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 "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_ElecCoul_VdwLJ_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_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 *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m256d          dummy_mask,cutoff_mask;
 107     __m128           tmpmask0,tmpmask1;
 108     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 109     __m256d          one     = _mm256_set1_pd(1.0);
 110     __m256d          two     = _mm256_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_pd(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 131     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 132     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 167
 168         fix0             = _mm256_setzero_pd();
 169         fiy0             = _mm256_setzero_pd();
 170         fiz0             = _mm256_setzero_pd();
 171         fix1             = _mm256_setzero_pd();
 172         fiy1             = _mm256_setzero_pd();
 173         fiz1             = _mm256_setzero_pd();
 174         fix2             = _mm256_setzero_pd();
 175         fiy2             = _mm256_setzero_pd();
 176         fiz2             = _mm256_setzero_pd();
 177         fix3             = _mm256_setzero_pd();
 178         fiy3             = _mm256_setzero_pd();
 179         fiz3             = _mm256_setzero_pd();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm256_setzero_pd();
 183         vvdwsum          = _mm256_setzero_pd();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 202                                                  &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm256_sub_pd(ix0,jx0);
 206             dy00             = _mm256_sub_pd(iy0,jy0);
 207             dz00             = _mm256_sub_pd(iz0,jz0);
 208             dx10             = _mm256_sub_pd(ix1,jx0);
 209             dy10             = _mm256_sub_pd(iy1,jy0);
 210             dz10             = _mm256_sub_pd(iz1,jz0);
 211             dx20             = _mm256_sub_pd(ix2,jx0);
 212             dy20             = _mm256_sub_pd(iy2,jy0);
 213             dz20             = _mm256_sub_pd(iz2,jz0);
 214             dx30             = _mm256_sub_pd(ix3,jx0);
 215             dy30             = _mm256_sub_pd(iy3,jy0);
 216             dz30             = _mm256_sub_pd(iz3,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 222             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 223
 224             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 225             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 226             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 227
 228             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 229             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 230             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 231             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 235                                                                  charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             fjx0             = _mm256_setzero_pd();
 242             fjy0             = _mm256_setzero_pd();
 243             fjz0             = _mm256_setzero_pd();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 251                                             vdwioffsetptr0+vdwjidx0B,
 252                                             vdwioffsetptr0+vdwjidx0C,
 253                                             vdwioffsetptr0+vdwjidx0D,
 254                                             &c6_00,&c12_00);
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 260             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 261             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 262             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = fvdw;
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm256_mul_pd(fscal,dx00);
 271             ty               = _mm256_mul_pd(fscal,dy00);
 272             tz               = _mm256_mul_pd(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm256_add_pd(fix0,tx);
 276             fiy0             = _mm256_add_pd(fiy0,ty);
 277             fiz0             = _mm256_add_pd(fiz0,tz);
 278
 279             fjx0             = _mm256_add_pd(fjx0,tx);
 280             fjy0             = _mm256_add_pd(fjy0,ty);
 281             fjz0             = _mm256_add_pd(fjz0,tz);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq10             = _mm256_mul_pd(iq1,jq0);
 289
 290             /* COULOMB ELECTROSTATICS */
 291             velec            = _mm256_mul_pd(qq10,rinv10);
 292             felec            = _mm256_mul_pd(velec,rinvsq10);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm256_add_pd(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm256_mul_pd(fscal,dx10);
 301             ty               = _mm256_mul_pd(fscal,dy10);
 302             tz               = _mm256_mul_pd(fscal,dz10);
 303
 304             /* Update vectorial force */
 305             fix1             = _mm256_add_pd(fix1,tx);
