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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m256d          dummy_mask,cutoff_mask;
 109     __m128           tmpmask0,tmpmask1;
 110     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 111     __m256d          one     = _mm256_set1_pd(1.0);
 112     __m256d          two     = _mm256_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     krf              = _mm256_set1_pd(fr->ic->k_rf);
 127     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 128     crf              = _mm256_set1_pd(fr->ic->c_rf);
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 136     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 137     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 138     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 143     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 144
 145     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 146     rvdw             = _mm256_set1_pd(fr->rvdw);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm256_setzero_pd();
 182         fiy0             = _mm256_setzero_pd();
 183         fiz0             = _mm256_setzero_pd();
 184         fix1             = _mm256_setzero_pd();
 185         fiy1             = _mm256_setzero_pd();
 186         fiz1             = _mm256_setzero_pd();
 187         fix2             = _mm256_setzero_pd();
 188         fiy2             = _mm256_setzero_pd();
 189         fiz2             = _mm256_setzero_pd();
 190         fix3             = _mm256_setzero_pd();
 191         fiy3             = _mm256_setzero_pd();
 192         fiz3             = _mm256_setzero_pd();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_pd();
 196         vvdwsum          = _mm256_setzero_pd();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm256_sub_pd(ix0,jx0);
 219             dy00             = _mm256_sub_pd(iy0,jy0);
 220             dz00             = _mm256_sub_pd(iz0,jz0);
 221             dx10             = _mm256_sub_pd(ix1,jx0);
 222             dy10             = _mm256_sub_pd(iy1,jy0);
 223             dz10             = _mm256_sub_pd(iz1,jz0);
 224             dx20             = _mm256_sub_pd(ix2,jx0);
 225             dy20             = _mm256_sub_pd(iy2,jy0);
 226             dz20             = _mm256_sub_pd(iz2,jz0);
 227             dx30             = _mm256_sub_pd(ix3,jx0);
 228             dy30             = _mm256_sub_pd(iy3,jy0);
 229             dz30             = _mm256_sub_pd(iz3,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 233             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 234             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 235             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 236
 237             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 238             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 239             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 240
 241             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 242             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 243             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 244             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 245
 246             /* Load parameters for j particles */
 247             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 248                                                                  charge+jnrC+0,charge+jnrD+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251             vdwjidx0C        = 2*vdwtype[jnrC+0];
 252             vdwjidx0D        = 2*vdwtype[jnrD+0];
 253
 254             fjx0             = _mm256_setzero_pd();
 255             fjy0             = _mm256_setzero_pd();
 256             fjz0             = _mm256_setzero_pd();
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 263             {
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 267                                             vdwioffsetptr0+vdwjidx0B,
 268                                             vdwioffsetptr0+vdwjidx0C,
 269                                             vdwioffsetptr0+vdwjidx0D,
 270                                             &c6_00,&c12_00);
 271
 272             /* LENNARD-JONES DISPERSION/REPULSION */
 273
 274             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 275             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 276             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 277             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) ,
 278                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 279             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 280
 281             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 285             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 286
 287             fscal            = fvdw;
 288
 289             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm256_mul_pd(fscal,dx00);
