src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_avx_256_single.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_single 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_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 109     real             rswitch_scalar,d_scalar;
 110     __m256           dummy_mask,cutoff_mask;
 111     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 112     __m256           one     = _mm256_set1_ps(1.0);
 113     __m256           two     = _mm256_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     krf              = _mm256_set1_ps(fr->ic->k_rf);
 128     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 129     crf              = _mm256_set1_ps(fr->ic->c_rf);
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 137     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 138     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 142     rcutoff_scalar   = fr->rcoulomb;
 143     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 144     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 145
 146     rswitch_scalar   = fr->rvdw_switch;
 147     rswitch          = _mm256_set1_ps(rswitch_scalar);
 148     /* Setup switch parameters */
 149     d_scalar         = rcutoff_scalar-rswitch_scalar;
 150     d                = _mm256_set1_ps(d_scalar);
 151     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 152     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 153     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 154     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 155     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 156     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 157
 158     /* Avoid stupid compiler warnings */
 159     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 160     j_coord_offsetA = 0;
 161     j_coord_offsetB = 0;
 162     j_coord_offsetC = 0;
 163     j_coord_offsetD = 0;
 164     j_coord_offsetE = 0;
 165     j_coord_offsetF = 0;
 166     j_coord_offsetG = 0;
 167     j_coord_offsetH = 0;
 168
 169     outeriter        = 0;
 170     inneriter        = 0;
 171
 172     for(iidx=0;iidx<4*DIM;iidx++)
 173     {
 174         scratch[iidx] = 0.0;
 175     }
 176
 177     /* Start outer loop over neighborlists */
 178     for(iidx=0; iidx<nri; iidx++)
 179     {
 180         /* Load shift vector for this list */
 181         i_shift_offset   = DIM*shiftidx[iidx];
 182
 183         /* Load limits for loop over neighbors */
 184         j_index_start    = jindex[iidx];
 185         j_index_end      = jindex[iidx+1];
 186
 187         /* Get outer coordinate index */
 188         inr              = iinr[iidx];
 189         i_coord_offset   = DIM*inr;
 190
 191         /* Load i particle coords and add shift vector */
 192         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 193                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 194
 195         fix0             = _mm256_setzero_ps();
 196         fiy0             = _mm256_setzero_ps();
 197         fiz0             = _mm256_setzero_ps();
 198         fix1             = _mm256_setzero_ps();
 199         fiy1             = _mm256_setzero_ps();
 200         fiz1             = _mm256_setzero_ps();
 201         fix2             = _mm256_setzero_ps();
 202         fiy2             = _mm256_setzero_ps();
 203         fiz2             = _mm256_setzero_ps();
 204         fix3             = _mm256_setzero_ps();
 205         fiy3             = _mm256_setzero_ps();
 206         fiz3             = _mm256_setzero_ps();
 207
 208         /* Reset potential sums */
 209         velecsum         = _mm256_setzero_ps();
 210         vvdwsum          = _mm256_setzero_ps();
 211
 212         /* Start inner kernel loop */
 213         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 214         {
 215
 216             /* Get j neighbor index, and coordinate index */
 217             jnrA             = jjnr[jidx];
 218             jnrB             = jjnr[jidx+1];
 219             jnrC             = jjnr[jidx+2];
 220             jnrD             = jjnr[jidx+3];
 221             jnrE             = jjnr[jidx+4];
 222             jnrF             = jjnr[jidx+5];
 223             jnrG             = jjnr[jidx+6];
 224             jnrH             = jjnr[jidx+7];
 225             j_coord_offsetA  = DIM*jnrA;
 226             j_coord_offsetB  = DIM*jnrB;
 227             j_coord_offsetC  = DIM*jnrC;
 228             j_coord_offsetD  = DIM*jnrD;
 229             j_coord_offsetE  = DIM*jnrE;
 230             j_coord_offsetF  = DIM*jnrF;
 231             j_coord_offsetG  = DIM*jnrG;
 232             j_coord_offsetH  = DIM*jnrH;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 237                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 238                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 239                                                  &jx0,&jy0,&jz0);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm256_sub_ps(ix0,jx0);
 243             dy00             = _mm256_sub_ps(iy0,jy0);
 244             dz00             = _mm256_sub_ps(iz0,jz0);
 245             dx10             = _mm256_sub_ps(ix1,jx0);
 246             dy10             = _mm256_sub_ps(iy1,jy0);
 247             dz10             = _mm256_sub_ps(iz1,jz0);
 248             dx20             = _mm256_sub_ps(ix2,jx0);
 249             dy20             = _mm256_sub_ps(iy2,jy0);
 250             dz20             = _mm256_sub_ps(iz2,jz0);
 251             dx30             = _mm256_sub_ps(ix3,jx0);
 252             dy30             = _mm256_sub_ps(iy3,jy0);
 253             dz30             = _mm256_sub_ps(iz3,jz0);
 254
 255             /* Calculate squared distance and things based on it */
 256             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 257             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 258             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 259             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 260
 261             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 262             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 263             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 264             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 265
 266             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 267             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 268             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 269             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 270
 271             /* Load parameters for j particles */
 272             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 273                                                                  charge+jnrC+0,charge+jnrD+0,
 274                                                                  charge+jnrE+0,charge+jnrF+0,
 275                                                                  charge+jnrG+0,charge+jnrH+0);
 276             vdwjidx0A        = 2*vdwtype[jnrA+0];
 277             vdwjidx0B        = 2*vdwtype[jnrB+0];
 278             vdwjidx0C        = 2*vdwtype[jnrC+0];
 279             vdwjidx0D        = 2*vdwtype[jnrD+0];
 280             vdwjidx0E        = 2*vdwtype[jnrE+0];
 281             vdwjidx0F        = 2*vdwtype[jnrF+0];
 282             vdwjidx0G        = 2*vdwtype[jnrG+0];
 283             vdwjidx0H        = 2*vdwtype[jnrH+0];
 284
 285             fjx0             = _mm256_setzero_ps();
 286             fjy0             = _mm256_setzero_ps();
 287             fjz0             = _mm256_setzero_ps();
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 294             {
 295
