src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_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_ElecCoul_VdwLJ_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_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           dummy_mask,cutoff_mask;
 109     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 110     __m256           one     = _mm256_set1_ps(1.0);
 111     __m256           two     = _mm256_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 132     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 133     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 134     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142     j_coord_offsetE = 0;
 143     j_coord_offsetF = 0;
 144     j_coord_offsetG = 0;
 145     j_coord_offsetH = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix0             = _mm256_setzero_ps();
 174         fiy0             = _mm256_setzero_ps();
 175         fiz0             = _mm256_setzero_ps();
 176         fix1             = _mm256_setzero_ps();
 177         fiy1             = _mm256_setzero_ps();
 178         fiz1             = _mm256_setzero_ps();
 179         fix2             = _mm256_setzero_ps();
 180         fiy2             = _mm256_setzero_ps();
 181         fiz2             = _mm256_setzero_ps();
 182         fix3             = _mm256_setzero_ps();
 183         fiy3             = _mm256_setzero_ps();
 184         fiz3             = _mm256_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm256_setzero_ps();
 188         vvdwsum          = _mm256_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             jnrE             = jjnr[jidx+4];
 200             jnrF             = jjnr[jidx+5];
 201             jnrG             = jjnr[jidx+6];
 202             jnrH             = jjnr[jidx+7];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207             j_coord_offsetE  = DIM*jnrE;
 208             j_coord_offsetF  = DIM*jnrF;
 209             j_coord_offsetG  = DIM*jnrG;
 210             j_coord_offsetH  = DIM*jnrH;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 216                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 217                                                  &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm256_sub_ps(ix0,jx0);
 221             dy00             = _mm256_sub_ps(iy0,jy0);
 222             dz00             = _mm256_sub_ps(iz0,jz0);
 223             dx10             = _mm256_sub_ps(ix1,jx0);
 224             dy10             = _mm256_sub_ps(iy1,jy0);
 225             dz10             = _mm256_sub_ps(iz1,jz0);
 226             dx20             = _mm256_sub_ps(ix2,jx0);
 227             dy20             = _mm256_sub_ps(iy2,jy0);
 228             dz20             = _mm256_sub_ps(iz2,jz0);
 229             dx30             = _mm256_sub_ps(ix3,jx0);
 230             dy30             = _mm256_sub_ps(iy3,jy0);
 231             dz30             = _mm256_sub_ps(iz3,jz0);
 232
 233             /* Calculate squared distance and things based on it */
 234             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 235             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 236             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 237             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 238
 239             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 240             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 241             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 242
 243             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 244             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 245             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 246             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 247
 248             /* Load parameters for j particles */
 249             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 250                                                                  charge+jnrC+0,charge+jnrD+0,
 251                                                                  charge+jnrE+0,charge+jnrF+0,
 252                                                                  charge+jnrG+0,charge+jnrH+0);
 253             vdwjidx0A        = 2*vdwtype[jnrA+0];
 254             vdwjidx0B        = 2*vdwtype[jnrB+0];
 255             vdwjidx0C        = 2*vdwtype[jnrC+0];
 256             vdwjidx0D        = 2*vdwtype[jnrD+0];
 257             vdwjidx0E        = 2*vdwtype[jnrE+0];
 258             vdwjidx0F        = 2*vdwtype[jnrF+0];
 259             vdwjidx0G        = 2*vdwtype[jnrG+0];
 260             vdwjidx0H        = 2*vdwtype[jnrH+0];
 261
 262             fjx0             = _mm256_setzero_ps();
 263             fjy0             = _mm256_setzero_ps();
 264             fjz0             = _mm256_setzero_ps();
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 272                                             vdwioffsetptr0+vdwjidx0B,
 273                                             vdwioffsetptr0+vdwjidx0C,
 274                                             vdwioffsetptr0+vdwjidx0D,
 275                                             vdwioffsetptr0+vdwjidx0E,
 276                                             vdwioffsetptr0+vdwjidx0F,
