src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_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_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_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     krf              = _mm256_set1_pd(fr->ic->k_rf);
 125     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 126     crf              = _mm256_set1_pd(fr->ic->c_rf);
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 134     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 135     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 139     rcutoff_scalar   = fr->rcoulomb;
 140     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 141     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 142
 143     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 144     rvdw             = _mm256_set1_pd(fr->rvdw);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     for(iidx=0;iidx<4*DIM;iidx++)
 157     {
 158         scratch[iidx] = 0.0;
 159     }
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 177                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 178
 179         fix0             = _mm256_setzero_pd();
 180         fiy0             = _mm256_setzero_pd();
 181         fiz0             = _mm256_setzero_pd();
 182         fix1             = _mm256_setzero_pd();
 183         fiy1             = _mm256_setzero_pd();
 184         fiz1             = _mm256_setzero_pd();
 185         fix2             = _mm256_setzero_pd();
 186         fiy2             = _mm256_setzero_pd();
 187         fiz2             = _mm256_setzero_pd();
 188         fix3             = _mm256_setzero_pd();
 189         fiy3             = _mm256_setzero_pd();
 190         fiz3             = _mm256_setzero_pd();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm256_setzero_pd();
 194         vvdwsum          = _mm256_setzero_pd();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209
 210             /* load j atom coordinates */
 211             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 212                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 213                                                  &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm256_sub_pd(ix0,jx0);
 217             dy00             = _mm256_sub_pd(iy0,jy0);
 218             dz00             = _mm256_sub_pd(iz0,jz0);
 219             dx10             = _mm256_sub_pd(ix1,jx0);
 220             dy10             = _mm256_sub_pd(iy1,jy0);
 221             dz10             = _mm256_sub_pd(iz1,jz0);
 222             dx20             = _mm256_sub_pd(ix2,jx0);
 223             dy20             = _mm256_sub_pd(iy2,jy0);
 224             dz20             = _mm256_sub_pd(iz2,jz0);
 225             dx30             = _mm256_sub_pd(ix3,jx0);
 226             dy30             = _mm256_sub_pd(iy3,jy0);
 227             dz30             = _mm256_sub_pd(iz3,jz0);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 231             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 232             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 233             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 234
 235             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 236             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 237             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 238
 239             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 240             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 241             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 242             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 246                                                                  charge+jnrC+0,charge+jnrD+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251
 252             fjx0             = _mm256_setzero_pd();
 253             fjy0             = _mm256_setzero_pd();
 254             fjz0             = _mm256_setzero_pd();
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 261             {
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 265                                             vdwioffsetptr0+vdwjidx0B,
 266                                             vdwioffsetptr0+vdwjidx0C,
 267                                             vdwioffsetptr0+vdwjidx0D,
 268                                             &c6_00,&c12_00);
 269
 270             /* LENNARD-JONES DISPERSION/REPULSION */
 271
 272             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 273             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 274             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 275             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 276                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 277             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 278
 279             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 283             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 284
 285             fscal            = fvdw;
 286
 287             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm256_mul_pd(fscal,dx00);
 291             ty               = _mm256_mul_pd(fscal,dy00);
 292             tz               = _mm256_mul_pd(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm256_add_pd(fix0,tx);
 296             fiy0             = _mm256_add_pd(fiy0,ty);
 297             fiz0             = _mm256_add_pd(fiz0,tz);
 298
 299             fjx0             = _mm256_add_pd(fjx0,tx);
 300             fjy0             = _mm256_add_pd(fjy0,ty);
 301             fjz0             = _mm256_add_pd(fjz0,tz);
 302
 303             }
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 310             {
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm256_mul_pd(iq1,jq0);
 314
 315             /* REACTION-FIELD ELECTROSTATICS */
 316             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 317             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 318