 296             r00              = _mm256_mul_ps(rsq00,rinv00);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 300                                             vdwioffsetptr0+vdwjidx0B,
 301                                             vdwioffsetptr0+vdwjidx0C,
 302                                             vdwioffsetptr0+vdwjidx0D,
 303                                             vdwioffsetptr0+vdwjidx0E,
 304                                             vdwioffsetptr0+vdwjidx0F,
 305                                             vdwioffsetptr0+vdwjidx0G,
 306                                             vdwioffsetptr0+vdwjidx0H,
 307                                             &c6_00,&c12_00);
 308
 309             /* LENNARD-JONES DISPERSION/REPULSION */
 310
 311             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 312             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 313             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 314             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 315             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 316
 317             d                = _mm256_sub_ps(r00,rswitch);
 318             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 319             d2               = _mm256_mul_ps(d,d);
 320             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 321
 322             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 323
 324             /* Evaluate switch function */
 325             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 326             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 327             vvdw             = _mm256_mul_ps(vvdw,sw);
 328             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 332             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 333
 334             fscal            = fvdw;
 335
 336             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm256_mul_ps(fscal,dx00);
 340             ty               = _mm256_mul_ps(fscal,dy00);
 341             tz               = _mm256_mul_ps(fscal,dz00);
 342
 343             /* Update vectorial force */
 344             fix0             = _mm256_add_ps(fix0,tx);
 345             fiy0             = _mm256_add_ps(fiy0,ty);
 346             fiz0             = _mm256_add_ps(fiz0,tz);
 347
 348             fjx0             = _mm256_add_ps(fjx0,tx);
 349             fjy0             = _mm256_add_ps(fjy0,ty);
 350             fjz0             = _mm256_add_ps(fjz0,tz);
 351
 352             }
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 359             {
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq10             = _mm256_mul_ps(iq1,jq0);
 363
 364             /* REACTION-FIELD ELECTROSTATICS */
 365             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 366             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 367
 368             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm256_and_ps(velec,cutoff_mask);
 372             velecsum         = _mm256_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm256_mul_ps(fscal,dx10);
 380             ty               = _mm256_mul_ps(fscal,dy10);
 381             tz               = _mm256_mul_ps(fscal,dz10);
 382
 383             /* Update vectorial force */
 384             fix1             = _mm256_add_ps(fix1,tx);
 385             fiy1             = _mm256_add_ps(fiy1,ty);
 386             fiz1             = _mm256_add_ps(fiz1,tz);
 387
 388             fjx0             = _mm256_add_ps(fjx0,tx);
 389             fjy0             = _mm256_add_ps(fjy0,ty);
 390             fjz0             = _mm256_add_ps(fjz0,tz);
 391
 392             }
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 399             {
 400
 401             /* Compute parameters for interactions between i and j atoms */
 402             qq20             = _mm256_mul_ps(iq2,jq0);
 403
 404             /* REACTION-FIELD ELECTROSTATICS */
 405             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 406             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 407
 408             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velec            = _mm256_and_ps(velec,cutoff_mask);
 412             velecsum         = _mm256_add_ps(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 417
 418             /* Calculate temporary vectorial force */
 419             tx               = _mm256_mul_ps(fscal,dx20);
 420             ty               = _mm256_mul_ps(fscal,dy20);
 421             tz               = _mm256_mul_ps(fscal,dz20);
 422
 423             /* Update vectorial force */
 424             fix2             = _mm256_add_ps(fix2,tx);
 425             fiy2             = _mm256_add_ps(fiy2,ty);
 426             fiz2             = _mm256_add_ps(fiz2,tz);
 427
 428             fjx0             = _mm256_add_ps(fjx0,tx);
 429             fjy0             = _mm256_add_ps(fjy0,ty);
 430             fjz0             = _mm256_add_ps(fjz0,tz);
 431
 432             }
 433
 434             /**************************
 435              * CALCULATE INTERACTIONS *
 436              **************************/
 437
 438             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 439             {
 440
 441             /* Compute parameters for interactions between i and j atoms */
 442             qq30             = _mm256_mul_ps(iq3,jq0);
 443
 444             /* REACTION-FIELD ELECTROSTATICS */
 445             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 446             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 447
 448             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velec            = _mm256_and_ps(velec,cutoff_mask);
 452             velecsum         = _mm256_add_ps(velecsum,velec);
 453
 454             fscal            = felec;
 455
 456             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 457
 458             /* Calculate temporary vectorial force */
 459             tx               = _mm256_mul_ps(fscal,dx30);
 460             ty               = _mm256_mul_ps(fscal,dy30);
 461             tz               = _mm256_mul_ps(fscal,dz30);
 462
 463             /* Update vectorial force */
 464             fix3             = _mm256_add_ps(fix3,tx);
 465             fiy3             = _mm256_add_ps(fiy3,ty);
 466             fiz3             = _mm256_add_ps(fiz3,tz);
 467
 468             fjx0             = _mm256_add_ps(fjx0,tx);
 469             fjy0             = _mm256_add_ps(fjy0,ty);
 470             fjz0             = _mm256_add_ps(fjz0,tz);
 471
 472             }
 473
 474             fjptrA             = f+j_coord_offsetA;
 475             fjptrB             = f+j_coord_offsetB;
 476             fjptrC             = f+j_coord_offsetC;
 477             fjptrD             = f+j_coord_offsetD;
 478             fjptrE             = f+j_coord_offsetE;
 479             fjptrF             = f+j_coord_offsetF;
 480             fjptrG             = f+j_coord_offsetG;
 481             fjptrH             = f+j_coord_offsetH;
 482
 483             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 484
 485             /* Inner loop uses 170 flops */
 486         }
 487
 488         if(jidx<j_index_end)
 489         {
 490
 491             /* Get j neighbor index, and coordinate index */
 492             jnrlistA         = jjnr[jidx];
 493             jnrlistB         = jjnr[jidx+1];
 494             jnrlistC         = jjnr[jidx+2];
 495             jnrlistD         = jjnr[jidx+3];
 496             jnrlistE         = jjnr[jidx+4];
 497             jnrlistF         = jjnr[jidx+5];
 498             jnrlistG         = jjnr[jidx+6];
 499             jnrlistH         = jjnr[jidx+7];
 500             /* Sign of each element will be negative for non-real atoms.