 277                                             vdwioffsetptr0+vdwjidx0G,
 278                                             vdwioffsetptr0+vdwjidx0H,
 279                                             &c6_00,&c12_00);
 280
 281             /* LENNARD-JONES DISPERSION/REPULSION */
 282
 283             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 284             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 285             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 286             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 287             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 291
 292             fscal            = fvdw;
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm256_mul_ps(fscal,dx00);
 296             ty               = _mm256_mul_ps(fscal,dy00);
 297             tz               = _mm256_mul_ps(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm256_add_ps(fix0,tx);
 301             fiy0             = _mm256_add_ps(fiy0,ty);
 302             fiz0             = _mm256_add_ps(fiz0,tz);
 303
 304             fjx0             = _mm256_add_ps(fjx0,tx);
 305             fjy0             = _mm256_add_ps(fjy0,ty);
 306             fjz0             = _mm256_add_ps(fjz0,tz);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm256_mul_ps(iq1,jq0);
 314
 315             /* COULOMB ELECTROSTATICS */
 316             velec            = _mm256_mul_ps(qq10,rinv10);
 317             felec            = _mm256_mul_ps(velec,rinvsq10);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velecsum         = _mm256_add_ps(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm256_mul_ps(fscal,dx10);
 326             ty               = _mm256_mul_ps(fscal,dy10);
 327             tz               = _mm256_mul_ps(fscal,dz10);
 328
 329             /* Update vectorial force */
 330             fix1             = _mm256_add_ps(fix1,tx);
 331             fiy1             = _mm256_add_ps(fiy1,ty);
 332             fiz1             = _mm256_add_ps(fiz1,tz);
 333
 334             fjx0             = _mm256_add_ps(fjx0,tx);
 335             fjy0             = _mm256_add_ps(fjy0,ty);
 336             fjz0             = _mm256_add_ps(fjz0,tz);
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq20             = _mm256_mul_ps(iq2,jq0);
 344
 345             /* COULOMB ELECTROSTATICS */
 346             velec            = _mm256_mul_ps(qq20,rinv20);
 347             felec            = _mm256_mul_ps(velec,rinvsq20);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm256_add_ps(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm256_mul_ps(fscal,dx20);
 356             ty               = _mm256_mul_ps(fscal,dy20);
 357             tz               = _mm256_mul_ps(fscal,dz20);
 358
 359             /* Update vectorial force */
 360             fix2             = _mm256_add_ps(fix2,tx);
 361             fiy2             = _mm256_add_ps(fiy2,ty);
 362             fiz2             = _mm256_add_ps(fiz2,tz);
 363
 364             fjx0             = _mm256_add_ps(fjx0,tx);
 365             fjy0             = _mm256_add_ps(fjy0,ty);
 366             fjz0             = _mm256_add_ps(fjz0,tz);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq30             = _mm256_mul_ps(iq3,jq0);
 374
 375             /* COULOMB ELECTROSTATICS */
 376             velec            = _mm256_mul_ps(qq30,rinv30);
 377             felec            = _mm256_mul_ps(velec,rinvsq30);
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm256_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm256_mul_ps(fscal,dx30);
 386             ty               = _mm256_mul_ps(fscal,dy30);
 387             tz               = _mm256_mul_ps(fscal,dz30);
 388
 389             /* Update vectorial force */
 390             fix3             = _mm256_add_ps(fix3,tx);
 391             fiy3             = _mm256_add_ps(fiy3,ty);
 392             fiz3             = _mm256_add_ps(fiz3,tz);
 393
 394             fjx0             = _mm256_add_ps(fjx0,tx);
 395             fjy0             = _mm256_add_ps(fjy0,ty);
 396             fjz0             = _mm256_add_ps(fjz0,tz);
 397
 398             fjptrA             = f+j_coord_offsetA;
 399             fjptrB             = f+j_coord_offsetB;
 400             fjptrC             = f+j_coord_offsetC;
 401             fjptrD             = f+j_coord_offsetD;
 402             fjptrE             = f+j_coord_offsetE;
 403             fjptrF             = f+j_coord_offsetF;
 404             fjptrG             = f+j_coord_offsetG;
 405             fjptrH             = f+j_coord_offsetH;
 406
 407             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 408
 409             /* Inner loop uses 116 flops */
 410         }
 411
 412         if(jidx<j_index_end)
 413         {
 414
 415             /* Get j neighbor index, and coordinate index */
 416             jnrlistA         = jjnr[jidx];
 417             jnrlistB         = jjnr[jidx+1];
 418             jnrlistC         = jjnr[jidx+2];
 419             jnrlistD         = jjnr[jidx+3];
 420             jnrlistE         = jjnr[jidx+4];
 421             jnrlistF         = jjnr[jidx+5];
 422             jnrlistG         = jjnr[jidx+6];
 423             jnrlistH         = jjnr[jidx+7];
 424             /* Sign of each element will be negative for non-real atoms.
