src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwLJ_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m256d          dummy_mask,cutoff_mask;
 109     __m128           tmpmask0,tmpmask1;
 110     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 111     __m256d          one     = _mm256_set1_pd(1.0);
 112     __m256d          two     = _mm256_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 133     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 134     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 135     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 169
 170         fix0             = _mm256_setzero_pd();
 171         fiy0             = _mm256_setzero_pd();
 172         fiz0             = _mm256_setzero_pd();
 173         fix1             = _mm256_setzero_pd();
 174         fiy1             = _mm256_setzero_pd();
 175         fiz1             = _mm256_setzero_pd();
 176         fix2             = _mm256_setzero_pd();
 177         fiy2             = _mm256_setzero_pd();
 178         fiz2             = _mm256_setzero_pd();
 179         fix3             = _mm256_setzero_pd();
 180         fiy3             = _mm256_setzero_pd();
 181         fiz3             = _mm256_setzero_pd();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm256_setzero_pd();
 185         vvdwsum          = _mm256_setzero_pd();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198             j_coord_offsetC  = DIM*jnrC;
 199             j_coord_offsetD  = DIM*jnrD;
 200
 201             /* load j atom coordinates */
 202             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 204                                                  &jx0,&jy0,&jz0);
 205
 206             /* Calculate displacement vector */
 207             dx00             = _mm256_sub_pd(ix0,jx0);
 208             dy00             = _mm256_sub_pd(iy0,jy0);
 209             dz00             = _mm256_sub_pd(iz0,jz0);
 210             dx10             = _mm256_sub_pd(ix1,jx0);
 211             dy10             = _mm256_sub_pd(iy1,jy0);
 212             dz10             = _mm256_sub_pd(iz1,jz0);
 213             dx20             = _mm256_sub_pd(ix2,jx0);
 214             dy20             = _mm256_sub_pd(iy2,jy0);
 215             dz20             = _mm256_sub_pd(iz2,jz0);
 216             dx30             = _mm256_sub_pd(ix3,jx0);
 217             dy30             = _mm256_sub_pd(iy3,jy0);
 218             dz30             = _mm256_sub_pd(iz3,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 222             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 223             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 224             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 225
 226             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 227             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 228             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 229
 230             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 231             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 232             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 233             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 234
 235             /* Load parameters for j particles */
 236             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 237                                                                  charge+jnrC+0,charge+jnrD+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242
 243             fjx0             = _mm256_setzero_pd();
 244             fjy0             = _mm256_setzero_pd();
 245             fjz0             = _mm256_setzero_pd();
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 253                                             vdwioffsetptr0+vdwjidx0B,
 254                                             vdwioffsetptr0+vdwjidx0C,
 255                                             vdwioffsetptr0+vdwjidx0D,
 256                                             &c6_00,&c12_00);
 257
 258             /* LENNARD-JONES DISPERSION/REPULSION */
 259
 260             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 261             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 262             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 263             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 264             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 268
 269             fscal            = fvdw;
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm256_mul_pd(fscal,dx00);
 273             ty               = _mm256_mul_pd(fscal,dy00);
 274             tz               = _mm256_mul_pd(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm256_add_pd(fix0,tx);
 278             fiy0             = _mm256_add_pd(fiy0,ty);
 279             fiz0             = _mm256_add_pd(fiz0,tz);
 280
 281             fjx0             = _mm256_add_pd(fjx0,tx);
 282             fjy0             = _mm256_add_pd(fjy0,ty);
 283             fjz0             = _mm256_add_pd(fjz0,tz);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq10             = _mm256_mul_pd(iq1,jq0);
 291
