src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m256d          dummy_mask,cutoff_mask;
 104     __m128           tmpmask0,tmpmask1;
 105     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 106     __m256d          one     = _mm256_set1_pd(1.0);
 107     __m256d          two     = _mm256_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm256_set1_pd(fr->ic->k_rf);
 122     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 123     crf              = _mm256_set1_pd(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 131     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 132     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->rcoulomb;
 137     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 138     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 139
 140     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 141     rvdw             = _mm256_set1_pd(fr->rvdw);
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 175
 176         fix0             = _mm256_setzero_pd();
 177         fiy0             = _mm256_setzero_pd();
 178         fiz0             = _mm256_setzero_pd();
 179         fix1             = _mm256_setzero_pd();
 180         fiy1             = _mm256_setzero_pd();
 181         fiz1             = _mm256_setzero_pd();
 182         fix2             = _mm256_setzero_pd();
 183         fiy2             = _mm256_setzero_pd();
 184         fiz2             = _mm256_setzero_pd();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm256_setzero_pd();
 188         vvdwsum          = _mm256_setzero_pd();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 207                                                  &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm256_sub_pd(ix0,jx0);
 211             dy00             = _mm256_sub_pd(iy0,jy0);
 212             dz00             = _mm256_sub_pd(iz0,jz0);
 213             dx10             = _mm256_sub_pd(ix1,jx0);
 214             dy10             = _mm256_sub_pd(iy1,jy0);
 215             dz10             = _mm256_sub_pd(iz1,jz0);
 216             dx20             = _mm256_sub_pd(ix2,jx0);
 217             dy20             = _mm256_sub_pd(iy2,jy0);
 218             dz20             = _mm256_sub_pd(iz2,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 222             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 223             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 224
 225             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 226             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 227             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 228
 229             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 230             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 231             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 235                                                                  charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             fjx0             = _mm256_setzero_pd();
 242             fjy0             = _mm256_setzero_pd();
 243             fjz0             = _mm256_setzero_pd();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 250             {
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm256_mul_pd(iq0,jq0);
 254             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 255                                             vdwioffsetptr0+vdwjidx0B,
 256                                             vdwioffsetptr0+vdwjidx0C,
 257                                             vdwioffsetptr0+vdwjidx0D,
 258                                             &c6_00,&c12_00);
 259
 260             /* REACTION-FIELD ELECTROSTATICS */
 261             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 262             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 268             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 269             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) ,
 270                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 271             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 272
 273             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm256_and_pd(velec,cutoff_mask);
 277             velecsum         = _mm256_add_pd(velecsum,velec);
 278             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 279             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 280
 281             fscal            = _mm256_add_pd(felec,fvdw);
 282
 283             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm256_mul_pd(fscal,dx00);
 287             ty               = _mm256_mul_pd(fscal,dy00);
 288             tz               = _mm256_mul_pd(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm256_add_pd(fix0,tx);
 292             fiy0             = _mm256_add_pd(fiy0,ty);
 293             fiz0             = _mm256_add_pd(fiz0,tz);
 294
 295             fjx0             = _mm256_add_pd(fjx0,tx);
 296             fjy0             = _mm256_add_pd(fjy0,ty);
 297             fjz0             = _mm256_add_pd(fjz0,tz);
 298
 299             }
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 306             {
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm256_mul_pd(iq1,jq0);
 310
 311             /* REACTION-FIELD ELECTROSTATICS */
 312             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 313             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 314
 315             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm256_and_pd(velec,cutoff_mask);
 319             velecsum         = _mm256_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm256_mul_pd(fscal,dx10);
 327             ty               = _mm256_mul_pd(fscal,dy10);
 328             tz               = _mm256_mul_pd(fscal,dz10);
 329
 330             /* Update vectorial force */
 331             fix1             = _mm256_add_pd(fix1,tx);
 332             fiy1             = _mm256_add_pd(fiy1,ty);
 333             fiz1             = _mm256_add_pd(fiz1,tz);