 293             ty               = _mm256_mul_pd(fscal,dy00);
 294             tz               = _mm256_mul_pd(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm256_add_pd(fix0,tx);
 298             fiy0             = _mm256_add_pd(fiy0,ty);
 299             fiz0             = _mm256_add_pd(fiz0,tz);
 300
 301             fjx0             = _mm256_add_pd(fjx0,tx);
 302             fjy0             = _mm256_add_pd(fjy0,ty);
 303             fjz0             = _mm256_add_pd(fjz0,tz);
 304
 305             }
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 312             {
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm256_mul_pd(iq1,jq0);
 316
 317             /* REACTION-FIELD ELECTROSTATICS */
 318             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 319             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 320
 321             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velec            = _mm256_and_pd(velec,cutoff_mask);
 325             velecsum         = _mm256_add_pd(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm256_mul_pd(fscal,dx10);
 333             ty               = _mm256_mul_pd(fscal,dy10);
 334             tz               = _mm256_mul_pd(fscal,dz10);
 335
 336             /* Update vectorial force */
 337             fix1             = _mm256_add_pd(fix1,tx);
 338             fiy1             = _mm256_add_pd(fiy1,ty);
 339             fiz1             = _mm256_add_pd(fiz1,tz);
 340
 341             fjx0             = _mm256_add_pd(fjx0,tx);
 342             fjy0             = _mm256_add_pd(fjy0,ty);
 343             fjz0             = _mm256_add_pd(fjz0,tz);
 344
 345             }
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 352             {
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq20             = _mm256_mul_pd(iq2,jq0);
 356
 357             /* REACTION-FIELD ELECTROSTATICS */
 358             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 359             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 360
 361             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm256_and_pd(velec,cutoff_mask);
 365             velecsum         = _mm256_add_pd(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm256_mul_pd(fscal,dx20);
 373             ty               = _mm256_mul_pd(fscal,dy20);
 374             tz               = _mm256_mul_pd(fscal,dz20);
 375
 376             /* Update vectorial force */
 377             fix2             = _mm256_add_pd(fix2,tx);
 378             fiy2             = _mm256_add_pd(fiy2,ty);
 379             fiz2             = _mm256_add_pd(fiz2,tz);
 380
 381             fjx0             = _mm256_add_pd(fjx0,tx);
 382             fjy0             = _mm256_add_pd(fjy0,ty);
 383             fjz0             = _mm256_add_pd(fjz0,tz);
 384
 385             }
 386
 387             /**************************
 388              * CALCULATE INTERACTIONS *
 389              **************************/
 390
 391             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 392             {
 393
 394             /* Compute parameters for interactions between i and j atoms */
 395             qq30             = _mm256_mul_pd(iq3,jq0);
 396
 397             /* REACTION-FIELD ELECTROSTATICS */
 398             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 399             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 400
 401             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velec            = _mm256_and_pd(velec,cutoff_mask);
 405             velecsum         = _mm256_add_pd(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm256_mul_pd(fscal,dx30);
 413             ty               = _mm256_mul_pd(fscal,dy30);
 414             tz               = _mm256_mul_pd(fscal,dz30);
 415
 416             /* Update vectorial force */
 417             fix3             = _mm256_add_pd(fix3,tx);
 418             fiy3             = _mm256_add_pd(fiy3,ty);
 419             fiz3             = _mm256_add_pd(fiz3,tz);
 420
 421             fjx0             = _mm256_add_pd(fjx0,tx);
 422             fjy0             = _mm256_add_pd(fjy0,ty);
 423             fjz0             = _mm256_add_pd(fjz0,tz);
 424
 425             }
 426
 427             fjptrA             = f+j_coord_offsetA;
 428             fjptrB             = f+j_coord_offsetB;
 429             fjptrC             = f+j_coord_offsetC;
 430             fjptrD             = f+j_coord_offsetD;
 431
 432             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 433
 434             /* Inner loop uses 152 flops */
 435         }
 436
 437         if(jidx<j_index_end)
 438         {
 439
 440             /* Get j neighbor index, and coordinate index */
 441             jnrlistA         = jjnr[jidx];
 442             jnrlistB         = jjnr[jidx+1];
 443             jnrlistC         = jjnr[jidx+2];
 444             jnrlistD         = jjnr[jidx+3];
 445             /* Sign of each element will be negative for non-real atoms.