 319             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velec            = _mm256_and_pd(velec,cutoff_mask);
 323             velecsum         = _mm256_add_pd(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm256_mul_pd(fscal,dx10);
 331             ty               = _mm256_mul_pd(fscal,dy10);
 332             tz               = _mm256_mul_pd(fscal,dz10);
 333
 334             /* Update vectorial force */
 335             fix1             = _mm256_add_pd(fix1,tx);
 336             fiy1             = _mm256_add_pd(fiy1,ty);
 337             fiz1             = _mm256_add_pd(fiz1,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
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 350             {
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm256_mul_pd(iq2,jq0);
 354
 355             /* REACTION-FIELD ELECTROSTATICS */
 356             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 357             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 358
 359             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm256_and_pd(velec,cutoff_mask);
 363             velecsum         = _mm256_add_pd(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm256_mul_pd(fscal,dx20);
 371             ty               = _mm256_mul_pd(fscal,dy20);
 372             tz               = _mm256_mul_pd(fscal,dz20);
 373
 374             /* Update vectorial force */
 375             fix2             = _mm256_add_pd(fix2,tx);
 376             fiy2             = _mm256_add_pd(fiy2,ty);
 377             fiz2             = _mm256_add_pd(fiz2,tz);
 378
 379             fjx0             = _mm256_add_pd(fjx0,tx);
 380             fjy0             = _mm256_add_pd(fjy0,ty);
 381             fjz0             = _mm256_add_pd(fjz0,tz);
 382
 383             }
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 390             {
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq30             = _mm256_mul_pd(iq3,jq0);
 394
 395             /* REACTION-FIELD ELECTROSTATICS */
 396             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 397             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 398
 399             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velec            = _mm256_and_pd(velec,cutoff_mask);
 403             velecsum         = _mm256_add_pd(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm256_mul_pd(fscal,dx30);
 411             ty               = _mm256_mul_pd(fscal,dy30);
 412             tz               = _mm256_mul_pd(fscal,dz30);
 413
 414             /* Update vectorial force */
 415             fix3             = _mm256_add_pd(fix3,tx);
 416             fiy3             = _mm256_add_pd(fiy3,ty);
 417             fiz3             = _mm256_add_pd(fiz3,tz);
 418
 419             fjx0             = _mm256_add_pd(fjx0,tx);
 420             fjy0             = _mm256_add_pd(fjy0,ty);
 421             fjz0             = _mm256_add_pd(fjz0,tz);
 422
 423             }
 424
 425             fjptrA             = f+j_coord_offsetA;
 426             fjptrB             = f+j_coord_offsetB;
 427             fjptrC             = f+j_coord_offsetC;
 428             fjptrD             = f+j_coord_offsetD;
 429
 430             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 431
 432             /* Inner loop uses 152 flops */
 433         }
 434
 435         if(jidx<j_index_end)
 436         {
 437
 438             /* Get j neighbor index, and coordinate index */
 439             jnrlistA         = jjnr[jidx];
 440             jnrlistB         = jjnr[jidx+1];
 441             jnrlistC         = jjnr[jidx+2];
 442             jnrlistD         = jjnr[jidx+3];
 443             /* Sign of each element will be negative for non-real atoms.
 444              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 445              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 446              */
 447             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 448
 449             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 450             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 451             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 452
 453             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 454             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 455             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 456             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 457             j_coord_offsetA  = DIM*jnrA;
 458             j_coord_offsetB  = DIM*jnrB;
 459             j_coord_offsetC  = DIM*jnrC;
 460             j_coord_offsetD  = DIM*jnrD;
 461
 462             /* load j atom coordinates */
 463             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 464                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 465                                                  &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 468             dx00             = _mm256_sub_pd(ix0,jx0);
 469             dy00             = _mm256_sub_pd(iy0,jy0);
 470             dz00             = _mm256_sub_pd(iz0,jz0);
 471             dx10             = _mm256_sub_pd(ix1,jx0);
 472             dy10             = _mm256_sub_pd(iy1,jy0);
 473             dz10             = _mm256_sub_pd(iz1,jz0);
 474             dx20             = _mm256_sub_pd(ix2,jx0);
 475             dy20             = _mm256_sub_pd(iy2,jy0);
 476             dz20             = _mm256_sub_pd(iz2,jz0);
 477             dx30             = _mm256_sub_pd(ix3,jx0);
 478             dy30             = _mm256_sub_pd(iy3,jy0);
 479             dz30             = _mm256_sub_pd(iz3,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 483             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 484             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 485             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 486
 487             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 488             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 489             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 490
