src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  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              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256           dummy_mask,cutoff_mask;
 108     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 109     __m256           one     = _mm256_set1_ps(1.0);
 110     __m256           two     = _mm256_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_ps(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 131     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 132     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141     j_coord_offsetE = 0;
 142     j_coord_offsetF = 0;
 143     j_coord_offsetG = 0;
 144     j_coord_offsetH = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix0             = _mm256_setzero_ps();
 173         fiy0             = _mm256_setzero_ps();
 174         fiz0             = _mm256_setzero_ps();
 175         fix1             = _mm256_setzero_ps();
 176         fiy1             = _mm256_setzero_ps();
 177         fiz1             = _mm256_setzero_ps();
 178         fix2             = _mm256_setzero_ps();
 179         fiy2             = _mm256_setzero_ps();
 180         fiz2             = _mm256_setzero_ps();
 181         fix3             = _mm256_setzero_ps();
 182         fiy3             = _mm256_setzero_ps();
 183         fiz3             = _mm256_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm256_setzero_ps();
 187         vvdwsum          = _mm256_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             jnrE             = jjnr[jidx+4];
 199             jnrF             = jjnr[jidx+5];
 200             jnrG             = jjnr[jidx+6];
 201             jnrH             = jjnr[jidx+7];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206             j_coord_offsetE  = DIM*jnrE;
 207             j_coord_offsetF  = DIM*jnrF;
 208             j_coord_offsetG  = DIM*jnrG;
 209             j_coord_offsetH  = DIM*jnrH;
 210
 211             /* load j atom coordinates */
 212             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 214                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 215                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 216                                                  &jx0,&jy0,&jz0);
 217
 218             /* Calculate displacement vector */
 219             dx00             = _mm256_sub_ps(ix0,jx0);
 220             dy00             = _mm256_sub_ps(iy0,jy0);
 221             dz00             = _mm256_sub_ps(iz0,jz0);
 222             dx10             = _mm256_sub_ps(ix1,jx0);
 223             dy10             = _mm256_sub_ps(iy1,jy0);
 224             dz10             = _mm256_sub_ps(iz1,jz0);
 225             dx20             = _mm256_sub_ps(ix2,jx0);
 226             dy20             = _mm256_sub_ps(iy2,jy0);
 227             dz20             = _mm256_sub_ps(iz2,jz0);
 228             dx30             = _mm256_sub_ps(ix3,jx0);
 229             dy30             = _mm256_sub_ps(iy3,jy0);
 230             dz30             = _mm256_sub_ps(iz3,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 237
 238             rinv10           = avx256_invsqrt_f(rsq10);
 239             rinv20           = avx256_invsqrt_f(rsq20);
 240             rinv30           = avx256_invsqrt_f(rsq30);
 241
 242             rinvsq00         = avx256_inv_f(rsq00);
 243             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 244             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 245             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 249                                                                  charge+jnrC+0,charge+jnrD+0,
 250                                                                  charge+jnrE+0,charge+jnrF+0,
 251                                                                  charge+jnrG+0,charge+jnrH+0);
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253             vdwjidx0B        = 2*vdwtype[jnrB+0];
 254             vdwjidx0C        = 2*vdwtype[jnrC+0];
 255             vdwjidx0D        = 2*vdwtype[jnrD+0];
 256             vdwjidx0E        = 2*vdwtype[jnrE+0];
 257             vdwjidx0F        = 2*vdwtype[jnrF+0];
 258             vdwjidx0G        = 2*vdwtype[jnrG+0];
 259             vdwjidx0H        = 2*vdwtype[jnrH+0];
 260
 261             fjx0             = _mm256_setzero_ps();
 262             fjy0             = _mm256_setzero_ps();
 263             fjz0             = _mm256_setzero_ps();
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 271                                             vdwioffsetptr0+vdwjidx0B,
 272                                             vdwioffsetptr0+vdwjidx0C,
 273                                             vdwioffsetptr0+vdwjidx0D,
 274                                             vdwioffsetptr0+vdwjidx0E,
 275                                             vdwioffsetptr0+vdwjidx0F,
 276                                             vdwioffsetptr0+vdwjidx0G,
 277                                             vdwioffsetptr0+vdwjidx0H,
 278                                             &c6_00,&c12_00);
 279
 280             /* LENNARD-JONES DISPERSION/REPULSION */
 281
