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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m128i          ewitab;
 106     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 108     real             *ewtab;
 109     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 110     real             rswitch_scalar,d_scalar;
 111     __m256d          dummy_mask,cutoff_mask;
 112     __m128           tmpmask0,tmpmask1;
 113     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 114     __m256d          one     = _mm256_set1_pd(1.0);
 115     __m256d          two     = _mm256_set1_pd(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_pd(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 134     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 135     beta2            = _mm256_mul_pd(beta,beta);
 136     beta3            = _mm256_mul_pd(beta,beta2);
 137
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 145     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 146     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 147     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 148
 149     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 150     rcutoff_scalar   = fr->rcoulomb;
 151     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 152     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 153
 154     rswitch_scalar   = fr->rcoulomb_switch;
 155     rswitch          = _mm256_set1_pd(rswitch_scalar);
 156     /* Setup switch parameters */
 157     d_scalar         = rcutoff_scalar-rswitch_scalar;
 158     d                = _mm256_set1_pd(d_scalar);
 159     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 160     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 161     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 162     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 163     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 164     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 165
 166     /* Avoid stupid compiler warnings */
 167     jnrA = jnrB = jnrC = jnrD = 0;
 168     j_coord_offsetA = 0;
 169     j_coord_offsetB = 0;
 170     j_coord_offsetC = 0;
 171     j_coord_offsetD = 0;
 172
 173     outeriter        = 0;
 174     inneriter        = 0;
 175
 176     for(iidx=0;iidx<4*DIM;iidx++)
 177     {
 178         scratch[iidx] = 0.0;
 179     }
 180
 181     /* Start outer loop over neighborlists */
 182     for(iidx=0; iidx<nri; iidx++)
 183     {
 184         /* Load shift vector for this list */
 185         i_shift_offset   = DIM*shiftidx[iidx];
 186
 187         /* Load limits for loop over neighbors */
 188         j_index_start    = jindex[iidx];
 189         j_index_end      = jindex[iidx+1];
 190
 191         /* Get outer coordinate index */
 192         inr              = iinr[iidx];
 193         i_coord_offset   = DIM*inr;
 194
 195         /* Load i particle coords and add shift vector */
 196         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 197                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 198
 199         fix0             = _mm256_setzero_pd();
 200         fiy0             = _mm256_setzero_pd();
 201         fiz0             = _mm256_setzero_pd();
 202         fix1             = _mm256_setzero_pd();
 203         fiy1             = _mm256_setzero_pd();
 204         fiz1             = _mm256_setzero_pd();
 205         fix2             = _mm256_setzero_pd();
 206         fiy2             = _mm256_setzero_pd();
 207         fiz2             = _mm256_setzero_pd();
 208
 209         /* Reset potential sums */
 210         velecsum         = _mm256_setzero_pd();
 211         vvdwsum          = _mm256_setzero_pd();
 212
 213         /* Start inner kernel loop */
 214         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 215         {
 216
 217             /* Get j neighbor index, and coordinate index */
 218             jnrA             = jjnr[jidx];
 219             jnrB             = jjnr[jidx+1];
 220             jnrC             = jjnr[jidx+2];
 221             jnrD             = jjnr[jidx+3];
 222             j_coord_offsetA  = DIM*jnrA;
 223             j_coord_offsetB  = DIM*jnrB;
 224             j_coord_offsetC  = DIM*jnrC;
 225             j_coord_offsetD  = DIM*jnrD;
 226
 227             /* load j atom coordinates */
 228             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 229                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 230                                                  &jx0,&jy0,&jz0);
 231
 232             /* Calculate displacement vector */
 233             dx00             = _mm256_sub_pd(ix0,jx0);
 234             dy00             = _mm256_sub_pd(iy0,jy0);
 235             dz00             = _mm256_sub_pd(iz0,jz0);
 236             dx10             = _mm256_sub_pd(ix1,jx0);
 237             dy10             = _mm256_sub_pd(iy1,jy0);
 238             dz10             = _mm256_sub_pd(iz1,jz0);
 239             dx20             = _mm256_sub_pd(ix2,jx0);
 240             dy20             = _mm256_sub_pd(iy2,jy0);
 241             dz20             = _mm256_sub_pd(iz2,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 245             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 246             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 247
 248             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 249             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 250             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 251
 252             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 253             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 254             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 258                                                                  charge+jnrC+0,charge+jnrD+0);
 259             vdwjidx0A        = 2*vdwtype[jnrA+0];
 260             vdwjidx0B        = 2*vdwtype[jnrB+0];
 261             vdwjidx0C        = 2*vdwtype[jnrC+0];
 262             vdwjidx0D        = 2*vdwtype[jnrD+0];
 263
 264             fjx0             = _mm256_setzero_pd();
 265             fjy0             = _mm256_setzero_pd();
 266             fjz0             = _mm256_setzero_pd();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 273             {
 274
 275             r00              = _mm256_mul_pd(rsq00,rinv00);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq00             = _mm256_mul_pd(iq0,jq0);
 279             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 280                                             vdwioffsetptr0+vdwjidx0B,
 281                                             vdwioffsetptr0+vdwjidx0C,
 282                                             vdwioffsetptr0+vdwjidx0D,
 283                                             &c6_00,&c12_00);
 284
 285             /* EWALD ELECTROSTATICS */
 286
 287             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 288             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 289             ewitab           = _mm256_cvttpd_epi32(ewrt);
 290             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 291             ewitab           = _mm_slli_epi32(ewitab,2);
 292             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 293             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 294             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 295             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 296             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 297             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 298             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 299             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 300             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 301
 302             /* LENNARD-JONES DISPERSION/REPULSION */
 303
 304             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 305             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 306             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 307             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 308             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 309