 334
 335             fjx0             = _mm256_add_pd(fjx0,tx);
 336             fjy0             = _mm256_add_pd(fjy0,ty);
 337             fjz0             = _mm256_add_pd(fjz0,tz);
 338
 339             }
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 346             {
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq20             = _mm256_mul_pd(iq2,jq0);
 350
 351             /* REACTION-FIELD ELECTROSTATICS */
 352             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 353             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 354
 355             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velec            = _mm256_and_pd(velec,cutoff_mask);
 359             velecsum         = _mm256_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm256_mul_pd(fscal,dx20);
 367             ty               = _mm256_mul_pd(fscal,dy20);
 368             tz               = _mm256_mul_pd(fscal,dz20);
 369
 370             /* Update vectorial force */
 371             fix2             = _mm256_add_pd(fix2,tx);
 372             fiy2             = _mm256_add_pd(fiy2,ty);
 373             fiz2             = _mm256_add_pd(fiz2,tz);
 374
 375             fjx0             = _mm256_add_pd(fjx0,tx);
 376             fjy0             = _mm256_add_pd(fjy0,ty);
 377             fjz0             = _mm256_add_pd(fjz0,tz);
 378
 379             }
 380
 381             fjptrA             = f+j_coord_offsetA;
 382             fjptrB             = f+j_coord_offsetB;
 383             fjptrC             = f+j_coord_offsetC;
 384             fjptrD             = f+j_coord_offsetD;
 385
 386             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 387
 388             /* Inner loop uses 129 flops */
 389         }
 390
 391         if(jidx<j_index_end)
 392         {
 393
 394             /* Get j neighbor index, and coordinate index */
 395             jnrlistA         = jjnr[jidx];
 396             jnrlistB         = jjnr[jidx+1];
 397             jnrlistC         = jjnr[jidx+2];
 398             jnrlistD         = jjnr[jidx+3];
 399             /* Sign of each element will be negative for non-real atoms.
 400              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 401              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 402              */
 403             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 404
 405             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 406             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 407             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 408
 409             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 410             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 411             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 412             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 413             j_coord_offsetA  = DIM*jnrA;
 414             j_coord_offsetB  = DIM*jnrB;
 415             j_coord_offsetC  = DIM*jnrC;
 416             j_coord_offsetD  = DIM*jnrD;
 417
 418             /* load j atom coordinates */
 419             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 420                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 421                                                  &jx0,&jy0,&jz0);
 422
 423             /* Calculate displacement vector */
 424             dx00             = _mm256_sub_pd(ix0,jx0);
 425             dy00             = _mm256_sub_pd(iy0,jy0);
 426             dz00             = _mm256_sub_pd(iz0,jz0);
 427             dx10             = _mm256_sub_pd(ix1,jx0);
 428             dy10             = _mm256_sub_pd(iy1,jy0);
 429             dz10             = _mm256_sub_pd(iz1,jz0);
 430             dx20             = _mm256_sub_pd(ix2,jx0);
 431             dy20             = _mm256_sub_pd(iy2,jy0);
 432             dz20             = _mm256_sub_pd(iz2,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 436             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 437             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 438
 439             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 440             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 441             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 442
 443             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 444             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 445             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 446
 447             /* Load parameters for j particles */
 448             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 449                                                                  charge+jnrC+0,charge+jnrD+0);
 450             vdwjidx0A        = 2*vdwtype[jnrA+0];
 451             vdwjidx0B        = 2*vdwtype[jnrB+0];
 452             vdwjidx0C        = 2*vdwtype[jnrC+0];
 453             vdwjidx0D        = 2*vdwtype[jnrD+0];
 454
 455             fjx0             = _mm256_setzero_pd();
 456             fjy0             = _mm256_setzero_pd();
 457             fjz0             = _mm256_setzero_pd();
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 464             {
 465
 466             /* Compute parameters for interactions between i and j atoms */
 467             qq00             = _mm256_mul_pd(iq0,jq0);
 468             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 469                                             vdwioffsetptr0+vdwjidx0B,
 470                                             vdwioffsetptr0+vdwjidx0C,
 471                                             vdwioffsetptr0+vdwjidx0D,
 472                                             &c6_00,&c12_00);
 473
 474             /* REACTION-FIELD ELECTROSTATICS */
 475             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 476             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 477
 478             /* LENNARD-JONES DISPERSION/REPULSION */
