src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSw_VdwNone_GeomW4P1_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_VdwNone_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_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 *           vdwioffsetptr1;
  85     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     real *           vdwioffsetptr2;
  87     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     real *           vdwioffsetptr3;
  89     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     __m128i          ewitab;
  98     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  99     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 100     real             *ewtab;
 101     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 102     real             rswitch_scalar,d_scalar;
 103     __m256d          dummy_mask,cutoff_mask;
 104     __m128           tmpmask0,tmpmask1;
 105     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 106     __m256d          one     = _mm256_set1_pd(1.0);
 107     __m256d          two     = _mm256_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121
 122     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 123     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 124     beta2            = _mm256_mul_pd(beta,beta);
 125     beta3            = _mm256_mul_pd(beta,beta2);
 126
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 134     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 135     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 141
 142     rswitch_scalar   = fr->rcoulomb_switch;
 143     rswitch          = _mm256_set1_pd(rswitch_scalar);
 144     /* Setup switch parameters */
 145     d_scalar         = rcutoff_scalar-rswitch_scalar;
 146     d                = _mm256_set1_pd(d_scalar);
 147     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 148     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 151     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     for(iidx=0;iidx<4*DIM;iidx++)
 165     {
 166         scratch[iidx] = 0.0;
 167     }
 168
 169     /* Start outer loop over neighborlists */
 170     for(iidx=0; iidx<nri; iidx++)
 171     {
 172         /* Load shift vector for this list */
 173         i_shift_offset   = DIM*shiftidx[iidx];
 174
 175         /* Load limits for loop over neighbors */
 176         j_index_start    = jindex[iidx];
 177         j_index_end      = jindex[iidx+1];
 178
 179         /* Get outer coordinate index */
 180         inr              = iinr[iidx];
 181         i_coord_offset   = DIM*inr;
 182
 183         /* Load i particle coords and add shift vector */
 184         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 185                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 186
 187         fix1             = _mm256_setzero_pd();
 188         fiy1             = _mm256_setzero_pd();
 189         fiz1             = _mm256_setzero_pd();
 190         fix2             = _mm256_setzero_pd();
 191         fiy2             = _mm256_setzero_pd();
 192         fiz2             = _mm256_setzero_pd();
 193         fix3             = _mm256_setzero_pd();
 194         fiy3             = _mm256_setzero_pd();
 195         fiz3             = _mm256_setzero_pd();
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm256_setzero_pd();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211             j_coord_offsetC  = DIM*jnrC;
 212             j_coord_offsetD  = DIM*jnrD;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx10             = _mm256_sub_pd(ix1,jx0);
 221             dy10             = _mm256_sub_pd(iy1,jy0);
 222             dz10             = _mm256_sub_pd(iz1,jz0);
 223             dx20             = _mm256_sub_pd(ix2,jx0);
 224             dy20             = _mm256_sub_pd(iy2,jy0);
 225             dz20             = _mm256_sub_pd(iz2,jz0);
 226             dx30             = _mm256_sub_pd(ix3,jx0);
 227             dy30             = _mm256_sub_pd(iy3,jy0);
 228             dz30             = _mm256_sub_pd(iz3,jz0);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 232             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 233             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 234
 235             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 236             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 237             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 238
 239             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 240             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 241             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 245                                                                  charge+jnrC+0,charge+jnrD+0);
 246
 247             fjx0             = _mm256_setzero_pd();
 248             fjy0             = _mm256_setzero_pd();
 249             fjz0             = _mm256_setzero_pd();
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 256             {
 257
 258             r10              = _mm256_mul_pd(rsq10,rinv10);
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq10             = _mm256_mul_pd(iq1,jq0);
 262
 263             /* EWALD ELECTROSTATICS */
 264
 265             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 266             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 267             ewitab           = _mm256_cvttpd_epi32(ewrt);
 268             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 269             ewitab           = _mm_slli_epi32(ewitab,2);
 270             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 271             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 272             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 273             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 274             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 275             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 276             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 277             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 278             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 279
 280             d                = _mm256_sub_pd(r10,rswitch);
 281             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 282             d2               = _mm256_mul_pd(d,d);
 283             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)))))));
 284
 285             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 286
 287             /* Evaluate switch function */
