src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m128i          ewitab;
 104     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 106     real             *ewtab;
 107     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 108     real             rswitch_scalar,d_scalar;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 132     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 133     beta2            = _mm256_mul_pd(beta,beta);
 134     beta3            = _mm256_mul_pd(beta,beta2);
 135
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 143     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 144     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 145     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 146
 147     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 148     rcutoff_scalar   = fr->rcoulomb;
 149     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 150     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 151
 152     rswitch_scalar   = fr->rcoulomb_switch;
 153     rswitch          = _mm256_set1_pd(rswitch_scalar);
 154     /* Setup switch parameters */
 155     d_scalar         = rcutoff_scalar-rswitch_scalar;
 156     d                = _mm256_set1_pd(d_scalar);
 157     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 158     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 159     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 160     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 161     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 162     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 163
 164     /* Avoid stupid compiler warnings */
 165     jnrA = jnrB = jnrC = jnrD = 0;
 166     j_coord_offsetA = 0;
 167     j_coord_offsetB = 0;
 168     j_coord_offsetC = 0;
 169     j_coord_offsetD = 0;
 170
 171     outeriter        = 0;
 172     inneriter        = 0;
 173
 174     for(iidx=0;iidx<4*DIM;iidx++)
 175     {
 176         scratch[iidx] = 0.0;
 177     }
 178
 179     /* Start outer loop over neighborlists */
 180     for(iidx=0; iidx<nri; iidx++)
 181     {
 182         /* Load shift vector for this list */
 183         i_shift_offset   = DIM*shiftidx[iidx];
 184
 185         /* Load limits for loop over neighbors */
 186         j_index_start    = jindex[iidx];
 187         j_index_end      = jindex[iidx+1];
 188
 189         /* Get outer coordinate index */
 190         inr              = iinr[iidx];
 191         i_coord_offset   = DIM*inr;
 192
 193         /* Load i particle coords and add shift vector */
 194         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 195                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 196
 197         fix0             = _mm256_setzero_pd();
 198         fiy0             = _mm256_setzero_pd();
 199         fiz0             = _mm256_setzero_pd();
 200         fix1             = _mm256_setzero_pd();
 201         fiy1             = _mm256_setzero_pd();
 202         fiz1             = _mm256_setzero_pd();
 203         fix2             = _mm256_setzero_pd();
 204         fiy2             = _mm256_setzero_pd();
 205         fiz2             = _mm256_setzero_pd();
 206
 207         /* Reset potential sums */
 208         velecsum         = _mm256_setzero_pd();
 209         vvdwsum          = _mm256_setzero_pd();
 210
 211         /* Start inner kernel loop */
 212         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 213         {
 214
 215             /* Get j neighbor index, and coordinate index */
 216             jnrA             = jjnr[jidx];
 217             jnrB             = jjnr[jidx+1];
 218             jnrC             = jjnr[jidx+2];
 219             jnrD             = jjnr[jidx+3];
 220             j_coord_offsetA  = DIM*jnrA;
 221             j_coord_offsetB  = DIM*jnrB;
 222             j_coord_offsetC  = DIM*jnrC;
 223             j_coord_offsetD  = DIM*jnrD;
 224
 225             /* load j atom coordinates */
 226             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 227                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 228                                                  &jx0,&jy0,&jz0);
 229
 230             /* Calculate displacement vector */
 231             dx00             = _mm256_sub_pd(ix0,jx0);
 232             dy00             = _mm256_sub_pd(iy0,jy0);
 233             dz00             = _mm256_sub_pd(iz0,jz0);
 234             dx10             = _mm256_sub_pd(ix1,jx0);
 235             dy10             = _mm256_sub_pd(iy1,jy0);
 236             dz10             = _mm256_sub_pd(iz1,jz0);
 237             dx20             = _mm256_sub_pd(ix2,jx0);
 238             dy20             = _mm256_sub_pd(iy2,jy0);
 239             dz20             = _mm256_sub_pd(iz2,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 243             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 244             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 245
 246             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 247             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 248             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 249
 250             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 251             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 252             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 253
 254             /* Load parameters for j particles */
 255             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 256                                                                  charge+jnrC+0,charge+jnrD+0);
 257             vdwjidx0A        = 2*vdwtype[jnrA+0];
 258             vdwjidx0B        = 2*vdwtype[jnrB+0];
 259             vdwjidx0C        = 2*vdwtype[jnrC+0];
 260             vdwjidx0D        = 2*vdwtype[jnrD+0];
 261
 262             fjx0             = _mm256_setzero_pd();
 263             fjy0             = _mm256_setzero_pd();
 264             fjz0             = _mm256_setzero_pd();
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 271             {
 272
 273             r00              = _mm256_mul_pd(rsq00,rinv00);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             qq00             = _mm256_mul_pd(iq0,jq0);
 277             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 278                                             vdwioffsetptr0+vdwjidx0B,
 279                                             vdwioffsetptr0+vdwjidx0C,
 280                                             vdwioffsetptr0+vdwjidx0D,
 281                                             &c6_00,&c12_00);
 282
 283             /* EWALD ELECTROSTATICS */
 284
 285             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 286             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 287             ewitab           = _mm256_cvttpd_epi32(ewrt);
 288             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 289             ewitab           = _mm_slli_epi32(ewitab,2);
 290             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 291             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 292             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 293             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 294             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 295             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 296             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 297             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 298             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 299
 300             /* LENNARD-JONES DISPERSION/REPULSION */
 301
 302             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 303             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 304             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 305             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 306             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 307
 308             d                = _mm256_sub_pd(r00,rswitch);