 282             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 283             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 284             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 285             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 286             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 290
 291             fscal            = fvdw;
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm256_mul_ps(fscal,dx00);
 295             ty               = _mm256_mul_ps(fscal,dy00);
 296             tz               = _mm256_mul_ps(fscal,dz00);
 297
 298             /* Update vectorial force */
 299             fix0             = _mm256_add_ps(fix0,tx);
 300             fiy0             = _mm256_add_ps(fiy0,ty);
 301             fiz0             = _mm256_add_ps(fiz0,tz);
 302
 303             fjx0             = _mm256_add_ps(fjx0,tx);
 304             fjy0             = _mm256_add_ps(fjy0,ty);
 305             fjz0             = _mm256_add_ps(fjz0,tz);
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq10             = _mm256_mul_ps(iq1,jq0);
 313
 314             /* COULOMB ELECTROSTATICS */
 315             velec            = _mm256_mul_ps(qq10,rinv10);
 316             felec            = _mm256_mul_ps(velec,rinvsq10);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velecsum         = _mm256_add_ps(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_ps(fscal,dx10);
 325             ty               = _mm256_mul_ps(fscal,dy10);
 326             tz               = _mm256_mul_ps(fscal,dz10);
 327
 328             /* Update vectorial force */
 329             fix1             = _mm256_add_ps(fix1,tx);
 330             fiy1             = _mm256_add_ps(fiy1,ty);
 331             fiz1             = _mm256_add_ps(fiz1,tz);
 332
 333             fjx0             = _mm256_add_ps(fjx0,tx);
 334             fjy0             = _mm256_add_ps(fjy0,ty);
 335             fjz0             = _mm256_add_ps(fjz0,tz);
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             /* Compute parameters for interactions between i and j atoms */
 342             qq20             = _mm256_mul_ps(iq2,jq0);
 343
 344             /* COULOMB ELECTROSTATICS */
 345             velec            = _mm256_mul_ps(qq20,rinv20);
 346             felec            = _mm256_mul_ps(velec,rinvsq20);
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm256_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_ps(fscal,dx20);
 355             ty               = _mm256_mul_ps(fscal,dy20);
 356             tz               = _mm256_mul_ps(fscal,dz20);
 357
 358             /* Update vectorial force */
 359             fix2             = _mm256_add_ps(fix2,tx);
 360             fiy2             = _mm256_add_ps(fiy2,ty);
 361             fiz2             = _mm256_add_ps(fiz2,tz);
 362
 363             fjx0             = _mm256_add_ps(fjx0,tx);
 364             fjy0             = _mm256_add_ps(fjy0,ty);
 365             fjz0             = _mm256_add_ps(fjz0,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq30             = _mm256_mul_ps(iq3,jq0);
 373
 374             /* COULOMB ELECTROSTATICS */
 375             velec            = _mm256_mul_ps(qq30,rinv30);
 376             felec            = _mm256_mul_ps(velec,rinvsq30);
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velecsum         = _mm256_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             /* Calculate temporary vectorial force */
 384             tx               = _mm256_mul_ps(fscal,dx30);
 385             ty               = _mm256_mul_ps(fscal,dy30);
 386             tz               = _mm256_mul_ps(fscal,dz30);
 387
 388             /* Update vectorial force */
 389             fix3             = _mm256_add_ps(fix3,tx);
 390             fiy3             = _mm256_add_ps(fiy3,ty);
 391             fiz3             = _mm256_add_ps(fiz3,tz);
 392
 393             fjx0             = _mm256_add_ps(fjx0,tx);
 394             fjy0             = _mm256_add_ps(fjy0,ty);
 395             fjz0             = _mm256_add_ps(fjz0,tz);
 396
 397             fjptrA             = f+j_coord_offsetA;
 398             fjptrB             = f+j_coord_offsetB;
 399             fjptrC             = f+j_coord_offsetC;
 400             fjptrD             = f+j_coord_offsetD;
 401             fjptrE             = f+j_coord_offsetE;
 402             fjptrF             = f+j_coord_offsetF;
 403             fjptrG             = f+j_coord_offsetG;
 404             fjptrH             = f+j_coord_offsetH;
 405
 406             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 407
 408             /* Inner loop uses 116 flops */
 409         }
 410
 411         if(jidx<j_index_end)
 412         {
 413
 414             /* Get j neighbor index, and coordinate index */
 415             jnrlistA         = jjnr[jidx];
 416             jnrlistB         = jjnr[jidx+1];
 417             jnrlistC         = jjnr[jidx+2];
 418             jnrlistD         = jjnr[jidx+3];
 419             jnrlistE         = jjnr[jidx+4];
 420             jnrlistF         = jjnr[jidx+5];
 421             jnrlistG         = jjnr[jidx+6];
 422             jnrlistH         = jjnr[jidx+7];
 423             /* Sign of each element will be negative for non-real atoms.