 288             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 289             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 290             velec            = _mm256_mul_pd(velec,sw);
 291             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velec            = _mm256_and_pd(velec,cutoff_mask);
 295             velecsum         = _mm256_add_pd(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm256_mul_pd(fscal,dx10);
 303             ty               = _mm256_mul_pd(fscal,dy10);
 304             tz               = _mm256_mul_pd(fscal,dz10);
 305
 306             /* Update vectorial force */
 307             fix1             = _mm256_add_pd(fix1,tx);
 308             fiy1             = _mm256_add_pd(fiy1,ty);
 309             fiz1             = _mm256_add_pd(fiz1,tz);
 310
 311             fjx0             = _mm256_add_pd(fjx0,tx);
 312             fjy0             = _mm256_add_pd(fjy0,ty);
 313             fjz0             = _mm256_add_pd(fjz0,tz);
 314
 315             }
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 322             {
 323
 324             r20              = _mm256_mul_pd(rsq20,rinv20);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq20             = _mm256_mul_pd(iq2,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 333             ewitab           = _mm256_cvttpd_epi32(ewrt);
 334             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 338             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 339             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 340             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 341             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 342             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 343             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 344             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 345
 346             d                = _mm256_sub_pd(r20,rswitch);
 347             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 348             d2               = _mm256_mul_pd(d,d);
 349             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)))))));
 350
 351             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 352
 353             /* Evaluate switch function */
 354             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 355             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 356             velec            = _mm256_mul_pd(velec,sw);
 357             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velec            = _mm256_and_pd(velec,cutoff_mask);
 361             velecsum         = _mm256_add_pd(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 366
 367             /* Calculate temporary vectorial force */
 368             tx               = _mm256_mul_pd(fscal,dx20);
 369             ty               = _mm256_mul_pd(fscal,dy20);
 370             tz               = _mm256_mul_pd(fscal,dz20);
 371
 372             /* Update vectorial force */
 373             fix2             = _mm256_add_pd(fix2,tx);
 374             fiy2             = _mm256_add_pd(fiy2,ty);
 375             fiz2             = _mm256_add_pd(fiz2,tz);
 376
 377             fjx0             = _mm256_add_pd(fjx0,tx);
 378             fjy0             = _mm256_add_pd(fjy0,ty);
 379             fjz0             = _mm256_add_pd(fjz0,tz);
 380
 381             }
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 388             {
 389
 390             r30              = _mm256_mul_pd(rsq30,rinv30);
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq30             = _mm256_mul_pd(iq3,jq0);
 394
 395             /* EWALD ELECTROSTATICS */
 396
 397             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 398             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 399             ewitab           = _mm256_cvttpd_epi32(ewrt);
 400             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 401             ewitab           = _mm_slli_epi32(ewitab,2);
 402             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 403             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 404             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 405             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 406             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 407             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 408             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 409             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 410             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 411
 412             d                = _mm256_sub_pd(r30,rswitch);
 413             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 414             d2               = _mm256_mul_pd(d,d);
 415             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)))))));
 416
 417             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 418
 419             /* Evaluate switch function */
 420             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 421             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
 422             velec            = _mm256_mul_pd(velec,sw);
 423             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velec            = _mm256_and_pd(velec,cutoff_mask);
 427             velecsum         = _mm256_add_pd(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_pd(fscal,dx30);
 435             ty               = _mm256_mul_pd(fscal,dy30);
 436             tz               = _mm256_mul_pd(fscal,dz30);
 437
 438             /* Update vectorial force */
 439             fix3             = _mm256_add_pd(fix3,tx);
 440             fiy3             = _mm256_add_pd(fiy3,ty);
 441             fiz3             = _mm256_add_pd(fiz3,tz);
 442
 443             fjx0             = _mm256_add_pd(fjx0,tx);
 444             fjy0             = _mm256_add_pd(fjy0,ty);
 445             fjz0             = _mm256_add_pd(fjz0,tz);
 446
 447             }
 448
 449             fjptrA             = f+j_coord_offsetA;
 450             fjptrB             = f+j_coord_offsetB;
 451             fjptrC             = f+j_coord_offsetC;
 452             fjptrD             = f+j_coord_offsetD;
 453
 454             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 455
 456             /* Inner loop uses 198 flops */
 457         }
 458
 459         if(jidx<j_index_end)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrlistA         = jjnr[jidx];
 464             jnrlistB         = jjnr[jidx+1];
 465             jnrlistC         = jjnr[jidx+2];
 466             jnrlistD         = jjnr[jidx+3];
 467             /* Sign of each element will be negative for non-real atoms.