 310             d                = _mm256_sub_pd(r00,rswitch);
 311             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 312             d2               = _mm256_mul_pd(d,d);
 313             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)))))));
 314
 315             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 316
 317             /* Evaluate switch function */
 318             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 319             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 320             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 321             velec            = _mm256_mul_pd(velec,sw);
 322             vvdw             = _mm256_mul_pd(vvdw,sw);
 323             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm256_and_pd(velec,cutoff_mask);
 327             velecsum         = _mm256_add_pd(velecsum,velec);
 328             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 329             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 330
 331             fscal            = _mm256_add_pd(felec,fvdw);
 332
 333             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_pd(fscal,dx00);
 337             ty               = _mm256_mul_pd(fscal,dy00);
 338             tz               = _mm256_mul_pd(fscal,dz00);
 339
 340             /* Update vectorial force */
 341             fix0             = _mm256_add_pd(fix0,tx);
 342             fiy0             = _mm256_add_pd(fiy0,ty);
 343             fiz0             = _mm256_add_pd(fiz0,tz);
 344
 345             fjx0             = _mm256_add_pd(fjx0,tx);
 346             fjy0             = _mm256_add_pd(fjy0,ty);
 347             fjz0             = _mm256_add_pd(fjz0,tz);
 348
 349             }
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 356             {
 357
 358             r10              = _mm256_mul_pd(rsq10,rinv10);
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq10             = _mm256_mul_pd(iq1,jq0);
 362
 363             /* EWALD ELECTROSTATICS */
 364
 365             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 366             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 367             ewitab           = _mm256_cvttpd_epi32(ewrt);
 368             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 369             ewitab           = _mm_slli_epi32(ewitab,2);
 370             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 371             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 372             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 373             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 374             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 375             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 376             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 377             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 378             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 379
 380             d                = _mm256_sub_pd(r10,rswitch);
 381             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 382             d2               = _mm256_mul_pd(d,d);
 383             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)))))));
 384
 385             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 386
 387             /* Evaluate switch function */
 388             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 389             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 390             velec            = _mm256_mul_pd(velec,sw);
 391             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velec            = _mm256_and_pd(velec,cutoff_mask);
 395             velecsum         = _mm256_add_pd(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm256_mul_pd(fscal,dx10);
 403             ty               = _mm256_mul_pd(fscal,dy10);
 404             tz               = _mm256_mul_pd(fscal,dz10);
 405
 406             /* Update vectorial force */
 407             fix1             = _mm256_add_pd(fix1,tx);
 408             fiy1             = _mm256_add_pd(fiy1,ty);
 409             fiz1             = _mm256_add_pd(fiz1,tz);
 410
 411             fjx0             = _mm256_add_pd(fjx0,tx);
 412             fjy0             = _mm256_add_pd(fjy0,ty);
 413             fjz0             = _mm256_add_pd(fjz0,tz);
 414
 415             }
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 422             {
 423
 424             r20              = _mm256_mul_pd(rsq20,rinv20);
 425
 426             /* Compute parameters for interactions between i and j atoms */
 427             qq20             = _mm256_mul_pd(iq2,jq0);
 428
 429             /* EWALD ELECTROSTATICS */
 430
 431             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 432             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 433             ewitab           = _mm256_cvttpd_epi32(ewrt);
 434             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 435             ewitab           = _mm_slli_epi32(ewitab,2);
 436             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 437             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 438             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 439             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 440             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 441             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 442             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 443             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 444             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 445
 446             d                = _mm256_sub_pd(r20,rswitch);
 447             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 448             d2               = _mm256_mul_pd(d,d);
 449             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)))))));
 450
 451             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 452
 453             /* Evaluate switch function */
 454             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 455             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 456             velec            = _mm256_mul_pd(velec,sw);
 457             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velec            = _mm256_and_pd(velec,cutoff_mask);
 461             velecsum         = _mm256_add_pd(velecsum,velec);
 462
 463             fscal            = felec;
 464
 465             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 466
 467             /* Calculate temporary vectorial force */
 468             tx               = _mm256_mul_pd(fscal,dx20);
 469             ty               = _mm256_mul_pd(fscal,dy20);
 470             tz               = _mm256_mul_pd(fscal,dz20);
 471
 472             /* Update vectorial force */
 473             fix2             = _mm256_add_pd(fix2,tx);
 474             fiy2             = _mm256_add_pd(fiy2,ty);
 475             fiz2             = _mm256_add_pd(fiz2,tz);
 476
 477             fjx0             = _mm256_add_pd(fjx0,tx);
 478             fjy0             = _mm256_add_pd(fjy0,ty);
 479             fjz0             = _mm256_add_pd(fjz0,tz);
 480
 481             }
 482
 483             fjptrA             = f+j_coord_offsetA;
 484             fjptrB             = f+j_coord_offsetB;
 485             fjptrC             = f+j_coord_offsetC;
 486             fjptrD             = f+j_coord_offsetD;
 487
 488             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 489
 490             /* Inner loop uses 216 flops */
 491         }
 492
 493         if(jidx<j_index_end)
 494         {
 495
 496             /* Get j neighbor index, and coordinate index */
 497             jnrlistA         = jjnr[jidx];
 498             jnrlistB         = jjnr[jidx+1];
 499             jnrlistC         = jjnr[jidx+2];
 500             jnrlistD         = jjnr[jidx+3];
 501             /* Sign of each element will be negative for non-real atoms.