 491             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 492             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 493             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 494             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 495
 496             /* Load parameters for j particles */
 497             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 498                                                                  charge+jnrC+0,charge+jnrD+0);
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501             vdwjidx0C        = 2*vdwtype[jnrC+0];
 502             vdwjidx0D        = 2*vdwtype[jnrD+0];
 503
 504             fjx0             = _mm256_setzero_pd();
 505             fjy0             = _mm256_setzero_pd();
 506             fjz0             = _mm256_setzero_pd();
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 513             {
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 517                                             vdwioffsetptr0+vdwjidx0B,
 518                                             vdwioffsetptr0+vdwjidx0C,
 519                                             vdwioffsetptr0+vdwjidx0D,
 520                                             &c6_00,&c12_00);
 521
 522             /* LENNARD-JONES DISPERSION/REPULSION */
 523
 524             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 525             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 526             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 527             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 528                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 529             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 530
 531             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 535             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 536             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 537
 538             fscal            = fvdw;
 539
 540             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 541
 542             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm256_mul_pd(fscal,dx00);
 546             ty               = _mm256_mul_pd(fscal,dy00);
 547             tz               = _mm256_mul_pd(fscal,dz00);
 548
 549             /* Update vectorial force */
 550             fix0             = _mm256_add_pd(fix0,tx);
 551             fiy0             = _mm256_add_pd(fiy0,ty);
 552             fiz0             = _mm256_add_pd(fiz0,tz);
 553
 554             fjx0             = _mm256_add_pd(fjx0,tx);
 555             fjy0             = _mm256_add_pd(fjy0,ty);
 556             fjz0             = _mm256_add_pd(fjz0,tz);
 557
 558             }
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 565             {
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq10             = _mm256_mul_pd(iq1,jq0);
 569
 570             /* REACTION-FIELD ELECTROSTATICS */
 571             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 572             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 573
 574             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velec            = _mm256_and_pd(velec,cutoff_mask);
 578             velec            = _mm256_andnot_pd(dummy_mask,velec);
 579             velecsum         = _mm256_add_pd(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 584
 585             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm256_mul_pd(fscal,dx10);
 589             ty               = _mm256_mul_pd(fscal,dy10);
 590             tz               = _mm256_mul_pd(fscal,dz10);
 591
 592             /* Update vectorial force */
 593             fix1             = _mm256_add_pd(fix1,tx);
 594             fiy1             = _mm256_add_pd(fiy1,ty);
 595             fiz1             = _mm256_add_pd(fiz1,tz);
 596
 597             fjx0             = _mm256_add_pd(fjx0,tx);
 598             fjy0             = _mm256_add_pd(fjy0,ty);
 599             fjz0             = _mm256_add_pd(fjz0,tz);
 600
 601             }
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 608             {
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq20             = _mm256_mul_pd(iq2,jq0);
 612
 613             /* REACTION-FIELD ELECTROSTATICS */
 614             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 615             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 616
 617             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm256_and_pd(velec,cutoff_mask);
 621             velec            = _mm256_andnot_pd(dummy_mask,velec);
 622             velecsum         = _mm256_add_pd(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 627
 628             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 629
 630             /* Calculate temporary vectorial force */
 631             tx               = _mm256_mul_pd(fscal,dx20);
 632             ty               = _mm256_mul_pd(fscal,dy20);
 633             tz               = _mm256_mul_pd(fscal,dz20);
 634
 635             /* Update vectorial force */
 636             fix2             = _mm256_add_pd(fix2,tx);
 637             fiy2             = _mm256_add_pd(fiy2,ty);
 638             fiz2             = _mm256_add_pd(fiz2,tz);
 639
 640             fjx0             = _mm256_add_pd(fjx0,tx);
 641             fjy0             = _mm256_add_pd(fjy0,ty);
 642             fjz0             = _mm256_add_pd(fjz0,tz);
 643
 644             }
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 651             {
 652
 653             /* Compute parameters for interactions between i and j atoms */
 654             qq30             = _mm256_mul_pd(iq3,jq0);
 655
 656             /* REACTION-FIELD ELECTROSTATICS */
 657             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 658             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 659
 660             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 661