 446              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 447              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 448              */
 449             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 450
 451             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 452             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 453             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 454
 455             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 456             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 457             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 458             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 459             j_coord_offsetA  = DIM*jnrA;
 460             j_coord_offsetB  = DIM*jnrB;
 461             j_coord_offsetC  = DIM*jnrC;
 462             j_coord_offsetD  = DIM*jnrD;
 463
 464             /* load j atom coordinates */
 465             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 466                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 467                                                  &jx0,&jy0,&jz0);
 468
 469             /* Calculate displacement vector */
 470             dx00             = _mm256_sub_pd(ix0,jx0);
 471             dy00             = _mm256_sub_pd(iy0,jy0);
 472             dz00             = _mm256_sub_pd(iz0,jz0);
 473             dx10             = _mm256_sub_pd(ix1,jx0);
 474             dy10             = _mm256_sub_pd(iy1,jy0);
 475             dz10             = _mm256_sub_pd(iz1,jz0);
 476             dx20             = _mm256_sub_pd(ix2,jx0);
 477             dy20             = _mm256_sub_pd(iy2,jy0);
 478             dz20             = _mm256_sub_pd(iz2,jz0);
 479             dx30             = _mm256_sub_pd(ix3,jx0);
 480             dy30             = _mm256_sub_pd(iy3,jy0);
 481             dz30             = _mm256_sub_pd(iz3,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 485             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 486             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 487             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 488
 489             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 490             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 491             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 492
 493             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 494             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 495             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 496             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 497
 498             /* Load parameters for j particles */
 499             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 500                                                                  charge+jnrC+0,charge+jnrD+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502             vdwjidx0B        = 2*vdwtype[jnrB+0];
 503             vdwjidx0C        = 2*vdwtype[jnrC+0];
 504             vdwjidx0D        = 2*vdwtype[jnrD+0];
 505
 506             fjx0             = _mm256_setzero_pd();
 507             fjy0             = _mm256_setzero_pd();
 508             fjz0             = _mm256_setzero_pd();
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 515             {
 516
 517             /* Compute parameters for interactions between i and j atoms */
 518             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 519                                             vdwioffsetptr0+vdwjidx0B,
 520                                             vdwioffsetptr0+vdwjidx0C,
 521                                             vdwioffsetptr0+vdwjidx0D,
 522                                             &c6_00,&c12_00);
 523
 524             /* LENNARD-JONES DISPERSION/REPULSION */
 525
 526             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 527             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 528             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 529             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) ,
 530                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 531             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 532
 533             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 537             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 538             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 539
 540             fscal            = fvdw;
 541
 542             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 543
 544             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 545
 546             /* Calculate temporary vectorial force */
 547             tx               = _mm256_mul_pd(fscal,dx00);
 548             ty               = _mm256_mul_pd(fscal,dy00);
 549             tz               = _mm256_mul_pd(fscal,dz00);
 550
 551             /* Update vectorial force */
 552             fix0             = _mm256_add_pd(fix0,tx);
 553             fiy0             = _mm256_add_pd(fiy0,ty);
 554             fiz0             = _mm256_add_pd(fiz0,tz);
 555
 556             fjx0             = _mm256_add_pd(fjx0,tx);
 557             fjy0             = _mm256_add_pd(fjy0,ty);
 558             fjz0             = _mm256_add_pd(fjz0,tz);
 559
 560             }
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 567             {
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq10             = _mm256_mul_pd(iq1,jq0);
 571
 572             /* REACTION-FIELD ELECTROSTATICS */
 573             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 574             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 575
 576             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             velec            = _mm256_and_pd(velec,cutoff_mask);
 580             velec            = _mm256_andnot_pd(dummy_mask,velec);
 581             velecsum         = _mm256_add_pd(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 586
 587             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 588
 589             /* Calculate temporary vectorial force */
 590             tx               = _mm256_mul_pd(fscal,dx10);
 591             ty               = _mm256_mul_pd(fscal,dy10);
 592             tz               = _mm256_mul_pd(fscal,dz10);
 593
 594             /* Update vectorial force */
 595             fix1             = _mm256_add_pd(fix1,tx);
 596             fiy1             = _mm256_add_pd(fiy1,ty);