 479
 480             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 481             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 482             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 483             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) ,
 484                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 485             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 486
 487             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm256_and_pd(velec,cutoff_mask);
 491             velec            = _mm256_andnot_pd(dummy_mask,velec);
 492             velecsum         = _mm256_add_pd(velecsum,velec);
 493             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 494             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 495             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 496
 497             fscal            = _mm256_add_pd(felec,fvdw);
 498
 499             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 500
 501             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 502
 503             /* Calculate temporary vectorial force */
 504             tx               = _mm256_mul_pd(fscal,dx00);
 505             ty               = _mm256_mul_pd(fscal,dy00);
 506             tz               = _mm256_mul_pd(fscal,dz00);
 507
 508             /* Update vectorial force */
 509             fix0             = _mm256_add_pd(fix0,tx);
 510             fiy0             = _mm256_add_pd(fiy0,ty);
 511             fiz0             = _mm256_add_pd(fiz0,tz);
 512
 513             fjx0             = _mm256_add_pd(fjx0,tx);
 514             fjy0             = _mm256_add_pd(fjy0,ty);
 515             fjz0             = _mm256_add_pd(fjz0,tz);
 516
 517             }
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 524             {
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq10             = _mm256_mul_pd(iq1,jq0);
 528
 529             /* REACTION-FIELD ELECTROSTATICS */
 530             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 531             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 532
 533             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             velec            = _mm256_and_pd(velec,cutoff_mask);
 537             velec            = _mm256_andnot_pd(dummy_mask,velec);
 538             velecsum         = _mm256_add_pd(velecsum,velec);
 539
 540             fscal            = felec;
 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,dx10);
 548             ty               = _mm256_mul_pd(fscal,dy10);
 549             tz               = _mm256_mul_pd(fscal,dz10);
 550
 551             /* Update vectorial force */
 552             fix1             = _mm256_add_pd(fix1,tx);
 553             fiy1             = _mm256_add_pd(fiy1,ty);
 554             fiz1             = _mm256_add_pd(fiz1,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(rsq20,rcutoff2))
 567             {
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq20             = _mm256_mul_pd(iq2,jq0);
 571
 572             /* REACTION-FIELD ELECTROSTATICS */
 573             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 574             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 575
 576             cutoff_mask      = _mm256_cmp_pd(rsq20,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,dx20);
 591             ty               = _mm256_mul_pd(fscal,dy20);
 592             tz               = _mm256_mul_pd(fscal,dz20);
 593
 594             /* Update vectorial force */
 595             fix2             = _mm256_add_pd(fix2,tx);
 596             fiy2             = _mm256_add_pd(fiy2,ty);
 597             fiz2             = _mm256_add_pd(fiz2,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 606             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 607             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 608             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 609
 610             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 611
 612             /* Inner loop uses 129 flops */
 613         }
 614
 615         /* End of innermost loop */
 616
 617         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 618                                                  f+i_coord_offset,fshift+i_shift_offset);
 619
 620         ggid                        = gid[iidx];
 621         /* Update potential energies */
 622         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 623         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 624
 625         /* Increment number of inner iterations */
 626         inneriter                  += j_index_end - j_index_start;
 627
 628         /* Outer loop uses 20 flops */
 629     }
 630
 631     /* Increment number of outer iterations */
 632     outeriter        += nri;
 633
 634     /* Update outer/inner flops */
 635
 636     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 637 }
 638 /*
 639  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_double
 640  * Electrostatics interaction: ReactionField
 641  * VdW interaction:            LennardJones
 642  * Geometry:                   Water3-Particle
 643  * Calculate force/pot:        Force
 644  */
 645 void
 646 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_double
 647                     (t_nblist                    * gmx_restrict       nlist,
 648                      rvec                        * gmx_restrict          xx,
 649                      rvec                        * gmx_restrict          ff,
 650                      t_forcerec                  * gmx_restrict          fr,
 651                      t_mdatoms                   * gmx_restrict     mdatoms,
 652                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 653                      t_nrnb                      * gmx_restrict        nrnb)
 654 {
 655     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 656      * just 0 for non-waters.