 425              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 426              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 427              */
 428             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 429                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 430
 431             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 432             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 433             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 434             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 435             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 436             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 437             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 438             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 439             j_coord_offsetA  = DIM*jnrA;
 440             j_coord_offsetB  = DIM*jnrB;
 441             j_coord_offsetC  = DIM*jnrC;
 442             j_coord_offsetD  = DIM*jnrD;
 443             j_coord_offsetE  = DIM*jnrE;
 444             j_coord_offsetF  = DIM*jnrF;
 445             j_coord_offsetG  = DIM*jnrG;
 446             j_coord_offsetH  = DIM*jnrH;
 447
 448             /* load j atom coordinates */
 449             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 450                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 451                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 452                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 453                                                  &jx0,&jy0,&jz0);
 454
 455             /* Calculate displacement vector */
 456             dx00             = _mm256_sub_ps(ix0,jx0);
 457             dy00             = _mm256_sub_ps(iy0,jy0);
 458             dz00             = _mm256_sub_ps(iz0,jz0);
 459             dx10             = _mm256_sub_ps(ix1,jx0);
 460             dy10             = _mm256_sub_ps(iy1,jy0);
 461             dz10             = _mm256_sub_ps(iz1,jz0);
 462             dx20             = _mm256_sub_ps(ix2,jx0);
 463             dy20             = _mm256_sub_ps(iy2,jy0);
 464             dz20             = _mm256_sub_ps(iz2,jz0);
 465             dx30             = _mm256_sub_ps(ix3,jx0);
 466             dy30             = _mm256_sub_ps(iy3,jy0);
 467             dz30             = _mm256_sub_ps(iz3,jz0);
 468
 469             /* Calculate squared distance and things based on it */
 470             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 471             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 472             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 473             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 474
 475             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 476             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 477             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 478
 479             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 480             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 481             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 482             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 483
 484             /* Load parameters for j particles */
 485             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 486                                                                  charge+jnrC+0,charge+jnrD+0,
 487                                                                  charge+jnrE+0,charge+jnrF+0,
 488                                                                  charge+jnrG+0,charge+jnrH+0);
 489             vdwjidx0A        = 2*vdwtype[jnrA+0];
 490             vdwjidx0B        = 2*vdwtype[jnrB+0];
 491             vdwjidx0C        = 2*vdwtype[jnrC+0];
 492             vdwjidx0D        = 2*vdwtype[jnrD+0];
 493             vdwjidx0E        = 2*vdwtype[jnrE+0];
 494             vdwjidx0F        = 2*vdwtype[jnrF+0];
 495             vdwjidx0G        = 2*vdwtype[jnrG+0];
 496             vdwjidx0H        = 2*vdwtype[jnrH+0];
 497
 498             fjx0             = _mm256_setzero_ps();
 499             fjy0             = _mm256_setzero_ps();
 500             fjz0             = _mm256_setzero_ps();
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             /* Compute parameters for interactions between i and j atoms */
 507             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 508                                             vdwioffsetptr0+vdwjidx0B,
 509                                             vdwioffsetptr0+vdwjidx0C,
 510                                             vdwioffsetptr0+vdwjidx0D,
 511                                             vdwioffsetptr0+vdwjidx0E,
 512                                             vdwioffsetptr0+vdwjidx0F,
 513                                             vdwioffsetptr0+vdwjidx0G,
 514                                             vdwioffsetptr0+vdwjidx0H,
 515                                             &c6_00,&c12_00);
 516
 517             /* LENNARD-JONES DISPERSION/REPULSION */
 518
 519             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 520             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 521             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 522             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 523             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 527             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 528
 529             fscal            = fvdw;
 530
 531             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 532
 533             /* Calculate temporary vectorial force */
 534             tx               = _mm256_mul_ps(fscal,dx00);
 535             ty               = _mm256_mul_ps(fscal,dy00);
 536             tz               = _mm256_mul_ps(fscal,dz00);
 537
 538             /* Update vectorial force */
 539             fix0             = _mm256_add_ps(fix0,tx);
 540             fiy0             = _mm256_add_ps(fiy0,ty);