 597             fiz1             = _mm256_add_pd(fiz1,tz);
 598
 599             fjx0             = _mm256_add_pd(fjx0,tx);
 600             fjy0             = _mm256_add_pd(fjy0,ty);
 601             fjz0             = _mm256_add_pd(fjz0,tz);
 602
 603             }
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 610             {
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq20             = _mm256_mul_pd(iq2,jq0);
 614
 615             /* REACTION-FIELD ELECTROSTATICS */
 616             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 617             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 618
 619             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm256_and_pd(velec,cutoff_mask);
 623             velec            = _mm256_andnot_pd(dummy_mask,velec);
 624             velecsum         = _mm256_add_pd(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 629
 630             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm256_mul_pd(fscal,dx20);
 634             ty               = _mm256_mul_pd(fscal,dy20);
 635             tz               = _mm256_mul_pd(fscal,dz20);
 636
 637             /* Update vectorial force */
 638             fix2             = _mm256_add_pd(fix2,tx);
 639             fiy2             = _mm256_add_pd(fiy2,ty);
 640             fiz2             = _mm256_add_pd(fiz2,tz);
 641
 642             fjx0             = _mm256_add_pd(fjx0,tx);
 643             fjy0             = _mm256_add_pd(fjy0,ty);
 644             fjz0             = _mm256_add_pd(fjz0,tz);
 645
 646             }
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 653             {
 654
 655             /* Compute parameters for interactions between i and j atoms */
 656             qq30             = _mm256_mul_pd(iq3,jq0);
 657
 658             /* REACTION-FIELD ELECTROSTATICS */
 659             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 660             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 661
 662             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 663
 664             /* Update potential sum for this i atom from the interaction with this j atom. */
 665             velec            = _mm256_and_pd(velec,cutoff_mask);
 666             velec            = _mm256_andnot_pd(dummy_mask,velec);
 667             velecsum         = _mm256_add_pd(velecsum,velec);
 668
 669             fscal            = felec;
 670
 671             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 672
 673             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm256_mul_pd(fscal,dx30);
 677             ty               = _mm256_mul_pd(fscal,dy30);
 678             tz               = _mm256_mul_pd(fscal,dz30);
 679
 680             /* Update vectorial force */
 681             fix3             = _mm256_add_pd(fix3,tx);
 682             fiy3             = _mm256_add_pd(fiy3,ty);
 683             fiz3             = _mm256_add_pd(fiz3,tz);
 684
 685             fjx0             = _mm256_add_pd(fjx0,tx);
 686             fjy0             = _mm256_add_pd(fjy0,ty);
 687             fjz0             = _mm256_add_pd(fjz0,tz);
 688
 689             }
 690
 691             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 692             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 693             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 694             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 695
 696             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 697
 698             /* Inner loop uses 152 flops */
 699         }
 700
 701         /* End of innermost loop */
 702
 703         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 704                                                  f+i_coord_offset,fshift+i_shift_offset);
 705
 706         ggid                        = gid[iidx];
 707         /* Update potential energies */
 708         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 709         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 710
 711         /* Increment number of inner iterations */
 712         inneriter                  += j_index_end - j_index_start;
 713
 714         /* Outer loop uses 26 flops */
 715     }
 716
 717     /* Increment number of outer iterations */
 718     outeriter        += nri;
 719
 720     /* Update outer/inner flops */
 721
 722     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*152);
 723 }
 724 /*
 725  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
 726  * Electrostatics interaction: ReactionField
 727  * VdW interaction:            LennardJones
 728  * Geometry:                   Water4-Particle
 729  * Calculate force/pot:        Force
 730  */
 731 void
 732 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
 733                     (t_nblist                    * gmx_restrict       nlist,
 734                      rvec                        * gmx_restrict          xx,
 735                      rvec                        * gmx_restrict          ff,
 736                      t_forcerec                  * gmx_restrict          fr,
 737                      t_mdatoms                   * gmx_restrict     mdatoms,
 738                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 739                      t_nrnb                      * gmx_restrict        nrnb)
 740 {
 741     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 742      * just 0 for non-waters.
 743      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 744      * jnr indices corresponding to data put in the four positions in the SIMD register.
 745      */
 746     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 747     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 748     int              jnrA,jnrB,jnrC,jnrD;
 749     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 750     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 751     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 752     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 753     real             rcutoff_scalar;
 754     real             *shiftvec,*fshift,*x,*f;
 755     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 756     real             scratch[4*DIM];
 757     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 758     real *           vdwioffsetptr0;
 759     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 760     real *           vdwioffsetptr1;
 761     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 762     real *           vdwioffsetptr2;