 501              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 502              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 503              */
 504             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 505                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 506
 507             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 508             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 509             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 510             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 511             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 512             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 513             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 514             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 515             j_coord_offsetA  = DIM*jnrA;
 516             j_coord_offsetB  = DIM*jnrB;
 517             j_coord_offsetC  = DIM*jnrC;
 518             j_coord_offsetD  = DIM*jnrD;
 519             j_coord_offsetE  = DIM*jnrE;
 520             j_coord_offsetF  = DIM*jnrF;
 521             j_coord_offsetG  = DIM*jnrG;
 522             j_coord_offsetH  = DIM*jnrH;
 523
 524             /* load j atom coordinates */
 525             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 526                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 527                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 528                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 529                                                  &jx0,&jy0,&jz0);
 530
 531             /* Calculate displacement vector */
 532             dx00             = _mm256_sub_ps(ix0,jx0);
 533             dy00             = _mm256_sub_ps(iy0,jy0);
 534             dz00             = _mm256_sub_ps(iz0,jz0);
 535             dx10             = _mm256_sub_ps(ix1,jx0);
 536             dy10             = _mm256_sub_ps(iy1,jy0);
 537             dz10             = _mm256_sub_ps(iz1,jz0);
 538             dx20             = _mm256_sub_ps(ix2,jx0);
 539             dy20             = _mm256_sub_ps(iy2,jy0);
 540             dz20             = _mm256_sub_ps(iz2,jz0);
 541             dx30             = _mm256_sub_ps(ix3,jx0);
 542             dy30             = _mm256_sub_ps(iy3,jy0);
 543             dz30             = _mm256_sub_ps(iz3,jz0);
 544
 545             /* Calculate squared distance and things based on it */
 546             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 547             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 548             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 549             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 550
 551             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 552             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 553             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 554             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 555
 556             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 557             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 558             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 559             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 560
 561             /* Load parameters for j particles */
 562             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 563                                                                  charge+jnrC+0,charge+jnrD+0,
 564                                                                  charge+jnrE+0,charge+jnrF+0,
 565                                                                  charge+jnrG+0,charge+jnrH+0);
 566             vdwjidx0A        = 2*vdwtype[jnrA+0];
 567             vdwjidx0B        = 2*vdwtype[jnrB+0];
 568             vdwjidx0C        = 2*vdwtype[jnrC+0];
 569             vdwjidx0D        = 2*vdwtype[jnrD+0];
 570             vdwjidx0E        = 2*vdwtype[jnrE+0];
 571             vdwjidx0F        = 2*vdwtype[jnrF+0];
 572             vdwjidx0G        = 2*vdwtype[jnrG+0];
 573             vdwjidx0H        = 2*vdwtype[jnrH+0];
 574
 575             fjx0             = _mm256_setzero_ps();
 576             fjy0             = _mm256_setzero_ps();
 577             fjz0             = _mm256_setzero_ps();
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 584             {
 585
 586             r00              = _mm256_mul_ps(rsq00,rinv00);
 587             r00              = _mm256_andnot_ps(dummy_mask,r00);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 591                                             vdwioffsetptr0+vdwjidx0B,
 592                                             vdwioffsetptr0+vdwjidx0C,
 593                                             vdwioffsetptr0+vdwjidx0D,
 594                                             vdwioffsetptr0+vdwjidx0E,
 595                                             vdwioffsetptr0+vdwjidx0F,
 596                                             vdwioffsetptr0+vdwjidx0G,
 597                                             vdwioffsetptr0+vdwjidx0H,
 598                                             &c6_00,&c12_00);
 599
 600             /* LENNARD-JONES DISPERSION/REPULSION */
 601
 602             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 603             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 604             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 605             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 606             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 607
 608             d                = _mm256_sub_ps(r00,rswitch);
 609             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 610             d2               = _mm256_mul_ps(d,d);
 611             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 612
 613             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 614
 615             /* Evaluate switch function */
 616             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 617             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 618             vvdw             = _mm256_mul_ps(vvdw,sw);
 619             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 623             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 624             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 625
 626             fscal            = fvdw;
 627
 628             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 629
 630             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm256_mul_ps(fscal,dx00);
 634             ty               = _mm256_mul_ps(fscal,dy00);
 635             tz               = _mm256_mul_ps(fscal,dz00);
 636
 637             /* Update vectorial force */
 638             fix0             = _mm256_add_ps(fix0,tx);
 639             fiy0             = _mm256_add_ps(fiy0,ty);
 640             fiz0             = _mm256_add_ps(fiz0,tz);
 641
 642             fjx0             = _mm256_add_ps(fjx0,tx);
 643             fjy0             = _mm256_add_ps(fjy0,ty);
 644             fjz0             = _mm256_add_ps(fjz0,tz);
 645
 646             }
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 653             {
 654
 655             /* Compute parameters for interactions between i and j atoms */
 656             qq10             = _mm256_mul_ps(iq1,jq0);
 657
 658             /* REACTION-FIELD ELECTROSTATICS */
 659             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 660             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 661