 306             fiy1             = _mm256_add_pd(fiy1,ty);
 307             fiz1             = _mm256_add_pd(fiz1,tz);
 308
 309             fjx0             = _mm256_add_pd(fjx0,tx);
 310             fjy0             = _mm256_add_pd(fjy0,ty);
 311             fjz0             = _mm256_add_pd(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq20             = _mm256_mul_pd(iq2,jq0);
 319
 320             /* COULOMB ELECTROSTATICS */
 321             velec            = _mm256_mul_pd(qq20,rinv20);
 322             felec            = _mm256_mul_pd(velec,rinvsq20);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm256_add_pd(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm256_mul_pd(fscal,dx20);
 331             ty               = _mm256_mul_pd(fscal,dy20);
 332             tz               = _mm256_mul_pd(fscal,dz20);
 333
 334             /* Update vectorial force */
 335             fix2             = _mm256_add_pd(fix2,tx);
 336             fiy2             = _mm256_add_pd(fiy2,ty);
 337             fiz2             = _mm256_add_pd(fiz2,tz);
 338
 339             fjx0             = _mm256_add_pd(fjx0,tx);
 340             fjy0             = _mm256_add_pd(fjy0,ty);
 341             fjz0             = _mm256_add_pd(fjz0,tz);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq30             = _mm256_mul_pd(iq3,jq0);
 349
 350             /* COULOMB ELECTROSTATICS */
 351             velec            = _mm256_mul_pd(qq30,rinv30);
 352             felec            = _mm256_mul_pd(velec,rinvsq30);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velecsum         = _mm256_add_pd(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm256_mul_pd(fscal,dx30);
 361             ty               = _mm256_mul_pd(fscal,dy30);
 362             tz               = _mm256_mul_pd(fscal,dz30);
 363
 364             /* Update vectorial force */
 365             fix3             = _mm256_add_pd(fix3,tx);
 366             fiy3             = _mm256_add_pd(fiy3,ty);
 367             fiz3             = _mm256_add_pd(fiz3,tz);
 368
 369             fjx0             = _mm256_add_pd(fjx0,tx);
 370             fjy0             = _mm256_add_pd(fjy0,ty);
 371             fjz0             = _mm256_add_pd(fjz0,tz);
 372
 373             fjptrA             = f+j_coord_offsetA;
 374             fjptrB             = f+j_coord_offsetB;
 375             fjptrC             = f+j_coord_offsetC;
 376             fjptrD             = f+j_coord_offsetD;
 377
 378             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 379
 380             /* Inner loop uses 116 flops */
 381         }
 382
 383         if(jidx<j_index_end)
 384         {
 385
 386             /* Get j neighbor index, and coordinate index */
 387             jnrlistA         = jjnr[jidx];
 388             jnrlistB         = jjnr[jidx+1];
 389             jnrlistC         = jjnr[jidx+2];
 390             jnrlistD         = jjnr[jidx+3];
 391             /* Sign of each element will be negative for non-real atoms.
 392              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 393              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 394              */
 395             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 396
 397             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 398             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 399             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 400
 401             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 402             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 403             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 404             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 405             j_coord_offsetA  = DIM*jnrA;
 406             j_coord_offsetB  = DIM*jnrB;
 407             j_coord_offsetC  = DIM*jnrC;
 408             j_coord_offsetD  = DIM*jnrD;
 409
 410             /* load j atom coordinates */
 411             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 412                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 413                                                  &jx0,&jy0,&jz0);
 414
 415             /* Calculate displacement vector */
 416             dx00             = _mm256_sub_pd(ix0,jx0);
 417             dy00             = _mm256_sub_pd(iy0,jy0);
 418             dz00             = _mm256_sub_pd(iz0,jz0);
 419             dx10             = _mm256_sub_pd(ix1,jx0);
 420             dy10             = _mm256_sub_pd(iy1,jy0);
 421             dz10             = _mm256_sub_pd(iz1,jz0);
 422             dx20             = _mm256_sub_pd(ix2,jx0);
 423             dy20             = _mm256_sub_pd(iy2,jy0);
 424             dz20             = _mm256_sub_pd(iz2,jz0);
 425             dx30             = _mm256_sub_pd(ix3,jx0);
 426             dy30             = _mm256_sub_pd(iy3,jy0);
 427             dz30             = _mm256_sub_pd(iz3,jz0);
 428
 429             /* Calculate squared distance and things based on it */
