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