 424              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 425              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 426              */
 427             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 428                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 429
 430             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 431             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 432             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 433             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 434             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 435             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 436             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 437             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 438             j_coord_offsetA  = DIM*jnrA;
 439             j_coord_offsetB  = DIM*jnrB;
 440             j_coord_offsetC  = DIM*jnrC;
 441             j_coord_offsetD  = DIM*jnrD;
 442             j_coord_offsetE  = DIM*jnrE;
 443             j_coord_offsetF  = DIM*jnrF;
 444             j_coord_offsetG  = DIM*jnrG;
 445             j_coord_offsetH  = DIM*jnrH;
 446
 447             /* load j atom coordinates */
 448             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 449                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 450                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 451                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 452                                                  &jx0,&jy0,&jz0);
 453
 454             /* Calculate displacement vector */
 455             dx00             = _mm256_sub_ps(ix0,jx0);
 456             dy00             = _mm256_sub_ps(iy0,jy0);
 457             dz00             = _mm256_sub_ps(iz0,jz0);
 458             dx10             = _mm256_sub_ps(ix1,jx0);
 459             dy10             = _mm256_sub_ps(iy1,jy0);
 460             dz10             = _mm256_sub_ps(iz1,jz0);
 461             dx20             = _mm256_sub_ps(ix2,jx0);
 462             dy20             = _mm256_sub_ps(iy2,jy0);
 463             dz20             = _mm256_sub_ps(iz2,jz0);
 464             dx30             = _mm256_sub_ps(ix3,jx0);
 465             dy30             = _mm256_sub_ps(iy3,jy0);
 466             dz30             = _mm256_sub_ps(iz3,jz0);
 467
 468             /* Calculate squared distance and things based on it */
 469             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 470             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 471             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 472             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 473
 474             rinv10           = avx256_invsqrt_f(rsq10);
 475             rinv20           = avx256_invsqrt_f(rsq20);
 476             rinv30           = avx256_invsqrt_f(rsq30);
 477
 478             rinvsq00         = avx256_inv_f(rsq00);
 479             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 480             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 481             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 482
 483             /* Load parameters for j particles */
 484             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 485                                                                  charge+jnrC+0,charge+jnrD+0,
 486                                                                  charge+jnrE+0,charge+jnrF+0,
 487                                                                  charge+jnrG+0,charge+jnrH+0);
 488             vdwjidx0A        = 2*vdwtype[jnrA+0];
 489             vdwjidx0B        = 2*vdwtype[jnrB+0];
 490             vdwjidx0C        = 2*vdwtype[jnrC+0];
 491             vdwjidx0D        = 2*vdwtype[jnrD+0];
 492             vdwjidx0E        = 2*vdwtype[jnrE+0];
 493             vdwjidx0F        = 2*vdwtype[jnrF+0];
 494             vdwjidx0G        = 2*vdwtype[jnrG+0];
 495             vdwjidx0H        = 2*vdwtype[jnrH+0];
 496
 497             fjx0             = _mm256_setzero_ps();
 498             fjy0             = _mm256_setzero_ps();
 499             fjz0             = _mm256_setzero_ps();
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 507                                             vdwioffsetptr0+vdwjidx0B,
 508                                             vdwioffsetptr0+vdwjidx0C,
 509                                             vdwioffsetptr0+vdwjidx0D,
 510                                             vdwioffsetptr0+vdwjidx0E,
 511                                             vdwioffsetptr0+vdwjidx0F,
 512                                             vdwioffsetptr0+vdwjidx0G,
 513                                             vdwioffsetptr0+vdwjidx0H,
 514                                             &c6_00,&c12_00);
 515
 516             /* LENNARD-JONES DISPERSION/REPULSION */
 517
 518             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 519             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 520             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 521             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 522             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 526             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 527
 528             fscal            = fvdw;
 529
 530             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm256_mul_ps(fscal,dx00);
 534             ty               = _mm256_mul_ps(fscal,dy00);
 535             tz               = _mm256_mul_ps(fscal,dz00);
 536
 537             /* Update vectorial force */
 538             fix0             = _mm256_add_ps(fix0,tx);
 539             fiy0             = _mm256_add_ps(fiy0,ty);
 540             fiz0             = _mm256_add_ps(fiz0,tz);
 541
 542             fjx0             = _mm256_add_ps(fjx0,tx);
 543             fjy0             = _mm256_add_ps(fjy0,ty);
 544             fjz0             = _mm256_add_ps(fjz0,tz);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq10             = _mm256_mul_ps(iq1,jq0);
 552
 553             /* COULOMB ELECTROSTATICS */
 554             velec            = _mm256_mul_ps(qq10,rinv10);
 555             felec            = _mm256_mul_ps(velec,rinvsq10);
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm256_andnot_ps(dummy_mask,velec);
 559             velecsum         = _mm256_add_ps(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm256_mul_ps(fscal,dx10);
 567             ty               = _mm256_mul_ps(fscal,dy10);
 568             tz               = _mm256_mul_ps(fscal,dz10);
 569
 570             /* Update vectorial force */