 657      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 658      * jnr indices corresponding to data put in the four positions in the SIMD register.
 659      */
 660     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 661     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 662     int              jnrA,jnrB,jnrC,jnrD;
 663     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 664     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 665     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 666     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 667     real             rcutoff_scalar;
 668     real             *shiftvec,*fshift,*x,*f;
 669     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 670     real             scratch[4*DIM];
 671     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 672     real *           vdwioffsetptr0;
 673     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 674     real *           vdwioffsetptr1;
 675     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 676     real *           vdwioffsetptr2;
 677     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 678     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 679     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 680     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 681     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 682     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 683     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 684     real             *charge;
 685     int              nvdwtype;
 686     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 687     int              *vdwtype;
 688     real             *vdwparam;
 689     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 690     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 691     __m256d          dummy_mask,cutoff_mask;
 692     __m128           tmpmask0,tmpmask1;
 693     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 694     __m256d          one     = _mm256_set1_pd(1.0);
 695     __m256d          two     = _mm256_set1_pd(2.0);
 696     x                = xx[0];
 697     f                = ff[0];
 698
 699     nri              = nlist->nri;
 700     iinr             = nlist->iinr;
 701     jindex           = nlist->jindex;
 702     jjnr             = nlist->jjnr;
 703     shiftidx         = nlist->shift;
 704     gid              = nlist->gid;
 705     shiftvec         = fr->shift_vec[0];
 706     fshift           = fr->fshift[0];
 707     facel            = _mm256_set1_pd(fr->epsfac);
 708     charge           = mdatoms->chargeA;
 709     krf              = _mm256_set1_pd(fr->ic->k_rf);
 710     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 711     crf              = _mm256_set1_pd(fr->ic->c_rf);
 712     nvdwtype         = fr->ntype;
 713     vdwparam         = fr->nbfp;
 714     vdwtype          = mdatoms->typeA;
 715
 716     /* Setup water-specific parameters */
 717     inr              = nlist->iinr[0];
 718     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 719     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 720     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 721     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 722
 723     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 724     rcutoff_scalar   = fr->rcoulomb;
 725     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 726     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 727
 728     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 729     rvdw             = _mm256_set1_pd(fr->rvdw);
 730
 731     /* Avoid stupid compiler warnings */
 732     jnrA = jnrB = jnrC = jnrD = 0;
 733     j_coord_offsetA = 0;
 734     j_coord_offsetB = 0;
 735     j_coord_offsetC = 0;
 736     j_coord_offsetD = 0;
 737
 738     outeriter        = 0;
 739     inneriter        = 0;
 740
 741     for(iidx=0;iidx<4*DIM;iidx++)
 742     {
 743         scratch[iidx] = 0.0;
 744     }
 745
 746     /* Start outer loop over neighborlists */
 747     for(iidx=0; iidx<nri; iidx++)
 748     {
 749         /* Load shift vector for this list */
 750         i_shift_offset   = DIM*shiftidx[iidx];
 751
 752         /* Load limits for loop over neighbors */
 753         j_index_start    = jindex[iidx];
 754         j_index_end      = jindex[iidx+1];
 755
 756         /* Get outer coordinate index */
 757         inr              = iinr[iidx];
 758         i_coord_offset   = DIM*inr;
 759
 760         /* Load i particle coords and add shift vector */
 761         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 762                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 763
 764         fix0             = _mm256_setzero_pd();
 765         fiy0             = _mm256_setzero_pd();
 766         fiz0             = _mm256_setzero_pd();
 767         fix1             = _mm256_setzero_pd();
 768         fiy1             = _mm256_setzero_pd();
 769         fiz1             = _mm256_setzero_pd();
 770         fix2             = _mm256_setzero_pd();
 771         fiy2             = _mm256_setzero_pd();
 772         fiz2             = _mm256_setzero_pd();
 773
 774         /* Start inner kernel loop */
 775         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 776         {
 777
 778             /* Get j neighbor index, and coordinate index */