 502              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 503              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 504              */
 505             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 506
 507             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 508             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 509             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 510
 511             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 512             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 513             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 514             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 515             j_coord_offsetA  = DIM*jnrA;
 516             j_coord_offsetB  = DIM*jnrB;
 517             j_coord_offsetC  = DIM*jnrC;
 518             j_coord_offsetD  = DIM*jnrD;
 519
 520             /* load j atom coordinates */
 521             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 522                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 523                                                  &jx0,&jy0,&jz0);
 524
 525             /* Calculate displacement vector */
 526             dx00             = _mm256_sub_pd(ix0,jx0);
 527             dy00             = _mm256_sub_pd(iy0,jy0);
 528             dz00             = _mm256_sub_pd(iz0,jz0);
 529             dx10             = _mm256_sub_pd(ix1,jx0);
 530             dy10             = _mm256_sub_pd(iy1,jy0);
 531             dz10             = _mm256_sub_pd(iz1,jz0);
 532             dx20             = _mm256_sub_pd(ix2,jx0);
 533             dy20             = _mm256_sub_pd(iy2,jy0);
 534             dz20             = _mm256_sub_pd(iz2,jz0);
 535
 536             /* Calculate squared distance and things based on it */
 537             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 538             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 539             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 540
 541             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 542             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 543             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 544
 545             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 546             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 547             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 548
 549             /* Load parameters for j particles */
 550             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 551                                                                  charge+jnrC+0,charge+jnrD+0);
 552             vdwjidx0A        = 2*vdwtype[jnrA+0];
 553             vdwjidx0B        = 2*vdwtype[jnrB+0];
 554             vdwjidx0C        = 2*vdwtype[jnrC+0];
 555             vdwjidx0D        = 2*vdwtype[jnrD+0];
 556
 557             fjx0             = _mm256_setzero_pd();
 558             fjy0             = _mm256_setzero_pd();
 559             fjz0             = _mm256_setzero_pd();
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 566             {
 567
 568             r00              = _mm256_mul_pd(rsq00,rinv00);
 569             r00              = _mm256_andnot_pd(dummy_mask,r00);
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             qq00             = _mm256_mul_pd(iq0,jq0);
 573             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 574                                             vdwioffsetptr0+vdwjidx0B,
 575                                             vdwioffsetptr0+vdwjidx0C,
 576                                             vdwioffsetptr0+vdwjidx0D,
 577                                             &c6_00,&c12_00);
 578
 579             /* EWALD ELECTROSTATICS */
 580
 581             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 582             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 583             ewitab           = _mm256_cvttpd_epi32(ewrt);
 584             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 585             ewitab           = _mm_slli_epi32(ewitab,2);
 586             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 587             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 588             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 589             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 590             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 591             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 592             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 593             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 594             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 595
 596             /* LENNARD-JONES DISPERSION/REPULSION */
 597
 598             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 599             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 600             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 601             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 602             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 603
 604             d                = _mm256_sub_pd(r00,rswitch);
 605             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 606             d2               = _mm256_mul_pd(d,d);
 607             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)))))));
 608
 609             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 610
 611             /* Evaluate switch function */
 612             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 613             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
 614             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 615             velec            = _mm256_mul_pd(velec,sw);
 616             vvdw             = _mm256_mul_pd(vvdw,sw);
 617             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm256_and_pd(velec,cutoff_mask);
 621             velec            = _mm256_andnot_pd(dummy_mask,velec);
 622             velecsum         = _mm256_add_pd(velecsum,velec);
 623             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 624             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 625             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 626
 627             fscal            = _mm256_add_pd(felec,fvdw);
 628
 629             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 630
 631             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm256_mul_pd(fscal,dx00);
 635             ty               = _mm256_mul_pd(fscal,dy00);
 636             tz               = _mm256_mul_pd(fscal,dz00);
 637
 638             /* Update vectorial force */
 639             fix0             = _mm256_add_pd(fix0,tx);
 640             fiy0             = _mm256_add_pd(fiy0,ty);
 641             fiz0             = _mm256_add_pd(fiz0,tz);
 642
 643             fjx0             = _mm256_add_pd(fjx0,tx);
 644             fjy0             = _mm256_add_pd(fjy0,ty);
 645             fjz0             = _mm256_add_pd(fjz0,tz);
 646
 647             }
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 654             {
 655
 656             r10              = _mm256_mul_pd(rsq10,rinv10);
 657             r10              = _mm256_andnot_pd(dummy_mask,r10);
 658
 659             /* Compute parameters for interactions between i and j atoms */
 660             qq10             = _mm256_mul_pd(iq1,jq0);
 661
 662             /* EWALD ELECTROSTATICS */
 663
 664             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 665             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 666             ewitab           = _mm256_cvttpd_epi32(ewrt);
 667             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 668             ewitab           = _mm_slli_epi32(ewitab,2);