 309             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 310             d2               = _mm256_mul_pd(d,d);
 311             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)))))));
 312
 313             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 314
 315             /* Evaluate switch function */
 316             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 317             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 318             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 319             velec            = _mm256_mul_pd(velec,sw);
 320             vvdw             = _mm256_mul_pd(vvdw,sw);
 321             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velec            = _mm256_and_pd(velec,cutoff_mask);
 325             velecsum         = _mm256_add_pd(velecsum,velec);
 326             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 327             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 328
 329             fscal            = _mm256_add_pd(felec,fvdw);
 330
 331             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm256_mul_pd(fscal,dx00);
 335             ty               = _mm256_mul_pd(fscal,dy00);
 336             tz               = _mm256_mul_pd(fscal,dz00);
 337
 338             /* Update vectorial force */
 339             fix0             = _mm256_add_pd(fix0,tx);
 340             fiy0             = _mm256_add_pd(fiy0,ty);
 341             fiz0             = _mm256_add_pd(fiz0,tz);
 342
 343             fjx0             = _mm256_add_pd(fjx0,tx);
 344             fjy0             = _mm256_add_pd(fjy0,ty);
 345             fjz0             = _mm256_add_pd(fjz0,tz);
 346
 347             }
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 354             {
 355
 356             r10              = _mm256_mul_pd(rsq10,rinv10);
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq10             = _mm256_mul_pd(iq1,jq0);
 360
 361             /* EWALD ELECTROSTATICS */
 362
 363             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 364             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 365             ewitab           = _mm256_cvttpd_epi32(ewrt);
 366             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 367             ewitab           = _mm_slli_epi32(ewitab,2);
 368             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 369             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 370             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 371             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 372             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 373             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 374             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 375             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 376             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 377
 378             d                = _mm256_sub_pd(r10,rswitch);
 379             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 380             d2               = _mm256_mul_pd(d,d);
 381             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)))))));
 382
 383             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 384
 385             /* Evaluate switch function */
 386             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 387             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 388             velec            = _mm256_mul_pd(velec,sw);
 389             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velec            = _mm256_and_pd(velec,cutoff_mask);
 393             velecsum         = _mm256_add_pd(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm256_mul_pd(fscal,dx10);
 401             ty               = _mm256_mul_pd(fscal,dy10);
 402             tz               = _mm256_mul_pd(fscal,dz10);
 403
 404             /* Update vectorial force */
 405             fix1             = _mm256_add_pd(fix1,tx);
 406             fiy1             = _mm256_add_pd(fiy1,ty);
 407             fiz1             = _mm256_add_pd(fiz1,tz);
 408
 409             fjx0             = _mm256_add_pd(fjx0,tx);
 410             fjy0             = _mm256_add_pd(fjy0,ty);
 411             fjz0             = _mm256_add_pd(fjz0,tz);
 412
 413             }
 414
 415             /**************************
 416              * CALCULATE INTERACTIONS *
 417              **************************/
 418
 419             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 420             {
 421
 422             r20              = _mm256_mul_pd(rsq20,rinv20);
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq20             = _mm256_mul_pd(iq2,jq0);
 426
 427             /* EWALD ELECTROSTATICS */
 428
 429             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 430             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 431             ewitab           = _mm256_cvttpd_epi32(ewrt);
 432             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 433             ewitab           = _mm_slli_epi32(ewitab,2);
 434             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 435             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 436             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 437             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 438             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 439             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 440             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 441             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 442             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 443
 444             d                = _mm256_sub_pd(r20,rswitch);
 445             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 446             d2               = _mm256_mul_pd(d,d);
 447             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)))))));
 448
 449             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 450
 451             /* Evaluate switch function */
 452             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 453             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 454             velec            = _mm256_mul_pd(velec,sw);
 455             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm256_and_pd(velec,cutoff_mask);
 459             velecsum         = _mm256_add_pd(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 464
 465             /* Calculate temporary vectorial force */
 466             tx               = _mm256_mul_pd(fscal,dx20);
 467             ty               = _mm256_mul_pd(fscal,dy20);
 468             tz               = _mm256_mul_pd(fscal,dz20);
 469
 470             /* Update vectorial force */
 471             fix2             = _mm256_add_pd(fix2,tx);
 472             fiy2             = _mm256_add_pd(fiy2,ty);
 473             fiz2             = _mm256_add_pd(fiz2,tz);
 474
 475             fjx0             = _mm256_add_pd(fjx0,tx);
 476             fjy0             = _mm256_add_pd(fjy0,ty);
 477             fjz0             = _mm256_add_pd(fjz0,tz);
 478
 479             }
 480
 481             fjptrA             = f+j_coord_offsetA;
 482             fjptrB             = f+j_coord_offsetB;
 483             fjptrC             = f+j_coord_offsetC;
 484             fjptrD             = f+j_coord_offsetD;
 485
 486             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 487
 488             /* Inner loop uses 216 flops */
 489         }
 490
 491         if(jidx<j_index_end)
 492         {
 493
 494             /* Get j neighbor index, and coordinate index */
 495             jnrlistA         = jjnr[jidx];
 496             jnrlistB         = jjnr[jidx+1];
 497             jnrlistC         = jjnr[jidx+2];
 498             jnrlistD         = jjnr[jidx+3];
 499             /* Sign of each element will be negative for non-real atoms.