 292             /* COULOMB ELECTROSTATICS */
 293             velec            = _mm256_mul_pd(qq10,rinv10);
 294             felec            = _mm256_mul_pd(velec,rinvsq10);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velecsum         = _mm256_add_pd(velecsum,velec);
 298
 299             fscal            = felec;
 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              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq20             = _mm256_mul_pd(iq2,jq0);
 321
 322             /* COULOMB ELECTROSTATICS */
 323             velec            = _mm256_mul_pd(qq20,rinv20);
 324             felec            = _mm256_mul_pd(velec,rinvsq20);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm256_add_pd(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm256_mul_pd(fscal,dx20);
 333             ty               = _mm256_mul_pd(fscal,dy20);
 334             tz               = _mm256_mul_pd(fscal,dz20);
 335
 336             /* Update vectorial force */
 337             fix2             = _mm256_add_pd(fix2,tx);
 338             fiy2             = _mm256_add_pd(fiy2,ty);
 339             fiz2             = _mm256_add_pd(fiz2,tz);
 340
 341             fjx0             = _mm256_add_pd(fjx0,tx);
 342             fjy0             = _mm256_add_pd(fjy0,ty);
 343             fjz0             = _mm256_add_pd(fjz0,tz);
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq30             = _mm256_mul_pd(iq3,jq0);
 351
 352             /* COULOMB ELECTROSTATICS */
 353             velec            = _mm256_mul_pd(qq30,rinv30);
 354             felec            = _mm256_mul_pd(velec,rinvsq30);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm256_add_pd(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm256_mul_pd(fscal,dx30);
 363             ty               = _mm256_mul_pd(fscal,dy30);
 364             tz               = _mm256_mul_pd(fscal,dz30);
 365
 366             /* Update vectorial force */
 367             fix3             = _mm256_add_pd(fix3,tx);
 368             fiy3             = _mm256_add_pd(fiy3,ty);
 369             fiz3             = _mm256_add_pd(fiz3,tz);
 370
 371             fjx0             = _mm256_add_pd(fjx0,tx);
 372             fjy0             = _mm256_add_pd(fjy0,ty);
 373             fjz0             = _mm256_add_pd(fjz0,tz);
 374
 375             fjptrA             = f+j_coord_offsetA;
 376             fjptrB             = f+j_coord_offsetB;
 377             fjptrC             = f+j_coord_offsetC;
 378             fjptrD             = f+j_coord_offsetD;
 379
 380             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 381
 382             /* Inner loop uses 116 flops */
 383         }
 384
 385         if(jidx<j_index_end)
 386         {
 387
 388             /* Get j neighbor index, and coordinate index */
 389             jnrlistA         = jjnr[jidx];
 390             jnrlistB         = jjnr[jidx+1];
 391             jnrlistC         = jjnr[jidx+2];
 392             jnrlistD         = jjnr[jidx+3];
 393             /* Sign of each element will be negative for non-real atoms.
 394              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 395              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 396              */
 397             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 398
 399             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 400             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 401             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 402
 403             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 404             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 405             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 406             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 407             j_coord_offsetA  = DIM*jnrA;
 408             j_coord_offsetB  = DIM*jnrB;
 409             j_coord_offsetC  = DIM*jnrC;
 410             j_coord_offsetD  = DIM*jnrD;
 411
 412             /* load j atom coordinates */
 413             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 414                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 415                                                  &jx0,&jy0,&jz0);
 416
 417             /* Calculate displacement vector */
 418             dx00             = _mm256_sub_pd(ix0,jx0);
 419             dy00             = _mm256_sub_pd(iy0,jy0);
 420             dz00             = _mm256_sub_pd(iz0,jz0);
 421             dx10             = _mm256_sub_pd(ix1,jx0);
 422             dy10             = _mm256_sub_pd(iy1,jy0);
 423             dz10             = _mm256_sub_pd(iz1,jz0);
 424             dx20             = _mm256_sub_pd(ix2,jx0);
 425             dy20             = _mm256_sub_pd(iy2,jy0);
 426             dz20             = _mm256_sub_pd(iz2,jz0);
 427             dx30             = _mm256_sub_pd(ix3,jx0);
 428             dy30             = _mm256_sub_pd(iy3,jy0);
 429             dz30             = _mm256_sub_pd(iz3,jz0);
 430
 431             /* Calculate squared distance and things based on it */
