src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256           dummy_mask,cutoff_mask;
 111     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 112     __m256           one     = _mm256_set1_ps(1.0);
 113     __m256           two     = _mm256_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     krf              = _mm256_set1_ps(fr->ic->k_rf);
 128     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 129     crf              = _mm256_set1_ps(fr->ic->c_rf);
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 137     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 138     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181         fix1             = _mm256_setzero_ps();
 182         fiy1             = _mm256_setzero_ps();
 183         fiz1             = _mm256_setzero_ps();
 184         fix2             = _mm256_setzero_ps();
 185         fiy2             = _mm256_setzero_ps();
 186         fiz2             = _mm256_setzero_ps();
 187         fix3             = _mm256_setzero_ps();
 188         fiy3             = _mm256_setzero_ps();
 189         fiz3             = _mm256_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm256_setzero_ps();
 193         vvdwsum          = _mm256_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             jnrE             = jjnr[jidx+4];
 205             jnrF             = jjnr[jidx+5];
 206             jnrG             = jjnr[jidx+6];
 207             jnrH             = jjnr[jidx+7];
 208             j_coord_offsetA  = DIM*jnrA;
 209             j_coord_offsetB  = DIM*jnrB;
 210             j_coord_offsetC  = DIM*jnrC;
 211             j_coord_offsetD  = DIM*jnrD;
 212             j_coord_offsetE  = DIM*jnrE;
 213             j_coord_offsetF  = DIM*jnrF;
 214             j_coord_offsetG  = DIM*jnrG;
 215             j_coord_offsetH  = DIM*jnrH;
 216
 217             /* load j atom coordinates */
 218             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 219                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 220                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 221                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 222                                                  &jx0,&jy0,&jz0);
 223
 224             /* Calculate displacement vector */
 225             dx00             = _mm256_sub_ps(ix0,jx0);
 226             dy00             = _mm256_sub_ps(iy0,jy0);
 227             dz00             = _mm256_sub_ps(iz0,jz0);
 228             dx10             = _mm256_sub_ps(ix1,jx0);
 229             dy10             = _mm256_sub_ps(iy1,jy0);
 230             dz10             = _mm256_sub_ps(iz1,jz0);
 231             dx20             = _mm256_sub_ps(ix2,jx0);
 232             dy20             = _mm256_sub_ps(iy2,jy0);
 233             dz20             = _mm256_sub_ps(iz2,jz0);
 234             dx30             = _mm256_sub_ps(ix3,jx0);
 235             dy30             = _mm256_sub_ps(iy3,jy0);
 236             dz30             = _mm256_sub_ps(iz3,jz0);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 240             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 241             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 242             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 243
 244             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 245             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 246             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 247
 248             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 249             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 250             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 251             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 252
 253             /* Load parameters for j particles */
 254             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 255                                                                  charge+jnrC+0,charge+jnrD+0,
 256                                                                  charge+jnrE+0,charge+jnrF+0,
 257                                                                  charge+jnrG+0,charge+jnrH+0);
 258             vdwjidx0A        = 2*vdwtype[jnrA+0];
 259             vdwjidx0B        = 2*vdwtype[jnrB+0];
 260             vdwjidx0C        = 2*vdwtype[jnrC+0];
 261             vdwjidx0D        = 2*vdwtype[jnrD+0];
 262             vdwjidx0E        = 2*vdwtype[jnrE+0];
 263             vdwjidx0F        = 2*vdwtype[jnrF+0];
 264             vdwjidx0G        = 2*vdwtype[jnrG+0];
 265             vdwjidx0H        = 2*vdwtype[jnrH+0];
 266
 267             fjx0             = _mm256_setzero_ps();
 268             fjy0             = _mm256_setzero_ps();
 269             fjz0             = _mm256_setzero_ps();
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 277                                             vdwioffsetptr0+vdwjidx0B,
 278                                             vdwioffsetptr0+vdwjidx0C,
 279                                             vdwioffsetptr0+vdwjidx0D,
 280                                             vdwioffsetptr0+vdwjidx0E,
 281                                             vdwioffsetptr0+vdwjidx0F,
 282                                             vdwioffsetptr0+vdwjidx0G,