 571             fix1             = _mm256_add_ps(fix1,tx);
 572             fiy1             = _mm256_add_ps(fiy1,ty);
 573             fiz1             = _mm256_add_ps(fiz1,tz);
 574
 575             fjx0             = _mm256_add_ps(fjx0,tx);
 576             fjy0             = _mm256_add_ps(fjy0,ty);
 577             fjz0             = _mm256_add_ps(fjz0,tz);
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq20             = _mm256_mul_ps(iq2,jq0);
 585
 586             /* COULOMB ELECTROSTATICS */
 587             velec            = _mm256_mul_ps(qq20,rinv20);
 588             felec            = _mm256_mul_ps(velec,rinvsq20);
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm256_andnot_ps(dummy_mask,velec);
 592             velecsum         = _mm256_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_ps(fscal,dx20);
 600             ty               = _mm256_mul_ps(fscal,dy20);
 601             tz               = _mm256_mul_ps(fscal,dz20);
 602
 603             /* Update vectorial force */
 604             fix2             = _mm256_add_ps(fix2,tx);
 605             fiy2             = _mm256_add_ps(fiy2,ty);
 606             fiz2             = _mm256_add_ps(fiz2,tz);
 607
 608             fjx0             = _mm256_add_ps(fjx0,tx);
 609             fjy0             = _mm256_add_ps(fjy0,ty);
 610             fjz0             = _mm256_add_ps(fjz0,tz);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq30             = _mm256_mul_ps(iq3,jq0);
 618
 619             /* COULOMB ELECTROSTATICS */
 620             velec            = _mm256_mul_ps(qq30,rinv30);
 621             felec            = _mm256_mul_ps(velec,rinvsq30);
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             velec            = _mm256_andnot_ps(dummy_mask,velec);
 625             velecsum         = _mm256_add_ps(velecsum,velec);
 626
 627             fscal            = felec;
 628
 629             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm256_mul_ps(fscal,dx30);
 633             ty               = _mm256_mul_ps(fscal,dy30);
 634             tz               = _mm256_mul_ps(fscal,dz30);
 635
 636             /* Update vectorial force */
 637             fix3             = _mm256_add_ps(fix3,tx);
 638             fiy3             = _mm256_add_ps(fiy3,ty);
 639             fiz3             = _mm256_add_ps(fiz3,tz);
 640
 641             fjx0             = _mm256_add_ps(fjx0,tx);
 642             fjy0             = _mm256_add_ps(fjy0,ty);
 643             fjz0             = _mm256_add_ps(fjz0,tz);
 644
 645             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 646             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 647             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 648             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 649             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 650             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 651             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 652             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 653
 654             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 655
 656             /* Inner loop uses 116 flops */
 657         }
 658
 659         /* End of innermost loop */
 660
 661         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 662                                                  f+i_coord_offset,fshift+i_shift_offset);
 663
 664         ggid                        = gid[iidx];
 665         /* Update potential energies */
 666         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 667         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 668
 669         /* Increment number of inner iterations */
 670         inneriter                  += j_index_end - j_index_start;
 671
 672         /* Outer loop uses 26 flops */
 673     }
 674
 675     /* Increment number of outer iterations */
 676     outeriter        += nri;
 677
 678     /* Update outer/inner flops */
 679
 680     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 681 }
 682 /*
 683  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_single
 684  * Electrostatics interaction: Coulomb
 685  * VdW interaction:            LennardJones
 686  * Geometry:                   Water4-Particle
 687  * Calculate force/pot:        Force
 688  */
 689 void
 690 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_single
 691                     (t_nblist                    * gmx_restrict       nlist,
 692                      rvec                        * gmx_restrict          xx,
 693                      rvec                        * gmx_restrict          ff,
 694                      struct t_forcerec           * gmx_restrict          fr,
 695                      t_mdatoms                   * gmx_restrict     mdatoms,
 696                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 697                      t_nrnb                      * gmx_restrict        nrnb)
 698 {
 699     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 700      * just 0 for non-waters.
 701      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 702      * jnr indices corresponding to data put in the four positions in the SIMD register.
 703      */
 704     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 705     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 706     int              jnrA,jnrB,jnrC,jnrD;
 707     int              jnrE,jnrF,jnrG,jnrH;
 708     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 709     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 710     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 711     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 712     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 713     real             rcutoff_scalar;
 714     real             *shiftvec,*fshift,*x,*f;
 715     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 716     real             scratch[4*DIM];
 717     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 718     real *           vdwioffsetptr0;
 719     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 720     real *           vdwioffsetptr1;
 721     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 722     real *           vdwioffsetptr2;
 723     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 724     real *           vdwioffsetptr3;
 725     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 726     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 727     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 728     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 729     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 730     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 731     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 732     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 733     real             *charge;