 432             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 433             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 434             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 435             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 436
 437             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 438             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 439             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 440
 441             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 442             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 443             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 444             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 445
 446             /* Load parameters for j particles */
 447             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 448                                                                  charge+jnrC+0,charge+jnrD+0);
 449             vdwjidx0A        = 2*vdwtype[jnrA+0];
 450             vdwjidx0B        = 2*vdwtype[jnrB+0];
 451             vdwjidx0C        = 2*vdwtype[jnrC+0];
 452             vdwjidx0D        = 2*vdwtype[jnrD+0];
 453
 454             fjx0             = _mm256_setzero_pd();
 455             fjy0             = _mm256_setzero_pd();
 456             fjz0             = _mm256_setzero_pd();
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             /* Compute parameters for interactions between i and j atoms */
 463             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 464                                             vdwioffsetptr0+vdwjidx0B,
 465                                             vdwioffsetptr0+vdwjidx0C,
 466                                             vdwioffsetptr0+vdwjidx0D,
 467                                             &c6_00,&c12_00);
 468
 469             /* LENNARD-JONES DISPERSION/REPULSION */
 470
 471             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 472             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 473             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 474             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 475             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 479             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 480
 481             fscal            = fvdw;
 482
 483             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm256_mul_pd(fscal,dx00);
 487             ty               = _mm256_mul_pd(fscal,dy00);
 488             tz               = _mm256_mul_pd(fscal,dz00);
 489
 490             /* Update vectorial force */
 491             fix0             = _mm256_add_pd(fix0,tx);
 492             fiy0             = _mm256_add_pd(fiy0,ty);
 493             fiz0             = _mm256_add_pd(fiz0,tz);
 494
 495             fjx0             = _mm256_add_pd(fjx0,tx);
 496             fjy0             = _mm256_add_pd(fjy0,ty);
 497             fjz0             = _mm256_add_pd(fjz0,tz);
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             qq10             = _mm256_mul_pd(iq1,jq0);
 505
 506             /* COULOMB ELECTROSTATICS */
 507             velec            = _mm256_mul_pd(qq10,rinv10);
 508             felec            = _mm256_mul_pd(velec,rinvsq10);
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velec            = _mm256_andnot_pd(dummy_mask,velec);
 512             velecsum         = _mm256_add_pd(velecsum,velec);
 513
 514             fscal            = felec;
 515
 516             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm256_mul_pd(fscal,dx10);
 520             ty               = _mm256_mul_pd(fscal,dy10);
 521             tz               = _mm256_mul_pd(fscal,dz10);
 522
 523             /* Update vectorial force */
 524             fix1             = _mm256_add_pd(fix1,tx);
 525             fiy1             = _mm256_add_pd(fiy1,ty);
 526             fiz1             = _mm256_add_pd(fiz1,tz);
 527
 528             fjx0             = _mm256_add_pd(fjx0,tx);
 529             fjy0             = _mm256_add_pd(fjy0,ty);
 530             fjz0             = _mm256_add_pd(fjz0,tz);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq20             = _mm256_mul_pd(iq2,jq0);
 538
 539             /* COULOMB ELECTROSTATICS */
 540             velec            = _mm256_mul_pd(qq20,rinv20);
 541             felec            = _mm256_mul_pd(velec,rinvsq20);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm256_andnot_pd(dummy_mask,velec);
 545             velecsum         = _mm256_add_pd(velecsum,velec);
 546
 547             fscal            = felec;
 548
 549             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 550
 551             /* Calculate temporary vectorial force */
 552             tx               = _mm256_mul_pd(fscal,dx20);
 553             ty               = _mm256_mul_pd(fscal,dy20);
 554             tz               = _mm256_mul_pd(fscal,dz20);
 555
 556             /* Update vectorial force */
 557             fix2             = _mm256_add_pd(fix2,tx);
 558             fiy2             = _mm256_add_pd(fiy2,ty);