 500              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 501              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 502              */
 503             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 504
 505             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 506             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 507             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 508
 509             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 510             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 511             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 512             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 513             j_coord_offsetA  = DIM*jnrA;
 514             j_coord_offsetB  = DIM*jnrB;
 515             j_coord_offsetC  = DIM*jnrC;
 516             j_coord_offsetD  = DIM*jnrD;
 517
 518             /* load j atom coordinates */
 519             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 520                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 521                                                  &jx0,&jy0,&jz0);
 522
 523             /* Calculate displacement vector */
 524             dx00             = _mm256_sub_pd(ix0,jx0);
 525             dy00             = _mm256_sub_pd(iy0,jy0);
 526             dz00             = _mm256_sub_pd(iz0,jz0);
 527             dx10             = _mm256_sub_pd(ix1,jx0);
 528             dy10             = _mm256_sub_pd(iy1,jy0);
 529             dz10             = _mm256_sub_pd(iz1,jz0);
 530             dx20             = _mm256_sub_pd(ix2,jx0);
 531             dy20             = _mm256_sub_pd(iy2,jy0);
 532             dz20             = _mm256_sub_pd(iz2,jz0);
 533
 534             /* Calculate squared distance and things based on it */
 535             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 536             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 537             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 538
 539             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 540             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 541             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 542
 543             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 544             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 545             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 546
 547             /* Load parameters for j particles */
 548             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 549                                                                  charge+jnrC+0,charge+jnrD+0);
 550             vdwjidx0A        = 2*vdwtype[jnrA+0];
 551             vdwjidx0B        = 2*vdwtype[jnrB+0];
 552             vdwjidx0C        = 2*vdwtype[jnrC+0];
 553             vdwjidx0D        = 2*vdwtype[jnrD+0];
 554
 555             fjx0             = _mm256_setzero_pd();
 556             fjy0             = _mm256_setzero_pd();
 557             fjz0             = _mm256_setzero_pd();
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 564             {
 565
 566             r00              = _mm256_mul_pd(rsq00,rinv00);
 567             r00              = _mm256_andnot_pd(dummy_mask,r00);
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq00             = _mm256_mul_pd(iq0,jq0);
 571             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 572                                             vdwioffsetptr0+vdwjidx0B,
 573                                             vdwioffsetptr0+vdwjidx0C,
 574                                             vdwioffsetptr0+vdwjidx0D,
 575                                             &c6_00,&c12_00);
 576
 577             /* EWALD ELECTROSTATICS */
 578
 579             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 580             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 581             ewitab           = _mm256_cvttpd_epi32(ewrt);
 582             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 583             ewitab           = _mm_slli_epi32(ewitab,2);
 584             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 585             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 586             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 587             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 588             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 589             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 590             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 591             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 592             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 593
 594             /* LENNARD-JONES DISPERSION/REPULSION */
 595
 596             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 597             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 598             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 599             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 600             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 601
 602             d                = _mm256_sub_pd(r00,rswitch);
 603             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 604             d2               = _mm256_mul_pd(d,d);
 605             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)))))));
 606
 607             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 608
 609             /* Evaluate switch function */
 610             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 611             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 612             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 613             velec            = _mm256_mul_pd(velec,sw);
 614             vvdw             = _mm256_mul_pd(vvdw,sw);
 615             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velec            = _mm256_and_pd(velec,cutoff_mask);
 619             velec            = _mm256_andnot_pd(dummy_mask,velec);
 620             velecsum         = _mm256_add_pd(velecsum,velec);
 621             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 622             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 623             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 624
 625             fscal            = _mm256_add_pd(felec,fvdw);
 626
 627             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 628
 629             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm256_mul_pd(fscal,dx00);
 633             ty               = _mm256_mul_pd(fscal,dy00);
 634             tz               = _mm256_mul_pd(fscal,dz00);
 635
 636             /* Update vectorial force */
 637             fix0             = _mm256_add_pd(fix0,tx);
 638             fiy0             = _mm256_add_pd(fiy0,ty);
 639             fiz0             = _mm256_add_pd(fiz0,tz);
 640
 641             fjx0             = _mm256_add_pd(fjx0,tx);
 642             fjy0             = _mm256_add_pd(fjy0,ty);
 643             fjz0             = _mm256_add_pd(fjz0,tz);
 644
 645             }
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 652             {
 653
 654             r10              = _mm256_mul_pd(rsq10,rinv10);
 655             r10              = _mm256_andnot_pd(dummy_mask,r10);
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             qq10             = _mm256_mul_pd(iq1,jq0);
 659
 660             /* EWALD ELECTROSTATICS */
 661
 662             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 663             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 664             ewitab           = _mm256_cvttpd_epi32(ewrt);
 665             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 666             ewitab           = _mm_slli_epi32(ewitab,2);