 541             fiz0             = _mm256_add_ps(fiz0,tz);
 542
 543             fjx0             = _mm256_add_ps(fjx0,tx);
 544             fjy0             = _mm256_add_ps(fjy0,ty);
 545             fjz0             = _mm256_add_ps(fjz0,tz);
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq10             = _mm256_mul_ps(iq1,jq0);
 553
 554             /* COULOMB ELECTROSTATICS */
 555             velec            = _mm256_mul_ps(qq10,rinv10);
 556             felec            = _mm256_mul_ps(velec,rinvsq10);
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velec            = _mm256_andnot_ps(dummy_mask,velec);
 560             velecsum         = _mm256_add_ps(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm256_mul_ps(fscal,dx10);
 568             ty               = _mm256_mul_ps(fscal,dy10);
 569             tz               = _mm256_mul_ps(fscal,dz10);
 570
 571             /* Update vectorial force */
 572             fix1             = _mm256_add_ps(fix1,tx);
 573             fiy1             = _mm256_add_ps(fiy1,ty);
 574             fiz1             = _mm256_add_ps(fiz1,tz);
 575
 576             fjx0             = _mm256_add_ps(fjx0,tx);
 577             fjy0             = _mm256_add_ps(fjy0,ty);
 578             fjz0             = _mm256_add_ps(fjz0,tz);
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq20             = _mm256_mul_ps(iq2,jq0);
 586
 587             /* COULOMB ELECTROSTATICS */
 588             velec            = _mm256_mul_ps(qq20,rinv20);
 589             felec            = _mm256_mul_ps(velec,rinvsq20);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm256_andnot_ps(dummy_mask,velec);
 593             velecsum         = _mm256_add_ps(velecsum,velec);
 594
 595             fscal            = felec;
 596
 597             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm256_mul_ps(fscal,dx20);
 601             ty               = _mm256_mul_ps(fscal,dy20);
 602             tz               = _mm256_mul_ps(fscal,dz20);
 603
 604             /* Update vectorial force */
 605             fix2             = _mm256_add_ps(fix2,tx);
 606             fiy2             = _mm256_add_ps(fiy2,ty);
 607             fiz2             = _mm256_add_ps(fiz2,tz);
 608
 609             fjx0             = _mm256_add_ps(fjx0,tx);
 610             fjy0             = _mm256_add_ps(fjy0,ty);
 611             fjz0             = _mm256_add_ps(fjz0,tz);
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq30             = _mm256_mul_ps(iq3,jq0);
 619
 620             /* COULOMB ELECTROSTATICS */
 621             velec            = _mm256_mul_ps(qq30,rinv30);
 622             felec            = _mm256_mul_ps(velec,rinvsq30);
 623
 624             /* Update potential sum for this i atom from the interaction with this j atom. */
 625             velec            = _mm256_andnot_ps(dummy_mask,velec);
 626             velecsum         = _mm256_add_ps(velecsum,velec);
 627
 628             fscal            = felec;
 629
 630             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm256_mul_ps(fscal,dx30);
 634             ty               = _mm256_mul_ps(fscal,dy30);
 635             tz               = _mm256_mul_ps(fscal,dz30);
 636
 637             /* Update vectorial force */
 638             fix3             = _mm256_add_ps(fix3,tx);
 639             fiy3             = _mm256_add_ps(fiy3,ty);
 640             fiz3             = _mm256_add_ps(fiz3,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 647             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 648             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 649             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 650             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 651             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 652             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 653             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 654
 655             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 656
 657             /* Inner loop uses 116 flops */
 658         }
 659
 660         /* End of innermost loop */
 661
 662         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 663                                                  f+i_coord_offset,fshift+i_shift_offset);
 664
 665         ggid                        = gid[iidx];
 666         /* Update potential energies */
 667         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 668         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 669
 670         /* Increment number of inner iterations */
 671         inneriter                  += j_index_end - j_index_start;
 672
 673         /* Outer loop uses 26 flops */
 674     }
 675
 676     /* Increment number of outer iterations */
 677     outeriter        += nri;
 678
 679     /* Update outer/inner flops */
 680
 681     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 682 }
 683 /*
 684  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_single
 685  * Electrostatics interaction: Coulomb
 686  * VdW interaction:            LennardJones
 687  * Geometry:                   Water4-Particle
 688  * Calculate force/pot:        Force
 689  */
 690 void
 691 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_single
 692                     (t_nblist                    * gmx_restrict       nlist,
 693                      rvec                        * gmx_restrict          xx,
 694                      rvec                        * gmx_restrict          ff,
 695                      t_forcerec                  * gmx_restrict          fr,
 696                      t_mdatoms                   * gmx_restrict     mdatoms,
 697                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 698                      t_nrnb                      * gmx_restrict        nrnb)
 699 {
 700     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 701      * just 0 for non-waters.
 702      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 703      * jnr indices corresponding to data put in the four positions in the SIMD register.