 662             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 663
 664             /* Update potential sum for this i atom from the interaction with this j atom. */
 665             velec            = _mm256_and_ps(velec,cutoff_mask);
 666             velec            = _mm256_andnot_ps(dummy_mask,velec);
 667             velecsum         = _mm256_add_ps(velecsum,velec);
 668
 669             fscal            = felec;
 670
 671             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 672
 673             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm256_mul_ps(fscal,dx10);
 677             ty               = _mm256_mul_ps(fscal,dy10);
 678             tz               = _mm256_mul_ps(fscal,dz10);
 679
 680             /* Update vectorial force */
 681             fix1             = _mm256_add_ps(fix1,tx);
 682             fiy1             = _mm256_add_ps(fiy1,ty);
 683             fiz1             = _mm256_add_ps(fiz1,tz);
 684
 685             fjx0             = _mm256_add_ps(fjx0,tx);
 686             fjy0             = _mm256_add_ps(fjy0,ty);
 687             fjz0             = _mm256_add_ps(fjz0,tz);
 688
 689             }
 690
 691             /**************************
 692              * CALCULATE INTERACTIONS *
 693              **************************/
 694
 695             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 696             {
 697
 698             /* Compute parameters for interactions between i and j atoms */
 699             qq20             = _mm256_mul_ps(iq2,jq0);
 700
 701             /* REACTION-FIELD ELECTROSTATICS */
 702             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 703             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 704
 705             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 706
 707             /* Update potential sum for this i atom from the interaction with this j atom. */
 708             velec            = _mm256_and_ps(velec,cutoff_mask);
 709             velec            = _mm256_andnot_ps(dummy_mask,velec);
 710             velecsum         = _mm256_add_ps(velecsum,velec);
 711
 712             fscal            = felec;
 713
 714             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 715
 716             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 717
 718             /* Calculate temporary vectorial force */
 719             tx               = _mm256_mul_ps(fscal,dx20);
 720             ty               = _mm256_mul_ps(fscal,dy20);
 721             tz               = _mm256_mul_ps(fscal,dz20);
 722
 723             /* Update vectorial force */
 724             fix2             = _mm256_add_ps(fix2,tx);
 725             fiy2             = _mm256_add_ps(fiy2,ty);
 726             fiz2             = _mm256_add_ps(fiz2,tz);
 727
 728             fjx0             = _mm256_add_ps(fjx0,tx);
 729             fjy0             = _mm256_add_ps(fjy0,ty);
 730             fjz0             = _mm256_add_ps(fjz0,tz);
 731
 732             }
 733
 734             /**************************
 735              * CALCULATE INTERACTIONS *
 736              **************************/
 737
 738             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 739             {
 740
 741             /* Compute parameters for interactions between i and j atoms */
 742             qq30             = _mm256_mul_ps(iq3,jq0);
 743
 744             /* REACTION-FIELD ELECTROSTATICS */
 745             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 746             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 747
 748             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 749
 750             /* Update potential sum for this i atom from the interaction with this j atom. */
 751             velec            = _mm256_and_ps(velec,cutoff_mask);
 752             velec            = _mm256_andnot_ps(dummy_mask,velec);
 753             velecsum         = _mm256_add_ps(velecsum,velec);
 754
 755             fscal            = felec;
 756
 757             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 758
 759             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 760
 761             /* Calculate temporary vectorial force */
 762             tx               = _mm256_mul_ps(fscal,dx30);
 763             ty               = _mm256_mul_ps(fscal,dy30);
 764             tz               = _mm256_mul_ps(fscal,dz30);
 765
 766             /* Update vectorial force */
 767             fix3             = _mm256_add_ps(fix3,tx);
 768             fiy3             = _mm256_add_ps(fiy3,ty);
 769             fiz3             = _mm256_add_ps(fiz3,tz);
 770
 771             fjx0             = _mm256_add_ps(fjx0,tx);
 772             fjy0             = _mm256_add_ps(fjy0,ty);
 773             fjz0             = _mm256_add_ps(fjz0,tz);
 774
 775             }
 776
 777             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 778             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 779             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 780             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 781             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 782             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 783             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 784             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 785
 786             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 787
 788             /* Inner loop uses 171 flops */
 789         }
 790
 791         /* End of innermost loop */
 792
 793         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 794                                                  f+i_coord_offset,fshift+i_shift_offset);
 795
 796         ggid                        = gid[iidx];
 797         /* Update potential energies */
 798         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 799         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 800
 801         /* Increment number of inner iterations */
 802         inneriter                  += j_index_end - j_index_start;
 803
 804         /* Outer loop uses 26 flops */
 805     }
 806
 807     /* Increment number of outer iterations */
 808     outeriter        += nri;
 809
 810     /* Update outer/inner flops */
 811
 812     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*171);
 813 }
 814 /*
 815  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_single
 816  * Electrostatics interaction: ReactionField
 817  * VdW interaction:            LennardJones
 818  * Geometry:                   Water4-Particle
 819  * Calculate force/pot:        Force
 820  */
 821 void
 822 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_single
 823                     (t_nblist                    * gmx_restrict       nlist,
 824                      rvec                        * gmx_restrict          xx,
 825                      rvec                        * gmx_restrict          ff,
 826                      t_forcerec                  * gmx_restrict          fr,
 827                      t_mdatoms                   * gmx_restrict     mdatoms,
 828                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 829                      t_nrnb                      * gmx_restrict        nrnb)
 830 {
 831     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 832      * just 0 for non-waters.
 833      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 834      * jnr indices corresponding to data put in the four positions in the SIMD register.