 667             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 668             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 669             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 670             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 671             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 672             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 673             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 674             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 675             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 676
 677             d                = _mm256_sub_pd(r10,rswitch);
 678             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 679             d2               = _mm256_mul_pd(d,d);
 680             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)))))));
 681
 682             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 683
 684             /* Evaluate switch function */
 685             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 686             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 687             velec            = _mm256_mul_pd(velec,sw);
 688             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 689
 690             /* Update potential sum for this i atom from the interaction with this j atom. */
 691             velec            = _mm256_and_pd(velec,cutoff_mask);
 692             velec            = _mm256_andnot_pd(dummy_mask,velec);
 693             velecsum         = _mm256_add_pd(velecsum,velec);
 694
 695             fscal            = felec;
 696
 697             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 698
 699             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 700
 701             /* Calculate temporary vectorial force */
 702             tx               = _mm256_mul_pd(fscal,dx10);
 703             ty               = _mm256_mul_pd(fscal,dy10);
 704             tz               = _mm256_mul_pd(fscal,dz10);
 705
 706             /* Update vectorial force */
 707             fix1             = _mm256_add_pd(fix1,tx);
 708             fiy1             = _mm256_add_pd(fiy1,ty);
 709             fiz1             = _mm256_add_pd(fiz1,tz);
 710
 711             fjx0             = _mm256_add_pd(fjx0,tx);
 712             fjy0             = _mm256_add_pd(fjy0,ty);
 713             fjz0             = _mm256_add_pd(fjz0,tz);
 714
 715             }
 716
 717             /**************************
 718              * CALCULATE INTERACTIONS *
 719              **************************/
 720
 721             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 722             {
 723
 724             r20              = _mm256_mul_pd(rsq20,rinv20);
 725             r20              = _mm256_andnot_pd(dummy_mask,r20);
 726
 727             /* Compute parameters for interactions between i and j atoms */
 728             qq20             = _mm256_mul_pd(iq2,jq0);
 729
 730             /* EWALD ELECTROSTATICS */
 731
 732             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 733             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 734             ewitab           = _mm256_cvttpd_epi32(ewrt);
 735             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 736             ewitab           = _mm_slli_epi32(ewitab,2);
 737             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 738             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 739             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 740             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 741             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 742             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 743             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 744             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 745             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 746
 747             d                = _mm256_sub_pd(r20,rswitch);
 748             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 749             d2               = _mm256_mul_pd(d,d);
 750             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)))))));
 751
 752             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 753
 754             /* Evaluate switch function */
 755             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 756             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 757             velec            = _mm256_mul_pd(velec,sw);
 758             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 759
 760             /* Update potential sum for this i atom from the interaction with this j atom. */
 761             velec            = _mm256_and_pd(velec,cutoff_mask);
 762             velec            = _mm256_andnot_pd(dummy_mask,velec);
 763             velecsum         = _mm256_add_pd(velecsum,velec);
 764
 765             fscal            = felec;
 766
 767             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 768
 769             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 770
 771             /* Calculate temporary vectorial force */
 772             tx               = _mm256_mul_pd(fscal,dx20);
 773             ty               = _mm256_mul_pd(fscal,dy20);
 774             tz               = _mm256_mul_pd(fscal,dz20);
 775
 776             /* Update vectorial force */
 777             fix2             = _mm256_add_pd(fix2,tx);
 778             fiy2             = _mm256_add_pd(fiy2,ty);
 779             fiz2             = _mm256_add_pd(fiz2,tz);
 780
 781             fjx0             = _mm256_add_pd(fjx0,tx);
 782             fjy0             = _mm256_add_pd(fjy0,ty);
 783             fjz0             = _mm256_add_pd(fjz0,tz);
 784
 785             }
 786
 787             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 788             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 789             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 790             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 791
 792             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 793
 794             /* Inner loop uses 219 flops */
 795         }
 796
 797         /* End of innermost loop */
 798
 799         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 800                                                  f+i_coord_offset,fshift+i_shift_offset);
 801
 802         ggid                        = gid[iidx];
 803         /* Update potential energies */
 804         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 805         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 806
 807         /* Increment number of inner iterations */
 808         inneriter                  += j_index_end - j_index_start;
 809
 810         /* Outer loop uses 20 flops */
 811     }
 812
 813     /* Increment number of outer iterations */
 814     outeriter        += nri;
 815
 816     /* Update outer/inner flops */
 817
 818     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*219);
 819 }
 820 /*
 821  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_256_double
 822  * Electrostatics interaction: Ewald
 823  * VdW interaction:            LennardJones
 824  * Geometry:                   Water3-Particle
 825  * Calculate force/pot:        Force
 826  */
 827 void
 828 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_256_double
 829                     (t_nblist                    * gmx_restrict       nlist,
 830                      rvec                        * gmx_restrict          xx,
 831                      rvec                        * gmx_restrict          ff,
 832                      t_forcerec                  * gmx_restrict          fr,
 833                      t_mdatoms                   * gmx_restrict     mdatoms,
 834                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 835                      t_nrnb                      * gmx_restrict        nrnb)
 836 {
 837     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 838      * just 0 for non-waters.