 734     int              nvdwtype;
 735     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 736     int              *vdwtype;
 737     real             *vdwparam;
 738     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 739     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 740     __m256           dummy_mask,cutoff_mask;
 741     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 742     __m256           one     = _mm256_set1_ps(1.0);
 743     __m256           two     = _mm256_set1_ps(2.0);
 744     x                = xx[0];
 745     f                = ff[0];
 746
 747     nri              = nlist->nri;
 748     iinr             = nlist->iinr;
 749     jindex           = nlist->jindex;
 750     jjnr             = nlist->jjnr;
 751     shiftidx         = nlist->shift;
 752     gid              = nlist->gid;
 753     shiftvec         = fr->shift_vec[0];
 754     fshift           = fr->fshift[0];
 755     facel            = _mm256_set1_ps(fr->ic->epsfac);
 756     charge           = mdatoms->chargeA;
 757     nvdwtype         = fr->ntype;
 758     vdwparam         = fr->nbfp;
 759     vdwtype          = mdatoms->typeA;
 760
 761     /* Setup water-specific parameters */
 762     inr              = nlist->iinr[0];
 763     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 764     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 765     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 766     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 767
 768     /* Avoid stupid compiler warnings */
 769     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 770     j_coord_offsetA = 0;
 771     j_coord_offsetB = 0;
 772     j_coord_offsetC = 0;
 773     j_coord_offsetD = 0;
 774     j_coord_offsetE = 0;
 775     j_coord_offsetF = 0;
 776     j_coord_offsetG = 0;
 777     j_coord_offsetH = 0;
 778
 779     outeriter        = 0;
 780     inneriter        = 0;
 781
 782     for(iidx=0;iidx<4*DIM;iidx++)
 783     {
 784         scratch[iidx] = 0.0;
 785     }
 786
 787     /* Start outer loop over neighborlists */
 788     for(iidx=0; iidx<nri; iidx++)
 789     {
 790         /* Load shift vector for this list */
 791         i_shift_offset   = DIM*shiftidx[iidx];
 792
 793         /* Load limits for loop over neighbors */
 794         j_index_start    = jindex[iidx];
 795         j_index_end      = jindex[iidx+1];
 796
 797         /* Get outer coordinate index */
 798         inr              = iinr[iidx];
 799         i_coord_offset   = DIM*inr;
 800
 801         /* Load i particle coords and add shift vector */
 802         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 803                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 804
 805         fix0             = _mm256_setzero_ps();
 806         fiy0             = _mm256_setzero_ps();
 807         fiz0             = _mm256_setzero_ps();
 808         fix1             = _mm256_setzero_ps();
 809         fiy1             = _mm256_setzero_ps();
 810         fiz1             = _mm256_setzero_ps();
 811         fix2             = _mm256_setzero_ps();
 812         fiy2             = _mm256_setzero_ps();
 813         fiz2             = _mm256_setzero_ps();
 814         fix3             = _mm256_setzero_ps();
 815         fiy3             = _mm256_setzero_ps();
 816         fiz3             = _mm256_setzero_ps();
 817
 818         /* Start inner kernel loop */
 819         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 820         {
 821
 822             /* Get j neighbor index, and coordinate index */
 823             jnrA             = jjnr[jidx];
 824             jnrB             = jjnr[jidx+1];
 825             jnrC             = jjnr[jidx+2];
 826             jnrD             = jjnr[jidx+3];
 827             jnrE             = jjnr[jidx+4];
 828             jnrF             = jjnr[jidx+5];
 829             jnrG             = jjnr[jidx+6];
 830             jnrH             = jjnr[jidx+7];
 831             j_coord_offsetA  = DIM*jnrA;
 832             j_coord_offsetB  = DIM*jnrB;
 833             j_coord_offsetC  = DIM*jnrC;
 834             j_coord_offsetD  = DIM*jnrD;
 835             j_coord_offsetE  = DIM*jnrE;
 836             j_coord_offsetF  = DIM*jnrF;
 837             j_coord_offsetG  = DIM*jnrG;
 838             j_coord_offsetH  = DIM*jnrH;
 839
 840             /* load j atom coordinates */
 841             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 842                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 843                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 844                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 845                                                  &jx0,&jy0,&jz0);
 846
 847             /* Calculate displacement vector */
 848             dx00             = _mm256_sub_ps(ix0,jx0);
 849             dy00             = _mm256_sub_ps(iy0,jy0);
 850             dz00             = _mm256_sub_ps(iz0,jz0);
 851             dx10             = _mm256_sub_ps(ix1,jx0);
 852             dy10             = _mm256_sub_ps(iy1,jy0);
 853             dz10             = _mm256_sub_ps(iz1,jz0);
 854             dx20             = _mm256_sub_ps(ix2,jx0);
 855             dy20             = _mm256_sub_ps(iy2,jy0);
 856             dz20             = _mm256_sub_ps(iz2,jz0);
 857             dx30             = _mm256_sub_ps(ix3,jx0);
 858             dy30             = _mm256_sub_ps(iy3,jy0);
 859             dz30             = _mm256_sub_ps(iz3,jz0);
 860
 861             /* Calculate squared distance and things based on it */
 862             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 863             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 864             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 865             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 866
 867             rinv10           = avx256_invsqrt_f(rsq10);
 868             rinv20           = avx256_invsqrt_f(rsq20);
 869             rinv30           = avx256_invsqrt_f(rsq30);
 870
 871             rinvsq00         = avx256_inv_f(rsq00);
 872             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 873             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 874             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 875
 876             /* Load parameters for j particles */
 877             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 878                                                                  charge+jnrC+0,charge+jnrD+0,