 559             fiz2             = _mm256_add_pd(fiz2,tz);
 560
 561             fjx0             = _mm256_add_pd(fjx0,tx);
 562             fjy0             = _mm256_add_pd(fjy0,ty);
 563             fjz0             = _mm256_add_pd(fjz0,tz);
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq30             = _mm256_mul_pd(iq3,jq0);
 571
 572             /* COULOMB ELECTROSTATICS */
 573             velec            = _mm256_mul_pd(qq30,rinv30);
 574             felec            = _mm256_mul_pd(velec,rinvsq30);
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velec            = _mm256_andnot_pd(dummy_mask,velec);
 578             velecsum         = _mm256_add_pd(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_pd(fscal,dx30);
 586             ty               = _mm256_mul_pd(fscal,dy30);
 587             tz               = _mm256_mul_pd(fscal,dz30);
 588
 589             /* Update vectorial force */
 590             fix3             = _mm256_add_pd(fix3,tx);
 591             fiy3             = _mm256_add_pd(fiy3,ty);
 592             fiz3             = _mm256_add_pd(fiz3,tz);
 593
 594             fjx0             = _mm256_add_pd(fjx0,tx);
 595             fjy0             = _mm256_add_pd(fjy0,ty);
 596             fjz0             = _mm256_add_pd(fjz0,tz);
 597
 598             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 599             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 600             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 601             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 602
 603             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 604
 605             /* Inner loop uses 116 flops */
 606         }
 607
 608         /* End of innermost loop */
 609
 610         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 611                                                  f+i_coord_offset,fshift+i_shift_offset);
 612
 613         ggid                        = gid[iidx];
 614         /* Update potential energies */
 615         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 616         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 617
 618         /* Increment number of inner iterations */
 619         inneriter                  += j_index_end - j_index_start;
 620
 621         /* Outer loop uses 26 flops */
 622     }
 623
 624     /* Increment number of outer iterations */
 625     outeriter        += nri;
 626
 627     /* Update outer/inner flops */
 628
 629     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 630 }
 631 /*
 632  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_double
 633  * Electrostatics interaction: Coulomb
 634  * VdW interaction:            LennardJones
 635  * Geometry:                   Water4-Particle
 636  * Calculate force/pot:        Force
 637  */
 638 void
 639 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_double
 640                     (t_nblist                    * gmx_restrict       nlist,
 641                      rvec                        * gmx_restrict          xx,
 642                      rvec                        * gmx_restrict          ff,
 643                      t_forcerec                  * gmx_restrict          fr,
 644                      t_mdatoms                   * gmx_restrict     mdatoms,
 645                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 646                      t_nrnb                      * gmx_restrict        nrnb)
 647 {
 648     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 649      * just 0 for non-waters.
 650      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 651      * jnr indices corresponding to data put in the four positions in the SIMD register.
 652      */
 653     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 654     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 655     int              jnrA,jnrB,jnrC,jnrD;
 656     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 657     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 658     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 659     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 660     real             rcutoff_scalar;
 661     real             *shiftvec,*fshift,*x,*f;
 662     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 663     real             scratch[4*DIM];
 664     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 665     real *           vdwioffsetptr0;
 666     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 667     real *           vdwioffsetptr1;
 668     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 669     real *           vdwioffsetptr2;
 670     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 671     real *           vdwioffsetptr3;
 672     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 673     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 674     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 675     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 676     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 677     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 678     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 679     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 680     real             *charge;