 283                                             vdwioffsetptr0+vdwjidx0H,
 284                                             &c6_00,&c12_00);
 285
 286             /* LENNARD-JONES DISPERSION/REPULSION */
 287
 288             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 289             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 290             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 291             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 292             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 296
 297             fscal            = fvdw;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm256_mul_ps(fscal,dx00);
 301             ty               = _mm256_mul_ps(fscal,dy00);
 302             tz               = _mm256_mul_ps(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm256_add_ps(fix0,tx);
 306             fiy0             = _mm256_add_ps(fiy0,ty);
 307             fiz0             = _mm256_add_ps(fiz0,tz);
 308
 309             fjx0             = _mm256_add_ps(fjx0,tx);
 310             fjy0             = _mm256_add_ps(fjy0,ty);
 311             fjz0             = _mm256_add_ps(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq10             = _mm256_mul_ps(iq1,jq0);
 319
 320             /* REACTION-FIELD ELECTROSTATICS */
 321             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 322             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm256_add_ps(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm256_mul_ps(fscal,dx10);
 331             ty               = _mm256_mul_ps(fscal,dy10);
 332             tz               = _mm256_mul_ps(fscal,dz10);
 333
 334             /* Update vectorial force */
 335             fix1             = _mm256_add_ps(fix1,tx);
 336             fiy1             = _mm256_add_ps(fiy1,ty);
 337             fiz1             = _mm256_add_ps(fiz1,tz);
 338
 339             fjx0             = _mm256_add_ps(fjx0,tx);
 340             fjy0             = _mm256_add_ps(fjy0,ty);
 341             fjz0             = _mm256_add_ps(fjz0,tz);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq20             = _mm256_mul_ps(iq2,jq0);
 349
 350             /* REACTION-FIELD ELECTROSTATICS */
 351             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 352             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velecsum         = _mm256_add_ps(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm256_mul_ps(fscal,dx20);
 361             ty               = _mm256_mul_ps(fscal,dy20);
 362             tz               = _mm256_mul_ps(fscal,dz20);
 363
 364             /* Update vectorial force */
 365             fix2             = _mm256_add_ps(fix2,tx);
 366             fiy2             = _mm256_add_ps(fiy2,ty);
 367             fiz2             = _mm256_add_ps(fiz2,tz);
 368
 369             fjx0             = _mm256_add_ps(fjx0,tx);
 370             fjy0             = _mm256_add_ps(fjy0,ty);
 371             fjz0             = _mm256_add_ps(fjz0,tz);
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             /* Compute parameters for interactions between i and j atoms */
 378             qq30             = _mm256_mul_ps(iq3,jq0);
 379
 380             /* REACTION-FIELD ELECTROSTATICS */
 381             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 382             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velecsum         = _mm256_add_ps(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             /* Calculate temporary vectorial force */
 390             tx               = _mm256_mul_ps(fscal,dx30);
 391             ty               = _mm256_mul_ps(fscal,dy30);
 392             tz               = _mm256_mul_ps(fscal,dz30);
 393
 394             /* Update vectorial force */
 395             fix3             = _mm256_add_ps(fix3,tx);
 396             fiy3             = _mm256_add_ps(fiy3,ty);
 397             fiz3             = _mm256_add_ps(fiz3,tz);
 398
 399             fjx0             = _mm256_add_ps(fjx0,tx);
 400             fjy0             = _mm256_add_ps(fjy0,ty);
 401             fjz0             = _mm256_add_ps(fjz0,tz);
 402
 403             fjptrA             = f+j_coord_offsetA;
 404             fjptrB             = f+j_coord_offsetB;
 405             fjptrC             = f+j_coord_offsetC;
 406             fjptrD             = f+j_coord_offsetD;
 407             fjptrE             = f+j_coord_offsetE;
 408             fjptrF             = f+j_coord_offsetF;
 409             fjptrG             = f+j_coord_offsetG;
 410             fjptrH             = f+j_coord_offsetH;
 411
 412             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 413
 414             /* Inner loop uses 131 flops */
 415         }
 416
 417         if(jidx<j_index_end)
 418         {
 419
 420             /* Get j neighbor index, and coordinate index */
 421             jnrlistA         = jjnr[jidx];
 422             jnrlistB         = jjnr[jidx+1];
 423             jnrlistC         = jjnr[jidx+2];
 424             jnrlistD         = jjnr[jidx+3];
 425             jnrlistE         = jjnr[jidx+4];
 426             jnrlistF         = jjnr[jidx+5];
 427             jnrlistG         = jjnr[jidx+6];
 428             jnrlistH         = jjnr[jidx+7];
 429             /* Sign of each element will be negative for non-real atoms.