 468              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 469              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 470              */
 471             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 472
 473             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 474             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 475             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 476
 477             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 478             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 479             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 480             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 481             j_coord_offsetA  = DIM*jnrA;
 482             j_coord_offsetB  = DIM*jnrB;
 483             j_coord_offsetC  = DIM*jnrC;
 484             j_coord_offsetD  = DIM*jnrD;
 485
 486             /* load j atom coordinates */
 487             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 488                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 489                                                  &jx0,&jy0,&jz0);
 490
 491             /* Calculate displacement vector */
 492             dx10             = _mm256_sub_pd(ix1,jx0);
 493             dy10             = _mm256_sub_pd(iy1,jy0);
 494             dz10             = _mm256_sub_pd(iz1,jz0);
 495             dx20             = _mm256_sub_pd(ix2,jx0);
 496             dy20             = _mm256_sub_pd(iy2,jy0);
 497             dz20             = _mm256_sub_pd(iz2,jz0);
 498             dx30             = _mm256_sub_pd(ix3,jx0);
 499             dy30             = _mm256_sub_pd(iy3,jy0);
 500             dz30             = _mm256_sub_pd(iz3,jz0);
 501
 502             /* Calculate squared distance and things based on it */
 503             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 504             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 505             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 506
 507             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 508             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 509             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 510
 511             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 512             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 513             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 514
 515             /* Load parameters for j particles */
 516             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 517                                                                  charge+jnrC+0,charge+jnrD+0);
 518
 519             fjx0             = _mm256_setzero_pd();
 520             fjy0             = _mm256_setzero_pd();
 521             fjz0             = _mm256_setzero_pd();
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 528             {
 529
 530             r10              = _mm256_mul_pd(rsq10,rinv10);
 531             r10              = _mm256_andnot_pd(dummy_mask,r10);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq10             = _mm256_mul_pd(iq1,jq0);
 535
 536             /* EWALD ELECTROSTATICS */
 537
 538             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 539             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 540             ewitab           = _mm256_cvttpd_epi32(ewrt);
 541             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 542             ewitab           = _mm_slli_epi32(ewitab,2);
 543             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 544             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 545             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 546             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 547             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 548             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 549             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 550             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 551             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 552
 553             d                = _mm256_sub_pd(r10,rswitch);
 554             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 555             d2               = _mm256_mul_pd(d,d);
 556             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)))))));
 557
 558             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 559
 560             /* Evaluate switch function */
 561             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 562             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
 563             velec            = _mm256_mul_pd(velec,sw);
 564             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velec            = _mm256_and_pd(velec,cutoff_mask);
 568             velec            = _mm256_andnot_pd(dummy_mask,velec);
 569             velecsum         = _mm256_add_pd(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 574
 575             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 576
 577             /* Calculate temporary vectorial force */
 578             tx               = _mm256_mul_pd(fscal,dx10);
 579             ty               = _mm256_mul_pd(fscal,dy10);
 580             tz               = _mm256_mul_pd(fscal,dz10);
 581
 582             /* Update vectorial force */
 583             fix1             = _mm256_add_pd(fix1,tx);
 584             fiy1             = _mm256_add_pd(fiy1,ty);
 585             fiz1             = _mm256_add_pd(fiz1,tz);
 586
 587             fjx0             = _mm256_add_pd(fjx0,tx);
 588             fjy0             = _mm256_add_pd(fjy0,ty);
 589             fjz0             = _mm256_add_pd(fjz0,tz);
 590
 591             }
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 598             {
 599
 600             r20              = _mm256_mul_pd(rsq20,rinv20);
 601             r20              = _mm256_andnot_pd(dummy_mask,r20);
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq20             = _mm256_mul_pd(iq2,jq0);
 605
 606             /* EWALD ELECTROSTATICS */
 607
 608             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 609             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 610             ewitab           = _mm256_cvttpd_epi32(ewrt);
 611             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 612             ewitab           = _mm_slli_epi32(ewitab,2);
 613             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 614             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 615             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 616             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 617             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 618             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 619             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 620             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 621             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 622