 681     int              nvdwtype;
 682     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 683     int              *vdwtype;
 684     real             *vdwparam;
 685     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 686     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 687     __m256d          dummy_mask,cutoff_mask;
 688     __m128           tmpmask0,tmpmask1;
 689     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 690     __m256d          one     = _mm256_set1_pd(1.0);
 691     __m256d          two     = _mm256_set1_pd(2.0);
 692     x                = xx[0];
 693     f                = ff[0];
 694
 695     nri              = nlist->nri;
 696     iinr             = nlist->iinr;
 697     jindex           = nlist->jindex;
 698     jjnr             = nlist->jjnr;
 699     shiftidx         = nlist->shift;
 700     gid              = nlist->gid;
 701     shiftvec         = fr->shift_vec[0];
 702     fshift           = fr->fshift[0];
 703     facel            = _mm256_set1_pd(fr->epsfac);
 704     charge           = mdatoms->chargeA;
 705     nvdwtype         = fr->ntype;
 706     vdwparam         = fr->nbfp;
 707     vdwtype          = mdatoms->typeA;
 708
 709     /* Setup water-specific parameters */
 710     inr              = nlist->iinr[0];
 711     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 712     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 713     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 714     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 715
 716     /* Avoid stupid compiler warnings */
 717     jnrA = jnrB = jnrC = jnrD = 0;
 718     j_coord_offsetA = 0;
 719     j_coord_offsetB = 0;
 720     j_coord_offsetC = 0;
 721     j_coord_offsetD = 0;
 722
 723     outeriter        = 0;
 724     inneriter        = 0;
 725
 726     for(iidx=0;iidx<4*DIM;iidx++)
 727     {
 728         scratch[iidx] = 0.0;
 729     }
 730
 731     /* Start outer loop over neighborlists */
 732     for(iidx=0; iidx<nri; iidx++)
 733     {
 734         /* Load shift vector for this list */
 735         i_shift_offset   = DIM*shiftidx[iidx];
 736
 737         /* Load limits for loop over neighbors */
 738         j_index_start    = jindex[iidx];
 739         j_index_end      = jindex[iidx+1];
 740
 741         /* Get outer coordinate index */
 742         inr              = iinr[iidx];
 743         i_coord_offset   = DIM*inr;
 744
 745         /* Load i particle coords and add shift vector */
 746         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 747                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 748
 749         fix0             = _mm256_setzero_pd();
 750         fiy0             = _mm256_setzero_pd();
 751         fiz0             = _mm256_setzero_pd();
 752         fix1             = _mm256_setzero_pd();
 753         fiy1             = _mm256_setzero_pd();
 754         fiz1             = _mm256_setzero_pd();
 755         fix2             = _mm256_setzero_pd();
 756         fiy2             = _mm256_setzero_pd();
 757         fiz2             = _mm256_setzero_pd();
 758         fix3             = _mm256_setzero_pd();
 759         fiy3             = _mm256_setzero_pd();
 760         fiz3             = _mm256_setzero_pd();
 761
 762         /* Start inner kernel loop */
 763         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 764         {
 765
 766             /* Get j neighbor index, and coordinate index */
 767             jnrA             = jjnr[jidx];
 768             jnrB             = jjnr[jidx+1];
 769             jnrC             = jjnr[jidx+2];
 770             jnrD             = jjnr[jidx+3];
 771             j_coord_offsetA  = DIM*jnrA;
 772             j_coord_offsetB  = DIM*jnrB;
 773             j_coord_offsetC  = DIM*jnrC;
 774             j_coord_offsetD  = DIM*jnrD;
 775
 776             /* load j atom coordinates */
 777             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 778                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 779                                                  &jx0,&jy0,&jz0);
 780
 781             /* Calculate displacement vector */
 782             dx00             = _mm256_sub_pd(ix0,jx0);
 783             dy00             = _mm256_sub_pd(iy0,jy0);
 784             dz00             = _mm256_sub_pd(iz0,jz0);
 785             dx10             = _mm256_sub_pd(ix1,jx0);
 786             dy10             = _mm256_sub_pd(iy1,jy0);
 787             dz10             = _mm256_sub_pd(iz1,jz0);
 788             dx20             = _mm256_sub_pd(ix2,jx0);
 789             dy20             = _mm256_sub_pd(iy2,jy0);
 790             dz20             = _mm256_sub_pd(iz2,jz0);
 791             dx30             = _mm256_sub_pd(ix3,jx0);
 792             dy30             = _mm256_sub_pd(iy3,jy0);
 793             dz30             = _mm256_sub_pd(iz3,jz0);
 794
 795             /* Calculate squared distance and things based on it */