 704      */
 705     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 706     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 707     int              jnrA,jnrB,jnrC,jnrD;
 708     int              jnrE,jnrF,jnrG,jnrH;
 709     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 710     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 711     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 712     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 713     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 714     real             rcutoff_scalar;
 715     real             *shiftvec,*fshift,*x,*f;
 716     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 717     real             scratch[4*DIM];
 718     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 719     real *           vdwioffsetptr0;
 720     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 721     real *           vdwioffsetptr1;
 722     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 723     real *           vdwioffsetptr2;
 724     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 725     real *           vdwioffsetptr3;
 726     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 727     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 728     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 729     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 730     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 731     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 732     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 733     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 734     real             *charge;
 735     int              nvdwtype;
 736     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 737     int              *vdwtype;
 738     real             *vdwparam;
 739     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 740     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 741     __m256           dummy_mask,cutoff_mask;
 742     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 743     __m256           one     = _mm256_set1_ps(1.0);
 744     __m256           two     = _mm256_set1_ps(2.0);
 745     x                = xx[0];
 746     f                = ff[0];
 747
 748     nri              = nlist->nri;
 749     iinr             = nlist->iinr;
 750     jindex           = nlist->jindex;
 751     jjnr             = nlist->jjnr;
 752     shiftidx         = nlist->shift;
 753     gid              = nlist->gid;
 754     shiftvec         = fr->shift_vec[0];
 755     fshift           = fr->fshift[0];
 756     facel            = _mm256_set1_ps(fr->epsfac);
 757     charge           = mdatoms->chargeA;
 758     nvdwtype         = fr->ntype;
 759     vdwparam         = fr->nbfp;
 760     vdwtype          = mdatoms->typeA;
 761
 762     /* Setup water-specific parameters */
 763     inr              = nlist->iinr[0];
 764     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 765     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 766     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 767     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 768
 769     /* Avoid stupid compiler warnings */
 770     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 771     j_coord_offsetA = 0;
 772     j_coord_offsetB = 0;
 773     j_coord_offsetC = 0;
 774     j_coord_offsetD = 0;
 775     j_coord_offsetE = 0;
 776     j_coord_offsetF = 0;
 777     j_coord_offsetG = 0;
 778     j_coord_offsetH = 0;
 779
 780     outeriter        = 0;
 781     inneriter        = 0;
 782
 783     for(iidx=0;iidx<4*DIM;iidx++)
 784     {
 785         scratch[iidx] = 0.0;
 786     }
 787
 788     /* Start outer loop over neighborlists */
 789     for(iidx=0; iidx<nri; iidx++)
 790     {
 791         /* Load shift vector for this list */
 792         i_shift_offset   = DIM*shiftidx[iidx];
 793
 794         /* Load limits for loop over neighbors */
 795         j_index_start    = jindex[iidx];
 796         j_index_end      = jindex[iidx+1];
 797
 798         /* Get outer coordinate index */
 799         inr              = iinr[iidx];
 800         i_coord_offset   = DIM*inr;
 801
 802         /* Load i particle coords and add shift vector */
 803         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 804                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 805
 806         fix0             = _mm256_setzero_ps();
 807         fiy0             = _mm256_setzero_ps();
 808         fiz0             = _mm256_setzero_ps();
 809         fix1             = _mm256_setzero_ps();
 810         fiy1             = _mm256_setzero_ps();
 811         fiz1             = _mm256_setzero_ps();
 812         fix2             = _mm256_setzero_ps();
 813         fiy2             = _mm256_setzero_ps();
 814         fiz2             = _mm256_setzero_ps();
 815         fix3             = _mm256_setzero_ps();
 816         fiy3             = _mm256_setzero_ps();
 817         fiz3             = _mm256_setzero_ps();
 818
 819         /* Start inner kernel loop */
 820         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 821         {
 822
 823             /* Get j neighbor index, and coordinate index */
 824             jnrA             = jjnr[jidx];
 825             jnrB             = jjnr[jidx+1];
 826             jnrC             = jjnr[jidx+2];
 827             jnrD             = jjnr[jidx+3];
 828             jnrE             = jjnr[jidx+4];
 829             jnrF             = jjnr[jidx+5];
 830             jnrG             = jjnr[jidx+6];
 831             jnrH             = jjnr[jidx+7];
 832             j_coord_offsetA  = DIM*jnrA;
 833             j_coord_offsetB  = DIM*jnrB;
 834             j_coord_offsetC  = DIM*jnrC;
 835             j_coord_offsetD  = DIM*jnrD;
 836             j_coord_offsetE  = DIM*jnrE;
 837             j_coord_offsetF  = DIM*jnrF;
 838             j_coord_offsetG  = DIM*jnrG;
 839             j_coord_offsetH  = DIM*jnrH;
 840
 841             /* load j atom coordinates */
 842             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 843                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 844                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 845                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 846                                                  &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm256_sub_ps(ix0,jx0);