 430             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 431             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 432             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 433             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 434
 435             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 436             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 437             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 438
 439             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 440             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 441             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 442             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 443
 444             /* Load parameters for j particles */
 445             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 446                                                                  charge+jnrC+0,charge+jnrD+0);
 447             vdwjidx0A        = 2*vdwtype[jnrA+0];
 448             vdwjidx0B        = 2*vdwtype[jnrB+0];
 449             vdwjidx0C        = 2*vdwtype[jnrC+0];
 450             vdwjidx0D        = 2*vdwtype[jnrD+0];
 451
 452             fjx0             = _mm256_setzero_pd();
 453             fjy0             = _mm256_setzero_pd();
 454             fjz0             = _mm256_setzero_pd();
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 462                                             vdwioffsetptr0+vdwjidx0B,
 463                                             vdwioffsetptr0+vdwjidx0C,
 464                                             vdwioffsetptr0+vdwjidx0D,
 465                                             &c6_00,&c12_00);
 466
 467             /* LENNARD-JONES DISPERSION/REPULSION */
 468
 469             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 470             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 471             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 472             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 473             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 477             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 478
 479             fscal            = fvdw;
 480
 481             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 482
 483             /* Calculate temporary vectorial force */
 484             tx               = _mm256_mul_pd(fscal,dx00);
 485             ty               = _mm256_mul_pd(fscal,dy00);
 486             tz               = _mm256_mul_pd(fscal,dz00);
 487
 488             /* Update vectorial force */
 489             fix0             = _mm256_add_pd(fix0,tx);
 490             fiy0             = _mm256_add_pd(fiy0,ty);
 491             fiz0             = _mm256_add_pd(fiz0,tz);
 492
 493             fjx0             = _mm256_add_pd(fjx0,tx);
 494             fjy0             = _mm256_add_pd(fjy0,ty);
 495             fjz0             = _mm256_add_pd(fjz0,tz);
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq10             = _mm256_mul_pd(iq1,jq0);
 503
 504             /* COULOMB ELECTROSTATICS */
 505             velec            = _mm256_mul_pd(qq10,rinv10);
 506             felec            = _mm256_mul_pd(velec,rinvsq10);
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm256_andnot_pd(dummy_mask,velec);
 510             velecsum         = _mm256_add_pd(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm256_mul_pd(fscal,dx10);
 518             ty               = _mm256_mul_pd(fscal,dy10);
 519             tz               = _mm256_mul_pd(fscal,dz10);
 520
 521             /* Update vectorial force */
 522             fix1             = _mm256_add_pd(fix1,tx);
 523             fiy1             = _mm256_add_pd(fiy1,ty);
 524             fiz1             = _mm256_add_pd(fiz1,tz);
 525
 526             fjx0             = _mm256_add_pd(fjx0,tx);
 527             fjy0             = _mm256_add_pd(fjy0,ty);
 528             fjz0             = _mm256_add_pd(fjz0,tz);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq20             = _mm256_mul_pd(iq2,jq0);
 536
 537             /* COULOMB ELECTROSTATICS */
 538             velec            = _mm256_mul_pd(qq20,rinv20);
 539             felec            = _mm256_mul_pd(velec,rinvsq20);
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm256_andnot_pd(dummy_mask,velec);
 543             velecsum         = _mm256_add_pd(velecsum,velec);
 544
 545             fscal            = felec;
 546
 547             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 548
 549             /* Calculate temporary vectorial force */
 550             tx               = _mm256_mul_pd(fscal,dx20);
 551             ty               = _mm256_mul_pd(fscal,dy20);
 552             tz               = _mm256_mul_pd(fscal,dz20);
 553
 554             /* Update vectorial force */
 555             fix2             = _mm256_add_pd(fix2,tx);
 556             fiy2             = _mm256_add_pd(fiy2,ty);
 557             fiz2             = _mm256_add_pd(fiz2,tz);
 558
 559             fjx0             = _mm256_add_pd(fjx0,tx);