 879                                                                  charge+jnrE+0,charge+jnrF+0,
 880                                                                  charge+jnrG+0,charge+jnrH+0);
 881             vdwjidx0A        = 2*vdwtype[jnrA+0];
 882             vdwjidx0B        = 2*vdwtype[jnrB+0];
 883             vdwjidx0C        = 2*vdwtype[jnrC+0];
 884             vdwjidx0D        = 2*vdwtype[jnrD+0];
 885             vdwjidx0E        = 2*vdwtype[jnrE+0];
 886             vdwjidx0F        = 2*vdwtype[jnrF+0];
 887             vdwjidx0G        = 2*vdwtype[jnrG+0];
 888             vdwjidx0H        = 2*vdwtype[jnrH+0];
 889
 890             fjx0             = _mm256_setzero_ps();
 891             fjy0             = _mm256_setzero_ps();
 892             fjz0             = _mm256_setzero_ps();
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 900                                             vdwioffsetptr0+vdwjidx0B,
 901                                             vdwioffsetptr0+vdwjidx0C,
 902                                             vdwioffsetptr0+vdwjidx0D,
 903                                             vdwioffsetptr0+vdwjidx0E,
 904                                             vdwioffsetptr0+vdwjidx0F,
 905                                             vdwioffsetptr0+vdwjidx0G,
 906                                             vdwioffsetptr0+vdwjidx0H,
 907                                             &c6_00,&c12_00);
 908
 909             /* LENNARD-JONES DISPERSION/REPULSION */
 910
 911             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 912             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 913
 914             fscal            = fvdw;
 915
 916             /* Calculate temporary vectorial force */
 917             tx               = _mm256_mul_ps(fscal,dx00);
 918             ty               = _mm256_mul_ps(fscal,dy00);
 919             tz               = _mm256_mul_ps(fscal,dz00);
 920
 921             /* Update vectorial force */
 922             fix0             = _mm256_add_ps(fix0,tx);
 923             fiy0             = _mm256_add_ps(fiy0,ty);
 924             fiz0             = _mm256_add_ps(fiz0,tz);
 925
 926             fjx0             = _mm256_add_ps(fjx0,tx);
 927             fjy0             = _mm256_add_ps(fjy0,ty);
 928             fjz0             = _mm256_add_ps(fjz0,tz);
 929
 930             /**************************
 931              * CALCULATE INTERACTIONS *
 932              **************************/
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             qq10             = _mm256_mul_ps(iq1,jq0);
 936
 937             /* COULOMB ELECTROSTATICS */
 938             velec            = _mm256_mul_ps(qq10,rinv10);
 939             felec            = _mm256_mul_ps(velec,rinvsq10);
 940
 941             fscal            = felec;
 942
 943             /* Calculate temporary vectorial force */
 944             tx               = _mm256_mul_ps(fscal,dx10);
 945             ty               = _mm256_mul_ps(fscal,dy10);
 946             tz               = _mm256_mul_ps(fscal,dz10);
 947
 948             /* Update vectorial force */
 949             fix1             = _mm256_add_ps(fix1,tx);
 950             fiy1             = _mm256_add_ps(fiy1,ty);
 951             fiz1             = _mm256_add_ps(fiz1,tz);
 952
 953             fjx0             = _mm256_add_ps(fjx0,tx);
 954             fjy0             = _mm256_add_ps(fjy0,ty);
 955             fjz0             = _mm256_add_ps(fjz0,tz);
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             qq20             = _mm256_mul_ps(iq2,jq0);
 963
 964             /* COULOMB ELECTROSTATICS */
 965             velec            = _mm256_mul_ps(qq20,rinv20);
 966             felec            = _mm256_mul_ps(velec,rinvsq20);
 967
 968             fscal            = felec;
 969
 970             /* Calculate temporary vectorial force */
 971             tx               = _mm256_mul_ps(fscal,dx20);
 972             ty               = _mm256_mul_ps(fscal,dy20);
 973             tz               = _mm256_mul_ps(fscal,dz20);
 974
 975             /* Update vectorial force */
 976             fix2             = _mm256_add_ps(fix2,tx);
 977             fiy2             = _mm256_add_ps(fiy2,ty);
 978             fiz2             = _mm256_add_ps(fiz2,tz);
 979
 980             fjx0             = _mm256_add_ps(fjx0,tx);
 981             fjy0             = _mm256_add_ps(fjy0,ty);
 982             fjz0             = _mm256_add_ps(fjz0,tz);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             /* Compute parameters for interactions between i and j atoms */
 989             qq30             = _mm256_mul_ps(iq3,jq0);
 990
 991             /* COULOMB ELECTROSTATICS */
 992             velec            = _mm256_mul_ps(qq30,rinv30);
 993             felec            = _mm256_mul_ps(velec,rinvsq30);
 994
 995             fscal            = felec;
 996
 997             /* Calculate temporary vectorial force */
 998             tx               = _mm256_mul_ps(fscal,dx30);
 999             ty               = _mm256_mul_ps(fscal,dy30);
1000             tz               = _mm256_mul_ps(fscal,dz30);
1001
1002             /* Update vectorial force */
1003             fix3             = _mm256_add_ps(fix3,tx);
1004             fiy3             = _mm256_add_ps(fiy3,ty);
1005             fiz3             = _mm256_add_ps(fiz3,tz);
1006
1007             fjx0             = _mm256_add_ps(fjx0,tx);
1008             fjy0             = _mm256_add_ps(fjy0,ty);
1009             fjz0             = _mm256_add_ps(fjz0,tz);
1010
1011             fjptrA             = f+j_coord_offsetA;
1012             fjptrB             = f+j_coord_offsetB;
1013             fjptrC             = f+j_coord_offsetC;
1014             fjptrD             = f+j_coord_offsetD;
1015             fjptrE             = f+j_coord_offsetE;
1016             fjptrF             = f+j_coord_offsetF;
1017             fjptrG             = f+j_coord_offsetG;
1018             fjptrH             = f+j_coord_offsetH;
1019
1020             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1021
1022             /* Inner loop uses 108 flops */
1023         }
1024
1025         if(jidx<j_index_end)
1026         {
1027
1028             /* Get j neighbor index, and coordinate index */
1029             jnrlistA         = jjnr[jidx];
1030             jnrlistB         = jjnr[jidx+1];
1031             jnrlistC         = jjnr[jidx+2];
1032             jnrlistD         = jjnr[jidx+3];
1033             jnrlistE         = jjnr[jidx+4];
1034             jnrlistF         = jjnr[jidx+5];
1035             jnrlistG         = jjnr[jidx+6];
1036             jnrlistH         = jjnr[jidx+7];
1037             /* Sign of each element will be negative for non-real atoms.