 623             d                = _mm256_sub_pd(r20,rswitch);
 624             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 625             d2               = _mm256_mul_pd(d,d);
 626             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)))))));
 627
 628             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 629
 630             /* Evaluate switch function */
 631             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 632             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
 633             velec            = _mm256_mul_pd(velec,sw);
 634             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 635
 636             /* Update potential sum for this i atom from the interaction with this j atom. */
 637             velec            = _mm256_and_pd(velec,cutoff_mask);
 638             velec            = _mm256_andnot_pd(dummy_mask,velec);
 639             velecsum         = _mm256_add_pd(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 644
 645             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm256_mul_pd(fscal,dx20);
 649             ty               = _mm256_mul_pd(fscal,dy20);
 650             tz               = _mm256_mul_pd(fscal,dz20);
 651
 652             /* Update vectorial force */
 653             fix2             = _mm256_add_pd(fix2,tx);
 654             fiy2             = _mm256_add_pd(fiy2,ty);
 655             fiz2             = _mm256_add_pd(fiz2,tz);
 656
 657             fjx0             = _mm256_add_pd(fjx0,tx);
 658             fjy0             = _mm256_add_pd(fjy0,ty);
 659             fjz0             = _mm256_add_pd(fjz0,tz);
 660
 661             }
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 668             {
 669
 670             r30              = _mm256_mul_pd(rsq30,rinv30);
 671             r30              = _mm256_andnot_pd(dummy_mask,r30);
 672
 673             /* Compute parameters for interactions between i and j atoms */
 674             qq30             = _mm256_mul_pd(iq3,jq0);
 675
 676             /* EWALD ELECTROSTATICS */
 677
 678             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 679             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 680             ewitab           = _mm256_cvttpd_epi32(ewrt);
 681             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 682             ewitab           = _mm_slli_epi32(ewitab,2);
 683             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 684             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 685             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 686             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 687             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 688             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 689             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 690             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 691             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 692
 693             d                = _mm256_sub_pd(r30,rswitch);
 694             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 695             d2               = _mm256_mul_pd(d,d);
 696             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)))))));
 697
 698             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 699
 700             /* Evaluate switch function */
 701             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 702             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
 703             velec            = _mm256_mul_pd(velec,sw);
 704             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 705
 706             /* Update potential sum for this i atom from the interaction with this j atom. */
 707             velec            = _mm256_and_pd(velec,cutoff_mask);
 708             velec            = _mm256_andnot_pd(dummy_mask,velec);
 709             velecsum         = _mm256_add_pd(velecsum,velec);
 710
 711             fscal            = felec;
 712
 713             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 714
 715             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm256_mul_pd(fscal,dx30);
 719             ty               = _mm256_mul_pd(fscal,dy30);
 720             tz               = _mm256_mul_pd(fscal,dz30);
 721
 722             /* Update vectorial force */
 723             fix3             = _mm256_add_pd(fix3,tx);
 724             fiy3             = _mm256_add_pd(fiy3,ty);
 725             fiz3             = _mm256_add_pd(fiz3,tz);
 726
 727             fjx0             = _mm256_add_pd(fjx0,tx);
 728             fjy0             = _mm256_add_pd(fjy0,ty);
 729             fjz0             = _mm256_add_pd(fjz0,tz);
 730
 731             }
 732
 733             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 734             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 735             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 736             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 737
 738             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 739
 740             /* Inner loop uses 201 flops */
 741         }
 742
 743         /* End of innermost loop */
 744
 745         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 746                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 747
 748         ggid                        = gid[iidx];
 749         /* Update potential energies */
 750         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 751
 752         /* Increment number of inner iterations */
 753         inneriter                  += j_index_end - j_index_start;
 754
 755         /* Outer loop uses 19 flops */
 756     }
 757
 758     /* Increment number of outer iterations */
 759     outeriter        += nri;
 760
 761     /* Update outer/inner flops */
 762
 763     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*201);
 764 }
 765 /*
 766  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_avx_256_double
 767  * Electrostatics interaction: Ewald
 768  * VdW interaction:            None
 769  * Geometry:                   Water4-Particle
 770  * Calculate force/pot:        Force
 771  */
 772 void
 773 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_avx_256_double
 774                     (t_nblist                    * gmx_restrict       nlist,
 775                      rvec                        * gmx_restrict          xx,
 776                      rvec                        * gmx_restrict          ff,
 777                      t_forcerec                  * gmx_restrict          fr,
 778                      t_mdatoms                   * gmx_restrict     mdatoms,
 779                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 780                      t_nrnb                      * gmx_restrict        nrnb)
 781 {
 782     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 783      * just 0 for non-waters.
 784      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 785      * jnr indices corresponding to data put in the four positions in the SIMD register.