 779             jnrA             = jjnr[jidx];
 780             jnrB             = jjnr[jidx+1];
 781             jnrC             = jjnr[jidx+2];
 782             jnrD             = jjnr[jidx+3];
 783             j_coord_offsetA  = DIM*jnrA;
 784             j_coord_offsetB  = DIM*jnrB;
 785             j_coord_offsetC  = DIM*jnrC;
 786             j_coord_offsetD  = DIM*jnrD;
 787
 788             /* load j atom coordinates */
 789             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 790                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 791                                                  &jx0,&jy0,&jz0);
 792
 793             /* Calculate displacement vector */
 794             dx00             = _mm256_sub_pd(ix0,jx0);
 795             dy00             = _mm256_sub_pd(iy0,jy0);
 796             dz00             = _mm256_sub_pd(iz0,jz0);
 797             dx10             = _mm256_sub_pd(ix1,jx0);
 798             dy10             = _mm256_sub_pd(iy1,jy0);
 799             dz10             = _mm256_sub_pd(iz1,jz0);
 800             dx20             = _mm256_sub_pd(ix2,jx0);
 801             dy20             = _mm256_sub_pd(iy2,jy0);
 802             dz20             = _mm256_sub_pd(iz2,jz0);
 803
 804             /* Calculate squared distance and things based on it */
 805             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 806             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 807             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 808
 809             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 810             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 811             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 812
 813             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 814             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 815             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 816
 817             /* Load parameters for j particles */
 818             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 819                                                                  charge+jnrC+0,charge+jnrD+0);
 820             vdwjidx0A        = 2*vdwtype[jnrA+0];
 821             vdwjidx0B        = 2*vdwtype[jnrB+0];
 822             vdwjidx0C        = 2*vdwtype[jnrC+0];
 823             vdwjidx0D        = 2*vdwtype[jnrD+0];
 824
 825             fjx0             = _mm256_setzero_pd();
 826             fjy0             = _mm256_setzero_pd();
 827             fjz0             = _mm256_setzero_pd();
 828
 829             /**************************
 830              * CALCULATE INTERACTIONS *
 831              **************************/
 832
 833             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 834             {
 835
 836             /* Compute parameters for interactions between i and j atoms */
 837             qq00             = _mm256_mul_pd(iq0,jq0);
 838             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 839                                             vdwioffsetptr0+vdwjidx0B,
 840                                             vdwioffsetptr0+vdwjidx0C,
 841                                             vdwioffsetptr0+vdwjidx0D,
 842                                             &c6_00,&c12_00);
 843
 844             /* REACTION-FIELD ELECTROSTATICS */
 845             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 846
 847             /* LENNARD-JONES DISPERSION/REPULSION */
 848
 849             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 850             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 851
 852             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 853
 854             fscal            = _mm256_add_pd(felec,fvdw);
 855
 856             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm256_mul_pd(fscal,dx00);
 860             ty               = _mm256_mul_pd(fscal,dy00);
 861             tz               = _mm256_mul_pd(fscal,dz00);
 862
 863             /* Update vectorial force */
 864             fix0             = _mm256_add_pd(fix0,tx);
 865             fiy0             = _mm256_add_pd(fiy0,ty);
 866             fiz0             = _mm256_add_pd(fiz0,tz);
 867
 868             fjx0             = _mm256_add_pd(fjx0,tx);
 869             fjy0             = _mm256_add_pd(fjy0,ty);
 870             fjz0             = _mm256_add_pd(fjz0,tz);
 871
 872             }
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 879             {
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             qq10             = _mm256_mul_pd(iq1,jq0);
 883
 884             /* REACTION-FIELD ELECTROSTATICS */
 885             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 886
 887             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 888
 889             fscal            = felec;
 890
 891             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 892
 893             /* Calculate temporary vectorial force */
 894             tx               = _mm256_mul_pd(fscal,dx10);
 895             ty               = _mm256_mul_pd(fscal,dy10);
 896             tz               = _mm256_mul_pd(fscal,dz10);
 897
 898             /* Update vectorial force */
 899             fix1             = _mm256_add_pd(fix1,tx);
 900             fiy1             = _mm256_add_pd(fiy1,ty);