 662             /* Update potential sum for this i atom from the interaction with this j atom. */
 663             velec            = _mm256_and_pd(velec,cutoff_mask);
 664             velec            = _mm256_andnot_pd(dummy_mask,velec);
 665             velecsum         = _mm256_add_pd(velecsum,velec);
 666
 667             fscal            = felec;
 668
 669             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 670
 671             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 672
 673             /* Calculate temporary vectorial force */
 674             tx               = _mm256_mul_pd(fscal,dx30);
 675             ty               = _mm256_mul_pd(fscal,dy30);
 676             tz               = _mm256_mul_pd(fscal,dz30);
 677
 678             /* Update vectorial force */
 679             fix3             = _mm256_add_pd(fix3,tx);
 680             fiy3             = _mm256_add_pd(fiy3,ty);
 681             fiz3             = _mm256_add_pd(fiz3,tz);
 682
 683             fjx0             = _mm256_add_pd(fjx0,tx);
 684             fjy0             = _mm256_add_pd(fjy0,ty);
 685             fjz0             = _mm256_add_pd(fjz0,tz);
 686
 687             }
 688
 689             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 690             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 691             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 692             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 693
 694             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 695
 696             /* Inner loop uses 152 flops */
 697         }
 698
 699         /* End of innermost loop */
 700
 701         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 702                                                  f+i_coord_offset,fshift+i_shift_offset);
 703
 704         ggid                        = gid[iidx];
 705         /* Update potential energies */
 706         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 707         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 708
 709         /* Increment number of inner iterations */
 710         inneriter                  += j_index_end - j_index_start;
 711
 712         /* Outer loop uses 26 flops */
 713     }
 714
 715     /* Increment number of outer iterations */
 716     outeriter        += nri;
 717
 718     /* Update outer/inner flops */
 719
 720     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*152);
 721 }
 722 /*
 723  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
 724  * Electrostatics interaction: ReactionField
 725  * VdW interaction:            LennardJones
 726  * Geometry:                   Water4-Particle
 727  * Calculate force/pot:        Force
 728  */
 729 void
 730 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
 731                     (t_nblist                    * gmx_restrict       nlist,
 732                      rvec                        * gmx_restrict          xx,
 733                      rvec                        * gmx_restrict          ff,
 734                      t_forcerec                  * gmx_restrict          fr,
 735                      t_mdatoms                   * gmx_restrict     mdatoms,
 736                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 737                      t_nrnb                      * gmx_restrict        nrnb)
 738 {
 739     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 740      * just 0 for non-waters.
 741      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 742      * jnr indices corresponding to data put in the four positions in the SIMD register.
 743      */
 744     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 745     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 746     int              jnrA,jnrB,jnrC,jnrD;
 747     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 748     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 749     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 750     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 751     real             rcutoff_scalar;
 752     real             *shiftvec,*fshift,*x,*f;
 753     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 754     real             scratch[4*DIM];
 755     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 756     real *           vdwioffsetptr0;
 757     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 758     real *           vdwioffsetptr1;
 759     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 760     real *           vdwioffsetptr2;
 761     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 762     real *           vdwioffsetptr3;
 763     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 764     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 765     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 766     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 767     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 768     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 769     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 770     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 771     real             *charge;
 772     int              nvdwtype;
 773     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 774     int              *vdwtype;
 775     real             *vdwparam;
 776     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 777     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 778     __m256d          dummy_mask,cutoff_mask;
 779     __m128           tmpmask0,tmpmask1;
 780     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 781     __m256d          one     = _mm256_set1_pd(1.0);
 782     __m256d          two     = _mm256_set1_pd(2.0);
 783     x                = xx[0];
 784     f                = ff[0];
 785
 786     nri              = nlist->nri;
 787     iinr             = nlist->iinr;
 788     jindex           = nlist->jindex;
 789     jjnr             = nlist->jjnr;
 790     shiftidx         = nlist->shift;