 560             fjy0             = _mm256_add_pd(fjy0,ty);
 561             fjz0             = _mm256_add_pd(fjz0,tz);
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq30             = _mm256_mul_pd(iq3,jq0);
 569
 570             /* COULOMB ELECTROSTATICS */
 571             velec            = _mm256_mul_pd(qq30,rinv30);
 572             felec            = _mm256_mul_pd(velec,rinvsq30);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm256_andnot_pd(dummy_mask,velec);
 576             velecsum         = _mm256_add_pd(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_pd(fscal,dx30);
 584             ty               = _mm256_mul_pd(fscal,dy30);
 585             tz               = _mm256_mul_pd(fscal,dz30);
 586
 587             /* Update vectorial force */
 588             fix3             = _mm256_add_pd(fix3,tx);
 589             fiy3             = _mm256_add_pd(fiy3,ty);
 590             fiz3             = _mm256_add_pd(fiz3,tz);
 591
 592             fjx0             = _mm256_add_pd(fjx0,tx);
 593             fjy0             = _mm256_add_pd(fjy0,ty);
 594             fjz0             = _mm256_add_pd(fjz0,tz);
 595
 596             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 597             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 598             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 599             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 600
 601             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 602
 603             /* Inner loop uses 116 flops */
 604         }
 605
 606         /* End of innermost loop */
 607
 608         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 609                                                  f+i_coord_offset,fshift+i_shift_offset);
 610
 611         ggid                        = gid[iidx];
 612         /* Update potential energies */
 613         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 614         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 615
 616         /* Increment number of inner iterations */
 617         inneriter                  += j_index_end - j_index_start;
 618
 619         /* Outer loop uses 26 flops */
 620     }
 621
 622     /* Increment number of outer iterations */
 623     outeriter        += nri;
 624
 625     /* Update outer/inner flops */
 626
 627     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 628 }
 629 /*
 630  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_double
 631  * Electrostatics interaction: Coulomb
 632  * VdW interaction:            LennardJones
 633  * Geometry:                   Water4-Particle
 634  * Calculate force/pot:        Force
 635  */
 636 void
 637 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_double
 638                     (t_nblist                    * gmx_restrict       nlist,
 639                      rvec                        * gmx_restrict          xx,
 640                      rvec                        * gmx_restrict          ff,
 641                      t_forcerec                  * gmx_restrict          fr,
 642                      t_mdatoms                   * gmx_restrict     mdatoms,
 643                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 644                      t_nrnb                      * gmx_restrict        nrnb)
 645 {
 646     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 647      * just 0 for non-waters.
 648      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 649      * jnr indices corresponding to data put in the four positions in the SIMD register.
 650      */
 651     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 652     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 653     int              jnrA,jnrB,jnrC,jnrD;
 654     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 655     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 656     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 657     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 658     real             rcutoff_scalar;
 659     real             *shiftvec,*fshift,*x,*f;
 660     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 661     real             scratch[4*DIM];
 662     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 663     real *           vdwioffsetptr0;
 664     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 665     real *           vdwioffsetptr1;
 666     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 667     real *           vdwioffsetptr2;
 668     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 669     real *           vdwioffsetptr3;
 670     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 671     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 672     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 673     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 674     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 675     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 676     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 677     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 678     real             *charge;