 835      */
 836     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 837     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 838     int              jnrA,jnrB,jnrC,jnrD;
 839     int              jnrE,jnrF,jnrG,jnrH;
 840     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 841     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 842     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 843     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 844     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 845     real             rcutoff_scalar;
 846     real             *shiftvec,*fshift,*x,*f;
 847     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 848     real             scratch[4*DIM];
 849     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 850     real *           vdwioffsetptr0;
 851     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 852     real *           vdwioffsetptr1;
 853     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 854     real *           vdwioffsetptr2;
 855     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 856     real *           vdwioffsetptr3;
 857     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 858     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 859     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 860     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 861     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 862     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 863     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 864     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 865     real             *charge;
 866     int              nvdwtype;
 867     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 868     int              *vdwtype;
 869     real             *vdwparam;
 870     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 871     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 872     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 873     real             rswitch_scalar,d_scalar;
 874     __m256           dummy_mask,cutoff_mask;
 875     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 876     __m256           one     = _mm256_set1_ps(1.0);
 877     __m256           two     = _mm256_set1_ps(2.0);
 878     x                = xx[0];
 879     f                = ff[0];
 880
 881     nri              = nlist->nri;
 882     iinr             = nlist->iinr;
 883     jindex           = nlist->jindex;
 884     jjnr             = nlist->jjnr;
 885     shiftidx         = nlist->shift;
 886     gid              = nlist->gid;
 887     shiftvec         = fr->shift_vec[0];
 888     fshift           = fr->fshift[0];
 889     facel            = _mm256_set1_ps(fr->epsfac);
 890     charge           = mdatoms->chargeA;
 891     krf              = _mm256_set1_ps(fr->ic->k_rf);
 892     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 893     crf              = _mm256_set1_ps(fr->ic->c_rf);
 894     nvdwtype         = fr->ntype;
 895     vdwparam         = fr->nbfp;
 896     vdwtype          = mdatoms->typeA;
 897
 898     /* Setup water-specific parameters */
 899     inr              = nlist->iinr[0];
 900     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 901     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 902     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 903     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 904
 905     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 906     rcutoff_scalar   = fr->rcoulomb;
 907     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 908     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 909
 910     rswitch_scalar   = fr->rvdw_switch;
 911     rswitch          = _mm256_set1_ps(rswitch_scalar);
 912     /* Setup switch parameters */
 913     d_scalar         = rcutoff_scalar-rswitch_scalar;
 914     d                = _mm256_set1_ps(d_scalar);
 915     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 916     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 917     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 918     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 919     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 920     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 921
 922     /* Avoid stupid compiler warnings */
 923     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 924     j_coord_offsetA = 0;
 925     j_coord_offsetB = 0;
 926     j_coord_offsetC = 0;
 927     j_coord_offsetD = 0;
 928     j_coord_offsetE = 0;
 929     j_coord_offsetF = 0;
 930     j_coord_offsetG = 0;
 931     j_coord_offsetH = 0;
 932
 933     outeriter        = 0;
 934     inneriter        = 0;
 935
 936     for(iidx=0;iidx<4*DIM;iidx++)
 937     {
 938         scratch[iidx] = 0.0;
 939     }
 940
 941     /* Start outer loop over neighborlists */
 942     for(iidx=0; iidx<nri; iidx++)
 943     {
 944         /* Load shift vector for this list */
 945         i_shift_offset   = DIM*shiftidx[iidx];
 946
 947         /* Load limits for loop over neighbors */
 948         j_index_start    = jindex[iidx];
 949         j_index_end      = jindex[iidx+1];
 950
 951         /* Get outer coordinate index */
 952         inr              = iinr[iidx];
 953         i_coord_offset   = DIM*inr;
 954
 955         /* Load i particle coords and add shift vector */
 956         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 957                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 958
 959         fix0             = _mm256_setzero_ps();
 960         fiy0             = _mm256_setzero_ps();
 961         fiz0             = _mm256_setzero_ps();
 962         fix1             = _mm256_setzero_ps();
 963         fiy1             = _mm256_setzero_ps();
 964         fiz1             = _mm256_setzero_ps();
 965         fix2             = _mm256_setzero_ps();
 966         fiy2             = _mm256_setzero_ps();
 967         fiz2             = _mm256_setzero_ps();
 968         fix3             = _mm256_setzero_ps();
 969         fiy3             = _mm256_setzero_ps();
 970         fiz3             = _mm256_setzero_ps();
 971
 972         /* Start inner kernel loop */
 973         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 974         {
 975
 976             /* Get j neighbor index, and coordinate index */
 977             jnrA             = jjnr[jidx];
 978             jnrB             = jjnr[jidx+1];
 979             jnrC             = jjnr[jidx+2];
 980             jnrD             = jjnr[jidx+3];
 981             jnrE             = jjnr[jidx+4];
 982             jnrF             = jjnr[jidx+5];
 983             jnrG             = jjnr[jidx+6];
 984             jnrH             = jjnr[jidx+7];
 985             j_coord_offsetA  = DIM*jnrA;
 986             j_coord_offsetB  = DIM*jnrB;
 987             j_coord_offsetC  = DIM*jnrC;
 988             j_coord_offsetD  = DIM*jnrD;
 989             j_coord_offsetE  = DIM*jnrE;
 990             j_coord_offsetF  = DIM*jnrF;
 991             j_coord_offsetG  = DIM*jnrG;
 992             j_coord_offsetH  = DIM*jnrH;
 993
 994             /* load j atom coordinates */
 995             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 996                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 997                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 998                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 999                                                  &jx0,&jy0,&jz0);
1000
1001             /* Calculate displacement vector */
1002             dx00             = _mm256_sub_ps(ix0,jx0);
1003             dy00             = _mm256_sub_ps(iy0,jy0);
1004             dz00             = _mm256_sub_ps(iz0,jz0);