 669             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 670             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 671             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 672             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 673             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 674             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 675             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 676             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 677             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 678
 679             d                = _mm256_sub_pd(r10,rswitch);
 680             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 681             d2               = _mm256_mul_pd(d,d);
 682             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)))))));
 683
 684             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 685
 686             /* Evaluate switch function */
 687             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 688             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 689             velec            = _mm256_mul_pd(velec,sw);
 690             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 691
 692             /* Update potential sum for this i atom from the interaction with this j atom. */
 693             velec            = _mm256_and_pd(velec,cutoff_mask);
 694             velec            = _mm256_andnot_pd(dummy_mask,velec);
 695             velecsum         = _mm256_add_pd(velecsum,velec);
 696
 697             fscal            = felec;
 698
 699             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 700
 701             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 702
 703             /* Calculate temporary vectorial force */
 704             tx               = _mm256_mul_pd(fscal,dx10);
 705             ty               = _mm256_mul_pd(fscal,dy10);
 706             tz               = _mm256_mul_pd(fscal,dz10);
 707
 708             /* Update vectorial force */
 709             fix1             = _mm256_add_pd(fix1,tx);
 710             fiy1             = _mm256_add_pd(fiy1,ty);
 711             fiz1             = _mm256_add_pd(fiz1,tz);
 712
 713             fjx0             = _mm256_add_pd(fjx0,tx);
 714             fjy0             = _mm256_add_pd(fjy0,ty);
 715             fjz0             = _mm256_add_pd(fjz0,tz);
 716
 717             }
 718
 719             /**************************
 720              * CALCULATE INTERACTIONS *
 721              **************************/
 722
 723             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 724             {
 725
 726             r20              = _mm256_mul_pd(rsq20,rinv20);
 727             r20              = _mm256_andnot_pd(dummy_mask,r20);
 728
 729             /* Compute parameters for interactions between i and j atoms */
 730             qq20             = _mm256_mul_pd(iq2,jq0);
 731
 732             /* EWALD ELECTROSTATICS */
 733
 734             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 735             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 736             ewitab           = _mm256_cvttpd_epi32(ewrt);
 737             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 738             ewitab           = _mm_slli_epi32(ewitab,2);
 739             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 740             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 741             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 742             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 743             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 744             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 745             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 746             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 747             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 748
 749             d                = _mm256_sub_pd(r20,rswitch);
 750             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 751             d2               = _mm256_mul_pd(d,d);
 752             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)))))));
 753
 754             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 755
 756             /* Evaluate switch function */
 757             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 758             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 759             velec            = _mm256_mul_pd(velec,sw);
 760             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 761
 762             /* Update potential sum for this i atom from the interaction with this j atom. */
 763             velec            = _mm256_and_pd(velec,cutoff_mask);
 764             velec            = _mm256_andnot_pd(dummy_mask,velec);
 765             velecsum         = _mm256_add_pd(velecsum,velec);
 766
 767             fscal            = felec;
 768
 769             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 770
 771             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 772
 773             /* Calculate temporary vectorial force */
 774             tx               = _mm256_mul_pd(fscal,dx20);
 775             ty               = _mm256_mul_pd(fscal,dy20);
 776             tz               = _mm256_mul_pd(fscal,dz20);
 777
 778             /* Update vectorial force */
 779             fix2             = _mm256_add_pd(fix2,tx);
 780             fiy2             = _mm256_add_pd(fiy2,ty);
 781             fiz2             = _mm256_add_pd(fiz2,tz);
 782
 783             fjx0             = _mm256_add_pd(fjx0,tx);
 784             fjy0             = _mm256_add_pd(fjy0,ty);
 785             fjz0             = _mm256_add_pd(fjz0,tz);
 786
 787             }
 788
 789             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 790             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 791             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 792             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 793
 794             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 795
 796             /* Inner loop uses 219 flops */
 797         }
 798
 799         /* End of innermost loop */
 800
 801         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 802                                                  f+i_coord_offset,fshift+i_shift_offset);
 803
 804         ggid                        = gid[iidx];
 805         /* Update potential energies */
 806         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 807         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 808
 809         /* Increment number of inner iterations */
 810         inneriter                  += j_index_end - j_index_start;
 811
 812         /* Outer loop uses 20 flops */
 813     }
 814
 815     /* Increment number of outer iterations */
 816     outeriter        += nri;
 817
 818     /* Update outer/inner flops */
 819
 820     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*219);
 821 }
 822 /*
 823  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_256_double
 824  * Electrostatics interaction: Ewald
 825  * VdW interaction:            LennardJones
 826  * Geometry:                   Water3-Particle
 827  * Calculate force/pot:        Force
 828  */
 829 void
 830 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_256_double
 831                     (t_nblist                    * gmx_restrict       nlist,
 832                      rvec                        * gmx_restrict          xx,
 833                      rvec                        * gmx_restrict          ff,
 834                      t_forcerec                  * gmx_restrict          fr,
 835                      t_mdatoms                   * gmx_restrict     mdatoms,
 836                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 837                      t_nrnb                      * gmx_restrict        nrnb)
 838 {
 839     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 840      * just 0 for non-waters.