 679     int              nvdwtype;
 680     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 681     int              *vdwtype;
 682     real             *vdwparam;
 683     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 684     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 685     __m256d          dummy_mask,cutoff_mask;
 686     __m128           tmpmask0,tmpmask1;
 687     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 688     __m256d          one     = _mm256_set1_pd(1.0);
 689     __m256d          two     = _mm256_set1_pd(2.0);
 690     x                = xx[0];
 691     f                = ff[0];
 692
 693     nri              = nlist->nri;
 694     iinr             = nlist->iinr;
 695     jindex           = nlist->jindex;
 696     jjnr             = nlist->jjnr;
 697     shiftidx         = nlist->shift;
 698     gid              = nlist->gid;
 699     shiftvec         = fr->shift_vec[0];
 700     fshift           = fr->fshift[0];
 701     facel            = _mm256_set1_pd(fr->epsfac);
 702     charge           = mdatoms->chargeA;
 703     nvdwtype         = fr->ntype;
 704     vdwparam         = fr->nbfp;
 705     vdwtype          = mdatoms->typeA;
 706
 707     /* Setup water-specific parameters */
 708     inr              = nlist->iinr[0];
 709     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 710     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 711     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 712     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 713
 714     /* Avoid stupid compiler warnings */
 715     jnrA = jnrB = jnrC = jnrD = 0;
 716     j_coord_offsetA = 0;
 717     j_coord_offsetB = 0;
 718     j_coord_offsetC = 0;
 719     j_coord_offsetD = 0;
 720
 721     outeriter        = 0;
 722     inneriter        = 0;
 723
 724     for(iidx=0;iidx<4*DIM;iidx++)
 725     {
 726         scratch[iidx] = 0.0;
 727     }
 728
 729     /* Start outer loop over neighborlists */
 730     for(iidx=0; iidx<nri; iidx++)
 731     {
 732         /* Load shift vector for this list */
 733         i_shift_offset   = DIM*shiftidx[iidx];
 734
 735         /* Load limits for loop over neighbors */
 736         j_index_start    = jindex[iidx];
 737         j_index_end      = jindex[iidx+1];
 738
 739         /* Get outer coordinate index */
 740         inr              = iinr[iidx];
 741         i_coord_offset   = DIM*inr;
 742
 743         /* Load i particle coords and add shift vector */
 744         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 745                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 746
 747         fix0             = _mm256_setzero_pd();
 748         fiy0             = _mm256_setzero_pd();
 749         fiz0             = _mm256_setzero_pd();
 750         fix1             = _mm256_setzero_pd();
 751         fiy1             = _mm256_setzero_pd();
 752         fiz1             = _mm256_setzero_pd();
 753         fix2             = _mm256_setzero_pd();
 754         fiy2             = _mm256_setzero_pd();
 755         fiz2             = _mm256_setzero_pd();
 756         fix3             = _mm256_setzero_pd();
 757         fiy3             = _mm256_setzero_pd();
 758         fiz3             = _mm256_setzero_pd();
 759
 760         /* Start inner kernel loop */
 761         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 762         {
 763
 764             /* Get j neighbor index, and coordinate index */
 765             jnrA             = jjnr[jidx];
 766             jnrB             = jjnr[jidx+1];
 767             jnrC             = jjnr[jidx+2];
 768             jnrD             = jjnr[jidx+3];
 769             j_coord_offsetA  = DIM*jnrA;
 770             j_coord_offsetB  = DIM*jnrB;
 771             j_coord_offsetC  = DIM*jnrC;
 772             j_coord_offsetD  = DIM*jnrD;
 773
 774             /* load j atom coordinates */
 775             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 776                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 777                                                  &jx0,&jy0,&jz0);
 778
 779             /* Calculate displacement vector */
 780             dx00             = _mm256_sub_pd(ix0,jx0);
 781             dy00             = _mm256_sub_pd(iy0,jy0);
 782             dz00             = _mm256_sub_pd(iz0,jz0);
 783             dx10             = _mm256_sub_pd(ix1,jx0);
 784             dy10             = _mm256_sub_pd(iy1,jy0);
 785             dz10             = _mm256_sub_pd(iz1,jz0);
 786             dx20             = _mm256_sub_pd(ix2,jx0);
 787             dy20             = _mm256_sub_pd(iy2,jy0);
 788             dz20             = _mm256_sub_pd(iz2,jz0);
 789             dx30             = _mm256_sub_pd(ix3,jx0);
 790             dy30             = _mm256_sub_pd(iy3,jy0);
 791             dz30             = _mm256_sub_pd(iz3,jz0);
 792
 793             /* Calculate squared distance and things based on it */