 791     gid              = nlist->gid;
 792     shiftvec         = fr->shift_vec[0];
 793     fshift           = fr->fshift[0];
 794     facel            = _mm256_set1_pd(fr->epsfac);
 795     charge           = mdatoms->chargeA;
 796     krf              = _mm256_set1_pd(fr->ic->k_rf);
 797     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 798     crf              = _mm256_set1_pd(fr->ic->c_rf);
 799     nvdwtype         = fr->ntype;
 800     vdwparam         = fr->nbfp;
 801     vdwtype          = mdatoms->typeA;
 802
 803     /* Setup water-specific parameters */
 804     inr              = nlist->iinr[0];
 805     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 806     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 807     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 808     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 809
 810     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 811     rcutoff_scalar   = fr->rcoulomb;
 812     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 813     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 814
 815     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 816     rvdw             = _mm256_set1_pd(fr->rvdw);
 817
 818     /* Avoid stupid compiler warnings */
 819     jnrA = jnrB = jnrC = jnrD = 0;
 820     j_coord_offsetA = 0;
 821     j_coord_offsetB = 0;
 822     j_coord_offsetC = 0;
 823     j_coord_offsetD = 0;
 824
 825     outeriter        = 0;
 826     inneriter        = 0;
 827
 828     for(iidx=0;iidx<4*DIM;iidx++)
 829     {
 830         scratch[iidx] = 0.0;
 831     }
 832
 833     /* Start outer loop over neighborlists */
 834     for(iidx=0; iidx<nri; iidx++)
 835     {
 836         /* Load shift vector for this list */
 837         i_shift_offset   = DIM*shiftidx[iidx];
 838
 839         /* Load limits for loop over neighbors */
 840         j_index_start    = jindex[iidx];
 841         j_index_end      = jindex[iidx+1];
 842
 843         /* Get outer coordinate index */
 844         inr              = iinr[iidx];
 845         i_coord_offset   = DIM*inr;
 846
 847         /* Load i particle coords and add shift vector */
 848         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 849                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 850
 851         fix0             = _mm256_setzero_pd();
 852         fiy0             = _mm256_setzero_pd();
 853         fiz0             = _mm256_setzero_pd();
 854         fix1             = _mm256_setzero_pd();
 855         fiy1             = _mm256_setzero_pd();
 856         fiz1             = _mm256_setzero_pd();
 857         fix2             = _mm256_setzero_pd();
 858         fiy2             = _mm256_setzero_pd();
 859         fiz2             = _mm256_setzero_pd();
 860         fix3             = _mm256_setzero_pd();
 861         fiy3             = _mm256_setzero_pd();
 862         fiz3             = _mm256_setzero_pd();
 863
 864         /* Start inner kernel loop */
 865         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 866         {
 867
 868             /* Get j neighbor index, and coordinate index */
 869             jnrA             = jjnr[jidx];
 870             jnrB             = jjnr[jidx+1];
 871             jnrC             = jjnr[jidx+2];
 872             jnrD             = jjnr[jidx+3];
 873             j_coord_offsetA  = DIM*jnrA;
 874             j_coord_offsetB  = DIM*jnrB;
 875             j_coord_offsetC  = DIM*jnrC;
 876             j_coord_offsetD  = DIM*jnrD;
 877
 878             /* load j atom coordinates */
 879             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 880                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 881                                                  &jx0,&jy0,&jz0);
 882
 883             /* Calculate displacement vector */
 884             dx00             = _mm256_sub_pd(ix0,jx0);
 885             dy00             = _mm256_sub_pd(iy0,jy0);
 886             dz00             = _mm256_sub_pd(iz0,jz0);
 887             dx10             = _mm256_sub_pd(ix1,jx0);
 888             dy10             = _mm256_sub_pd(iy1,jy0);
 889             dz10             = _mm256_sub_pd(iz1,jz0);
 890             dx20             = _mm256_sub_pd(ix2,jx0);
 891             dy20             = _mm256_sub_pd(iy2,jy0);
 892             dz20             = _mm256_sub_pd(iz2,jz0);
 893             dx30             = _mm256_sub_pd(ix3,jx0);
 894             dy30             = _mm256_sub_pd(iy3,jy0);
 895             dz30             = _mm256_sub_pd(iz3,jz0);
 896
 897             /* Calculate squared distance and things based on it */
 898             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 899             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 900             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 901             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 902
 903             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 904             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 905             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 906
 907             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 908             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 909             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 910             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 911
 912             /* Load parameters for j particles */
 913             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 914                                                                  charge+jnrC+0,charge+jnrD+0);
 915             vdwjidx0A        = 2*vdwtype[jnrA+0];
 916             vdwjidx0B        = 2*vdwtype[jnrB+0];
 917             vdwjidx0C        = 2*vdwtype[jnrC+0];
 918             vdwjidx0D        = 2*vdwtype[jnrD+0];
 919
 920             fjx0             = _mm256_setzero_pd();
 921             fjy0             = _mm256_setzero_pd();
 922             fjz0             = _mm256_setzero_pd();