 841      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 842      * jnr indices corresponding to data put in the four positions in the SIMD register.
 843      */
 844     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 845     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 846     int              jnrA,jnrB,jnrC,jnrD;
 847     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 848     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 849     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 850     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 851     real             rcutoff_scalar;
 852     real             *shiftvec,*fshift,*x,*f;
 853     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 854     real             scratch[4*DIM];
 855     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 856     real *           vdwioffsetptr0;
 857     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 858     real *           vdwioffsetptr1;
 859     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 860     real *           vdwioffsetptr2;
 861     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 862     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 863     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 864     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 865     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 866     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 867     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 868     real             *charge;
 869     int              nvdwtype;
 870     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 871     int              *vdwtype;
 872     real             *vdwparam;
 873     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 874     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 875     __m128i          ewitab;
 876     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 877     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 878     real             *ewtab;
 879     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 880     real             rswitch_scalar,d_scalar;
 881     __m256d          dummy_mask,cutoff_mask;
 882     __m128           tmpmask0,tmpmask1;
 883     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 884     __m256d          one     = _mm256_set1_pd(1.0);
 885     __m256d          two     = _mm256_set1_pd(2.0);
 886     x                = xx[0];
 887     f                = ff[0];
 888
 889     nri              = nlist->nri;
 890     iinr             = nlist->iinr;
 891     jindex           = nlist->jindex;
 892     jjnr             = nlist->jjnr;
 893     shiftidx         = nlist->shift;
 894     gid              = nlist->gid;
 895     shiftvec         = fr->shift_vec[0];
 896     fshift           = fr->fshift[0];
 897     facel            = _mm256_set1_pd(fr->epsfac);
 898     charge           = mdatoms->chargeA;
 899     nvdwtype         = fr->ntype;
 900     vdwparam         = fr->nbfp;
 901     vdwtype          = mdatoms->typeA;
 902
 903     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 904     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 905     beta2            = _mm256_mul_pd(beta,beta);
 906     beta3            = _mm256_mul_pd(beta,beta2);
 907
 908     ewtab            = fr->ic->tabq_coul_FDV0;
 909     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 910     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 911
 912     /* Setup water-specific parameters */
 913     inr              = nlist->iinr[0];
 914     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 915     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 916     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 917     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 918
 919     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 920     rcutoff_scalar   = fr->rcoulomb;
 921     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 922     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 923
 924     rswitch_scalar   = fr->rcoulomb_switch;
 925     rswitch          = _mm256_set1_pd(rswitch_scalar);
 926     /* Setup switch parameters */
 927     d_scalar         = rcutoff_scalar-rswitch_scalar;
 928     d                = _mm256_set1_pd(d_scalar);
 929     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 930     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 931     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 932     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 933     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 934     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 935
 936     /* Avoid stupid compiler warnings */
 937     jnrA = jnrB = jnrC = jnrD = 0;
 938     j_coord_offsetA = 0;
 939     j_coord_offsetB = 0;
 940     j_coord_offsetC = 0;
 941     j_coord_offsetD = 0;
 942
 943     outeriter        = 0;
 944     inneriter        = 0;
 945
 946     for(iidx=0;iidx<4*DIM;iidx++)
 947     {
 948         scratch[iidx] = 0.0;
 949     }
 950
 951     /* Start outer loop over neighborlists */
 952     for(iidx=0; iidx<nri; iidx++)
 953     {
 954         /* Load shift vector for this list */
 955         i_shift_offset   = DIM*shiftidx[iidx];
 956
 957         /* Load limits for loop over neighbors */
 958         j_index_start    = jindex[iidx];
 959         j_index_end      = jindex[iidx+1];
 960
 961         /* Get outer coordinate index */
 962         inr              = iinr[iidx];
 963         i_coord_offset   = DIM*inr;
 964
 965         /* Load i particle coords and add shift vector */
 966         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 967                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 968
 969         fix0             = _mm256_setzero_pd();
 970         fiy0             = _mm256_setzero_pd();
 971         fiz0             = _mm256_setzero_pd();
 972         fix1             = _mm256_setzero_pd();
 973         fiy1             = _mm256_setzero_pd();
 974         fiz1             = _mm256_setzero_pd();
 975         fix2             = _mm256_setzero_pd();
 976         fiy2             = _mm256_setzero_pd();
 977         fiz2             = _mm256_setzero_pd();
 978
 979         /* Start inner kernel loop */
 980         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 981         {
 982
 983             /* Get j neighbor index, and coordinate index */
 984             jnrA             = jjnr[jidx];
 985             jnrB             = jjnr[jidx+1];
 986             jnrC             = jjnr[jidx+2];
 987             jnrD             = jjnr[jidx+3];
 988             j_coord_offsetA  = DIM*jnrA;
 989             j_coord_offsetB  = DIM*jnrB;
 990             j_coord_offsetC  = DIM*jnrC;
 991             j_coord_offsetD  = DIM*jnrD;
 992
 993             /* load j atom coordinates */
 994             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 995                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 996                                                  &jx0,&jy0,&jz0);
 997
 998             /* Calculate displacement vector */
 999             dx00             = _mm256_sub_pd(ix0,jx0);
1000             dy00             = _mm256_sub_pd(iy0,jy0);
1001             dz00             = _mm256_sub_pd(iz0,jz0);
1002             dx10             = _mm256_sub_pd(ix1,jx0);
1003             dy10             = _mm256_sub_pd(iy1,jy0);
1004             dz10             = _mm256_sub_pd(iz1,jz0);
1005             dx20             = _mm256_sub_pd(ix2,jx0);
1006             dy20             = _mm256_sub_pd(iy2,jy0);
1007             dz20             = _mm256_sub_pd(iz2,jz0);
1008
1009             /* Calculate squared distance and things based on it */
1010             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1011             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1012             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1013
1014             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1015             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1016             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1017