1038              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1039              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1040              */
1041             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1042                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1043
1044             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1045             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1046             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1047             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1048             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1049             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1050             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1051             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1052             j_coord_offsetA  = DIM*jnrA;
1053             j_coord_offsetB  = DIM*jnrB;
1054             j_coord_offsetC  = DIM*jnrC;
1055             j_coord_offsetD  = DIM*jnrD;
1056             j_coord_offsetE  = DIM*jnrE;
1057             j_coord_offsetF  = DIM*jnrF;
1058             j_coord_offsetG  = DIM*jnrG;
1059             j_coord_offsetH  = DIM*jnrH;
1060
1061             /* load j atom coordinates */
1062             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1063                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1064                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1065                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1066                                                  &jx0,&jy0,&jz0);
1067
1068             /* Calculate displacement vector */
1069             dx00             = _mm256_sub_ps(ix0,jx0);
1070             dy00             = _mm256_sub_ps(iy0,jy0);
1071             dz00             = _mm256_sub_ps(iz0,jz0);
1072             dx10             = _mm256_sub_ps(ix1,jx0);
1073             dy10             = _mm256_sub_ps(iy1,jy0);
1074             dz10             = _mm256_sub_ps(iz1,jz0);
1075             dx20             = _mm256_sub_ps(ix2,jx0);
1076             dy20             = _mm256_sub_ps(iy2,jy0);
1077             dz20             = _mm256_sub_ps(iz2,jz0);
1078             dx30             = _mm256_sub_ps(ix3,jx0);
1079             dy30             = _mm256_sub_ps(iy3,jy0);
1080             dz30             = _mm256_sub_ps(iz3,jz0);
1081
1082             /* Calculate squared distance and things based on it */
1083             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1084             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1085             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1086             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1087
1088             rinv10           = avx256_invsqrt_f(rsq10);
1089             rinv20           = avx256_invsqrt_f(rsq20);
1090             rinv30           = avx256_invsqrt_f(rsq30);
1091
1092             rinvsq00         = avx256_inv_f(rsq00);
1093             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1094             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1095             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1096
1097             /* Load parameters for j particles */
1098             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1099                                                                  charge+jnrC+0,charge+jnrD+0,
1100                                                                  charge+jnrE+0,charge+jnrF+0,
1101                                                                  charge+jnrG+0,charge+jnrH+0);
1102             vdwjidx0A        = 2*vdwtype[jnrA+0];
1103             vdwjidx0B        = 2*vdwtype[jnrB+0];
1104             vdwjidx0C        = 2*vdwtype[jnrC+0];
1105             vdwjidx0D        = 2*vdwtype[jnrD+0];
1106             vdwjidx0E        = 2*vdwtype[jnrE+0];
1107             vdwjidx0F        = 2*vdwtype[jnrF+0];
1108             vdwjidx0G        = 2*vdwtype[jnrG+0];
1109             vdwjidx0H        = 2*vdwtype[jnrH+0];
1110
1111             fjx0             = _mm256_setzero_ps();
1112             fjy0             = _mm256_setzero_ps();
1113             fjz0             = _mm256_setzero_ps();
1114
1115             /**************************
1116              * CALCULATE INTERACTIONS *
1117              **************************/
1118
1119             /* Compute parameters for interactions between i and j atoms */
1120             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1121                                             vdwioffsetptr0+vdwjidx0B,
1122                                             vdwioffsetptr0+vdwjidx0C,
1123                                             vdwioffsetptr0+vdwjidx0D,
1124                                             vdwioffsetptr0+vdwjidx0E,
1125                                             vdwioffsetptr0+vdwjidx0F,
1126                                             vdwioffsetptr0+vdwjidx0G,
1127                                             vdwioffsetptr0+vdwjidx0H,
1128                                             &c6_00,&c12_00);
1129
1130             /* LENNARD-JONES DISPERSION/REPULSION */
1131
1132             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1133             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1134
1135             fscal            = fvdw;
1136
