
 320
 321             /* Inner loop uses 83 flops */
 322         }
 323
 324         if(jidx<j_index_end)
 325         {
 326
 327             /* Get j neighbor index, and coordinate index */
 328             jnrlistA         = jjnr[jidx];
 329             jnrlistB         = jjnr[jidx+1];
 330             jnrlistC         = jjnr[jidx+2];
 331             jnrlistD         = jjnr[jidx+3];
 332             /* Sign of each element will be negative for non-real atoms.
 333              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 334              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 335              */
 336             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 337
 338             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 339             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 340             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 341
 342             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 343             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 344             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 345             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 346             j_coord_offsetA  = DIM*jnrA;
 347             j_coord_offsetB  = DIM*jnrB;
 348             j_coord_offsetC  = DIM*jnrC;
 349             j_coord_offsetD  = DIM*jnrD;
 350
 351             /* load j atom coordinates */
 352             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 353                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 354                                                  &jx0,&jy0,&jz0);
 355
 356             /* Calculate displacement vector */
 357             dx00             = _mm256_sub_pd(ix0,jx0);
 358             dy00             = _mm256_sub_pd(iy0,jy0);
 359             dz00             = _mm256_sub_pd(iz0,jz0);
 360
 361             /* Calculate squared distance and things based on it */
 362             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 363
 364             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 365
 366             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 367
 368             /* Load parameters for j particles */
 369             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 370                                                                  charge+jnrC+0,charge+jnrD+0);
 371             vdwjidx0A        = 2*vdwtype[jnrA+0];
 372             vdwjidx0B        = 2*vdwtype[jnrB+0];
 373             vdwjidx0C        = 2*vdwtype[jnrC+0];
 374             vdwjidx0D        = 2*vdwtype[jnrD+0];
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 381             {
 382
 383             r00              = _mm256_mul_pd(rsq00,rinv00);
 384             r00              = _mm256_andnot_pd(dummy_mask,r00);
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq00             = _mm256_mul_pd(iq0,jq0);
 388             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 389                                             vdwioffsetptr0+vdwjidx0B,
 390                                             vdwioffsetptr0+vdwjidx0C,
 391                                             vdwioffsetptr0+vdwjidx0D,
 392                                             &c6_00,&c12_00);
 393
 394             /* EWALD ELECTROSTATICS */
 395
 396             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 397             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 398             ewitab           = _mm256_cvttpd_epi32(ewrt);
 399             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 400             ewitab           = _mm_slli_epi32(ewitab,2);
 401             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 402             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 403             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 404             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 405             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 406             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 407             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 408             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 409             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 410
 411             /* LENNARD-JONES DISPERSION/REPULSION */
 412
 413             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 414             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 415             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 416             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 417             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 418
 419             d                = _mm256_sub_pd(r00,rswitch);
 420             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 421             d2               = _mm256_mul_pd(d,d);
 422             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 423
 424             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 425
 426             /* Evaluate switch function */
 427             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 428             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 429             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 430             velec            = _mm256_mul_pd(velec,sw);
 431             vvdw             = _mm256_mul_pd(vvdw,sw);
 432             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velec            = _mm256_and_pd(velec,cutoff_mask);
 436             velec            = _mm256_andnot_pd(dummy_mask,velec);
 437             velecsum         = _mm256_add_pd(velecsum,velec);
 438             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 439             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 440             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 441
 442             fscal            = _mm256_add_pd(felec,fvdw);
 443
 444             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 445
 446             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 447
 448             /* Calculate temporary vectorial force */
 449             tx               = _mm256_mul_pd(fscal,dx00);
 450             ty               = _mm256_mul_pd(fscal,dy00);
 451             tz               = _mm256_mul_pd(fscal,dz00);
 452
 453             /* Update vectorial force */
 454             fix0             = _mm256_add_pd(fix0,tx);
 455             fiy0             = _mm256_add_pd(fiy0,ty);
 456             fiz0             = _mm256_add_pd(fiz0,tz);
 457
 458             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 459             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 460             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 461             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 462             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 463
 464             }
 465
 466             /* Inner loop uses 84 flops */
 467         }
 468
 469         /* End of innermost loop */
 470
 471         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 472                                                  f+i_coord_offset,fshift+i_shift_offset);
 473
 474         ggid                        = gid[iidx];
 475         /* Update potential energies */
 476         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 477         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 478
 479         /* Increment number of inner iterations */
 480         inneriter                  += j_index_end - j_index_start;
 481
 482         /* Outer loop uses 9 flops */
 483     }
 484
 485     /* Increment number of outer iterations */
 486     outeriter        += nri;
 487
 488     /* Update outer/inner flops */
 489
 490     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*84);
 491 }
 492 /*
 493  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_avx_256_double
 494  * Electrostatics interaction: Ewald
 495  * VdW interaction:            LennardJones
 496  * Geometry:                   Particle-Particle
 497  * Calculate force/pot:        Force
 498  */
 499 void
 500 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_avx_256_double
 501                     (t_nblist                    * gmx_restrict       nlist,
 502                      rvec                        * gmx_restrict          xx,
 503                      rvec                        * gmx_restrict          ff,
 504                      t_forcerec                  * gmx_restrict          fr,
 505                      t_mdatoms                   * gmx_restrict     mdatoms,
 506                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 507                      t_nrnb                      * gmx_restrict        nrnb)
 508 {
 509     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 510      * just 0 for non-waters.