 923
 924             /**************************
 925              * CALCULATE INTERACTIONS *
 926              **************************/
 927
 928             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 929             {
 930
 931             /* Compute parameters for interactions between i and j atoms */
 932             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 933                                             vdwioffsetptr0+vdwjidx0B,
 934                                             vdwioffsetptr0+vdwjidx0C,
 935                                             vdwioffsetptr0+vdwjidx0D,
 936                                             &c6_00,&c12_00);
 937
 938             /* LENNARD-JONES DISPERSION/REPULSION */
 939
 940             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 941             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 942
 943             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 944
 945             fscal            = fvdw;
 946
 947             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm256_mul_pd(fscal,dx00);
 951             ty               = _mm256_mul_pd(fscal,dy00);
 952             tz               = _mm256_mul_pd(fscal,dz00);
 953
 954             /* Update vectorial force */
 955             fix0             = _mm256_add_pd(fix0,tx);
 956             fiy0             = _mm256_add_pd(fiy0,ty);
 957             fiz0             = _mm256_add_pd(fiz0,tz);
 958
 959             fjx0             = _mm256_add_pd(fjx0,tx);
 960             fjy0             = _mm256_add_pd(fjy0,ty);
 961             fjz0             = _mm256_add_pd(fjz0,tz);
 962
 963             }
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 970             {
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq10             = _mm256_mul_pd(iq1,jq0);
 974
 975             /* REACTION-FIELD ELECTROSTATICS */
 976             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 977
 978             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 979
 980             fscal            = felec;
 981
 982             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 983
 984             /* Calculate temporary vectorial force */
 985             tx               = _mm256_mul_pd(fscal,dx10);
 986             ty               = _mm256_mul_pd(fscal,dy10);
 987             tz               = _mm256_mul_pd(fscal,dz10);
 988
 989             /* Update vectorial force */
 990             fix1             = _mm256_add_pd(fix1,tx);
 991             fiy1             = _mm256_add_pd(fiy1,ty);
 992             fiz1             = _mm256_add_pd(fiz1,tz);
 993
 994             fjx0             = _mm256_add_pd(fjx0,tx);
 995             fjy0             = _mm256_add_pd(fjy0,ty);
 996             fjz0             = _mm256_add_pd(fjz0,tz);
 997
 998             }
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1005             {
1006
1007             /* Compute parameters for interactions between i and j atoms */
1008             qq20             = _mm256_mul_pd(iq2,jq0);
1009
1010             /* REACTION-FIELD ELECTROSTATICS */
1011             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1012
1013             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1014
1015             fscal            = felec;
1016
1017             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1018
1019             /* Calculate temporary vectorial force */
1020             tx               = _mm256_mul_pd(fscal,dx20);
1021             ty               = _mm256_mul_pd(fscal,dy20);
1022             tz               = _mm256_mul_pd(fscal,dz20);
1023
1024             /* Update vectorial force */
1025             fix2             = _mm256_add_pd(fix2,tx);
1026             fiy2             = _mm256_add_pd(fiy2,ty);
1027             fiz2             = _mm256_add_pd(fiz2,tz);
1028
1029             fjx0             = _mm256_add_pd(fjx0,tx);
1030             fjy0             = _mm256_add_pd(fjy0,ty);
1031             fjz0             = _mm256_add_pd(fjz0,tz);
1032
1033             }
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1040             {
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             qq30             = _mm256_mul_pd(iq3,jq0);
1044
1045             /* REACTION-FIELD ELECTROSTATICS */
1046             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1047
1048             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1049
1050             fscal            = felec;
1051
1052             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1053
1054             /* Calculate temporary vectorial force */
1055             tx               = _mm256_mul_pd(fscal,dx30);
1056             ty               = _mm256_mul_pd(fscal,dy30);
1057             tz               = _mm256_mul_pd(fscal,dz30);
1058
1059             /* Update vectorial force */
1060             fix3             = _mm256_add_pd(fix3,tx);
1061             fiy3             = _mm256_add_pd(fiy3,ty);
1062             fiz3             = _mm256_add_pd(fiz3,tz);
1063
1064             fjx0             = _mm256_add_pd(fjx0,tx);
1065             fjy0             = _mm256_add_pd(fjy0,ty);
1066             fjz0             = _mm256_add_pd(fjz0,tz);
1067
1068             }
1069
1070             fjptrA             = f+j_coord_offsetA;
1071             fjptrB             = f+j_coord_offsetB;
1072             fjptrC             = f+j_coord_offsetC;
1073             fjptrD             = f+j_coord_offsetD;
1074
1075             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1076
1077             /* Inner loop uses 123 flops */
1078         }
1079
1080         if(jidx<j_index_end)
1081         {
1082
1083             /* Get j neighbor index, and coordinate index */
1084             jnrlistA         = jjnr[jidx];
1085             jnrlistB         = jjnr[jidx+1];
1086             jnrlistC         = jjnr[jidx+2];
1087             jnrlistD         = jjnr[jidx+3];
1088             /* Sign of each element will be negative for non-real atoms.