 839      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 840      * jnr indices corresponding to data put in the four positions in the SIMD register.
 841      */
 842     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 843     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 844     int              jnrA,jnrB,jnrC,jnrD;
 845     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 846     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 847     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 848     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 849     real             rcutoff_scalar;
 850     real             *shiftvec,*fshift,*x,*f;
 851     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 852     real             scratch[4*DIM];
 853     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 854     real *           vdwioffsetptr0;
 855     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 856     real *           vdwioffsetptr1;
 857     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 858     real *           vdwioffsetptr2;
 859     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 860     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 861     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 862     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 863     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 864     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 865     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 866     real             *charge;
 867     int              nvdwtype;
 868     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 869     int              *vdwtype;
 870     real             *vdwparam;
 871     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 872     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 873     __m128i          ewitab;
 874     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 875     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 876     real             *ewtab;
 877     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 878     real             rswitch_scalar,d_scalar;
 879     __m256d          dummy_mask,cutoff_mask;
 880     __m128           tmpmask0,tmpmask1;
 881     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 882     __m256d          one     = _mm256_set1_pd(1.0);
 883     __m256d          two     = _mm256_set1_pd(2.0);
 884     x                = xx[0];
 885     f                = ff[0];
 886
 887     nri              = nlist->nri;
 888     iinr             = nlist->iinr;
 889     jindex           = nlist->jindex;
 890     jjnr             = nlist->jjnr;
 891     shiftidx         = nlist->shift;
 892     gid              = nlist->gid;
 893     shiftvec         = fr->shift_vec[0];
 894     fshift           = fr->fshift[0];
 895     facel            = _mm256_set1_pd(fr->epsfac);
 896     charge           = mdatoms->chargeA;
 897     nvdwtype         = fr->ntype;
 898     vdwparam         = fr->nbfp;
 899     vdwtype          = mdatoms->typeA;
 900
 901     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 902     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 903     beta2            = _mm256_mul_pd(beta,beta);
 904     beta3            = _mm256_mul_pd(beta,beta2);
 905
 906     ewtab            = fr->ic->tabq_coul_FDV0;
 907     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 908     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 909
 910     /* Setup water-specific parameters */
 911     inr              = nlist->iinr[0];
 912     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 913     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 914     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 915     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 916
 917     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 918     rcutoff_scalar   = fr->rcoulomb;
 919     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 920     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 921
 922     rswitch_scalar   = fr->rcoulomb_switch;
 923     rswitch          = _mm256_set1_pd(rswitch_scalar);
 924     /* Setup switch parameters */
 925     d_scalar         = rcutoff_scalar-rswitch_scalar;
 926     d                = _mm256_set1_pd(d_scalar);
 927     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 928     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 929     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 930     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 931     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 932     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 933
 934     /* Avoid stupid compiler warnings */
 935     jnrA = jnrB = jnrC = jnrD = 0;
 936     j_coord_offsetA = 0;
 937     j_coord_offsetB = 0;
 938     j_coord_offsetC = 0;
 939     j_coord_offsetD = 0;
 940
 941     outeriter        = 0;
 942     inneriter        = 0;
 943
 944     for(iidx=0;iidx<4*DIM;iidx++)
 945     {
 946         scratch[iidx] = 0.0;
 947     }
 948
 949     /* Start outer loop over neighborlists */
 950     for(iidx=0; iidx<nri; iidx++)
 951     {
 952         /* Load shift vector for this list */
 953         i_shift_offset   = DIM*shiftidx[iidx];
 954
 955         /* Load limits for loop over neighbors */
 956         j_index_start    = jindex[iidx];
 957         j_index_end      = jindex[iidx+1];
 958
 959         /* Get outer coordinate index */
 960         inr              = iinr[iidx];
 961         i_coord_offset   = DIM*inr;
 962
 963         /* Load i particle coords and add shift vector */
 964         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 965                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 966
 967         fix0             = _mm256_setzero_pd();
 968         fiy0             = _mm256_setzero_pd();
 969         fiz0             = _mm256_setzero_pd();
 970         fix1             = _mm256_setzero_pd();
 971         fiy1             = _mm256_setzero_pd();
 972         fiz1             = _mm256_setzero_pd();
 973         fix2             = _mm256_setzero_pd();
 974         fiy2             = _mm256_setzero_pd();
 975         fiz2             = _mm256_setzero_pd();
 976
 977         /* Start inner kernel loop */
 978         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 979         {
 980
 981             /* Get j neighbor index, and coordinate index */
 982             jnrA             = jjnr[jidx];
 983             jnrB             = jjnr[jidx+1];
 984             jnrC             = jjnr[jidx+2];
 985             jnrD             = jjnr[jidx+3];
 986             j_coord_offsetA  = DIM*jnrA;
 987             j_coord_offsetB  = DIM*jnrB;
 988             j_coord_offsetC  = DIM*jnrC;
 989             j_coord_offsetD  = DIM*jnrD;
 990
 991             /* load j atom coordinates */
 992             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 993                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 994                                                  &jx0,&jy0,&jz0);
 995
 996             /* Calculate displacement vector */
 997             dx00             = _mm256_sub_pd(ix0,jx0);
 998             dy00             = _mm256_sub_pd(iy0,jy0);
 999             dz00             = _mm256_sub_pd(iz0,jz0);
1000             dx10             = _mm256_sub_pd(ix1,jx0);
1001             dy10             = _mm256_sub_pd(iy1,jy0);
1002             dz10             = _mm256_sub_pd(iz1,jz0);
1003             dx20             = _mm256_sub_pd(ix2,jx0);
1004             dy20             = _mm256_sub_pd(iy2,jy0);
1005             dz20             = _mm256_sub_pd(iz2,jz0);
1006
1007             /* Calculate squared distance and things based on it */
1008             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1009             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1010             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1011
1012             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1013             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1014             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1015