 796             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 797             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 798             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 799             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 800
 801             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 802             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 803             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 804
 805             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 806             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 807             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 808             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 809
 810             /* Load parameters for j particles */
 811             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 812                                                                  charge+jnrC+0,charge+jnrD+0);
 813             vdwjidx0A        = 2*vdwtype[jnrA+0];
 814             vdwjidx0B        = 2*vdwtype[jnrB+0];
 815             vdwjidx0C        = 2*vdwtype[jnrC+0];
 816             vdwjidx0D        = 2*vdwtype[jnrD+0];
 817
 818             fjx0             = _mm256_setzero_pd();
 819             fjy0             = _mm256_setzero_pd();
 820             fjz0             = _mm256_setzero_pd();
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             /* Compute parameters for interactions between i and j atoms */
 827             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 828                                             vdwioffsetptr0+vdwjidx0B,
 829                                             vdwioffsetptr0+vdwjidx0C,
 830                                             vdwioffsetptr0+vdwjidx0D,
 831                                             &c6_00,&c12_00);
 832
 833             /* LENNARD-JONES DISPERSION/REPULSION */
 834
 835             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 836             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 837
 838             fscal            = fvdw;
 839
 840             /* Calculate temporary vectorial force */
 841             tx               = _mm256_mul_pd(fscal,dx00);
 842             ty               = _mm256_mul_pd(fscal,dy00);
 843             tz               = _mm256_mul_pd(fscal,dz00);
 844
 845             /* Update vectorial force */
 846             fix0             = _mm256_add_pd(fix0,tx);
 847             fiy0             = _mm256_add_pd(fiy0,ty);
 848             fiz0             = _mm256_add_pd(fiz0,tz);
 849
 850             fjx0             = _mm256_add_pd(fjx0,tx);
 851             fjy0             = _mm256_add_pd(fjy0,ty);
 852             fjz0             = _mm256_add_pd(fjz0,tz);
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             /* Compute parameters for interactions between i and j atoms */
 859             qq10             = _mm256_mul_pd(iq1,jq0);
 860
 861             /* COULOMB ELECTROSTATICS */
 862             velec            = _mm256_mul_pd(qq10,rinv10);
 863             felec            = _mm256_mul_pd(velec,rinvsq10);
 864
 865             fscal            = felec;
 866
 867             /* Calculate temporary vectorial force */
 868             tx               = _mm256_mul_pd(fscal,dx10);
 869             ty               = _mm256_mul_pd(fscal,dy10);
 870             tz               = _mm256_mul_pd(fscal,dz10);
 871
 872             /* Update vectorial force */
 873             fix1             = _mm256_add_pd(fix1,tx);
 874             fiy1             = _mm256_add_pd(fiy1,ty);
 875             fiz1             = _mm256_add_pd(fiz1,tz);
 876
 877             fjx0             = _mm256_add_pd(fjx0,tx);
 878             fjy0             = _mm256_add_pd(fjy0,ty);
 879             fjz0             = _mm256_add_pd(fjz0,tz);
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq20             = _mm256_mul_pd(iq2,jq0);
 887
 888             /* COULOMB ELECTROSTATICS */
 889             velec            = _mm256_mul_pd(qq20,rinv20);
 890             felec            = _mm256_mul_pd(velec,rinvsq20);
 891
 892             fscal            = felec;
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm256_mul_pd(fscal,dx20);
 896             ty               = _mm256_mul_pd(fscal,dy20);
 897             tz               = _mm256_mul_pd(fscal,dz20);
 898
 899             /* Update vectorial force */
 900             fix2             = _mm256_add_pd(fix2,tx);
 901             fiy2             = _mm256_add_pd(fiy2,ty);
 902             fiz2             = _mm256_add_pd(fiz2,tz);
 903
 904             fjx0             = _mm256_add_pd(fjx0,tx);
 905             fjy0             = _mm256_add_pd(fjy0,ty);
 906             fjz0             = _mm256_add_pd(fjz0,tz);
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             /* Compute parameters for interactions between i and j atoms */
 913             qq30             = _mm256_mul_pd(iq3,jq0);
 914
 915             /* COULOMB ELECTROSTATICS */
 916             velec            = _mm256_mul_pd(qq30,rinv30);
 917             felec            = _mm256_mul_pd(velec,rinvsq30);
 918
 919             fscal            = felec;