1089              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1090              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1091              */
1092             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1093
1094             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1095             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1096             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1097
1098             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1099             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1100             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1101             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1102             j_coord_offsetA  = DIM*jnrA;
1103             j_coord_offsetB  = DIM*jnrB;
1104             j_coord_offsetC  = DIM*jnrC;
1105             j_coord_offsetD  = DIM*jnrD;
1106
1107             /* load j atom coordinates */
1108             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1109                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1110                                                  &jx0,&jy0,&jz0);
1111
1112             /* Calculate displacement vector */
1113             dx00             = _mm256_sub_pd(ix0,jx0);
1114             dy00             = _mm256_sub_pd(iy0,jy0);
1115             dz00             = _mm256_sub_pd(iz0,jz0);
1116             dx10             = _mm256_sub_pd(ix1,jx0);
1117             dy10             = _mm256_sub_pd(iy1,jy0);
1118             dz10             = _mm256_sub_pd(iz1,jz0);
1119             dx20             = _mm256_sub_pd(ix2,jx0);
1120             dy20             = _mm256_sub_pd(iy2,jy0);
1121             dz20             = _mm256_sub_pd(iz2,jz0);
1122             dx30             = _mm256_sub_pd(ix3,jx0);
1123             dy30             = _mm256_sub_pd(iy3,jy0);
1124             dz30             = _mm256_sub_pd(iz3,jz0);
1125
1126             /* Calculate squared distance and things based on it */
1127             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1128             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1129             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1130             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1131
1132             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1133             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1134             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1135
1136             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1137             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1138             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1139             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1140
1141             /* Load parameters for j particles */
1142             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1143                                                                  charge+jnrC+0,charge+jnrD+0);
1144             vdwjidx0A        = 2*vdwtype[jnrA+0];
1145             vdwjidx0B        = 2*vdwtype[jnrB+0];
1146             vdwjidx0C        = 2*vdwtype[jnrC+0];
1147             vdwjidx0D        = 2*vdwtype[jnrD+0];
1148
1149             fjx0             = _mm256_setzero_pd();
1150             fjy0             = _mm256_setzero_pd();
1151             fjz0             = _mm256_setzero_pd();
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1158             {
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1162                                             vdwioffsetptr0+vdwjidx0B,
1163                                             vdwioffsetptr0+vdwjidx0C,
1164                                             vdwioffsetptr0+vdwjidx0D,
1165                                             &c6_00,&c12_00);
1166
1167             /* LENNARD-JONES DISPERSION/REPULSION */
1168
1169             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1170             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1171
1172             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1173
1174             fscal            = fvdw;
1175
1176             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1177
1178             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm256_mul_pd(fscal,dx00);
1182             ty               = _mm256_mul_pd(fscal,dy00);
1183             tz               = _mm256_mul_pd(fscal,dz00);
1184
1185             /* Update vectorial force */
1186             fix0             = _mm256_add_pd(fix0,tx);
1187             fiy0             = _mm256_add_pd(fiy0,ty);
1188             fiz0             = _mm256_add_pd(fiz0,tz);
1189
1190             fjx0             = _mm256_add_pd(fjx0,tx);
1191             fjy0             = _mm256_add_pd(fjy0,ty);
1192             fjz0             = _mm256_add_pd(fjz0,tz);
1193
1194             }
1195
1196             /**************************
1197              * CALCULATE INTERACTIONS *
1198              **************************/
1199