 786      */
 787     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 788     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 789     int              jnrA,jnrB,jnrC,jnrD;
 790     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 791     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 792     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 793     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 794     real             rcutoff_scalar;
 795     real             *shiftvec,*fshift,*x,*f;
 796     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 797     real             scratch[4*DIM];
 798     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 799     real *           vdwioffsetptr1;
 800     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 801     real *           vdwioffsetptr2;
 802     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 803     real *           vdwioffsetptr3;
 804     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 805     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 806     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 807     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 808     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 809     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 810     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 811     real             *charge;
 812     __m128i          ewitab;
 813     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 814     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 815     real             *ewtab;
 816     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 817     real             rswitch_scalar,d_scalar;
 818     __m256d          dummy_mask,cutoff_mask;
 819     __m128           tmpmask0,tmpmask1;
 820     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 821     __m256d          one     = _mm256_set1_pd(1.0);
 822     __m256d          two     = _mm256_set1_pd(2.0);
 823     x                = xx[0];
 824     f                = ff[0];
 825
 826     nri              = nlist->nri;
 827     iinr             = nlist->iinr;
 828     jindex           = nlist->jindex;
 829     jjnr             = nlist->jjnr;
 830     shiftidx         = nlist->shift;
 831     gid              = nlist->gid;
 832     shiftvec         = fr->shift_vec[0];
 833     fshift           = fr->fshift[0];
 834     facel            = _mm256_set1_pd(fr->epsfac);
 835     charge           = mdatoms->chargeA;
 836
 837     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 838     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 839     beta2            = _mm256_mul_pd(beta,beta);
 840     beta3            = _mm256_mul_pd(beta,beta2);
 841
 842     ewtab            = fr->ic->tabq_coul_FDV0;
 843     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 844     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 845
 846     /* Setup water-specific parameters */
 847     inr              = nlist->iinr[0];
 848     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 849     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 850     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 851
 852     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 853     rcutoff_scalar   = fr->rcoulomb;
 854     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 855     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 856
 857     rswitch_scalar   = fr->rcoulomb_switch;
 858     rswitch          = _mm256_set1_pd(rswitch_scalar);
 859     /* Setup switch parameters */
 860     d_scalar         = rcutoff_scalar-rswitch_scalar;
 861     d                = _mm256_set1_pd(d_scalar);
 862     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 863     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 864     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 865     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 866     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 867     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 868
 869     /* Avoid stupid compiler warnings */
 870     jnrA = jnrB = jnrC = jnrD = 0;
 871     j_coord_offsetA = 0;
 872     j_coord_offsetB = 0;
 873     j_coord_offsetC = 0;
 874     j_coord_offsetD = 0;
 875
 876     outeriter        = 0;
 877     inneriter        = 0;
 878
 879     for(iidx=0;iidx<4*DIM;iidx++)
 880     {
 881         scratch[iidx] = 0.0;
 882     }
 883
 884     /* Start outer loop over neighborlists */
 885     for(iidx=0; iidx<nri; iidx++)
 886     {
 887         /* Load shift vector for this list */
 888         i_shift_offset   = DIM*shiftidx[iidx];
 889
 890         /* Load limits for loop over neighbors */
 891         j_index_start    = jindex[iidx];
 892         j_index_end      = jindex[iidx+1];
 893
 894         /* Get outer coordinate index */
 895         inr              = iinr[iidx];
 896         i_coord_offset   = DIM*inr;
 897
 898         /* Load i particle coords and add shift vector */
 899         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 900                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 901
 902         fix1             = _mm256_setzero_pd();
 903         fiy1             = _mm256_setzero_pd();
 904         fiz1             = _mm256_setzero_pd();
 905         fix2             = _mm256_setzero_pd();
 906         fiy2             = _mm256_setzero_pd();
 907         fiz2             = _mm256_setzero_pd();
 908         fix3             = _mm256_setzero_pd();
 909         fiy3             = _mm256_setzero_pd();
 910         fiz3             = _mm256_setzero_pd();
 911
 912         /* Start inner kernel loop */
 913         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrA             = jjnr[jidx];
 918             jnrB             = jjnr[jidx+1];
 919             jnrC             = jjnr[jidx+2];
 920             jnrD             = jjnr[jidx+3];
 921             j_coord_offsetA  = DIM*jnrA;
 922             j_coord_offsetB  = DIM*jnrB;
 923             j_coord_offsetC  = DIM*jnrC;
 924             j_coord_offsetD  = DIM*jnrD;
 925
 926             /* load j atom coordinates */
 927             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 928                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 929                                                  &jx0,&jy0,&jz0);
 930
 931             /* Calculate displacement vector */
 932             dx10             = _mm256_sub_pd(ix1,jx0);
 933             dy10             = _mm256_sub_pd(iy1,jy0);
 934             dz10             = _mm256_sub_pd(iz1,jz0);
 935             dx20             = _mm256_sub_pd(ix2,jx0);
 936             dy20             = _mm256_sub_pd(iy2,jy0);
 937             dz20             = _mm256_sub_pd(iz2,jz0);
 938             dx30             = _mm256_sub_pd(ix3,jx0);
 939             dy30             = _mm256_sub_pd(iy3,jy0);
 940             dz30             = _mm256_sub_pd(iz3,jz0);
 941
 942             /* Calculate squared distance and things based on it */
 943             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 944             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 945             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 946
 947             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 948             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 949             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 950
 951             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 952             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 953             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 954
 955             /* Load parameters for j particles */
 956             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 957                                                                  charge+jnrC+0,charge+jnrD+0);
 958
 959             fjx0             = _mm256_setzero_pd();
 960             fjy0             = _mm256_setzero_pd();
 961             fjz0             = _mm256_setzero_pd();
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 968             {
 969
 970             r10              = _mm256_mul_pd(rsq10,rinv10);
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq10             = _mm256_mul_pd(iq1,jq0);
 974