1005             dx10             = _mm256_sub_ps(ix1,jx0);
1006             dy10             = _mm256_sub_ps(iy1,jy0);
1007             dz10             = _mm256_sub_ps(iz1,jz0);
1008             dx20             = _mm256_sub_ps(ix2,jx0);
1009             dy20             = _mm256_sub_ps(iy2,jy0);
1010             dz20             = _mm256_sub_ps(iz2,jz0);
1011             dx30             = _mm256_sub_ps(ix3,jx0);
1012             dy30             = _mm256_sub_ps(iy3,jy0);
1013             dz30             = _mm256_sub_ps(iz3,jz0);
1014
1015             /* Calculate squared distance and things based on it */
1016             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1017             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1018             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1019             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1020
1021             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1022             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1023             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1024             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1025
1026             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1027             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1028             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1029             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1030
1031             /* Load parameters for j particles */
1032             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1033                                                                  charge+jnrC+0,charge+jnrD+0,
1034                                                                  charge+jnrE+0,charge+jnrF+0,
1035                                                                  charge+jnrG+0,charge+jnrH+0);
1036             vdwjidx0A        = 2*vdwtype[jnrA+0];
1037             vdwjidx0B        = 2*vdwtype[jnrB+0];
1038             vdwjidx0C        = 2*vdwtype[jnrC+0];
1039             vdwjidx0D        = 2*vdwtype[jnrD+0];
1040             vdwjidx0E        = 2*vdwtype[jnrE+0];
1041             vdwjidx0F        = 2*vdwtype[jnrF+0];
1042             vdwjidx0G        = 2*vdwtype[jnrG+0];
1043             vdwjidx0H        = 2*vdwtype[jnrH+0];
1044
1045             fjx0             = _mm256_setzero_ps();
1046             fjy0             = _mm256_setzero_ps();
1047             fjz0             = _mm256_setzero_ps();
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1054             {
1055
1056             r00              = _mm256_mul_ps(rsq00,rinv00);
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1060                                             vdwioffsetptr0+vdwjidx0B,
1061                                             vdwioffsetptr0+vdwjidx0C,
1062                                             vdwioffsetptr0+vdwjidx0D,
1063                                             vdwioffsetptr0+vdwjidx0E,
1064                                             vdwioffsetptr0+vdwjidx0F,
1065                                             vdwioffsetptr0+vdwjidx0G,
1066                                             vdwioffsetptr0+vdwjidx0H,
1067                                             &c6_00,&c12_00);
1068
1069             /* LENNARD-JONES DISPERSION/REPULSION */
1070
1071             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1072             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1073             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1074             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1075             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1076
1077             d                = _mm256_sub_ps(r00,rswitch);
1078             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1079             d2               = _mm256_mul_ps(d,d);
1080             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1081
1082             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1083
1084             /* Evaluate switch function */
1085             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1086             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1087             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1088
1089             fscal            = fvdw;
1090
1091             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1092
1093             /* Calculate temporary vectorial force */
1094             tx               = _mm256_mul_ps(fscal,dx00);
1095             ty               = _mm256_mul_ps(fscal,dy00);
1096             tz               = _mm256_mul_ps(fscal,dz00);
1097
1098             /* Update vectorial force */
1099             fix0             = _mm256_add_ps(fix0,tx);
1100             fiy0             = _mm256_add_ps(fiy0,ty);
1101             fiz0             = _mm256_add_ps(fiz0,tz);
1102
1103             fjx0             = _mm256_add_ps(fjx0,tx);
1104             fjy0             = _mm256_add_ps(fjy0,ty);
1105             fjz0             = _mm256_add_ps(fjz0,tz);
1106
1107             }
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1114             {
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq10             = _mm256_mul_ps(iq1,jq0);
1118
1119             /* REACTION-FIELD ELECTROSTATICS */
1120             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1121
1122             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1123
1124             fscal            = felec;
1125
1126             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1127
1128             /* Calculate temporary vectorial force */
1129             tx               = _mm256_mul_ps(fscal,dx10);
1130             ty               = _mm256_mul_ps(fscal,dy10);
1131             tz               = _mm256_mul_ps(fscal,dz10);
1132
1133             /* Update vectorial force */
1134             fix1             = _mm256_add_ps(fix1,tx);
1135             fiy1             = _mm256_add_ps(fiy1,ty);
1136             fiz1             = _mm256_add_ps(fiz1,tz);
1137
1138             fjx0             = _mm256_add_ps(fjx0,tx);
1139             fjy0             = _mm256_add_ps(fjy0,ty);
1140             fjz0             = _mm256_add_ps(fjz0,tz);
1141
1142             }
1143
1144             /**************************
1145              * CALCULATE INTERACTIONS *
1146              **************************/
1147
1148             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1149             {
1150
1151             /* Compute parameters for interactions between i and j atoms */
1152             qq20             = _mm256_mul_ps(iq2,jq0);
1153
1154             /* REACTION-FIELD ELECTROSTATICS */
1155             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1156
1157             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1162
1163             /* Calculate temporary vectorial force */
1164             tx               = _mm256_mul_ps(fscal,dx20);
1165             ty               = _mm256_mul_ps(fscal,dy20);
1166             tz               = _mm256_mul_ps(fscal,dz20);
1167
1168             /* Update vectorial force */
1169             fix2             = _mm256_add_ps(fix2,tx);
1170             fiy2             = _mm256_add_ps(fiy2,ty);
1171             fiz2             = _mm256_add_ps(fiz2,tz);
1172
1173             fjx0             = _mm256_add_ps(fjx0,tx);
1174             fjy0             = _mm256_add_ps(fjy0,ty);
1175             fjz0             = _mm256_add_ps(fjz0,tz);
1176
1177             }
1178
1179             /**************************
1180              * CALCULATE INTERACTIONS *
1181              **************************/
1182