 850             dy00             = _mm256_sub_ps(iy0,jy0);
 851             dz00             = _mm256_sub_ps(iz0,jz0);
 852             dx10             = _mm256_sub_ps(ix1,jx0);
 853             dy10             = _mm256_sub_ps(iy1,jy0);
 854             dz10             = _mm256_sub_ps(iz1,jz0);
 855             dx20             = _mm256_sub_ps(ix2,jx0);
 856             dy20             = _mm256_sub_ps(iy2,jy0);
 857             dz20             = _mm256_sub_ps(iz2,jz0);
 858             dx30             = _mm256_sub_ps(ix3,jx0);
 859             dy30             = _mm256_sub_ps(iy3,jy0);
 860             dz30             = _mm256_sub_ps(iz3,jz0);
 861
 862             /* Calculate squared distance and things based on it */
 863             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 864             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 866             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 867
 868             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 869             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 870             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 871
 872             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 873             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 874             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 875             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 876
 877             /* Load parameters for j particles */
 878             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 879                                                                  charge+jnrC+0,charge+jnrD+0,
 880                                                                  charge+jnrE+0,charge+jnrF+0,
 881                                                                  charge+jnrG+0,charge+jnrH+0);
 882             vdwjidx0A        = 2*vdwtype[jnrA+0];
 883             vdwjidx0B        = 2*vdwtype[jnrB+0];
 884             vdwjidx0C        = 2*vdwtype[jnrC+0];
 885             vdwjidx0D        = 2*vdwtype[jnrD+0];
 886             vdwjidx0E        = 2*vdwtype[jnrE+0];
 887             vdwjidx0F        = 2*vdwtype[jnrF+0];
 888             vdwjidx0G        = 2*vdwtype[jnrG+0];
 889             vdwjidx0H        = 2*vdwtype[jnrH+0];
 890
 891             fjx0             = _mm256_setzero_ps();
 892             fjy0             = _mm256_setzero_ps();
 893             fjz0             = _mm256_setzero_ps();
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 901                                             vdwioffsetptr0+vdwjidx0B,
 902                                             vdwioffsetptr0+vdwjidx0C,
 903                                             vdwioffsetptr0+vdwjidx0D,
 904                                             vdwioffsetptr0+vdwjidx0E,
 905                                             vdwioffsetptr0+vdwjidx0F,
 906                                             vdwioffsetptr0+vdwjidx0G,
 907                                             vdwioffsetptr0+vdwjidx0H,
 908                                             &c6_00,&c12_00);
 909
 910             /* LENNARD-JONES DISPERSION/REPULSION */
 911
 912             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 913             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 914
 915             fscal            = fvdw;
 916
 917             /* Calculate temporary vectorial force */
 918             tx               = _mm256_mul_ps(fscal,dx00);
 919             ty               = _mm256_mul_ps(fscal,dy00);
 920             tz               = _mm256_mul_ps(fscal,dz00);
 921
 922             /* Update vectorial force */
 923             fix0             = _mm256_add_ps(fix0,tx);
 924             fiy0             = _mm256_add_ps(fiy0,ty);
 925             fiz0             = _mm256_add_ps(fiz0,tz);
 926
 927             fjx0             = _mm256_add_ps(fjx0,tx);
 928             fjy0             = _mm256_add_ps(fjy0,ty);
 929             fjz0             = _mm256_add_ps(fjz0,tz);
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             /* Compute parameters for interactions between i and j atoms */
 936             qq10             = _mm256_mul_ps(iq1,jq0);
 937
 938             /* COULOMB ELECTROSTATICS */
 939             velec            = _mm256_mul_ps(qq10,rinv10);
 940             felec            = _mm256_mul_ps(velec,rinvsq10);
 941
 942             fscal            = felec;
 943
 944             /* Calculate temporary vectorial force */
 945             tx               = _mm256_mul_ps(fscal,dx10);
 946             ty               = _mm256_mul_ps(fscal,dy10);
 947             tz               = _mm256_mul_ps(fscal,dz10);
 948
 949             /* Update vectorial force */
 950             fix1             = _mm256_add_ps(fix1,tx);
 951             fiy1             = _mm256_add_ps(fiy1,ty);
 952             fiz1             = _mm256_add_ps(fiz1,tz);
 953
 954             fjx0             = _mm256_add_ps(fjx0,tx);
 955             fjy0             = _mm256_add_ps(fjy0,ty);
 956             fjz0             = _mm256_add_ps(fjz0,tz);
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq20             = _mm256_mul_ps(iq2,jq0);
 964
 965             /* COULOMB ELECTROSTATICS */
 966             velec            = _mm256_mul_ps(qq20,rinv20);
 967             felec            = _mm256_mul_ps(velec,rinvsq20);
 968
 969             fscal            = felec;
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm256_mul_ps(fscal,dx20);
 973             ty               = _mm256_mul_ps(fscal,dy20);
 974             tz               = _mm256_mul_ps(fscal,dz20);
 975
 976             /* Update vectorial force */
 977             fix2             = _mm256_add_ps(fix2,tx);
 978             fiy2             = _mm256_add_ps(fiy2,ty);
 979             fiz2             = _mm256_add_ps(fiz2,tz);
 980
 981             fjx0             = _mm256_add_ps(fjx0,tx);
 982             fjy0             = _mm256_add_ps(fjy0,ty);
 983             fjz0             = _mm256_add_ps(fjz0,tz);
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             /* Compute parameters for interactions between i and j atoms */