 794             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 795             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 796             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 797             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 798
 799             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 800             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 801             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 802
 803             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 804             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 805             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 806             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 807
 808             /* Load parameters for j particles */
 809             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 810                                                                  charge+jnrC+0,charge+jnrD+0);
 811             vdwjidx0A        = 2*vdwtype[jnrA+0];
 812             vdwjidx0B        = 2*vdwtype[jnrB+0];
 813             vdwjidx0C        = 2*vdwtype[jnrC+0];
 814             vdwjidx0D        = 2*vdwtype[jnrD+0];
 815
 816             fjx0             = _mm256_setzero_pd();
 817             fjy0             = _mm256_setzero_pd();
 818             fjz0             = _mm256_setzero_pd();
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 826                                             vdwioffsetptr0+vdwjidx0B,
 827                                             vdwioffsetptr0+vdwjidx0C,
 828                                             vdwioffsetptr0+vdwjidx0D,
 829                                             &c6_00,&c12_00);
 830
 831             /* LENNARD-JONES DISPERSION/REPULSION */
 832
 833             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 834             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 835
 836             fscal            = fvdw;
 837
 838             /* Calculate temporary vectorial force */
 839             tx               = _mm256_mul_pd(fscal,dx00);
 840             ty               = _mm256_mul_pd(fscal,dy00);
 841             tz               = _mm256_mul_pd(fscal,dz00);
 842
 843             /* Update vectorial force */
 844             fix0             = _mm256_add_pd(fix0,tx);
 845             fiy0             = _mm256_add_pd(fiy0,ty);
 846             fiz0             = _mm256_add_pd(fiz0,tz);
 847
 848             fjx0             = _mm256_add_pd(fjx0,tx);
 849             fjy0             = _mm256_add_pd(fjy0,ty);
 850             fjz0             = _mm256_add_pd(fjz0,tz);
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             qq10             = _mm256_mul_pd(iq1,jq0);
 858
 859             /* COULOMB ELECTROSTATICS */
 860             velec            = _mm256_mul_pd(qq10,rinv10);
 861             felec            = _mm256_mul_pd(velec,rinvsq10);
 862
 863             fscal            = felec;
 864
 865             /* Calculate temporary vectorial force */
 866             tx               = _mm256_mul_pd(fscal,dx10);
 867             ty               = _mm256_mul_pd(fscal,dy10);
 868             tz               = _mm256_mul_pd(fscal,dz10);
 869
 870             /* Update vectorial force */
 871             fix1             = _mm256_add_pd(fix1,tx);
 872             fiy1             = _mm256_add_pd(fiy1,ty);
 873             fiz1             = _mm256_add_pd(fiz1,tz);
 874
 875             fjx0             = _mm256_add_pd(fjx0,tx);
 876             fjy0             = _mm256_add_pd(fjy0,ty);
 877             fjz0             = _mm256_add_pd(fjz0,tz);
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq20             = _mm256_mul_pd(iq2,jq0);
 885
 886             /* COULOMB ELECTROSTATICS */
 887             velec            = _mm256_mul_pd(qq20,rinv20);
 888             felec            = _mm256_mul_pd(velec,rinvsq20);
 889
 890             fscal            = felec;
 891
 892             /* Calculate temporary vectorial force */
 893             tx               = _mm256_mul_pd(fscal,dx20);
 894             ty               = _mm256_mul_pd(fscal,dy20);
 895             tz               = _mm256_mul_pd(fscal,dz20);
 896
 897             /* Update vectorial force */
 898             fix2             = _mm256_add_pd(fix2,tx);
 899             fiy2             = _mm256_add_pd(fiy2,ty);
 900             fiz2             = _mm256_add_pd(fiz2,tz);
 901
 902             fjx0             = _mm256_add_pd(fjx0,tx);
 903             fjy0             = _mm256_add_pd(fjy0,ty);
 904             fjz0             = _mm256_add_pd(fjz0,tz);
 905
 906             /**************************
 907              * CALCULATE INTERACTIONS *
 908              **************************/
 909
 910             /* Compute parameters for interactions between i and j atoms */
 911             qq30             = _mm256_mul_pd(iq3,jq0);
 912
 913             /* COULOMB ELECTROSTATICS */
 914             velec            = _mm256_mul_pd(qq30,rinv30);
 915             felec            = _mm256_mul_pd(velec,rinvsq30);
 916