 430              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 431              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 432              */
 433             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 434                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 435
 436             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 437             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 438             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 439             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 440             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 441             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 442             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 443             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 444             j_coord_offsetA  = DIM*jnrA;
 445             j_coord_offsetB  = DIM*jnrB;
 446             j_coord_offsetC  = DIM*jnrC;
 447             j_coord_offsetD  = DIM*jnrD;
 448             j_coord_offsetE  = DIM*jnrE;
 449             j_coord_offsetF  = DIM*jnrF;
 450             j_coord_offsetG  = DIM*jnrG;
 451             j_coord_offsetH  = DIM*jnrH;
 452
 453             /* load j atom coordinates */
 454             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 455                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 456                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 457                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 458                                                  &jx0,&jy0,&jz0);
 459
 460             /* Calculate displacement vector */
 461             dx00             = _mm256_sub_ps(ix0,jx0);
 462             dy00             = _mm256_sub_ps(iy0,jy0);
 463             dz00             = _mm256_sub_ps(iz0,jz0);
 464             dx10             = _mm256_sub_ps(ix1,jx0);
 465             dy10             = _mm256_sub_ps(iy1,jy0);
 466             dz10             = _mm256_sub_ps(iz1,jz0);
 467             dx20             = _mm256_sub_ps(ix2,jx0);
 468             dy20             = _mm256_sub_ps(iy2,jy0);
 469             dz20             = _mm256_sub_ps(iz2,jz0);
 470             dx30             = _mm256_sub_ps(ix3,jx0);
 471             dy30             = _mm256_sub_ps(iy3,jy0);
 472             dz30             = _mm256_sub_ps(iz3,jz0);
 473
 474             /* Calculate squared distance and things based on it */
 475             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 476             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 477             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 478             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 479
 480             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 481             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 482             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 483
 484             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 485             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 486             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 487             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 491                                                                  charge+jnrC+0,charge+jnrD+0,
 492                                                                  charge+jnrE+0,charge+jnrF+0,
 493                                                                  charge+jnrG+0,charge+jnrH+0);
 494             vdwjidx0A        = 2*vdwtype[jnrA+0];
 495             vdwjidx0B        = 2*vdwtype[jnrB+0];
 496             vdwjidx0C        = 2*vdwtype[jnrC+0];
 497             vdwjidx0D        = 2*vdwtype[jnrD+0];
 498             vdwjidx0E        = 2*vdwtype[jnrE+0];
 499             vdwjidx0F        = 2*vdwtype[jnrF+0];
 500             vdwjidx0G        = 2*vdwtype[jnrG+0];
 501             vdwjidx0H        = 2*vdwtype[jnrH+0];
 502
 503             fjx0             = _mm256_setzero_ps();
 504             fjy0             = _mm256_setzero_ps();
 505             fjz0             = _mm256_setzero_ps();
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 513                                             vdwioffsetptr0+vdwjidx0B,
 514                                             vdwioffsetptr0+vdwjidx0C,
 515                                             vdwioffsetptr0+vdwjidx0D,
 516                                             vdwioffsetptr0+vdwjidx0E,
 517                                             vdwioffsetptr0+vdwjidx0F,
 518                                             vdwioffsetptr0+vdwjidx0G,
 519                                             vdwioffsetptr0+vdwjidx0H,
 520                                             &c6_00,&c12_00);
 521
 522             /* LENNARD-JONES DISPERSION/REPULSION */
 523
 524             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 525             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 526             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 527             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 528             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 532             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 533
 534             fscal            = fvdw;
 535
 536             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm256_mul_ps(fscal,dx00);
 540             ty               = _mm256_mul_ps(fscal,dy00);
 541             tz               = _mm256_mul_ps(fscal,dz00);
 542
 543             /* Update vectorial force */
 544             fix0             = _mm256_add_ps(fix0,tx);
 545             fiy0             = _mm256_add_ps(fiy0,ty);
 546             fiz0             = _mm256_add_ps(fiz0,tz);
 547
 548             fjx0             = _mm256_add_ps(fjx0,tx);