 763     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 764     real *           vdwioffsetptr3;
 765     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 766     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 767     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 768     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 769     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 770     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 771     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 772     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 773     real             *charge;
 774     int              nvdwtype;
 775     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 776     int              *vdwtype;
 777     real             *vdwparam;
 778     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 779     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 780     __m256d          dummy_mask,cutoff_mask;
 781     __m128           tmpmask0,tmpmask1;
 782     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 783     __m256d          one     = _mm256_set1_pd(1.0);
 784     __m256d          two     = _mm256_set1_pd(2.0);
 785     x                = xx[0];
 786     f                = ff[0];
 787
 788     nri              = nlist->nri;
 789     iinr             = nlist->iinr;
 790     jindex           = nlist->jindex;
 791     jjnr             = nlist->jjnr;
 792     shiftidx         = nlist->shift;
 793     gid              = nlist->gid;
 794     shiftvec         = fr->shift_vec[0];
 795     fshift           = fr->fshift[0];
 796     facel            = _mm256_set1_pd(fr->epsfac);
 797     charge           = mdatoms->chargeA;
 798     krf              = _mm256_set1_pd(fr->ic->k_rf);
 799     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 800     crf              = _mm256_set1_pd(fr->ic->c_rf);
 801     nvdwtype         = fr->ntype;
 802     vdwparam         = fr->nbfp;
 803     vdwtype          = mdatoms->typeA;
 804
 805     /* Setup water-specific parameters */
 806     inr              = nlist->iinr[0];
 807     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 808     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 809     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 810     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 811
 812     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 813     rcutoff_scalar   = fr->rcoulomb;
 814     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 815     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 816
 817     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 818     rvdw             = _mm256_set1_pd(fr->rvdw);
 819
 820     /* Avoid stupid compiler warnings */
 821     jnrA = jnrB = jnrC = jnrD = 0;
 822     j_coord_offsetA = 0;
 823     j_coord_offsetB = 0;
 824     j_coord_offsetC = 0;
 825     j_coord_offsetD = 0;
 826
 827     outeriter        = 0;
 828     inneriter        = 0;
 829
 830     for(iidx=0;iidx<4*DIM;iidx++)
 831     {
 832         scratch[iidx] = 0.0;
 833     }
 834
 835     /* Start outer loop over neighborlists */
 836     for(iidx=0; iidx<nri; iidx++)
 837     {
 838         /* Load shift vector for this list */
 839         i_shift_offset   = DIM*shiftidx[iidx];
 840
 841         /* Load limits for loop over neighbors */
 842         j_index_start    = jindex[iidx];
 843         j_index_end      = jindex[iidx+1];
 844
 845         /* Get outer coordinate index */
 846         inr              = iinr[iidx];
 847         i_coord_offset   = DIM*inr;
 848
 849         /* Load i particle coords and add shift vector */
 850         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 851                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 852
 853         fix0             = _mm256_setzero_pd();
 854         fiy0             = _mm256_setzero_pd();
 855         fiz0             = _mm256_setzero_pd();
 856         fix1             = _mm256_setzero_pd();
 857         fiy1             = _mm256_setzero_pd();
 858         fiz1             = _mm256_setzero_pd();
 859         fix2             = _mm256_setzero_pd();
 860         fiy2             = _mm256_setzero_pd();
 861         fiz2             = _mm256_setzero_pd();
 862         fix3             = _mm256_setzero_pd();
 863         fiy3             = _mm256_setzero_pd();
 864         fiz3             = _mm256_setzero_pd();
 865
 866         /* Start inner kernel loop */
 867         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 868         {
 869
 870             /* Get j neighbor index, and coordinate index */
 871             jnrA             = jjnr[jidx];
 872             jnrB             = jjnr[jidx+1];
 873             jnrC             = jjnr[jidx+2];
 874             jnrD             = jjnr[jidx+3];
 875             j_coord_offsetA  = DIM*jnrA;
 876             j_coord_offsetB  = DIM*jnrB;
 877             j_coord_offsetC  = DIM*jnrC;
 878             j_coord_offsetD  = DIM*jnrD;
 879
 880             /* load j atom coordinates */
 881             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 882                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 883                                                  &jx0,&jy0,&jz0);
 884
 885             /* Calculate displacement vector */
 886             dx00             = _mm256_sub_pd(ix0,jx0);
 887             dy00             = _mm256_sub_pd(iy0,jy0);
 888             dz00             = _mm256_sub_pd(iz0,jz0);
 889             dx10             = _mm256_sub_pd(ix1,jx0);
 890             dy10             = _mm256_sub_pd(iy1,jy0);
 891             dz10             = _mm256_sub_pd(iz1,jz0);
 892             dx20             = _mm256_sub_pd(ix2,jx0);
 893             dy20             = _mm256_sub_pd(iy2,jy0);
 894             dz20             = _mm256_sub_pd(iz2,jz0);
 895             dx30             = _mm256_sub_pd(ix3,jx0);
 896             dy30             = _mm256_sub_pd(iy3,jy0);
 897             dz30             = _mm256_sub_pd(iz3,jz0);
 898
 899             /* Calculate squared distance and things based on it */
 900             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 901             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 902             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 903             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 904
 905             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 906             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 907             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 908