1137             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm256_mul_ps(fscal,dx00);
1141             ty               = _mm256_mul_ps(fscal,dy00);
1142             tz               = _mm256_mul_ps(fscal,dz00);
1143
1144             /* Update vectorial force */
1145             fix0             = _mm256_add_ps(fix0,tx);
1146             fiy0             = _mm256_add_ps(fiy0,ty);
1147             fiz0             = _mm256_add_ps(fiz0,tz);
1148
1149             fjx0             = _mm256_add_ps(fjx0,tx);
1150             fjy0             = _mm256_add_ps(fjy0,ty);
1151             fjz0             = _mm256_add_ps(fjz0,tz);
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             /* Compute parameters for interactions between i and j atoms */
1158             qq10             = _mm256_mul_ps(iq1,jq0);
1159
1160             /* COULOMB ELECTROSTATICS */
1161             velec            = _mm256_mul_ps(qq10,rinv10);
1162             felec            = _mm256_mul_ps(velec,rinvsq10);
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1167
1168             /* Calculate temporary vectorial force */
1169             tx               = _mm256_mul_ps(fscal,dx10);
1170             ty               = _mm256_mul_ps(fscal,dy10);
1171             tz               = _mm256_mul_ps(fscal,dz10);
1172
1173             /* Update vectorial force */
1174             fix1             = _mm256_add_ps(fix1,tx);
1175             fiy1             = _mm256_add_ps(fiy1,ty);
1176             fiz1             = _mm256_add_ps(fiz1,tz);
1177
1178             fjx0             = _mm256_add_ps(fjx0,tx);
1179             fjy0             = _mm256_add_ps(fjy0,ty);
1180             fjz0             = _mm256_add_ps(fjz0,tz);
1181
1182             /**************************
1183              * CALCULATE INTERACTIONS *
1184              **************************/
1185
1186             /* Compute parameters for interactions between i and j atoms */
1187             qq20             = _mm256_mul_ps(iq2,jq0);
1188
1189             /* COULOMB ELECTROSTATICS */
1190             velec            = _mm256_mul_ps(qq20,rinv20);
1191             felec            = _mm256_mul_ps(velec,rinvsq20);
1192
1193             fscal            = felec;
1194
1195             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1196
1197             /* Calculate temporary vectorial force */
1198             tx               = _mm256_mul_ps(fscal,dx20);
1199             ty               = _mm256_mul_ps(fscal,dy20);
1200             tz               = _mm256_mul_ps(fscal,dz20);
1201
1202             /* Update vectorial force */
1203             fix2             = _mm256_add_ps(fix2,tx);
1204             fiy2             = _mm256_add_ps(fiy2,ty);
1205             fiz2             = _mm256_add_ps(fiz2,tz);
1206
1207             fjx0             = _mm256_add_ps(fjx0,tx);
1208             fjy0             = _mm256_add_ps(fjy0,ty);
1209             fjz0             = _mm256_add_ps(fjz0,tz);
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             /* Compute parameters for interactions between i and j atoms */
1216             qq30             = _mm256_mul_ps(iq3,jq0);
1217
1218             /* COULOMB ELECTROSTATICS */
1219             velec            = _mm256_mul_ps(qq30,rinv30);
1220             felec            = _mm256_mul_ps(velec,rinvsq30);
1221
1222             fscal            = felec;
1223
1224             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1225
1226             /* Calculate temporary vectorial force */
1227             tx               = _mm256_mul_ps(fscal,dx30);
1228             ty               = _mm256_mul_ps(fscal,dy30);
1229             tz               = _mm256_mul_ps(fscal,dz30);
1230
1231             /* Update vectorial force */
1232             fix3             = _mm256_add_ps(fix3,tx);
1233             fiy3             = _mm256_add_ps(fiy3,ty);
1234             fiz3             = _mm256_add_ps(fiz3,tz);
1235
1236             fjx0             = _mm256_add_ps(fjx0,tx);
1237             fjy0             = _mm256_add_ps(fjy0,ty);
1238             fjz0             = _mm256_add_ps(fjz0,tz);
1239
1240             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1241             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1242             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1243             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1244             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1245             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1246             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1247             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1248
1249             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1250
1251             /* Inner loop uses 108 flops */
1252         }
1253
1254         /* End of innermost loop */
1255
1256         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1257                                                  f+i_coord_offset,fshift+i_shift_offset);
1258
1259         /* Increment number of inner iterations */
1260         inneriter                  += j_index_end - j_index_start;
1261
1262         /* Outer loop uses 24 flops */
1263     }
1264
1265     /* Increment number of outer iterations */
1266     outeriter        += nri;
1267
1268     /* Update outer/inner flops */
1269
1270     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1271 }