1018             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1019             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1020             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1021
1022             /* Load parameters for j particles */
1023             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1024                                                                  charge+jnrC+0,charge+jnrD+0);
1025             vdwjidx0A        = 2*vdwtype[jnrA+0];
1026             vdwjidx0B        = 2*vdwtype[jnrB+0];
1027             vdwjidx0C        = 2*vdwtype[jnrC+0];
1028             vdwjidx0D        = 2*vdwtype[jnrD+0];
1029
1030             fjx0             = _mm256_setzero_pd();
1031             fjy0             = _mm256_setzero_pd();
1032             fjz0             = _mm256_setzero_pd();
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1039             {
1040
1041             r00              = _mm256_mul_pd(rsq00,rinv00);
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq00             = _mm256_mul_pd(iq0,jq0);
1045             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1046                                             vdwioffsetptr0+vdwjidx0B,
1047                                             vdwioffsetptr0+vdwjidx0C,
1048                                             vdwioffsetptr0+vdwjidx0D,
1049                                             &c6_00,&c12_00);
1050
1051             /* EWALD ELECTROSTATICS */
1052
1053             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1054             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1055             ewitab           = _mm256_cvttpd_epi32(ewrt);
1056             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1057             ewitab           = _mm_slli_epi32(ewitab,2);
1058             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1059             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1060             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1061             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1062             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1063             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1064             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1065             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
1066             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1067
1068             /* LENNARD-JONES DISPERSION/REPULSION */
1069
1070             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1071             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1072             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1073             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1074             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1075
1076             d                = _mm256_sub_pd(r00,rswitch);
1077             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1078             d2               = _mm256_mul_pd(d,d);
1079             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)))))));
1080
1081             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1082
1083             /* Evaluate switch function */
1084             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1085             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
1086             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1087             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1088
1089             fscal            = _mm256_add_pd(felec,fvdw);
1090
1091             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1092
1093             /* Calculate temporary vectorial force */
1094             tx               = _mm256_mul_pd(fscal,dx00);
1095             ty               = _mm256_mul_pd(fscal,dy00);
1096             tz               = _mm256_mul_pd(fscal,dz00);
1097
1098             /* Update vectorial force */
1099             fix0             = _mm256_add_pd(fix0,tx);
1100             fiy0             = _mm256_add_pd(fiy0,ty);
1101             fiz0             = _mm256_add_pd(fiz0,tz);
1102
1103             fjx0             = _mm256_add_pd(fjx0,tx);
1104             fjy0             = _mm256_add_pd(fjy0,ty);
1105             fjz0             = _mm256_add_pd(fjz0,tz);
1106
1107             }
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1114             {
1115
1116             r10              = _mm256_mul_pd(rsq10,rinv10);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq10             = _mm256_mul_pd(iq1,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1124             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1125             ewitab           = _mm256_cvttpd_epi32(ewrt);
1126             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1127             ewitab           = _mm_slli_epi32(ewitab,2);
1128             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1129             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1130             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1131             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1132             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1133             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1134             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1135             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1136             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1137
1138             d                = _mm256_sub_pd(r10,rswitch);
1139             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1140             d2               = _mm256_mul_pd(d,d);
1141             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)))))));
1142
1143             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1144
1145             /* Evaluate switch function */
1146             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1147             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1148             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1149
1150             fscal            = felec;
1151
1152             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm256_mul_pd(fscal,dx10);
1156             ty               = _mm256_mul_pd(fscal,dy10);
1157             tz               = _mm256_mul_pd(fscal,dz10);
1158
1159             /* Update vectorial force */
1160             fix1             = _mm256_add_pd(fix1,tx);
1161             fiy1             = _mm256_add_pd(fiy1,ty);
1162             fiz1             = _mm256_add_pd(fiz1,tz);
1163
1164             fjx0             = _mm256_add_pd(fjx0,tx);
1165             fjy0             = _mm256_add_pd(fjy0,ty);
1166             fjz0             = _mm256_add_pd(fjz0,tz);
1167
1168             }
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1175             {
1176
1177             r20              = _mm256_mul_pd(rsq20,rinv20);
1178
1179             /* Compute parameters for interactions between i and j atoms */
1180             qq20             = _mm256_mul_pd(iq2,jq0);
1181
1182             /* EWALD ELECTROSTATICS */
1183
1184             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1185             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1186             ewitab           = _mm256_cvttpd_epi32(ewrt);
1187             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1188             ewitab           = _mm_slli_epi32(ewitab,2);
1189             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1190             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1191             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1192             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1193             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1194             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1195             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1196             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1197             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1198