 909             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 910             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 911             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 912             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 913
 914             /* Load parameters for j particles */
 915             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 916                                                                  charge+jnrC+0,charge+jnrD+0);
 917             vdwjidx0A        = 2*vdwtype[jnrA+0];
 918             vdwjidx0B        = 2*vdwtype[jnrB+0];
 919             vdwjidx0C        = 2*vdwtype[jnrC+0];
 920             vdwjidx0D        = 2*vdwtype[jnrD+0];
 921
 922             fjx0             = _mm256_setzero_pd();
 923             fjy0             = _mm256_setzero_pd();
 924             fjz0             = _mm256_setzero_pd();
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 931             {
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 935                                             vdwioffsetptr0+vdwjidx0B,
 936                                             vdwioffsetptr0+vdwjidx0C,
 937                                             vdwioffsetptr0+vdwjidx0D,
 938                                             &c6_00,&c12_00);
 939
 940             /* LENNARD-JONES DISPERSION/REPULSION */
 941
 942             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 943             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 944
 945             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 946
 947             fscal            = fvdw;
 948
 949             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 950
 951             /* Calculate temporary vectorial force */
 952             tx               = _mm256_mul_pd(fscal,dx00);
 953             ty               = _mm256_mul_pd(fscal,dy00);
 954             tz               = _mm256_mul_pd(fscal,dz00);
 955
 956             /* Update vectorial force */
 957             fix0             = _mm256_add_pd(fix0,tx);
 958             fiy0             = _mm256_add_pd(fiy0,ty);
 959             fiz0             = _mm256_add_pd(fiz0,tz);
 960
 961             fjx0             = _mm256_add_pd(fjx0,tx);
 962             fjy0             = _mm256_add_pd(fjy0,ty);
 963             fjz0             = _mm256_add_pd(fjz0,tz);
 964
 965             }
 966
 967             /**************************
 968              * CALCULATE INTERACTIONS *
 969              **************************/
 970
 971             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 972             {
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             qq10             = _mm256_mul_pd(iq1,jq0);
 976
 977             /* REACTION-FIELD ELECTROSTATICS */
 978             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 979
 980             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm256_mul_pd(fscal,dx10);
 988             ty               = _mm256_mul_pd(fscal,dy10);
 989             tz               = _mm256_mul_pd(fscal,dz10);
 990
 991             /* Update vectorial force */
 992             fix1             = _mm256_add_pd(fix1,tx);
 993             fiy1             = _mm256_add_pd(fiy1,ty);
 994             fiz1             = _mm256_add_pd(fiz1,tz);
 995
 996             fjx0             = _mm256_add_pd(fjx0,tx);
 997             fjy0             = _mm256_add_pd(fjy0,ty);
 998             fjz0             = _mm256_add_pd(fjz0,tz);
 999
1000             }
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1007             {
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             qq20             = _mm256_mul_pd(iq2,jq0);
1011
1012             /* REACTION-FIELD ELECTROSTATICS */
1013             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1014
1015             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1016
1017             fscal            = felec;
1018
1019             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1020
1021             /* Calculate temporary vectorial force */
1022             tx               = _mm256_mul_pd(fscal,dx20);
1023             ty               = _mm256_mul_pd(fscal,dy20);
1024             tz               = _mm256_mul_pd(fscal,dz20);
1025
1026             /* Update vectorial force */
1027             fix2             = _mm256_add_pd(fix2,tx);
1028             fiy2             = _mm256_add_pd(fiy2,ty);
1029             fiz2             = _mm256_add_pd(fiz2,tz);
1030
1031             fjx0             = _mm256_add_pd(fjx0,tx);
1032             fjy0             = _mm256_add_pd(fjy0,ty);
1033             fjz0             = _mm256_add_pd(fjz0,tz);
1034
1035             }
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1042             {
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq30             = _mm256_mul_pd(iq3,jq0);
1046
1047             /* REACTION-FIELD ELECTROSTATICS */
1048             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1049
1050             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1055
1056             /* Calculate temporary vectorial force */
1057             tx               = _mm256_mul_pd(fscal,dx30);
1058             ty               = _mm256_mul_pd(fscal,dy30);
1059             tz               = _mm256_mul_pd(fscal,dz30);
1060
1061             /* Update vectorial force */
1062             fix3             = _mm256_add_pd(fix3,tx);
1063             fiy3             = _mm256_add_pd(fiy3,ty);
1064             fiz3             = _mm256_add_pd(fiz3,tz);
1065
1066             fjx0             = _mm256_add_pd(fjx0,tx);
1067             fjy0             = _mm256_add_pd(fjy0,ty);
1068             fjz0             = _mm256_add_pd(fjz0,tz);
1069
1070             }
1071
1072             fjptrA             = f+j_coord_offsetA;
1073             fjptrB             = f+j_coord_offsetB;
1074             fjptrC             = f+j_coord_offsetC;
1075             fjptrD             = f+j_coord_offsetD;
1076
1077             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1078
1079             /* Inner loop uses 123 flops */
1080         }
1081
1082         if(jidx<j_index_end)
1083         {
1084
1085             /* Get j neighbor index, and coordinate index */
1086             jnrlistA         = jjnr[jidx];
1087             jnrlistB         = jjnr[jidx+1];
1088             jnrlistC         = jjnr[jidx+2];
1089             jnrlistD         = jjnr[jidx+3];
1090             /* Sign of each element will be negative for non-real atoms.