 549             fjy0             = _mm256_add_ps(fjy0,ty);
 550             fjz0             = _mm256_add_ps(fjz0,tz);
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq10             = _mm256_mul_ps(iq1,jq0);
 558
 559             /* REACTION-FIELD ELECTROSTATICS */
 560             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 561             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _mm256_andnot_ps(dummy_mask,velec);
 565             velecsum         = _mm256_add_ps(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm256_mul_ps(fscal,dx10);
 573             ty               = _mm256_mul_ps(fscal,dy10);
 574             tz               = _mm256_mul_ps(fscal,dz10);
 575
 576             /* Update vectorial force */
 577             fix1             = _mm256_add_ps(fix1,tx);
 578             fiy1             = _mm256_add_ps(fiy1,ty);
 579             fiz1             = _mm256_add_ps(fiz1,tz);
 580
 581             fjx0             = _mm256_add_ps(fjx0,tx);
 582             fjy0             = _mm256_add_ps(fjy0,ty);
 583             fjz0             = _mm256_add_ps(fjz0,tz);
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq20             = _mm256_mul_ps(iq2,jq0);
 591
 592             /* REACTION-FIELD ELECTROSTATICS */
 593             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 594             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm256_andnot_ps(dummy_mask,velec);
 598             velecsum         = _mm256_add_ps(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm256_mul_ps(fscal,dx20);
 606             ty               = _mm256_mul_ps(fscal,dy20);
 607             tz               = _mm256_mul_ps(fscal,dz20);
 608
 609             /* Update vectorial force */
 610             fix2             = _mm256_add_ps(fix2,tx);
 611             fiy2             = _mm256_add_ps(fiy2,ty);
 612             fiz2             = _mm256_add_ps(fiz2,tz);
 613
 614             fjx0             = _mm256_add_ps(fjx0,tx);
 615             fjy0             = _mm256_add_ps(fjy0,ty);
 616             fjz0             = _mm256_add_ps(fjz0,tz);
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq30             = _mm256_mul_ps(iq3,jq0);
 624
 625             /* REACTION-FIELD ELECTROSTATICS */
 626             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 627             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             velec            = _mm256_andnot_ps(dummy_mask,velec);
 631             velecsum         = _mm256_add_ps(velecsum,velec);
 632
 633             fscal            = felec;
 634
 635             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 636
 637             /* Calculate temporary vectorial force */
 638             tx               = _mm256_mul_ps(fscal,dx30);
 639             ty               = _mm256_mul_ps(fscal,dy30);
 640             tz               = _mm256_mul_ps(fscal,dz30);
 641
 642             /* Update vectorial force */
 643             fix3             = _mm256_add_ps(fix3,tx);
 644             fiy3             = _mm256_add_ps(fiy3,ty);
 645             fiz3             = _mm256_add_ps(fiz3,tz);
 646
 647             fjx0             = _mm256_add_ps(fjx0,tx);
 648             fjy0             = _mm256_add_ps(fjy0,ty);
 649             fjz0             = _mm256_add_ps(fjz0,tz);
 650
 651             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 652             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 653             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 654             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 655             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 656             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 657             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 658             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 659
 660             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 661
 662             /* Inner loop uses 131 flops */
 663         }
 664
 665         /* End of innermost loop */
 666
 667         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 668                                                  f+i_coord_offset,fshift+i_shift_offset);
 669
 670         ggid                        = gid[iidx];
 671         /* Update potential energies */
 672         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 673         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 674
 675         /* Increment number of inner iterations */
 676         inneriter                  += j_index_end - j_index_start;
 677
 678         /* Outer loop uses 26 flops */
 679     }
 680
 681     /* Increment number of outer iterations */
 682     outeriter        += nri;
 683
 684     /* Update outer/inner flops */
 685
 686     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
 687 }
 688 /*
 689  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_single
 690  * Electrostatics interaction: ReactionField
 691  * VdW interaction:            LennardJones
 692  * Geometry:                   Water4-Particle
 693  * Calculate force/pot:        Force
 694  */
 695 void
 696 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_single
 697                     (t_nblist                    * gmx_restrict       nlist,
 698                      rvec                        * gmx_restrict          xx,
 699                      rvec                        * gmx_restrict          ff,
 700                      t_forcerec                  * gmx_restrict          fr,
 701                      t_mdatoms                   * gmx_restrict     mdatoms,
 702                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 703                      t_nrnb                      * gmx_restrict        nrnb)
 704 {
 705     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 706      * just 0 for non-waters.