1199             d                = _mm256_sub_pd(r20,rswitch);
1200             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1201             d2               = _mm256_mul_pd(d,d);
1202             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)))))));
1203
1204             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1205
1206             /* Evaluate switch function */
1207             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1208             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1209             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1210
1211             fscal            = felec;
1212
1213             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1214
1215             /* Calculate temporary vectorial force */
1216             tx               = _mm256_mul_pd(fscal,dx20);
1217             ty               = _mm256_mul_pd(fscal,dy20);
1218             tz               = _mm256_mul_pd(fscal,dz20);
1219
1220             /* Update vectorial force */
1221             fix2             = _mm256_add_pd(fix2,tx);
1222             fiy2             = _mm256_add_pd(fiy2,ty);
1223             fiz2             = _mm256_add_pd(fiz2,tz);
1224
1225             fjx0             = _mm256_add_pd(fjx0,tx);
1226             fjy0             = _mm256_add_pd(fjy0,ty);
1227             fjz0             = _mm256_add_pd(fjz0,tz);
1228
1229             }
1230
1231             fjptrA             = f+j_coord_offsetA;
1232             fjptrB             = f+j_coord_offsetB;
1233             fjptrC             = f+j_coord_offsetC;
1234             fjptrD             = f+j_coord_offsetD;
1235
1236             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1237
1238             /* Inner loop uses 204 flops */
1239         }
1240
1241         if(jidx<j_index_end)
1242         {
1243
1244             /* Get j neighbor index, and coordinate index */
1245             jnrlistA         = jjnr[jidx];
1246             jnrlistB         = jjnr[jidx+1];
1247             jnrlistC         = jjnr[jidx+2];
1248             jnrlistD         = jjnr[jidx+3];
1249             /* Sign of each element will be negative for non-real atoms.
1250              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1251              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1252              */
1253             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1254
1255             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1256             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1257             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1258
1259             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1260             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1261             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1262             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1263             j_coord_offsetA  = DIM*jnrA;
1264             j_coord_offsetB  = DIM*jnrB;
1265             j_coord_offsetC  = DIM*jnrC;
1266             j_coord_offsetD  = DIM*jnrD;
1267
1268             /* load j atom coordinates */
1269             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1270                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1271                                                  &jx0,&jy0,&jz0);
1272
1273             /* Calculate displacement vector */
1274             dx00             = _mm256_sub_pd(ix0,jx0);
1275             dy00             = _mm256_sub_pd(iy0,jy0);
1276             dz00             = _mm256_sub_pd(iz0,jz0);
1277             dx10             = _mm256_sub_pd(ix1,jx0);
1278             dy10             = _mm256_sub_pd(iy1,jy0);
1279             dz10             = _mm256_sub_pd(iz1,jz0);
1280             dx20             = _mm256_sub_pd(ix2,jx0);
1281             dy20             = _mm256_sub_pd(iy2,jy0);
1282             dz20             = _mm256_sub_pd(iz2,jz0);
1283
1284             /* Calculate squared distance and things based on it */
1285             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1286             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1287             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1288
1289             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1290             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1291             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1292
1293             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1294             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1295             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1296
1297             /* Load parameters for j particles */
1298             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1299                                                                  charge+jnrC+0,charge+jnrD+0);
1300             vdwjidx0A        = 2*vdwtype[jnrA+0];
1301             vdwjidx0B        = 2*vdwtype[jnrB+0];
1302             vdwjidx0C        = 2*vdwtype[jnrC+0];
1303             vdwjidx0D        = 2*vdwtype[jnrD+0];
1304
1305             fjx0             = _mm256_setzero_pd();
1306             fjy0             = _mm256_setzero_pd();
1307             fjz0             = _mm256_setzero_pd();
1308
1309             /**************************
1310              * CALCULATE INTERACTIONS *
1311              **************************/
1312
1313             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1314             {
1315
1316             r00              = _mm256_mul_pd(rsq00,rinv00);
1317             r00              = _mm256_andnot_pd(dummy_mask,r00);
1318
1319             /* Compute parameters for interactions between i and j atoms */
1320             qq00             = _mm256_mul_pd(iq0,jq0);
1321             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1322                                             vdwioffsetptr0+vdwjidx0B,
1323                                             vdwioffsetptr0+vdwjidx0C,
1324                                             vdwioffsetptr0+vdwjidx0D,
1325                                             &c6_00,&c12_00);
1326
1327             /* EWALD ELECTROSTATICS */
1328
1329             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1330             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1331             ewitab           = _mm256_cvttpd_epi32(ewrt);
1332             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1333             ewitab           = _mm_slli_epi32(ewitab,2);
1334             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1335             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1336             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1337             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1338             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1339             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1340             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1341             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
1342             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1343
1344             /* LENNARD-JONES DISPERSION/REPULSION */
1345
1346             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1347             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1348             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1349             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1350             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1351
1352             d                = _mm256_sub_pd(r00,rswitch);
1353             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1354             d2               = _mm256_mul_pd(d,d);
1355             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)))))));
1356
1357             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1358
1359             /* Evaluate switch function */
1360             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1361             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