1091              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1092              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1093              */
1094             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1095
1096             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1097             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1098             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1099
1100             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1101             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1102             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1103             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1104             j_coord_offsetA  = DIM*jnrA;
1105             j_coord_offsetB  = DIM*jnrB;
1106             j_coord_offsetC  = DIM*jnrC;
1107             j_coord_offsetD  = DIM*jnrD;
1108
1109             /* load j atom coordinates */
1110             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1111                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1112                                                  &jx0,&jy0,&jz0);
1113
1114             /* Calculate displacement vector */
1115             dx00             = _mm256_sub_pd(ix0,jx0);
1116             dy00             = _mm256_sub_pd(iy0,jy0);
1117             dz00             = _mm256_sub_pd(iz0,jz0);
1118             dx10             = _mm256_sub_pd(ix1,jx0);
1119             dy10             = _mm256_sub_pd(iy1,jy0);
1120             dz10             = _mm256_sub_pd(iz1,jz0);
1121             dx20             = _mm256_sub_pd(ix2,jx0);
1122             dy20             = _mm256_sub_pd(iy2,jy0);
1123             dz20             = _mm256_sub_pd(iz2,jz0);
1124             dx30             = _mm256_sub_pd(ix3,jx0);
1125             dy30             = _mm256_sub_pd(iy3,jy0);
1126             dz30             = _mm256_sub_pd(iz3,jz0);
1127
1128             /* Calculate squared distance and things based on it */
1129             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1130             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1131             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1132             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1133
1134             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1135             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1136             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1137
1138             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1139             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1140             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1141             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1142
1143             /* Load parameters for j particles */
1144             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1145                                                                  charge+jnrC+0,charge+jnrD+0);
1146             vdwjidx0A        = 2*vdwtype[jnrA+0];
1147             vdwjidx0B        = 2*vdwtype[jnrB+0];
1148             vdwjidx0C        = 2*vdwtype[jnrC+0];
1149             vdwjidx0D        = 2*vdwtype[jnrD+0];
1150
1151             fjx0             = _mm256_setzero_pd();
1152             fjy0             = _mm256_setzero_pd();
1153             fjz0             = _mm256_setzero_pd();
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1160             {
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1164                                             vdwioffsetptr0+vdwjidx0B,
1165                                             vdwioffsetptr0+vdwjidx0C,
1166                                             vdwioffsetptr0+vdwjidx0D,
1167                                             &c6_00,&c12_00);
1168
1169             /* LENNARD-JONES DISPERSION/REPULSION */
1170
1171             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1172             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1173
1174             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1175
1176             fscal            = fvdw;
1177
1178             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1179
1180             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1181
1182             /* Calculate temporary vectorial force */
1183             tx               = _mm256_mul_pd(fscal,dx00);
1184             ty               = _mm256_mul_pd(fscal,dy00);
1185             tz               = _mm256_mul_pd(fscal,dz00);
1186
1187             /* Update vectorial force */
1188             fix0             = _mm256_add_pd(fix0,tx);
1189             fiy0             = _mm256_add_pd(fiy0,ty);
1190             fiz0             = _mm256_add_pd(fiz0,tz);
1191
1192             fjx0             = _mm256_add_pd(fjx0,tx);
1193             fjy0             = _mm256_add_pd(fjy0,ty);
1194             fjz0             = _mm256_add_pd(fjz0,tz);
1195
1196             }
1197
1198             /**************************
1199              * CALCULATE INTERACTIONS *
1200              **************************/
1201
1202             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1203             {