1016             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1017             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1018             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1019
1020             /* Load parameters for j particles */
1021             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1022                                                                  charge+jnrC+0,charge+jnrD+0);
1023             vdwjidx0A        = 2*vdwtype[jnrA+0];
1024             vdwjidx0B        = 2*vdwtype[jnrB+0];
1025             vdwjidx0C        = 2*vdwtype[jnrC+0];
1026             vdwjidx0D        = 2*vdwtype[jnrD+0];
1027
1028             fjx0             = _mm256_setzero_pd();
1029             fjy0             = _mm256_setzero_pd();
1030             fjz0             = _mm256_setzero_pd();
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1037             {
1038
1039             r00              = _mm256_mul_pd(rsq00,rinv00);
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq00             = _mm256_mul_pd(iq0,jq0);
1043             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1044                                             vdwioffsetptr0+vdwjidx0B,
1045                                             vdwioffsetptr0+vdwjidx0C,
1046                                             vdwioffsetptr0+vdwjidx0D,
1047                                             &c6_00,&c12_00);
1048
1049             /* EWALD ELECTROSTATICS */
1050
1051             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1052             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1053             ewitab           = _mm256_cvttpd_epi32(ewrt);
1054             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1055             ewitab           = _mm_slli_epi32(ewitab,2);
1056             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1057             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1058             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1059             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1060             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1061             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1062             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1063             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
1064             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1065
1066             /* LENNARD-JONES DISPERSION/REPULSION */
1067
1068             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1069             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1070             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1071             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1072             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1073
1074             d                = _mm256_sub_pd(r00,rswitch);
1075             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1076             d2               = _mm256_mul_pd(d,d);
1077             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)))))));
1078
1079             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1080
1081             /* Evaluate switch function */
1082             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1083             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
1084             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1085             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1086
1087             fscal            = _mm256_add_pd(felec,fvdw);
1088
1089             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1090
1091             /* Calculate temporary vectorial force */
1092             tx               = _mm256_mul_pd(fscal,dx00);
1093             ty               = _mm256_mul_pd(fscal,dy00);
1094             tz               = _mm256_mul_pd(fscal,dz00);
1095
1096             /* Update vectorial force */
1097             fix0             = _mm256_add_pd(fix0,tx);
1098             fiy0             = _mm256_add_pd(fiy0,ty);
1099             fiz0             = _mm256_add_pd(fiz0,tz);
1100
1101             fjx0             = _mm256_add_pd(fjx0,tx);
1102             fjy0             = _mm256_add_pd(fjy0,ty);
1103             fjz0             = _mm256_add_pd(fjz0,tz);
1104
1105             }
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1112             {
1113
1114             r10              = _mm256_mul_pd(rsq10,rinv10);
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq10             = _mm256_mul_pd(iq1,jq0);
1118
1119             /* EWALD ELECTROSTATICS */
1120
1121             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1122             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1123             ewitab           = _mm256_cvttpd_epi32(ewrt);
1124             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1125             ewitab           = _mm_slli_epi32(ewitab,2);
1126             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1127             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1128             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1129             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1130             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1131             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1132             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1133             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1134             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1135
1136             d                = _mm256_sub_pd(r10,rswitch);
1137             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1138             d2               = _mm256_mul_pd(d,d);
1139             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)))))));
1140
1141             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1142
1143             /* Evaluate switch function */
1144             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1145             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1146             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1151
1152             /* Calculate temporary vectorial force */
1153             tx               = _mm256_mul_pd(fscal,dx10);
1154             ty               = _mm256_mul_pd(fscal,dy10);
1155             tz               = _mm256_mul_pd(fscal,dz10);
1156
1157             /* Update vectorial force */
1158             fix1             = _mm256_add_pd(fix1,tx);
1159             fiy1             = _mm256_add_pd(fiy1,ty);
1160             fiz1             = _mm256_add_pd(fiz1,tz);
1161
1162             fjx0             = _mm256_add_pd(fjx0,tx);
1163             fjy0             = _mm256_add_pd(fjy0,ty);
1164             fjz0             = _mm256_add_pd(fjz0,tz);
1165
1166             }
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1173             {
1174
1175             r20              = _mm256_mul_pd(rsq20,rinv20);
1176
1177             /* Compute parameters for interactions between i and j atoms */
1178             qq20             = _mm256_mul_pd(iq2,jq0);
1179
1180             /* EWALD ELECTROSTATICS */
1181
1182             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1183             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1184             ewitab           = _mm256_cvttpd_epi32(ewrt);
1185             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1186             ewitab           = _mm_slli_epi32(ewitab,2);
1187             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1188             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1189             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1190             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1191             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1192             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1193             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1194             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1195             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1196