1362             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1363             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1364
1365             fscal            = _mm256_add_pd(felec,fvdw);
1366
1367             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1368
1369             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1370
1371             /* Calculate temporary vectorial force */
1372             tx               = _mm256_mul_pd(fscal,dx00);
1373             ty               = _mm256_mul_pd(fscal,dy00);
1374             tz               = _mm256_mul_pd(fscal,dz00);
1375
1376             /* Update vectorial force */
1377             fix0             = _mm256_add_pd(fix0,tx);
1378             fiy0             = _mm256_add_pd(fiy0,ty);
1379             fiz0             = _mm256_add_pd(fiz0,tz);
1380
1381             fjx0             = _mm256_add_pd(fjx0,tx);
1382             fjy0             = _mm256_add_pd(fjy0,ty);
1383             fjz0             = _mm256_add_pd(fjz0,tz);
1384
1385             }
1386
1387             /**************************
1388              * CALCULATE INTERACTIONS *
1389              **************************/
1390
1391             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1392             {
1393
1394             r10              = _mm256_mul_pd(rsq10,rinv10);
1395             r10              = _mm256_andnot_pd(dummy_mask,r10);
1396
1397             /* Compute parameters for interactions between i and j atoms */
1398             qq10             = _mm256_mul_pd(iq1,jq0);
1399
1400             /* EWALD ELECTROSTATICS */
1401
1402             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1403             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1404             ewitab           = _mm256_cvttpd_epi32(ewrt);
1405             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1406             ewitab           = _mm_slli_epi32(ewitab,2);
1407             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1408             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1409             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1410             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1411             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1412             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1413             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1414             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1415             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1416
1417             d                = _mm256_sub_pd(r10,rswitch);
1418             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1419             d2               = _mm256_mul_pd(d,d);
1420             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)))))));
1421
1422             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1423
1424             /* Evaluate switch function */
1425             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1426             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1427             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1428
1429             fscal            = felec;
1430
1431             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1432
1433             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1434
1435             /* Calculate temporary vectorial force */
1436             tx               = _mm256_mul_pd(fscal,dx10);
1437             ty               = _mm256_mul_pd(fscal,dy10);
1438             tz               = _mm256_mul_pd(fscal,dz10);
1439
1440             /* Update vectorial force */
1441             fix1             = _mm256_add_pd(fix1,tx);
1442             fiy1             = _mm256_add_pd(fiy1,ty);
1443             fiz1             = _mm256_add_pd(fiz1,tz);
1444
1445             fjx0             = _mm256_add_pd(fjx0,tx);
1446             fjy0             = _mm256_add_pd(fjy0,ty);
1447             fjz0             = _mm256_add_pd(fjz0,tz);
1448
1449             }
1450
1451             /**************************
1452              * CALCULATE INTERACTIONS *
1453              **************************/
1454
1455             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1456             {
1457
1458             r20              = _mm256_mul_pd(rsq20,rinv20);
1459             r20              = _mm256_andnot_pd(dummy_mask,r20);
1460
1461             /* Compute parameters for interactions between i and j atoms */
1462             qq20             = _mm256_mul_pd(iq2,jq0);
1463
1464             /* EWALD ELECTROSTATICS */
1465
1466             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1467             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1468             ewitab           = _mm256_cvttpd_epi32(ewrt);
1469             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1470             ewitab           = _mm_slli_epi32(ewitab,2);
1471             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1472             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1473             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1474             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1475             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1476             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1477             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1478             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1479             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1480
1481             d                = _mm256_sub_pd(r20,rswitch);
1482             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1483             d2               = _mm256_mul_pd(d,d);
1484             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)))))));
1485
1486             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1487
1488             /* Evaluate switch function */
1489             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1490             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1491             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1492
1493             fscal            = felec;
1494
1495             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1496
1497             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1498
1499             /* Calculate temporary vectorial force */
1500             tx               = _mm256_mul_pd(fscal,dx20);
1501             ty               = _mm256_mul_pd(fscal,dy20);
1502             tz               = _mm256_mul_pd(fscal,dz20);
1503
1504             /* Update vectorial force */
1505             fix2             = _mm256_add_pd(fix2,tx);
1506             fiy2             = _mm256_add_pd(fiy2,ty);
1507             fiz2             = _mm256_add_pd(fiz2,tz);
1508
1509             fjx0             = _mm256_add_pd(fjx0,tx);
1510             fjy0             = _mm256_add_pd(fjy0,ty);
1511             fjz0             = _mm256_add_pd(fjz0,tz);
1512
1513             }
1514
1515             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1516             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1517             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1518             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1519
1520             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1521
1522             /* Inner loop uses 207 flops */
1523         }
1524
1525         /* End of innermost loop */
1526
1527         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1528                                                  f+i_coord_offset,fshift+i_shift_offset);
1529
1530         /* Increment number of inner iterations */
1531         inneriter                  += j_index_end - j_index_start;
1532
1533         /* Outer loop uses 18 flops */
1534     }
1535
1536     /* Increment number of outer iterations */
1537     outeriter        += nri;
1538
1539     /* Update outer/inner flops */
1540
1541     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*207);
1542 }