1183             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1184             {
1185
1186             /* Compute parameters for interactions between i and j atoms */
1187             qq30             = _mm256_mul_ps(iq3,jq0);
1188
1189             /* REACTION-FIELD ELECTROSTATICS */
1190             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1191
1192             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1193
1194             fscal            = felec;
1195
1196             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1197
1198             /* Calculate temporary vectorial force */
1199             tx               = _mm256_mul_ps(fscal,dx30);
1200             ty               = _mm256_mul_ps(fscal,dy30);
1201             tz               = _mm256_mul_ps(fscal,dz30);
1202
1203             /* Update vectorial force */
1204             fix3             = _mm256_add_ps(fix3,tx);
1205             fiy3             = _mm256_add_ps(fiy3,ty);
1206             fiz3             = _mm256_add_ps(fiz3,tz);
1207
1208             fjx0             = _mm256_add_ps(fjx0,tx);
1209             fjy0             = _mm256_add_ps(fjy0,ty);
1210             fjz0             = _mm256_add_ps(fjz0,tz);
1211
1212             }
1213
1214             fjptrA             = f+j_coord_offsetA;
1215             fjptrB             = f+j_coord_offsetB;
1216             fjptrC             = f+j_coord_offsetC;
1217             fjptrD             = f+j_coord_offsetD;
1218             fjptrE             = f+j_coord_offsetE;
1219             fjptrF             = f+j_coord_offsetF;
1220             fjptrG             = f+j_coord_offsetG;
1221             fjptrH             = f+j_coord_offsetH;
1222
1223             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1224
1225             /* Inner loop uses 149 flops */
1226         }
1227
1228         if(jidx<j_index_end)
1229         {
1230
1231             /* Get j neighbor index, and coordinate index */
1232             jnrlistA         = jjnr[jidx];
1233             jnrlistB         = jjnr[jidx+1];
1234             jnrlistC         = jjnr[jidx+2];
1235             jnrlistD         = jjnr[jidx+3];
1236             jnrlistE         = jjnr[jidx+4];
1237             jnrlistF         = jjnr[jidx+5];
1238             jnrlistG         = jjnr[jidx+6];
1239             jnrlistH         = jjnr[jidx+7];
1240             /* Sign of each element will be negative for non-real atoms.
1241              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1242              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1243              */
1244             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1245                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1246
1247             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1248             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1249             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1250             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1251             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1252             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1253             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1254             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1255             j_coord_offsetA  = DIM*jnrA;
1256             j_coord_offsetB  = DIM*jnrB;
1257             j_coord_offsetC  = DIM*jnrC;
1258             j_coord_offsetD  = DIM*jnrD;
1259             j_coord_offsetE  = DIM*jnrE;
1260             j_coord_offsetF  = DIM*jnrF;
1261             j_coord_offsetG  = DIM*jnrG;
1262             j_coord_offsetH  = DIM*jnrH;
1263
1264             /* load j atom coordinates */
1265             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1266                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1267                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1268                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1269                                                  &jx0,&jy0,&jz0);
1270
1271             /* Calculate displacement vector */
1272             dx00             = _mm256_sub_ps(ix0,jx0);
1273             dy00             = _mm256_sub_ps(iy0,jy0);
1274             dz00             = _mm256_sub_ps(iz0,jz0);
1275             dx10             = _mm256_sub_ps(ix1,jx0);
1276             dy10             = _mm256_sub_ps(iy1,jy0);
1277             dz10             = _mm256_sub_ps(iz1,jz0);
1278             dx20             = _mm256_sub_ps(ix2,jx0);
1279             dy20             = _mm256_sub_ps(iy2,jy0);
1280             dz20             = _mm256_sub_ps(iz2,jz0);
1281             dx30             = _mm256_sub_ps(ix3,jx0);
1282             dy30             = _mm256_sub_ps(iy3,jy0);
1283             dz30             = _mm256_sub_ps(iz3,jz0);
1284
1285             /* Calculate squared distance and things based on it */
1286             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1287             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1288             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1289             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1290
1291             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1292             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1293             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1294             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1295
1296             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1297             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1298             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1299             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1300
1301             /* Load parameters for j particles */
1302             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1303                                                                  charge+jnrC+0,charge+jnrD+0,
1304                                                                  charge+jnrE+0,charge+jnrF+0,
1305                                                                  charge+jnrG+0,charge+jnrH+0);
1306             vdwjidx0A        = 2*vdwtype[jnrA+0];
1307             vdwjidx0B        = 2*vdwtype[jnrB+0];
1308             vdwjidx0C        = 2*vdwtype[jnrC+0];
1309             vdwjidx0D        = 2*vdwtype[jnrD+0];
1310             vdwjidx0E        = 2*vdwtype[jnrE+0];
1311             vdwjidx0F        = 2*vdwtype[jnrF+0];
1312             vdwjidx0G        = 2*vdwtype[jnrG+0];
1313             vdwjidx0H        = 2*vdwtype[jnrH+0];
1314
1315             fjx0             = _mm256_setzero_ps();
1316             fjy0             = _mm256_setzero_ps();
1317             fjz0             = _mm256_setzero_ps();
1318
1319             /**************************
1320              * CALCULATE INTERACTIONS *
1321              **************************/
1322
1323             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1324             {
1325
1326             r00              = _mm256_mul_ps(rsq00,rinv00);
1327             r00              = _mm256_andnot_ps(dummy_mask,r00);
1328
1329             /* Compute parameters for interactions between i and j atoms */
1330             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1331                                             vdwioffsetptr0+vdwjidx0B,
1332                                             vdwioffsetptr0+vdwjidx0C,
1333                                             vdwioffsetptr0+vdwjidx0D,
1334                                             vdwioffsetptr0+vdwjidx0E,
1335                                             vdwioffsetptr0+vdwjidx0F,
1336                                             vdwioffsetptr0+vdwjidx0G,