1197             d                = _mm256_sub_pd(r20,rswitch);
1198             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1199             d2               = _mm256_mul_pd(d,d);
1200             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)))))));
1201
1202             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1203
1204             /* Evaluate switch function */
1205             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1206             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1207             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1208
1209             fscal            = felec;
1210
1211             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1212
1213             /* Calculate temporary vectorial force */
1214             tx               = _mm256_mul_pd(fscal,dx20);
1215             ty               = _mm256_mul_pd(fscal,dy20);
1216             tz               = _mm256_mul_pd(fscal,dz20);
1217
1218             /* Update vectorial force */
1219             fix2             = _mm256_add_pd(fix2,tx);
1220             fiy2             = _mm256_add_pd(fiy2,ty);
1221             fiz2             = _mm256_add_pd(fiz2,tz);
1222
1223             fjx0             = _mm256_add_pd(fjx0,tx);
1224             fjy0             = _mm256_add_pd(fjy0,ty);
1225             fjz0             = _mm256_add_pd(fjz0,tz);
1226
1227             }
1228
1229             fjptrA             = f+j_coord_offsetA;
1230             fjptrB             = f+j_coord_offsetB;
1231             fjptrC             = f+j_coord_offsetC;
1232             fjptrD             = f+j_coord_offsetD;
1233
1234             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1235
1236             /* Inner loop uses 204 flops */
1237         }
1238
1239         if(jidx<j_index_end)
1240         {
1241
1242             /* Get j neighbor index, and coordinate index */
1243             jnrlistA         = jjnr[jidx];
1244             jnrlistB         = jjnr[jidx+1];
1245             jnrlistC         = jjnr[jidx+2];
1246             jnrlistD         = jjnr[jidx+3];
1247             /* Sign of each element will be negative for non-real atoms.
1248              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1249              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1250              */
1251             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1252
1253             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1254             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1255             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1256
1257             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1258             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1259             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1260             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1261             j_coord_offsetA  = DIM*jnrA;
1262             j_coord_offsetB  = DIM*jnrB;
1263             j_coord_offsetC  = DIM*jnrC;
1264             j_coord_offsetD  = DIM*jnrD;
1265
1266             /* load j atom coordinates */
1267             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1268                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1269                                                  &jx0,&jy0,&jz0);
1270
1271             /* Calculate displacement vector */
1272             dx00             = _mm256_sub_pd(ix0,jx0);
1273             dy00             = _mm256_sub_pd(iy0,jy0);
1274             dz00             = _mm256_sub_pd(iz0,jz0);
1275             dx10             = _mm256_sub_pd(ix1,jx0);
1276             dy10             = _mm256_sub_pd(iy1,jy0);
1277             dz10             = _mm256_sub_pd(iz1,jz0);
1278             dx20             = _mm256_sub_pd(ix2,jx0);
1279             dy20             = _mm256_sub_pd(iy2,jy0);
1280             dz20             = _mm256_sub_pd(iz2,jz0);
1281
1282             /* Calculate squared distance and things based on it */
1283             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1284             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1285             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1286
1287             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1288             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1289             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1290
1291             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1292             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1293             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1294
1295             /* Load parameters for j particles */
1296             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1297                                                                  charge+jnrC+0,charge+jnrD+0);
1298             vdwjidx0A        = 2*vdwtype[jnrA+0];
1299             vdwjidx0B        = 2*vdwtype[jnrB+0];
1300             vdwjidx0C        = 2*vdwtype[jnrC+0];
1301             vdwjidx0D        = 2*vdwtype[jnrD+0];
1302
1303             fjx0             = _mm256_setzero_pd();
1304             fjy0             = _mm256_setzero_pd();
1305             fjz0             = _mm256_setzero_pd();
1306
1307             /**************************
1308              * CALCULATE INTERACTIONS *
1309              **************************/
1310
1311             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1312             {
1313
1314             r00              = _mm256_mul_pd(rsq00,rinv00);
1315             r00              = _mm256_andnot_pd(dummy_mask,r00);
1316
1317             /* Compute parameters for interactions between i and j atoms */
1318             qq00             = _mm256_mul_pd(iq0,jq0);
1319             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1320                                             vdwioffsetptr0+vdwjidx0B,
1321                                             vdwioffsetptr0+vdwjidx0C,
1322                                             vdwioffsetptr0+vdwjidx0D,
1323                                             &c6_00,&c12_00);
1324
1325             /* EWALD ELECTROSTATICS */
1326
1327             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1328             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1329             ewitab           = _mm256_cvttpd_epi32(ewrt);
1330             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1331             ewitab           = _mm_slli_epi32(ewitab,2);
1332             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1333             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1334             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1335             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1336             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1337             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1338             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1339             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
1340             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1341
1342             /* LENNARD-JONES DISPERSION/REPULSION */
1343
1344             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1345             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1346             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1347             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1348             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1349
1350             d                = _mm256_sub_pd(r00,rswitch);
1351             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1352             d2               = _mm256_mul_pd(d,d);
1353             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)))))));
1354
1355             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1356
1357             /* Evaluate switch function */
1358             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1359             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