 990             qq30             = _mm256_mul_ps(iq3,jq0);
 991
 992             /* COULOMB ELECTROSTATICS */
 993             velec            = _mm256_mul_ps(qq30,rinv30);
 994             felec            = _mm256_mul_ps(velec,rinvsq30);
 995
 996             fscal            = felec;
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm256_mul_ps(fscal,dx30);
1000             ty               = _mm256_mul_ps(fscal,dy30);
1001             tz               = _mm256_mul_ps(fscal,dz30);
1002
1003             /* Update vectorial force */
1004             fix3             = _mm256_add_ps(fix3,tx);
1005             fiy3             = _mm256_add_ps(fiy3,ty);
1006             fiz3             = _mm256_add_ps(fiz3,tz);
1007
1008             fjx0             = _mm256_add_ps(fjx0,tx);
1009             fjy0             = _mm256_add_ps(fjy0,ty);
1010             fjz0             = _mm256_add_ps(fjz0,tz);
1011
1012             fjptrA             = f+j_coord_offsetA;
1013             fjptrB             = f+j_coord_offsetB;
1014             fjptrC             = f+j_coord_offsetC;
1015             fjptrD             = f+j_coord_offsetD;
1016             fjptrE             = f+j_coord_offsetE;
1017             fjptrF             = f+j_coord_offsetF;
1018             fjptrG             = f+j_coord_offsetG;
1019             fjptrH             = f+j_coord_offsetH;
1020
1021             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1022
1023             /* Inner loop uses 108 flops */
1024         }
1025
1026         if(jidx<j_index_end)
1027         {
1028
1029             /* Get j neighbor index, and coordinate index */
1030             jnrlistA         = jjnr[jidx];
1031             jnrlistB         = jjnr[jidx+1];
1032             jnrlistC         = jjnr[jidx+2];
1033             jnrlistD         = jjnr[jidx+3];
1034             jnrlistE         = jjnr[jidx+4];
1035             jnrlistF         = jjnr[jidx+5];
1036             jnrlistG         = jjnr[jidx+6];
1037             jnrlistH         = jjnr[jidx+7];
1038             /* Sign of each element will be negative for non-real atoms.
1039              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1040              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1041              */
1042             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1043                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1044
1045             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1046             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1047             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1048             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1049             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1050             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1051             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1052             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1053             j_coord_offsetA  = DIM*jnrA;
1054             j_coord_offsetB  = DIM*jnrB;
1055             j_coord_offsetC  = DIM*jnrC;
1056             j_coord_offsetD  = DIM*jnrD;
1057             j_coord_offsetE  = DIM*jnrE;
1058             j_coord_offsetF  = DIM*jnrF;
1059             j_coord_offsetG  = DIM*jnrG;
1060             j_coord_offsetH  = DIM*jnrH;
1061
1062             /* load j atom coordinates */
1063             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1064                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1065                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1066                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1067                                                  &jx0,&jy0,&jz0);
1068
1069             /* Calculate displacement vector */
1070             dx00             = _mm256_sub_ps(ix0,jx0);
1071             dy00             = _mm256_sub_ps(iy0,jy0);
1072             dz00             = _mm256_sub_ps(iz0,jz0);
1073             dx10             = _mm256_sub_ps(ix1,jx0);
1074             dy10             = _mm256_sub_ps(iy1,jy0);
1075             dz10             = _mm256_sub_ps(iz1,jz0);
1076             dx20             = _mm256_sub_ps(ix2,jx0);
1077             dy20             = _mm256_sub_ps(iy2,jy0);
1078             dz20             = _mm256_sub_ps(iz2,jz0);
1079             dx30             = _mm256_sub_ps(ix3,jx0);
1080             dy30             = _mm256_sub_ps(iy3,jy0);
1081             dz30             = _mm256_sub_ps(iz3,jz0);
1082
1083             /* Calculate squared distance and things based on it */
1084             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1085             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1086             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1087             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1088
1089             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1090             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1091             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1092
1093             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1094             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1095             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1096             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1097
1098             /* Load parameters for j particles */
1099             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1100                                                                  charge+jnrC+0,charge+jnrD+0,
1101                                                                  charge+jnrE+0,charge+jnrF+0,
1102                                                                  charge+jnrG+0,charge+jnrH+0);
1103             vdwjidx0A        = 2*vdwtype[jnrA+0];
1104             vdwjidx0B        = 2*vdwtype[jnrB+0];
1105             vdwjidx0C        = 2*vdwtype[jnrC+0];
1106             vdwjidx0D        = 2*vdwtype[jnrD+0];
1107             vdwjidx0E        = 2*vdwtype[jnrE+0];
1108             vdwjidx0F        = 2*vdwtype[jnrF+0];
1109             vdwjidx0G        = 2*vdwtype[jnrG+0];
1110             vdwjidx0H        = 2*vdwtype[jnrH+0];
1111