1337                                             vdwioffsetptr0+vdwjidx0H,
1338                                             &c6_00,&c12_00);
1339
1340             /* LENNARD-JONES DISPERSION/REPULSION */
1341
1342             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1343             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1344             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1345             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1346             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1347
1348             d                = _mm256_sub_ps(r00,rswitch);
1349             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1350             d2               = _mm256_mul_ps(d,d);
1351             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1352
1353             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1354
1355             /* Evaluate switch function */
1356             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1357             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1358             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1359
1360             fscal            = fvdw;
1361
1362             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1363
1364             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1365
1366             /* Calculate temporary vectorial force */
1367             tx               = _mm256_mul_ps(fscal,dx00);
1368             ty               = _mm256_mul_ps(fscal,dy00);
1369             tz               = _mm256_mul_ps(fscal,dz00);
1370
1371             /* Update vectorial force */
1372             fix0             = _mm256_add_ps(fix0,tx);
1373             fiy0             = _mm256_add_ps(fiy0,ty);
1374             fiz0             = _mm256_add_ps(fiz0,tz);
1375
1376             fjx0             = _mm256_add_ps(fjx0,tx);
1377             fjy0             = _mm256_add_ps(fjy0,ty);
1378             fjz0             = _mm256_add_ps(fjz0,tz);
1379
1380             }
1381
1382             /**************************
1383              * CALCULATE INTERACTIONS *
1384              **************************/
1385
1386             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1387             {
1388
1389             /* Compute parameters for interactions between i and j atoms */
1390             qq10             = _mm256_mul_ps(iq1,jq0);
1391
1392             /* REACTION-FIELD ELECTROSTATICS */
1393             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1394
1395             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1396
1397             fscal            = felec;
1398
1399             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1400
1401             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1402
1403             /* Calculate temporary vectorial force */
1404             tx               = _mm256_mul_ps(fscal,dx10);
1405             ty               = _mm256_mul_ps(fscal,dy10);
1406             tz               = _mm256_mul_ps(fscal,dz10);
1407
1408             /* Update vectorial force */
1409             fix1             = _mm256_add_ps(fix1,tx);
1410             fiy1             = _mm256_add_ps(fiy1,ty);
1411             fiz1             = _mm256_add_ps(fiz1,tz);
1412
1413             fjx0             = _mm256_add_ps(fjx0,tx);
1414             fjy0             = _mm256_add_ps(fjy0,ty);
1415             fjz0             = _mm256_add_ps(fjz0,tz);
1416
1417             }
1418
1419             /**************************
1420              * CALCULATE INTERACTIONS *
1421              **************************/
1422
1423             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1424             {
1425
1426             /* Compute parameters for interactions between i and j atoms */
1427             qq20             = _mm256_mul_ps(iq2,jq0);
1428
1429             /* REACTION-FIELD ELECTROSTATICS */
1430             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1431
1432             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1433
1434             fscal            = felec;
1435
1436             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1437
1438             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1439
1440             /* Calculate temporary vectorial force */
1441             tx               = _mm256_mul_ps(fscal,dx20);
1442             ty               = _mm256_mul_ps(fscal,dy20);
1443             tz               = _mm256_mul_ps(fscal,dz20);
1444
1445             /* Update vectorial force */
1446             fix2             = _mm256_add_ps(fix2,tx);
1447             fiy2             = _mm256_add_ps(fiy2,ty);
1448             fiz2             = _mm256_add_ps(fiz2,tz);
1449
1450             fjx0             = _mm256_add_ps(fjx0,tx);
1451             fjy0             = _mm256_add_ps(fjy0,ty);
1452             fjz0             = _mm256_add_ps(fjz0,tz);
1453
1454             }
1455
1456             /**************************
1457              * CALCULATE INTERACTIONS *
1458              **************************/
1459
1460             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1461             {
1462
1463             /* Compute parameters for interactions between i and j atoms */
1464             qq30             = _mm256_mul_ps(iq3,jq0);
1465
1466             /* REACTION-FIELD ELECTROSTATICS */
1467             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1468
1469             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1470
1471             fscal            = felec;
1472
1473             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1474
1475             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1476
1477             /* Calculate temporary vectorial force */
1478             tx               = _mm256_mul_ps(fscal,dx30);
1479             ty               = _mm256_mul_ps(fscal,dy30);
1480             tz               = _mm256_mul_ps(fscal,dz30);
1481
1482             /* Update vectorial force */
1483             fix3             = _mm256_add_ps(fix3,tx);
1484             fiy3             = _mm256_add_ps(fiy3,ty);
1485             fiz3             = _mm256_add_ps(fiz3,tz);
1486
1487             fjx0             = _mm256_add_ps(fjx0,tx);
1488             fjy0             = _mm256_add_ps(fjy0,ty);
1489             fjz0             = _mm256_add_ps(fjz0,tz);
1490
1491             }
1492
1493             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1494             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1495             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1496             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1497             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1498             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1499             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1500             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1501
1502             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1503
1504             /* Inner loop uses 150 flops */
1505         }
1506
1507         /* End of innermost loop */
1508
1509         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1510                                                  f+i_coord_offset,fshift+i_shift_offset);
1511
1512         /* Increment number of inner iterations */
1513         inneriter                  += j_index_end - j_index_start;
1514
1515         /* Outer loop uses 24 flops */
1516     }
1517
1518     /* Increment number of outer iterations */
1519     outeriter        += nri;
1520
1521     /* Update outer/inner flops */
1522
1523     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1524 }