 707      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 708      * jnr indices corresponding to data put in the four positions in the SIMD register.
 709      */
 710     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 711     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 712     int              jnrA,jnrB,jnrC,jnrD;
 713     int              jnrE,jnrF,jnrG,jnrH;
 714     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 715     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 716     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 717     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 718     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 719     real             rcutoff_scalar;
 720     real             *shiftvec,*fshift,*x,*f;
 721     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 722     real             scratch[4*DIM];
 723     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 724     real *           vdwioffsetptr0;
 725     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 726     real *           vdwioffsetptr1;
 727     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 728     real *           vdwioffsetptr2;
 729     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 730     real *           vdwioffsetptr3;
 731     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 732     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 733     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 734     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 735     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 736     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 737     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 738     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 739     real             *charge;
 740     int              nvdwtype;
 741     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 742     int              *vdwtype;
 743     real             *vdwparam;
 744     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 745     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 746     __m256           dummy_mask,cutoff_mask;
 747     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 748     __m256           one     = _mm256_set1_ps(1.0);
 749     __m256           two     = _mm256_set1_ps(2.0);
 750     x                = xx[0];
 751     f                = ff[0];
 752
 753     nri              = nlist->nri;
 754     iinr             = nlist->iinr;
 755     jindex           = nlist->jindex;
 756     jjnr             = nlist->jjnr;
 757     shiftidx         = nlist->shift;
 758     gid              = nlist->gid;
 759     shiftvec         = fr->shift_vec[0];
 760     fshift           = fr->fshift[0];
 761     facel            = _mm256_set1_ps(fr->epsfac);
 762     charge           = mdatoms->chargeA;
 763     krf              = _mm256_set1_ps(fr->ic->k_rf);
 764     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 765     crf              = _mm256_set1_ps(fr->ic->c_rf);
 766     nvdwtype         = fr->ntype;
 767     vdwparam         = fr->nbfp;
 768     vdwtype          = mdatoms->typeA;
 769
 770     /* Setup water-specific parameters */
 771     inr              = nlist->iinr[0];
 772     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 773     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 774     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 775     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 776
 777     /* Avoid stupid compiler warnings */
 778     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 779     j_coord_offsetA = 0;
 780     j_coord_offsetB = 0;
 781     j_coord_offsetC = 0;
 782     j_coord_offsetD = 0;
 783     j_coord_offsetE = 0;
 784     j_coord_offsetF = 0;
 785     j_coord_offsetG = 0;
 786     j_coord_offsetH = 0;
 787
 788     outeriter        = 0;
 789     inneriter        = 0;
 790
 791     for(iidx=0;iidx<4*DIM;iidx++)
 792     {
 793         scratch[iidx] = 0.0;
 794     }
 795
 796     /* Start outer loop over neighborlists */
 797     for(iidx=0; iidx<nri; iidx++)
 798     {
 799         /* Load shift vector for this list */
 800         i_shift_offset   = DIM*shiftidx[iidx];
 801
 802         /* Load limits for loop over neighbors */
 803         j_index_start    = jindex[iidx];
 804         j_index_end      = jindex[iidx+1];
 805
 806         /* Get outer coordinate index */
 807         inr              = iinr[iidx];
 808         i_coord_offset   = DIM*inr;
 809
 810         /* Load i particle coords and add shift vector */
 811         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 812                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 813
 814         fix0             = _mm256_setzero_ps();
 815         fiy0             = _mm256_setzero_ps();
 816         fiz0             = _mm256_setzero_ps();