1112             fjx0             = _mm256_setzero_ps();
1113             fjy0             = _mm256_setzero_ps();
1114             fjz0             = _mm256_setzero_ps();
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1122                                             vdwioffsetptr0+vdwjidx0B,
1123                                             vdwioffsetptr0+vdwjidx0C,
1124                                             vdwioffsetptr0+vdwjidx0D,
1125                                             vdwioffsetptr0+vdwjidx0E,
1126                                             vdwioffsetptr0+vdwjidx0F,
1127                                             vdwioffsetptr0+vdwjidx0G,
1128                                             vdwioffsetptr0+vdwjidx0H,
1129                                             &c6_00,&c12_00);
1130
1131             /* LENNARD-JONES DISPERSION/REPULSION */
1132
1133             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1134             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1135
1136             fscal            = fvdw;
1137
1138             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm256_mul_ps(fscal,dx00);
1142             ty               = _mm256_mul_ps(fscal,dy00);
1143             tz               = _mm256_mul_ps(fscal,dz00);
1144
1145             /* Update vectorial force */
1146             fix0             = _mm256_add_ps(fix0,tx);
1147             fiy0             = _mm256_add_ps(fiy0,ty);
1148             fiz0             = _mm256_add_ps(fiz0,tz);
1149
1150             fjx0             = _mm256_add_ps(fjx0,tx);
1151             fjy0             = _mm256_add_ps(fjy0,ty);
1152             fjz0             = _mm256_add_ps(fjz0,tz);
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             /* Compute parameters for interactions between i and j atoms */
1159             qq10             = _mm256_mul_ps(iq1,jq0);
1160
1161             /* COULOMB ELECTROSTATICS */
1162             velec            = _mm256_mul_ps(qq10,rinv10);
1163             felec            = _mm256_mul_ps(velec,rinvsq10);
1164
1165             fscal            = felec;
1166
1167             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm256_mul_ps(fscal,dx10);
1171             ty               = _mm256_mul_ps(fscal,dy10);
1172             tz               = _mm256_mul_ps(fscal,dz10);
1173
1174             /* Update vectorial force */
1175             fix1             = _mm256_add_ps(fix1,tx);
1176             fiy1             = _mm256_add_ps(fiy1,ty);
1177             fiz1             = _mm256_add_ps(fiz1,tz);
1178
1179             fjx0             = _mm256_add_ps(fjx0,tx);
1180             fjy0             = _mm256_add_ps(fjy0,ty);
1181             fjz0             = _mm256_add_ps(fjz0,tz);
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             /* Compute parameters for interactions between i and j atoms */
1188             qq20             = _mm256_mul_ps(iq2,jq0);
1189
1190             /* COULOMB ELECTROSTATICS */
1191             velec            = _mm256_mul_ps(qq20,rinv20);
1192             felec            = _mm256_mul_ps(velec,rinvsq20);
1193
1194             fscal            = felec;
1195
1196             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1197
1198             /* Calculate temporary vectorial force */
1199             tx               = _mm256_mul_ps(fscal,dx20);
1200             ty               = _mm256_mul_ps(fscal,dy20);
1201             tz               = _mm256_mul_ps(fscal,dz20);
1202
1203             /* Update vectorial force */
1204             fix2             = _mm256_add_ps(fix2,tx);
1205             fiy2             = _mm256_add_ps(fiy2,ty);
1206             fiz2             = _mm256_add_ps(fiz2,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              * CALCULATE INTERACTIONS *
1214              **************************/
1215
1216             /* Compute parameters for interactions between i and j atoms */
1217             qq30             = _mm256_mul_ps(iq3,jq0);
1218
1219             /* COULOMB ELECTROSTATICS */
1220             velec            = _mm256_mul_ps(qq30,rinv30);
1221             felec            = _mm256_mul_ps(velec,rinvsq30);
1222
1223             fscal            = felec;
1224
1225             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1226
1227             /* Calculate temporary vectorial force */
1228             tx               = _mm256_mul_ps(fscal,dx30);
1229             ty               = _mm256_mul_ps(fscal,dy30);
1230             tz               = _mm256_mul_ps(fscal,dz30);
1231
1232             /* Update vectorial force */
1233             fix3             = _mm256_add_ps(fix3,tx);
1234             fiy3             = _mm256_add_ps(fiy3,ty);
1235             fiz3             = _mm256_add_ps(fiz3,tz);
1236
1237             fjx0             = _mm256_add_ps(fjx0,tx);
1238             fjy0             = _mm256_add_ps(fjy0,ty);
1239             fjz0             = _mm256_add_ps(fjz0,tz);
1240
1241             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1242             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1243             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1244             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1245             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1246             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1247             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1248             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1249
1250             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1251
1252             /* Inner loop uses 108 flops */
1253         }
1254
1255         /* End of innermost loop */
1256
1257         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1258                                                  f+i_coord_offset,fshift+i_shift_offset);
1259
1260         /* Increment number of inner iterations */
1261         inneriter                  += j_index_end - j_index_start;
1262
1263         /* Outer loop uses 24 flops */
1264     }
1265
1266     /* Increment number of outer iterations */
1267     outeriter        += nri;
1268
1269     /* Update outer/inner flops */
1270
1271     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1272 }