 511      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 512      * jnr indices corresponding to data put in the four positions in the SIMD register.
 513      */
 514     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 515     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 516     int              jnrA,jnrB,jnrC,jnrD;
 517     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 518     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 519     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 520     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 521     real             rcutoff_scalar;
 522     real             *shiftvec,*fshift,*x,*f;
 523     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 524     real             scratch[4*DIM];
 525     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 526     real *           vdwioffsetptr0;
 527     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 528     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 529     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 530     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 531     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 532     real             *charge;
 533     int              nvdwtype;
 534     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 535     int              *vdwtype;
 536     real             *vdwparam;
 537     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 538     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 539     __m128i          ewitab;
 540     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 541     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 542     real             *ewtab;
 543     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 544     real             rswitch_scalar,d_scalar;
 545     __m256d          dummy_mask,cutoff_mask;
 546     __m128           tmpmask0,tmpmask1;
 547     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 548     __m256d          one     = _mm256_set1_pd(1.0);
 549     __m256d          two     = _mm256_set1_pd(2.0);
 550     x                = xx[0];
 551     f                = ff[0];
 552
 553     nri              = nlist->nri;
 554     iinr             = nlist->iinr;
 555     jindex           = nlist->jindex;
 556     jjnr             = nlist->jjnr;
 557     shiftidx         = nlist->shift;
 558     gid              = nlist->gid;
 559     shiftvec         = fr->shift_vec[0];
 560     fshift           = fr->fshift[0];
 561     facel            = _mm256_set1_pd(fr->epsfac);
 562     charge           = mdatoms->chargeA;
 563     nvdwtype         = fr->ntype;
 564     vdwparam         = fr->nbfp;
 565     vdwtype          = mdatoms->typeA;
 566
 567     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 568     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 569     beta2            = _mm256_mul_pd(beta,beta);
 570     beta3            = _mm256_mul_pd(beta,beta2);
 571
 572     ewtab            = fr->ic->tabq_coul_FDV0;
 573     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 574     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 575
 576     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 577     rcutoff_scalar   = fr->rcoulomb;
 578     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 579     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 580
 581     rswitch_scalar   = fr->rcoulomb_switch;
 582     rswitch          = _mm256_set1_pd(rswitch_scalar);
 583     /* Setup switch parameters */
 584     d_scalar         = rcutoff_scalar-rswitch_scalar;
 585     d                = _mm256_set1_pd(d_scalar);
 586     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 587     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 588     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 589     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 590     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 591     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 592
 593     /* Avoid stupid compiler warnings */
 594     jnrA = jnrB = jnrC = jnrD = 0;
 595     j_coord_offsetA = 0;
 596     j_coord_offsetB = 0;
 597     j_coord_offsetC = 0;
 598     j_coord_offsetD = 0;
 599
 600     outeriter        = 0;
 601     inneriter        = 0;
 602
 603     for(iidx=0;iidx<4*DIM;iidx++)
 604     {
 605         scratch[iidx] = 0.0;
 606     }
 607
 608     /* Start outer loop over neighborlists */
 609     for(iidx=0; iidx<nri; iidx++)
 610     {
 611         /* Load shift vector for this list */
 612         i_shift_offset   = DIM*shiftidx[iidx];
 613
 614         /* Load limits for loop over neighbors */
 615         j_index_start    = jindex[iidx];
 616         j_index_end      = jindex[iidx+1];
 617
 618         /* Get outer coordinate index */
 619         inr              = iinr[iidx];
 620         i_coord_offset   = DIM*inr;
 621
 622         /* Load i particle coords and add shift vector */
 623         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 624
 625         fix0             = _mm256_setzero_pd();
 626         fiy0             = _mm256_setzero_pd();
 627         fiz0             = _mm256_setzero_pd();
 628
 629         /* Load parameters for i particles */
 630         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 631         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 632
 633         /* Start inner kernel loop */
 634         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 635         {
 636
 637             /* Get j neighbor index, and coordinate index */
 638             jnrA             = jjnr[jidx];
 639             jnrB             = jjnr[jidx+1];