 920
 921             /* Calculate temporary vectorial force */
 922             tx               = _mm256_mul_pd(fscal,dx30);
 923             ty               = _mm256_mul_pd(fscal,dy30);
 924             tz               = _mm256_mul_pd(fscal,dz30);
 925
 926             /* Update vectorial force */
 927             fix3             = _mm256_add_pd(fix3,tx);
 928             fiy3             = _mm256_add_pd(fiy3,ty);
 929             fiz3             = _mm256_add_pd(fiz3,tz);
 930
 931             fjx0             = _mm256_add_pd(fjx0,tx);
 932             fjy0             = _mm256_add_pd(fjy0,ty);
 933             fjz0             = _mm256_add_pd(fjz0,tz);
 934
 935             fjptrA             = f+j_coord_offsetA;
 936             fjptrB             = f+j_coord_offsetB;
 937             fjptrC             = f+j_coord_offsetC;
 938             fjptrD             = f+j_coord_offsetD;
 939
 940             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 941
 942             /* Inner loop uses 108 flops */
 943         }
 944
 945         if(jidx<j_index_end)
 946         {
 947
 948             /* Get j neighbor index, and coordinate index */
 949             jnrlistA         = jjnr[jidx];
 950             jnrlistB         = jjnr[jidx+1];
 951             jnrlistC         = jjnr[jidx+2];
 952             jnrlistD         = jjnr[jidx+3];
 953             /* Sign of each element will be negative for non-real atoms.
 954              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 955              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 956              */
 957             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 958
 959             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 960             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 961             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 962
 963             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 964             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 965             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 966             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 967             j_coord_offsetA  = DIM*jnrA;
 968             j_coord_offsetB  = DIM*jnrB;
 969             j_coord_offsetC  = DIM*jnrC;
 970             j_coord_offsetD  = DIM*jnrD;
 971
 972             /* load j atom coordinates */
 973             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 974                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 975                                                  &jx0,&jy0,&jz0);
 976
 977             /* Calculate displacement vector */
 978             dx00             = _mm256_sub_pd(ix0,jx0);
 979             dy00             = _mm256_sub_pd(iy0,jy0);
 980             dz00             = _mm256_sub_pd(iz0,jz0);
 981             dx10             = _mm256_sub_pd(ix1,jx0);
 982             dy10             = _mm256_sub_pd(iy1,jy0);
 983             dz10             = _mm256_sub_pd(iz1,jz0);
 984             dx20             = _mm256_sub_pd(ix2,jx0);
 985             dy20             = _mm256_sub_pd(iy2,jy0);
 986             dz20             = _mm256_sub_pd(iz2,jz0);
 987             dx30             = _mm256_sub_pd(ix3,jx0);
 988             dy30             = _mm256_sub_pd(iy3,jy0);
 989             dz30             = _mm256_sub_pd(iz3,jz0);
 990
 991             /* Calculate squared distance and things based on it */
 992             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 993             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 994             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 995             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 996
 997             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 998             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 999             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1000
1001             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1002             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1003             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1004             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1005
1006             /* Load parameters for j particles */
1007             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1008                                                                  charge+jnrC+0,charge+jnrD+0);
1009             vdwjidx0A        = 2*vdwtype[jnrA+0];
1010             vdwjidx0B        = 2*vdwtype[jnrB+0];
1011             vdwjidx0C        = 2*vdwtype[jnrC+0];
1012             vdwjidx0D        = 2*vdwtype[jnrD+0];
1013
1014             fjx0             = _mm256_setzero_pd();
1015             fjy0             = _mm256_setzero_pd();
1016             fjz0             = _mm256_setzero_pd();
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             /* Compute parameters for interactions between i and j atoms */
1023             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1024                                             vdwioffsetptr0+vdwjidx0B,