 817         fix1             = _mm256_setzero_ps();
 818         fiy1             = _mm256_setzero_ps();
 819         fiz1             = _mm256_setzero_ps();
 820         fix2             = _mm256_setzero_ps();
 821         fiy2             = _mm256_setzero_ps();
 822         fiz2             = _mm256_setzero_ps();
 823         fix3             = _mm256_setzero_ps();
 824         fiy3             = _mm256_setzero_ps();
 825         fiz3             = _mm256_setzero_ps();
 826
 827         /* Start inner kernel loop */
 828         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 829         {
 830
 831             /* Get j neighbor index, and coordinate index */
 832             jnrA             = jjnr[jidx];
 833             jnrB             = jjnr[jidx+1];
 834             jnrC             = jjnr[jidx+2];
 835             jnrD             = jjnr[jidx+3];
 836             jnrE             = jjnr[jidx+4];
 837             jnrF             = jjnr[jidx+5];
 838             jnrG             = jjnr[jidx+6];
 839             jnrH             = jjnr[jidx+7];
 840             j_coord_offsetA  = DIM*jnrA;
 841             j_coord_offsetB  = DIM*jnrB;
 842             j_coord_offsetC  = DIM*jnrC;
 843             j_coord_offsetD  = DIM*jnrD;
 844             j_coord_offsetE  = DIM*jnrE;
 845             j_coord_offsetF  = DIM*jnrF;
 846             j_coord_offsetG  = DIM*jnrG;
 847             j_coord_offsetH  = DIM*jnrH;
 848
 849             /* load j atom coordinates */
 850             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 851                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 852                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 853                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 854                                                  &jx0,&jy0,&jz0);
 855
 856             /* Calculate displacement vector */
 857             dx00             = _mm256_sub_ps(ix0,jx0);
 858             dy00             = _mm256_sub_ps(iy0,jy0);
 859             dz00             = _mm256_sub_ps(iz0,jz0);
 860             dx10             = _mm256_sub_ps(ix1,jx0);
 861             dy10             = _mm256_sub_ps(iy1,jy0);
 862             dz10             = _mm256_sub_ps(iz1,jz0);
 863             dx20             = _mm256_sub_ps(ix2,jx0);
 864             dy20             = _mm256_sub_ps(iy2,jy0);
 865             dz20             = _mm256_sub_ps(iz2,jz0);
 866             dx30             = _mm256_sub_ps(ix3,jx0);
 867             dy30             = _mm256_sub_ps(iy3,jy0);
 868             dz30             = _mm256_sub_ps(iz3,jz0);
 869
 870             /* Calculate squared distance and things based on it */
 871             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 872             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 873             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 874             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 875
 876             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 877             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 878             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 879
 880             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 881             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 882             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 883             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 884
 885             /* Load parameters for j particles */
 886             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 887                                                                  charge+jnrC+0,charge+jnrD+0,
 888                                                                  charge+jnrE+0,charge+jnrF+0,
 889                                                                  charge+jnrG+0,charge+jnrH+0);
 890             vdwjidx0A        = 2*vdwtype[jnrA+0];
 891             vdwjidx0B        = 2*vdwtype[jnrB+0];
 892             vdwjidx0C        = 2*vdwtype[jnrC+0];
 893             vdwjidx0D        = 2*vdwtype[jnrD+0];
 894             vdwjidx0E        = 2*vdwtype[jnrE+0];
 895             vdwjidx0F        = 2*vdwtype[jnrF+0];
 896             vdwjidx0G        = 2*vdwtype[jnrG+0];
 897             vdwjidx0H        = 2*vdwtype[jnrH+0];
 898
 899             fjx0             = _mm256_setzero_ps();
 900             fjy0             = _mm256_setzero_ps();
 901             fjz0             = _mm256_setzero_ps();
 902
 903             /**************************
 904              * CALCULATE INTERACTIONS *
 905              **************************/
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 909                                             vdwioffsetptr0+vdwjidx0B,
 910                                             vdwioffsetptr0+vdwjidx0C,