 640             jnrC             = jjnr[jidx+2];
 641             jnrD             = jjnr[jidx+3];
 642             j_coord_offsetA  = DIM*jnrA;
 643             j_coord_offsetB  = DIM*jnrB;
 644             j_coord_offsetC  = DIM*jnrC;
 645             j_coord_offsetD  = DIM*jnrD;
 646
 647             /* load j atom coordinates */
 648             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 649                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 650                                                  &jx0,&jy0,&jz0);
 651
 652             /* Calculate displacement vector */
 653             dx00             = _mm256_sub_pd(ix0,jx0);
 654             dy00             = _mm256_sub_pd(iy0,jy0);
 655             dz00             = _mm256_sub_pd(iz0,jz0);
 656
 657             /* Calculate squared distance and things based on it */
 658             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 659
 660             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 661
 662             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 663
 664             /* Load parameters for j particles */
 665             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 666                                                                  charge+jnrC+0,charge+jnrD+0);
 667             vdwjidx0A        = 2*vdwtype[jnrA+0];
 668             vdwjidx0B        = 2*vdwtype[jnrB+0];
 669             vdwjidx0C        = 2*vdwtype[jnrC+0];
 670             vdwjidx0D        = 2*vdwtype[jnrD+0];
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 677             {
 678
 679             r00              = _mm256_mul_pd(rsq00,rinv00);
 680
 681             /* Compute parameters for interactions between i and j atoms */
 682             qq00             = _mm256_mul_pd(iq0,jq0);
 683             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 684                                             vdwioffsetptr0+vdwjidx0B,
 685                                             vdwioffsetptr0+vdwjidx0C,
 686                                             vdwioffsetptr0+vdwjidx0D,
 687                                             &c6_00,&c12_00);
 688
 689             /* EWALD ELECTROSTATICS */
 690
 691             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 692             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 693             ewitab           = _mm256_cvttpd_epi32(ewrt);
 694             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 695             ewitab           = _mm_slli_epi32(ewitab,2);
 696             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 697             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 698             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 699             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 700             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 701             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 702             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 703             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 704             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 705
 706             /* LENNARD-JONES DISPERSION/REPULSION */
 707
 708             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 709             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 710             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 711             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 712             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 713
 714             d                = _mm256_sub_pd(r00,rswitch);
 715             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 716             d2               = _mm256_mul_pd(d,d);
 717             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 718
 719             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 720
 721             /* Evaluate switch function */
 722             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 723             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 724             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 725             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 726
 727             fscal            = _mm256_add_pd(felec,fvdw);
 728
 729             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 730
 731             /* Calculate temporary vectorial force */
 732             tx               = _mm256_mul_pd(fscal,dx00);
 733             ty               = _mm256_mul_pd(fscal,dy00);
 734             tz               = _mm256_mul_pd(fscal,dz00);
 735
 736             /* Update vectorial force */
 737             fix0             = _mm256_add_pd(fix0,tx);
 738             fiy0             = _mm256_add_pd(fiy0,ty);
 739             fiz0             = _mm256_add_pd(fiz0,tz);
 740
 741             fjptrA             = f+j_coord_offsetA;
 742             fjptrB             = f+j_coord_offsetB;
 743             fjptrC             = f+j_coord_offsetC;
 744             fjptrD             = f+j_coord_offsetD;
 745             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 746
 747             }
 748
 749             /* Inner loop uses 77 flops */
 750         }
 751
 752         if(jidx<j_index_end)
 753         {
 754
 755             /* Get j neighbor index, and coordinate index */
 756             jnrlistA         = jjnr[jidx];
 757             jnrlistB         = jjnr[jidx+1];
 758             jnrlistC         = jjnr[jidx+2];
 759             jnrlistD         = jjnr[jidx+3];
 760             /* Sign of each element will be negative for non-real atoms.