1025                                             vdwioffsetptr0+vdwjidx0C,
1026                                             vdwioffsetptr0+vdwjidx0D,
1027                                             &c6_00,&c12_00);
1028
1029             /* LENNARD-JONES DISPERSION/REPULSION */
1030
1031             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1032             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1033
1034             fscal            = fvdw;
1035
1036             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm256_mul_pd(fscal,dx00);
1040             ty               = _mm256_mul_pd(fscal,dy00);
1041             tz               = _mm256_mul_pd(fscal,dz00);
1042
1043             /* Update vectorial force */
1044             fix0             = _mm256_add_pd(fix0,tx);
1045             fiy0             = _mm256_add_pd(fiy0,ty);
1046             fiz0             = _mm256_add_pd(fiz0,tz);
1047
1048             fjx0             = _mm256_add_pd(fjx0,tx);
1049             fjy0             = _mm256_add_pd(fjy0,ty);
1050             fjz0             = _mm256_add_pd(fjz0,tz);
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             /* Compute parameters for interactions between i and j atoms */
1057             qq10             = _mm256_mul_pd(iq1,jq0);
1058
1059             /* COULOMB ELECTROSTATICS */
1060             velec            = _mm256_mul_pd(qq10,rinv10);
1061             felec            = _mm256_mul_pd(velec,rinvsq10);
1062
1063             fscal            = felec;
1064
1065             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm256_mul_pd(fscal,dx10);
1069             ty               = _mm256_mul_pd(fscal,dy10);
1070             tz               = _mm256_mul_pd(fscal,dz10);
1071
1072             /* Update vectorial force */
1073             fix1             = _mm256_add_pd(fix1,tx);
1074             fiy1             = _mm256_add_pd(fiy1,ty);
1075             fiz1             = _mm256_add_pd(fiz1,tz);
1076
1077             fjx0             = _mm256_add_pd(fjx0,tx);
1078             fjy0             = _mm256_add_pd(fjy0,ty);
1079             fjz0             = _mm256_add_pd(fjz0,tz);
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq20             = _mm256_mul_pd(iq2,jq0);
1087
1088             /* COULOMB ELECTROSTATICS */
1089             velec            = _mm256_mul_pd(qq20,rinv20);
1090             felec            = _mm256_mul_pd(velec,rinvsq20);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1095
1096             /* Calculate temporary vectorial force */
1097             tx               = _mm256_mul_pd(fscal,dx20);
1098             ty               = _mm256_mul_pd(fscal,dy20);
1099             tz               = _mm256_mul_pd(fscal,dz20);
1100
1101             /* Update vectorial force */
1102             fix2             = _mm256_add_pd(fix2,tx);
1103             fiy2             = _mm256_add_pd(fiy2,ty);
1104             fiz2             = _mm256_add_pd(fiz2,tz);
1105
1106             fjx0             = _mm256_add_pd(fjx0,tx);
1107             fjy0             = _mm256_add_pd(fjy0,ty);
1108             fjz0             = _mm256_add_pd(fjz0,tz);
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq30             = _mm256_mul_pd(iq3,jq0);
1116
1117             /* COULOMB ELECTROSTATICS */
1118             velec            = _mm256_mul_pd(qq30,rinv30);
1119             felec            = _mm256_mul_pd(velec,rinvsq30);
1120
1121             fscal            = felec;
1122
1123             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm256_mul_pd(fscal,dx30);
1127             ty               = _mm256_mul_pd(fscal,dy30);
1128             tz               = _mm256_mul_pd(fscal,dz30);
1129
1130             /* Update vectorial force */
1131             fix3             = _mm256_add_pd(fix3,tx);
1132             fiy3             = _mm256_add_pd(fiy3,ty);
1133             fiz3             = _mm256_add_pd(fiz3,tz);
1134
1135             fjx0             = _mm256_add_pd(fjx0,tx);
1136             fjy0             = _mm256_add_pd(fjy0,ty);
1137             fjz0             = _mm256_add_pd(fjz0,tz);
1138
1139             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1140             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1141             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1142             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1143
1144             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1145
1146             /* Inner loop uses 108 flops */
1147         }
1148
1149         /* End of innermost loop */
1150
1151         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1152                                                  f+i_coord_offset,fshift+i_shift_offset);
1153
1154         /* Increment number of inner iterations */
1155         inneriter                  += j_index_end - j_index_start;
1156
1157         /* Outer loop uses 24 flops */
1158     }
1159
1160     /* Increment number of outer iterations */
1161     outeriter        += nri;
1162
1163     /* Update outer/inner flops */
1164
1165     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1166 }