 917             fscal            = felec;
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm256_mul_pd(fscal,dx30);
 921             ty               = _mm256_mul_pd(fscal,dy30);
 922             tz               = _mm256_mul_pd(fscal,dz30);
 923
 924             /* Update vectorial force */
 925             fix3             = _mm256_add_pd(fix3,tx);
 926             fiy3             = _mm256_add_pd(fiy3,ty);
 927             fiz3             = _mm256_add_pd(fiz3,tz);
 928
 929             fjx0             = _mm256_add_pd(fjx0,tx);
 930             fjy0             = _mm256_add_pd(fjy0,ty);
 931             fjz0             = _mm256_add_pd(fjz0,tz);
 932
 933             fjptrA             = f+j_coord_offsetA;
 934             fjptrB             = f+j_coord_offsetB;
 935             fjptrC             = f+j_coord_offsetC;
 936             fjptrD             = f+j_coord_offsetD;
 937
 938             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 939
 940             /* Inner loop uses 108 flops */
 941         }
 942
 943         if(jidx<j_index_end)
 944         {
 945
 946             /* Get j neighbor index, and coordinate index */
 947             jnrlistA         = jjnr[jidx];
 948             jnrlistB         = jjnr[jidx+1];
 949             jnrlistC         = jjnr[jidx+2];
 950             jnrlistD         = jjnr[jidx+3];
 951             /* Sign of each element will be negative for non-real atoms.
 952              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 953              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 954              */
 955             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 956
 957             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 958             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 959             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 960
 961             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 962             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 963             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 964             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 965             j_coord_offsetA  = DIM*jnrA;
 966             j_coord_offsetB  = DIM*jnrB;
 967             j_coord_offsetC  = DIM*jnrC;
 968             j_coord_offsetD  = DIM*jnrD;
 969
 970             /* load j atom coordinates */
 971             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 972                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 973                                                  &jx0,&jy0,&jz0);
 974
 975             /* Calculate displacement vector */
 976             dx00             = _mm256_sub_pd(ix0,jx0);
 977             dy00             = _mm256_sub_pd(iy0,jy0);
 978             dz00             = _mm256_sub_pd(iz0,jz0);
 979             dx10             = _mm256_sub_pd(ix1,jx0);
 980             dy10             = _mm256_sub_pd(iy1,jy0);
 981             dz10             = _mm256_sub_pd(iz1,jz0);
 982             dx20             = _mm256_sub_pd(ix2,jx0);
 983             dy20             = _mm256_sub_pd(iy2,jy0);
 984             dz20             = _mm256_sub_pd(iz2,jz0);
 985             dx30             = _mm256_sub_pd(ix3,jx0);
 986             dy30             = _mm256_sub_pd(iy3,jy0);
 987             dz30             = _mm256_sub_pd(iz3,jz0);
 988
 989             /* Calculate squared distance and things based on it */
 990             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 991             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 992             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 993             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 994
 995             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 996             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 997             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 998
 999             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1000             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1001             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1002             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1003
1004             /* Load parameters for j particles */
1005             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1006                                                                  charge+jnrC+0,charge+jnrD+0);
1007             vdwjidx0A        = 2*vdwtype[jnrA+0];
1008             vdwjidx0B        = 2*vdwtype[jnrB+0];
1009             vdwjidx0C        = 2*vdwtype[jnrC+0];
1010             vdwjidx0D        = 2*vdwtype[jnrD+0];
1011
1012             fjx0             = _mm256_setzero_pd();
1013             fjy0             = _mm256_setzero_pd();
1014             fjz0             = _mm256_setzero_pd();
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1022                                             vdwioffsetptr0+vdwjidx0B,
1023                                             vdwioffsetptr0+vdwjidx0C,