1200             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1201             {
1202
1203             /* Compute parameters for interactions between i and j atoms */
1204             qq10             = _mm256_mul_pd(iq1,jq0);
1205
1206             /* REACTION-FIELD ELECTROSTATICS */
1207             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1208
1209             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1210
1211             fscal            = felec;
1212
1213             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1214
1215             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1216
1217             /* Calculate temporary vectorial force */
1218             tx               = _mm256_mul_pd(fscal,dx10);
1219             ty               = _mm256_mul_pd(fscal,dy10);
1220             tz               = _mm256_mul_pd(fscal,dz10);
1221
1222             /* Update vectorial force */
1223             fix1             = _mm256_add_pd(fix1,tx);
1224             fiy1             = _mm256_add_pd(fiy1,ty);
1225             fiz1             = _mm256_add_pd(fiz1,tz);
1226
1227             fjx0             = _mm256_add_pd(fjx0,tx);
1228             fjy0             = _mm256_add_pd(fjy0,ty);
1229             fjz0             = _mm256_add_pd(fjz0,tz);
1230
1231             }
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1238             {
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq20             = _mm256_mul_pd(iq2,jq0);
1242
1243             /* REACTION-FIELD ELECTROSTATICS */
1244             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1245
1246             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1247
1248             fscal            = felec;
1249
1250             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1251
1252             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1253
1254             /* Calculate temporary vectorial force */
1255             tx               = _mm256_mul_pd(fscal,dx20);
1256             ty               = _mm256_mul_pd(fscal,dy20);
1257             tz               = _mm256_mul_pd(fscal,dz20);
1258
1259             /* Update vectorial force */
1260             fix2             = _mm256_add_pd(fix2,tx);
1261             fiy2             = _mm256_add_pd(fiy2,ty);
1262             fiz2             = _mm256_add_pd(fiz2,tz);
1263
1264             fjx0             = _mm256_add_pd(fjx0,tx);
1265             fjy0             = _mm256_add_pd(fjy0,ty);
1266             fjz0             = _mm256_add_pd(fjz0,tz);
1267
1268             }
1269
1270             /**************************
1271              * CALCULATE INTERACTIONS *
1272              **************************/
1273
1274             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1275             {
1276
1277             /* Compute parameters for interactions between i and j atoms */
1278             qq30             = _mm256_mul_pd(iq3,jq0);
1279
1280             /* REACTION-FIELD ELECTROSTATICS */
1281             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1282
1283             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1284
1285             fscal            = felec;
1286
1287             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1288
1289             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1290
1291             /* Calculate temporary vectorial force */
1292             tx               = _mm256_mul_pd(fscal,dx30);
1293             ty               = _mm256_mul_pd(fscal,dy30);
1294             tz               = _mm256_mul_pd(fscal,dz30);
1295
1296             /* Update vectorial force */
1297             fix3             = _mm256_add_pd(fix3,tx);
1298             fiy3             = _mm256_add_pd(fiy3,ty);
1299             fiz3             = _mm256_add_pd(fiz3,tz);
1300
1301             fjx0             = _mm256_add_pd(fjx0,tx);
1302             fjy0             = _mm256_add_pd(fjy0,ty);
1303             fjz0             = _mm256_add_pd(fjz0,tz);
1304
1305             }
1306
1307             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1308             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1309             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1310             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1311
1312             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1313
1314             /* Inner loop uses 123 flops */
1315         }
1316
1317         /* End of innermost loop */
1318
1319         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1320                                                  f+i_coord_offset,fshift+i_shift_offset);
1321
1322         /* Increment number of inner iterations */
1323         inneriter                  += j_index_end - j_index_start;
1324
1325         /* Outer loop uses 24 flops */
1326     }
1327
1328     /* Increment number of outer iterations */
1329     outeriter        += nri;
1330
1331     /* Update outer/inner flops */
1332
1333     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1334 }