1360             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1361             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1362
1363             fscal            = _mm256_add_pd(felec,fvdw);
1364
1365             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1366
1367             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1368
1369             /* Calculate temporary vectorial force */
1370             tx               = _mm256_mul_pd(fscal,dx00);
1371             ty               = _mm256_mul_pd(fscal,dy00);
1372             tz               = _mm256_mul_pd(fscal,dz00);
1373
1374             /* Update vectorial force */
1375             fix0             = _mm256_add_pd(fix0,tx);
1376             fiy0             = _mm256_add_pd(fiy0,ty);
1377             fiz0             = _mm256_add_pd(fiz0,tz);
1378
1379             fjx0             = _mm256_add_pd(fjx0,tx);
1380             fjy0             = _mm256_add_pd(fjy0,ty);
1381             fjz0             = _mm256_add_pd(fjz0,tz);
1382
1383             }
1384
1385             /**************************
1386              * CALCULATE INTERACTIONS *
1387              **************************/
1388
1389             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1390             {
1391
1392             r10              = _mm256_mul_pd(rsq10,rinv10);
1393             r10              = _mm256_andnot_pd(dummy_mask,r10);
1394
1395             /* Compute parameters for interactions between i and j atoms */
1396             qq10             = _mm256_mul_pd(iq1,jq0);
1397
1398             /* EWALD ELECTROSTATICS */
1399
1400             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1401             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1402             ewitab           = _mm256_cvttpd_epi32(ewrt);
1403             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1404             ewitab           = _mm_slli_epi32(ewitab,2);
1405             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1406             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1407             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1408             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1409             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1410             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1411             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1412             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1413             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1414
1415             d                = _mm256_sub_pd(r10,rswitch);
1416             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1417             d2               = _mm256_mul_pd(d,d);
1418             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)))))));
1419
1420             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1421
1422             /* Evaluate switch function */
1423             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1424             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1425             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1426
1427             fscal            = felec;
1428
1429             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1430
1431             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1432
1433             /* Calculate temporary vectorial force */
1434             tx               = _mm256_mul_pd(fscal,dx10);
1435             ty               = _mm256_mul_pd(fscal,dy10);
1436             tz               = _mm256_mul_pd(fscal,dz10);
1437
1438             /* Update vectorial force */
1439             fix1             = _mm256_add_pd(fix1,tx);
1440             fiy1             = _mm256_add_pd(fiy1,ty);
1441             fiz1             = _mm256_add_pd(fiz1,tz);
1442
1443             fjx0             = _mm256_add_pd(fjx0,tx);
1444             fjy0             = _mm256_add_pd(fjy0,ty);
1445             fjz0             = _mm256_add_pd(fjz0,tz);
1446
1447             }
1448
1449             /**************************
1450              * CALCULATE INTERACTIONS *
1451              **************************/
1452
1453             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1454             {
1455
1456             r20              = _mm256_mul_pd(rsq20,rinv20);
1457             r20              = _mm256_andnot_pd(dummy_mask,r20);
1458
1459             /* Compute parameters for interactions between i and j atoms */
1460             qq20             = _mm256_mul_pd(iq2,jq0);
1461
1462             /* EWALD ELECTROSTATICS */
1463
1464             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1465             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1466             ewitab           = _mm256_cvttpd_epi32(ewrt);
1467             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1468             ewitab           = _mm_slli_epi32(ewitab,2);
1469             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1470             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1471             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1472             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1473             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1474             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1475             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1476             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1477             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1478
1479             d                = _mm256_sub_pd(r20,rswitch);
1480             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1481             d2               = _mm256_mul_pd(d,d);
1482             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)))))));
1483
1484             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1485
1486             /* Evaluate switch function */
1487             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1488             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1489             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1490
1491             fscal            = felec;
1492
1493             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1494
1495             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1496
1497             /* Calculate temporary vectorial force */
1498             tx               = _mm256_mul_pd(fscal,dx20);
1499             ty               = _mm256_mul_pd(fscal,dy20);
1500             tz               = _mm256_mul_pd(fscal,dz20);
1501
1502             /* Update vectorial force */
1503             fix2             = _mm256_add_pd(fix2,tx);
1504             fiy2             = _mm256_add_pd(fiy2,ty);
1505             fiz2             = _mm256_add_pd(fiz2,tz);
1506
1507             fjx0             = _mm256_add_pd(fjx0,tx);
1508             fjy0             = _mm256_add_pd(fjy0,ty);
1509             fjz0             = _mm256_add_pd(fjz0,tz);
1510
1511             }
1512
1513             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1514             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1515             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1516             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1517
1518             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1519
1520             /* Inner loop uses 207 flops */
1521         }
1522
1523         /* End of innermost loop */
1524
1525         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1526                                                  f+i_coord_offset,fshift+i_shift_offset);
1527
1528         /* Increment number of inner iterations */
1529         inneriter                  += j_index_end - j_index_start;
1530
1531         /* Outer loop uses 18 flops */
1532     }
1533
1534     /* Increment number of outer iterations */
1535     outeriter        += nri;
1536
1537     /* Update outer/inner flops */
1538
1539     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*207);
1540 }