1024                                             vdwioffsetptr0+vdwjidx0D,
1025                                             &c6_00,&c12_00);
1026
1027             /* LENNARD-JONES DISPERSION/REPULSION */
1028
1029             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1030             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1031
1032             fscal            = fvdw;
1033
1034             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_pd(fscal,dx00);
1038             ty               = _mm256_mul_pd(fscal,dy00);
1039             tz               = _mm256_mul_pd(fscal,dz00);
1040
1041             /* Update vectorial force */
1042             fix0             = _mm256_add_pd(fix0,tx);
1043             fiy0             = _mm256_add_pd(fiy0,ty);
1044             fiz0             = _mm256_add_pd(fiz0,tz);
1045
1046             fjx0             = _mm256_add_pd(fjx0,tx);
1047             fjy0             = _mm256_add_pd(fjy0,ty);
1048             fjz0             = _mm256_add_pd(fjz0,tz);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq10             = _mm256_mul_pd(iq1,jq0);
1056
1057             /* COULOMB ELECTROSTATICS */
1058             velec            = _mm256_mul_pd(qq10,rinv10);
1059             felec            = _mm256_mul_pd(velec,rinvsq10);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1064
1065             /* Calculate temporary vectorial force */
1066             tx               = _mm256_mul_pd(fscal,dx10);
1067             ty               = _mm256_mul_pd(fscal,dy10);
1068             tz               = _mm256_mul_pd(fscal,dz10);
1069
1070             /* Update vectorial force */
1071             fix1             = _mm256_add_pd(fix1,tx);
1072             fiy1             = _mm256_add_pd(fiy1,ty);
1073             fiz1             = _mm256_add_pd(fiz1,tz);
1074
1075             fjx0             = _mm256_add_pd(fjx0,tx);
1076             fjy0             = _mm256_add_pd(fjy0,ty);
1077             fjz0             = _mm256_add_pd(fjz0,tz);
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             /* Compute parameters for interactions between i and j atoms */
1084             qq20             = _mm256_mul_pd(iq2,jq0);
1085
1086             /* COULOMB ELECTROSTATICS */
1087             velec            = _mm256_mul_pd(qq20,rinv20);
1088             felec            = _mm256_mul_pd(velec,rinvsq20);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm256_mul_pd(fscal,dx20);
1096             ty               = _mm256_mul_pd(fscal,dy20);
1097             tz               = _mm256_mul_pd(fscal,dz20);
1098
1099             /* Update vectorial force */
1100             fix2             = _mm256_add_pd(fix2,tx);
1101             fiy2             = _mm256_add_pd(fiy2,ty);
1102             fiz2             = _mm256_add_pd(fiz2,tz);
1103
1104             fjx0             = _mm256_add_pd(fjx0,tx);
1105             fjy0             = _mm256_add_pd(fjy0,ty);
1106             fjz0             = _mm256_add_pd(fjz0,tz);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             /* Compute parameters for interactions between i and j atoms */
1113             qq30             = _mm256_mul_pd(iq3,jq0);
1114
1115             /* COULOMB ELECTROSTATICS */
1116             velec            = _mm256_mul_pd(qq30,rinv30);
1117             felec            = _mm256_mul_pd(velec,rinvsq30);
1118
1119             fscal            = felec;
1120
1121             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm256_mul_pd(fscal,dx30);
1125             ty               = _mm256_mul_pd(fscal,dy30);
1126             tz               = _mm256_mul_pd(fscal,dz30);
1127
1128             /* Update vectorial force */
1129             fix3             = _mm256_add_pd(fix3,tx);
1130             fiy3             = _mm256_add_pd(fiy3,ty);
1131             fiz3             = _mm256_add_pd(fiz3,tz);
1132
1133             fjx0             = _mm256_add_pd(fjx0,tx);
1134             fjy0             = _mm256_add_pd(fjy0,ty);
1135             fjz0             = _mm256_add_pd(fjz0,tz);
1136
1137             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1138             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1139             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1140             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1141
1142             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1143
1144             /* Inner loop uses 108 flops */
1145         }
1146
1147         /* End of innermost loop */
1148
1149         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1150                                                  f+i_coord_offset,fshift+i_shift_offset);
1151
1152         /* Increment number of inner iterations */
1153         inneriter                  += j_index_end - j_index_start;
1154
1155         /* Outer loop uses 24 flops */
1156     }
1157
1158     /* Increment number of outer iterations */
1159     outeriter        += nri;
1160
1161     /* Update outer/inner flops */
1162
1163     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1164 }