1204
1205             /* Compute parameters for interactions between i and j atoms */
1206             qq10             = _mm256_mul_pd(iq1,jq0);
1207
1208             /* REACTION-FIELD ELECTROSTATICS */
1209             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1210
1211             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1212
1213             fscal            = felec;
1214
1215             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1216
1217             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1218
1219             /* Calculate temporary vectorial force */
1220             tx               = _mm256_mul_pd(fscal,dx10);
1221             ty               = _mm256_mul_pd(fscal,dy10);
1222             tz               = _mm256_mul_pd(fscal,dz10);
1223
1224             /* Update vectorial force */
1225             fix1             = _mm256_add_pd(fix1,tx);
1226             fiy1             = _mm256_add_pd(fiy1,ty);
1227             fiz1             = _mm256_add_pd(fiz1,tz);
1228
1229             fjx0             = _mm256_add_pd(fjx0,tx);
1230             fjy0             = _mm256_add_pd(fjy0,ty);
1231             fjz0             = _mm256_add_pd(fjz0,tz);
1232
1233             }
1234
1235             /**************************
1236              * CALCULATE INTERACTIONS *
1237              **************************/
1238
1239             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1240             {
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             qq20             = _mm256_mul_pd(iq2,jq0);
1244
1245             /* REACTION-FIELD ELECTROSTATICS */
1246             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1247
1248             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1249
1250             fscal            = felec;
1251
1252             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1253
1254             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1255
1256             /* Calculate temporary vectorial force */
1257             tx               = _mm256_mul_pd(fscal,dx20);
1258             ty               = _mm256_mul_pd(fscal,dy20);
1259             tz               = _mm256_mul_pd(fscal,dz20);
1260
1261             /* Update vectorial force */
1262             fix2             = _mm256_add_pd(fix2,tx);
1263             fiy2             = _mm256_add_pd(fiy2,ty);
1264             fiz2             = _mm256_add_pd(fiz2,tz);
1265
1266             fjx0             = _mm256_add_pd(fjx0,tx);
1267             fjy0             = _mm256_add_pd(fjy0,ty);
1268             fjz0             = _mm256_add_pd(fjz0,tz);
1269
1270             }
1271
1272             /**************************
1273              * CALCULATE INTERACTIONS *
1274              **************************/
1275
1276             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1277             {
1278
1279             /* Compute parameters for interactions between i and j atoms */
1280             qq30             = _mm256_mul_pd(iq3,jq0);
1281
1282             /* REACTION-FIELD ELECTROSTATICS */
1283             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1284
1285             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1286
1287             fscal            = felec;
1288
1289             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1290
1291             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1292
1293             /* Calculate temporary vectorial force */
1294             tx               = _mm256_mul_pd(fscal,dx30);
1295             ty               = _mm256_mul_pd(fscal,dy30);
1296             tz               = _mm256_mul_pd(fscal,dz30);
1297
1298             /* Update vectorial force */
1299             fix3             = _mm256_add_pd(fix3,tx);
1300             fiy3             = _mm256_add_pd(fiy3,ty);
1301             fiz3             = _mm256_add_pd(fiz3,tz);
1302
1303             fjx0             = _mm256_add_pd(fjx0,tx);
1304             fjy0             = _mm256_add_pd(fjy0,ty);
1305             fjz0             = _mm256_add_pd(fjz0,tz);
1306
1307             }
1308
1309             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1310             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1311             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1312             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1313
1314             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1315
1316             /* Inner loop uses 123 flops */
1317         }
1318
1319         /* End of innermost loop */
1320
1321         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1322                                                  f+i_coord_offset,fshift+i_shift_offset);
1323
1324         /* Increment number of inner iterations */
1325         inneriter                  += j_index_end - j_index_start;
1326
1327         /* Outer loop uses 24 flops */
1328     }
1329
1330     /* Increment number of outer iterations */
1331     outeriter        += nri;
1332
1333     /* Update outer/inner flops */
1334
1335     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1336 }