 975             /* EWALD ELECTROSTATICS */
 976
 977             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 978             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 979             ewitab           = _mm256_cvttpd_epi32(ewrt);
 980             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 981             ewitab           = _mm_slli_epi32(ewitab,2);
 982             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 983             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 984             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 985             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 986             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 987             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 988             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 989             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 990             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 991
 992             d                = _mm256_sub_pd(r10,rswitch);
 993             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 994             d2               = _mm256_mul_pd(d,d);
 995             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)))))));
 996
 997             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 998
 999             /* Evaluate switch function */
1000             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1001             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1002             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm256_mul_pd(fscal,dx10);
1010             ty               = _mm256_mul_pd(fscal,dy10);
1011             tz               = _mm256_mul_pd(fscal,dz10);
1012
1013             /* Update vectorial force */
1014             fix1             = _mm256_add_pd(fix1,tx);
1015             fiy1             = _mm256_add_pd(fiy1,ty);
1016             fiz1             = _mm256_add_pd(fiz1,tz);
1017
1018             fjx0             = _mm256_add_pd(fjx0,tx);
1019             fjy0             = _mm256_add_pd(fjy0,ty);
1020             fjz0             = _mm256_add_pd(fjz0,tz);
1021
1022             }
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1029             {
1030
1031             r20              = _mm256_mul_pd(rsq20,rinv20);
1032
1033             /* Compute parameters for interactions between i and j atoms */
1034             qq20             = _mm256_mul_pd(iq2,jq0);
1035
1036             /* EWALD ELECTROSTATICS */
1037
1038             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1039             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1040             ewitab           = _mm256_cvttpd_epi32(ewrt);
1041             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1042             ewitab           = _mm_slli_epi32(ewitab,2);
1043             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1044             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1045             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1046             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1047             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1048             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1049             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1050             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1051             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1052
1053             d                = _mm256_sub_pd(r20,rswitch);
1054             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1055             d2               = _mm256_mul_pd(d,d);
1056             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)))))));
1057
1058             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1059
1060             /* Evaluate switch function */
1061             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1062             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1063             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1064
1065             fscal            = felec;
1066
1067             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm256_mul_pd(fscal,dx20);
1071             ty               = _mm256_mul_pd(fscal,dy20);
1072             tz               = _mm256_mul_pd(fscal,dz20);
1073
1074             /* Update vectorial force */
1075             fix2             = _mm256_add_pd(fix2,tx);
1076             fiy2             = _mm256_add_pd(fiy2,ty);
1077             fiz2             = _mm256_add_pd(fiz2,tz);
1078
1079             fjx0             = _mm256_add_pd(fjx0,tx);
1080             fjy0             = _mm256_add_pd(fjy0,ty);
1081             fjz0             = _mm256_add_pd(fjz0,tz);
1082
1083             }
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1090             {
1091
1092             r30              = _mm256_mul_pd(rsq30,rinv30);
1093
1094             /* Compute parameters for interactions between i and j atoms */
1095             qq30             = _mm256_mul_pd(iq3,jq0);
1096
1097             /* EWALD ELECTROSTATICS */
1098
1099             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1100             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1101             ewitab           = _mm256_cvttpd_epi32(ewrt);
1102             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1103             ewitab           = _mm_slli_epi32(ewitab,2);
1104             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1105             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1106             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1107             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1108             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1109             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1110             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1111             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
1112             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1113
1114             d                = _mm256_sub_pd(r30,rswitch);
1115             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1116             d2               = _mm256_mul_pd(d,d);
1117             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)))))));
1118
1119             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1120
1121             /* Evaluate switch function */
1122             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1123             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
1124             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1125
1126             fscal            = felec;
1127
1128             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1129
1130             /* Calculate temporary vectorial force */
1131             tx               = _mm256_mul_pd(fscal,dx30);
1132             ty               = _mm256_mul_pd(fscal,dy30);
1133             tz               = _mm256_mul_pd(fscal,dz30);
1134
1135             /* Update vectorial force */
1136             fix3             = _mm256_add_pd(fix3,tx);
1137             fiy3             = _mm256_add_pd(fiy3,ty);
1138             fiz3             = _mm256_add_pd(fiz3,tz);
1139
1140             fjx0             = _mm256_add_pd(fjx0,tx);
1141             fjy0             = _mm256_add_pd(fjy0,ty);
1142             fjz0             = _mm256_add_pd(fjz0,tz);
1143
1144             }
1145
1146             fjptrA             = f+j_coord_offsetA;
1147             fjptrB             = f+j_coord_offsetB;
1148             fjptrC             = f+j_coord_offsetC;
1149             fjptrD             = f+j_coord_offsetD;
1150
1151             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1152
1153             /* Inner loop uses 189 flops */
1154         }
1155
1156         if(jidx<j_index_end)
1157         {
1158
1159             /* Get j neighbor index, and coordinate index */
1160             jnrlistA         = jjnr[jidx];
1161             jnrlistB         = jjnr[jidx+1];
1162             jnrlistC         = jjnr[jidx+2];
1163             jnrlistD         = jjnr[jidx+3];
1164             /* Sign of each element will be negative for non-real atoms.