 901             fiz1             = _mm256_add_pd(fiz1,tz);
 902
 903             fjx0             = _mm256_add_pd(fjx0,tx);
 904             fjy0             = _mm256_add_pd(fjy0,ty);
 905             fjz0             = _mm256_add_pd(fjz0,tz);
 906
 907             }
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 914             {
 915
 916             /* Compute parameters for interactions between i and j atoms */
 917             qq20             = _mm256_mul_pd(iq2,jq0);
 918
 919             /* REACTION-FIELD ELECTROSTATICS */
 920             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 921
 922             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 923
 924             fscal            = felec;
 925
 926             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 927
 928             /* Calculate temporary vectorial force */
 929             tx               = _mm256_mul_pd(fscal,dx20);
 930             ty               = _mm256_mul_pd(fscal,dy20);
 931             tz               = _mm256_mul_pd(fscal,dz20);
 932
 933             /* Update vectorial force */
 934             fix2             = _mm256_add_pd(fix2,tx);
 935             fiy2             = _mm256_add_pd(fiy2,ty);
 936             fiz2             = _mm256_add_pd(fiz2,tz);
 937
 938             fjx0             = _mm256_add_pd(fjx0,tx);
 939             fjy0             = _mm256_add_pd(fjy0,ty);
 940             fjz0             = _mm256_add_pd(fjz0,tz);
 941
 942             }
 943
 944             fjptrA             = f+j_coord_offsetA;
 945             fjptrB             = f+j_coord_offsetB;
 946             fjptrC             = f+j_coord_offsetC;
 947             fjptrD             = f+j_coord_offsetD;
 948
 949             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 950
 951             /* Inner loop uses 100 flops */
 952         }
 953
 954         if(jidx<j_index_end)
 955         {
 956
 957             /* Get j neighbor index, and coordinate index */
 958             jnrlistA         = jjnr[jidx];
 959             jnrlistB         = jjnr[jidx+1];
 960             jnrlistC         = jjnr[jidx+2];
 961             jnrlistD         = jjnr[jidx+3];
 962             /* Sign of each element will be negative for non-real atoms.
 963              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 964              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 965              */
 966             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 967
 968             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 969             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 970             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 971
 972             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 973             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 974             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 975             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 976             j_coord_offsetA  = DIM*jnrA;
 977             j_coord_offsetB  = DIM*jnrB;
 978             j_coord_offsetC  = DIM*jnrC;
 979             j_coord_offsetD  = DIM*jnrD;
 980
 981             /* load j atom coordinates */
 982             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 983                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 984                                                  &jx0,&jy0,&jz0);
 985
 986             /* Calculate displacement vector */
 987             dx00             = _mm256_sub_pd(ix0,jx0);
 988             dy00             = _mm256_sub_pd(iy0,jy0);
 989             dz00             = _mm256_sub_pd(iz0,jz0);
 990             dx10             = _mm256_sub_pd(ix1,jx0);
 991             dy10             = _mm256_sub_pd(iy1,jy0);
 992             dz10             = _mm256_sub_pd(iz1,jz0);
 993             dx20             = _mm256_sub_pd(ix2,jx0);
 994             dy20             = _mm256_sub_pd(iy2,jy0);
 995             dz20             = _mm256_sub_pd(iz2,jz0);
 996
 997             /* Calculate squared distance and things based on it */
 998             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 999             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1000             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1001
1002             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1003             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1004             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1005
1006             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1007             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1008             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1009
1010             /* Load parameters for j particles */
1011             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1012                                                                  charge+jnrC+0,charge+jnrD+0);
1013             vdwjidx0A        = 2*vdwtype[jnrA+0];
1014             vdwjidx0B        = 2*vdwtype[jnrB+0];
1015             vdwjidx0C        = 2*vdwtype[jnrC+0];
1016             vdwjidx0D        = 2*vdwtype[jnrD+0];
1017
1018             fjx0             = _mm256_setzero_pd();
1019             fjy0             = _mm256_setzero_pd();
1020             fjz0             = _mm256_setzero_pd();
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1027             {