 761              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 762              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 763              */
 764             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 765
 766             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 767             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 768             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 769
 770             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 771             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 772             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 773             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 774             j_coord_offsetA  = DIM*jnrA;
 775             j_coord_offsetB  = DIM*jnrB;
 776             j_coord_offsetC  = DIM*jnrC;
 777             j_coord_offsetD  = DIM*jnrD;
 778
 779             /* load j atom coordinates */
 780             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 781                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 782                                                  &jx0,&jy0,&jz0);
 783
 784             /* Calculate displacement vector */
 785             dx00             = _mm256_sub_pd(ix0,jx0);
 786             dy00             = _mm256_sub_pd(iy0,jy0);
 787             dz00             = _mm256_sub_pd(iz0,jz0);
 788
 789             /* Calculate squared distance and things based on it */
 790             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 791
 792             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 793
 794             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 795
 796             /* Load parameters for j particles */
 797             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 798                                                                  charge+jnrC+0,charge+jnrD+0);
 799             vdwjidx0A        = 2*vdwtype[jnrA+0];
 800             vdwjidx0B        = 2*vdwtype[jnrB+0];
 801             vdwjidx0C        = 2*vdwtype[jnrC+0];
 802             vdwjidx0D        = 2*vdwtype[jnrD+0];
 803
 804             /**************************
 805              * CALCULATE INTERACTIONS *
 806              **************************/
 807
 808             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 809             {
 810
 811             r00              = _mm256_mul_pd(rsq00,rinv00);
 812             r00              = _mm256_andnot_pd(dummy_mask,r00);
 813
 814             /* Compute parameters for interactions between i and j atoms */
 815             qq00             = _mm256_mul_pd(iq0,jq0);
 816             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 817                                             vdwioffsetptr0+vdwjidx0B,
 818                                             vdwioffsetptr0+vdwjidx0C,
 819                                             vdwioffsetptr0+vdwjidx0D,
 820                                             &c6_00,&c12_00);
 821
 822             /* EWALD ELECTROSTATICS */
 823
 824             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 825             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 826             ewitab           = _mm256_cvttpd_epi32(ewrt);
 827             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 828             ewitab           = _mm_slli_epi32(ewitab,2);
 829             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 830             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 831             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 832             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 833             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 834             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 835             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 836             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 837             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 838
 839             /* LENNARD-JONES DISPERSION/REPULSION */
 840
 841             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 842             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 843             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 844             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 845             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 846
 847             d                = _mm256_sub_pd(r00,rswitch);
 848             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 849             d2               = _mm256_mul_pd(d,d);
 850             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 851
 852             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 853
 854             /* Evaluate switch function */
 855             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 856             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 857             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 858             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 859
 860             fscal            = _mm256_add_pd(felec,fvdw);
 861
 862             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 863
 864             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 865
 866             /* Calculate temporary vectorial force */
 867             tx               = _mm256_mul_pd(fscal,dx00);
 868             ty               = _mm256_mul_pd(fscal,dy00);
 869             tz               = _mm256_mul_pd(fscal,dz00);
 870
 871             /* Update vectorial force */
 872             fix0             = _mm256_add_pd(fix0,tx);
 873             fiy0             = _mm256_add_pd(fiy0,ty);
 874             fiz0             = _mm256_add_pd(fiz0,tz);
 875
 876             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 877             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 878             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 879             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 880             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 881
 882             }
 883
 884             /* Inner loop uses 78 flops */
 885         }
 886
 887         /* End of innermost loop */
 888
 889         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 890                                                  f+i_coord_offset,fshift+i_shift_offset);
 891
 892         /* Increment number of inner iterations */
 893         inneriter                  += j_index_end - j_index_start;
 894
 895         /* Outer loop uses 7 flops */
 896     }
 897
 898     /* Increment number of outer iterations */
 899     outeriter        += nri;
 900
 901     /* Update outer/inner flops */
 902
 903     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*78);
 904 }