1165              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1166              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1167              */
1168             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1169
1170             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1171             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1172             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1173
1174             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1175             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1176             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1177             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1178             j_coord_offsetA  = DIM*jnrA;
1179             j_coord_offsetB  = DIM*jnrB;
1180             j_coord_offsetC  = DIM*jnrC;
1181             j_coord_offsetD  = DIM*jnrD;
1182
1183             /* load j atom coordinates */
1184             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1185                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1186                                                  &jx0,&jy0,&jz0);
1187
1188             /* Calculate displacement vector */
1189             dx10             = _mm256_sub_pd(ix1,jx0);
1190             dy10             = _mm256_sub_pd(iy1,jy0);
1191             dz10             = _mm256_sub_pd(iz1,jz0);
1192             dx20             = _mm256_sub_pd(ix2,jx0);
1193             dy20             = _mm256_sub_pd(iy2,jy0);
1194             dz20             = _mm256_sub_pd(iz2,jz0);
1195             dx30             = _mm256_sub_pd(ix3,jx0);
1196             dy30             = _mm256_sub_pd(iy3,jy0);
1197             dz30             = _mm256_sub_pd(iz3,jz0);
1198
1199             /* Calculate squared distance and things based on it */
1200             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1201             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1202             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1203
1204             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1205             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1206             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1207
1208             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1209             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1210             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1211
1212             /* Load parameters for j particles */
1213             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1214                                                                  charge+jnrC+0,charge+jnrD+0);
1215
1216             fjx0             = _mm256_setzero_pd();
1217             fjy0             = _mm256_setzero_pd();
1218             fjz0             = _mm256_setzero_pd();
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1225             {
1226
1227             r10              = _mm256_mul_pd(rsq10,rinv10);
1228             r10              = _mm256_andnot_pd(dummy_mask,r10);
1229
1230             /* Compute parameters for interactions between i and j atoms */
1231             qq10             = _mm256_mul_pd(iq1,jq0);
1232
1233             /* EWALD ELECTROSTATICS */
1234
1235             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1236             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1237             ewitab           = _mm256_cvttpd_epi32(ewrt);
1238             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1239             ewitab           = _mm_slli_epi32(ewitab,2);
1240             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1241             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1242             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1243             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1244             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1245             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1246             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1247             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
1248             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1249
1250             d                = _mm256_sub_pd(r10,rswitch);
1251             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1252             d2               = _mm256_mul_pd(d,d);
1253             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)))))));
1254
1255             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1256
1257             /* Evaluate switch function */
1258             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1259             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv10,_mm256_mul_pd(velec,dsw)) );
1260             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1261
1262             fscal            = felec;
1263
1264             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1265
1266             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1267
1268             /* Calculate temporary vectorial force */
1269             tx               = _mm256_mul_pd(fscal,dx10);
1270             ty               = _mm256_mul_pd(fscal,dy10);
1271             tz               = _mm256_mul_pd(fscal,dz10);
1272
1273             /* Update vectorial force */
1274             fix1             = _mm256_add_pd(fix1,tx);
1275             fiy1             = _mm256_add_pd(fiy1,ty);
1276             fiz1             = _mm256_add_pd(fiz1,tz);
1277
1278             fjx0             = _mm256_add_pd(fjx0,tx);
1279             fjy0             = _mm256_add_pd(fjy0,ty);
1280             fjz0             = _mm256_add_pd(fjz0,tz);
1281
1282             }
1283
1284             /**************************
1285              * CALCULATE INTERACTIONS *
1286              **************************/
1287
1288             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1289             {
1290
1291             r20              = _mm256_mul_pd(rsq20,rinv20);
1292             r20              = _mm256_andnot_pd(dummy_mask,r20);
1293
1294             /* Compute parameters for interactions between i and j atoms */
1295             qq20             = _mm256_mul_pd(iq2,jq0);
1296
1297             /* EWALD ELECTROSTATICS */
1298
1299             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1300             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1301             ewitab           = _mm256_cvttpd_epi32(ewrt);