1028
1029             /* Compute parameters for interactions between i and j atoms */
1030             qq00             = _mm256_mul_pd(iq0,jq0);
1031             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1032                                             vdwioffsetptr0+vdwjidx0B,
1033                                             vdwioffsetptr0+vdwjidx0C,
1034                                             vdwioffsetptr0+vdwjidx0D,
1035                                             &c6_00,&c12_00);
1036
1037             /* REACTION-FIELD ELECTROSTATICS */
1038             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1039
1040             /* LENNARD-JONES DISPERSION/REPULSION */
1041
1042             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1043             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1044
1045             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1046
1047             fscal            = _mm256_add_pd(felec,fvdw);
1048
1049             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1050
1051             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm256_mul_pd(fscal,dx00);
1055             ty               = _mm256_mul_pd(fscal,dy00);
1056             tz               = _mm256_mul_pd(fscal,dz00);
1057
1058             /* Update vectorial force */
1059             fix0             = _mm256_add_pd(fix0,tx);
1060             fiy0             = _mm256_add_pd(fiy0,ty);
1061             fiz0             = _mm256_add_pd(fiz0,tz);
1062
1063             fjx0             = _mm256_add_pd(fjx0,tx);
1064             fjy0             = _mm256_add_pd(fjy0,ty);
1065             fjz0             = _mm256_add_pd(fjz0,tz);
1066
1067             }
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1074             {
1075
1076             /* Compute parameters for interactions between i and j atoms */
1077             qq10             = _mm256_mul_pd(iq1,jq0);
1078
1079             /* REACTION-FIELD ELECTROSTATICS */
1080             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1081
1082             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1083
1084             fscal            = felec;
1085
1086             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1087
1088             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1089
1090             /* Calculate temporary vectorial force */
1091             tx               = _mm256_mul_pd(fscal,dx10);
1092             ty               = _mm256_mul_pd(fscal,dy10);
1093             tz               = _mm256_mul_pd(fscal,dz10);
1094
1095             /* Update vectorial force */
1096             fix1             = _mm256_add_pd(fix1,tx);
1097             fiy1             = _mm256_add_pd(fiy1,ty);
1098             fiz1             = _mm256_add_pd(fiz1,tz);
1099
1100             fjx0             = _mm256_add_pd(fjx0,tx);
1101             fjy0             = _mm256_add_pd(fjy0,ty);
1102             fjz0             = _mm256_add_pd(fjz0,tz);
1103
1104             }
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1111             {
1112
1113             /* Compute parameters for interactions between i and j atoms */
1114             qq20             = _mm256_mul_pd(iq2,jq0);
1115
1116             /* REACTION-FIELD ELECTROSTATICS */
1117             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1118
1119             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1120
1121             fscal            = felec;
1122
1123             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1124
1125             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1126
1127             /* Calculate temporary vectorial force */
1128             tx               = _mm256_mul_pd(fscal,dx20);
1129             ty               = _mm256_mul_pd(fscal,dy20);
1130             tz               = _mm256_mul_pd(fscal,dz20);
1131
1132             /* Update vectorial force */
1133             fix2             = _mm256_add_pd(fix2,tx);
1134             fiy2             = _mm256_add_pd(fiy2,ty);
1135             fiz2             = _mm256_add_pd(fiz2,tz);
1136
1137             fjx0             = _mm256_add_pd(fjx0,tx);
1138             fjy0             = _mm256_add_pd(fjy0,ty);
1139             fjz0             = _mm256_add_pd(fjz0,tz);
1140
1141             }
1142
1143             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1144             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1145             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1146             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1147
1148             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1149
1150             /* Inner loop uses 100 flops */
1151         }
1152
1153         /* End of innermost loop */
1154
1155         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1156                                                  f+i_coord_offset,fshift+i_shift_offset);
1157
1158         /* Increment number of inner iterations */
1159         inneriter                  += j_index_end - j_index_start;
1160
1161         /* Outer loop uses 18 flops */
1162     }
1163
1164     /* Increment number of outer iterations */
1165     outeriter        += nri;
1166
1167     /* Update outer/inner flops */
1168
1169     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1170 }