1302             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1303             ewitab           = _mm_slli_epi32(ewitab,2);
1304             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1305             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1306             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1307             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1308             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1309             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1310             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1311             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1312             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1313
1314             d                = _mm256_sub_pd(r20,rswitch);
1315             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1316             d2               = _mm256_mul_pd(d,d);
1317             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)))))));
1318
1319             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1320
1321             /* Evaluate switch function */
1322             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1323             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv20,_mm256_mul_pd(velec,dsw)) );
1324             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1325
1326             fscal            = felec;
1327
1328             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1329
1330             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1331
1332             /* Calculate temporary vectorial force */
1333             tx               = _mm256_mul_pd(fscal,dx20);
1334             ty               = _mm256_mul_pd(fscal,dy20);
1335             tz               = _mm256_mul_pd(fscal,dz20);
1336
1337             /* Update vectorial force */
1338             fix2             = _mm256_add_pd(fix2,tx);
1339             fiy2             = _mm256_add_pd(fiy2,ty);
1340             fiz2             = _mm256_add_pd(fiz2,tz);
1341
1342             fjx0             = _mm256_add_pd(fjx0,tx);
1343             fjy0             = _mm256_add_pd(fjy0,ty);
1344             fjz0             = _mm256_add_pd(fjz0,tz);
1345
1346             }
1347
1348             /**************************
1349              * CALCULATE INTERACTIONS *
1350              **************************/
1351
1352             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1353             {
1354
1355             r30              = _mm256_mul_pd(rsq30,rinv30);
1356             r30              = _mm256_andnot_pd(dummy_mask,r30);
1357
1358             /* Compute parameters for interactions between i and j atoms */
1359             qq30             = _mm256_mul_pd(iq3,jq0);
1360
1361             /* EWALD ELECTROSTATICS */
1362
1363             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1364             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1365             ewitab           = _mm256_cvttpd_epi32(ewrt);
1366             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1367             ewitab           = _mm_slli_epi32(ewitab,2);
1368             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1369             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1370             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1371             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1372             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1373             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1374             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1375             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
1376             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1377
1378             d                = _mm256_sub_pd(r30,rswitch);
1379             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1380             d2               = _mm256_mul_pd(d,d);
1381             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)))))));
1382
1383             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1384
1385             /* Evaluate switch function */
1386             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1387             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv30,_mm256_mul_pd(velec,dsw)) );
1388             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1389
1390             fscal            = felec;
1391
1392             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1393
1394             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1395
1396             /* Calculate temporary vectorial force */
1397             tx               = _mm256_mul_pd(fscal,dx30);
1398             ty               = _mm256_mul_pd(fscal,dy30);
1399             tz               = _mm256_mul_pd(fscal,dz30);
1400
1401             /* Update vectorial force */
1402             fix3             = _mm256_add_pd(fix3,tx);
1403             fiy3             = _mm256_add_pd(fiy3,ty);
1404             fiz3             = _mm256_add_pd(fiz3,tz);
1405
1406             fjx0             = _mm256_add_pd(fjx0,tx);
1407             fjy0             = _mm256_add_pd(fjy0,ty);
1408             fjz0             = _mm256_add_pd(fjz0,tz);
1409
1410             }
1411
1412             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1413             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1414             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1415             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1416
1417             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1418
1419             /* Inner loop uses 192 flops */
1420         }
1421
1422         /* End of innermost loop */
1423
1424         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1425                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1426
1427         /* Increment number of inner iterations */
1428         inneriter                  += j_index_end - j_index_start;
1429
1430         /* Outer loop uses 18 flops */
1431     }
1432
1433     /* Increment number of outer iterations */
1434     outeriter        += nri;
1435
1436     /* Update outer/inner flops */
1437
1438     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*192);
1439 }