 911                                             vdwioffsetptr0+vdwjidx0D,
 912                                             vdwioffsetptr0+vdwjidx0E,
 913                                             vdwioffsetptr0+vdwjidx0F,
 914                                             vdwioffsetptr0+vdwjidx0G,
 915                                             vdwioffsetptr0+vdwjidx0H,
 916                                             &c6_00,&c12_00);
 917
 918             /* LENNARD-JONES DISPERSION/REPULSION */
 919
 920             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 921             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 922
 923             fscal            = fvdw;
 924
 925             /* Calculate temporary vectorial force */
 926             tx               = _mm256_mul_ps(fscal,dx00);
 927             ty               = _mm256_mul_ps(fscal,dy00);
 928             tz               = _mm256_mul_ps(fscal,dz00);
 929
 930             /* Update vectorial force */
 931             fix0             = _mm256_add_ps(fix0,tx);
 932             fiy0             = _mm256_add_ps(fiy0,ty);
 933             fiz0             = _mm256_add_ps(fiz0,tz);
 934
 935             fjx0             = _mm256_add_ps(fjx0,tx);
 936             fjy0             = _mm256_add_ps(fjy0,ty);
 937             fjz0             = _mm256_add_ps(fjz0,tz);
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             qq10             = _mm256_mul_ps(iq1,jq0);
 945
 946             /* REACTION-FIELD ELECTROSTATICS */
 947             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 948
 949             fscal            = felec;
 950
 951             /* Calculate temporary vectorial force */
 952             tx               = _mm256_mul_ps(fscal,dx10);
 953             ty               = _mm256_mul_ps(fscal,dy10);
 954             tz               = _mm256_mul_ps(fscal,dz10);
 955
 956             /* Update vectorial force */
 957             fix1             = _mm256_add_ps(fix1,tx);
 958             fiy1             = _mm256_add_ps(fiy1,ty);
 959             fiz1             = _mm256_add_ps(fiz1,tz);
 960
 961             fjx0             = _mm256_add_ps(fjx0,tx);
 962             fjy0             = _mm256_add_ps(fjy0,ty);
 963             fjz0             = _mm256_add_ps(fjz0,tz);
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq20             = _mm256_mul_ps(iq2,jq0);
 971
 972             /* REACTION-FIELD ELECTROSTATICS */
 973             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 974
 975             fscal            = felec;
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_ps(fscal,dx20);
 979             ty               = _mm256_mul_ps(fscal,dy20);
 980             tz               = _mm256_mul_ps(fscal,dz20);
 981
 982             /* Update vectorial force */
 983             fix2             = _mm256_add_ps(fix2,tx);
 984             fiy2             = _mm256_add_ps(fiy2,ty);
 985             fiz2             = _mm256_add_ps(fiz2,tz);
 986
 987             fjx0             = _mm256_add_ps(fjx0,tx);
 988             fjy0             = _mm256_add_ps(fjy0,ty);
 989             fjz0             = _mm256_add_ps(fjz0,tz);
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             /* Compute parameters for interactions between i and j atoms */
 996             qq30             = _mm256_mul_ps(iq3,jq0);
 997
 998             /* REACTION-FIELD ELECTROSTATICS */
 999             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1000
1001             fscal            = felec;
1002
1003             /* Calculate temporary vectorial force */
1004             tx               = _mm256_mul_ps(fscal,dx30);
1005             ty               = _mm256_mul_ps(fscal,dy30);
1006             tz               = _mm256_mul_ps(fscal,dz30);
1007
1008             /* Update vectorial force */
1009             fix3             = _mm256_add_ps(fix3,tx);
1010             fiy3             = _mm256_add_ps(fiy3,ty);
1011             fiz3             = _mm256_add_ps(fiz3,tz);
1012
1013             fjx0             = _mm256_add_ps(fjx0,tx);
1014             fjy0             = _mm256_add_ps(fjy0,ty);
1015             fjz0             = _mm256_add_ps(fjz0,tz);
1016
1017             fjptrA             = f+j_coord_offsetA;
1018             fjptrB             = f+j_coord_offsetB;
1019             fjptrC             = f+j_coord_offsetC;
1020             fjptrD             = f+j_coord_offsetD;
1021             fjptrE             = f+j_coord_offsetE;
1022             fjptrF             = f+j_coord_offsetF;
1023             fjptrG             = f+j_coord_offsetG;
1024             fjptrH             = f+j_coord_offsetH;
1025
1026             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1027
1028             /* Inner loop uses 111 flops */
1029         }
1030
1031         if(jidx<j_index_end)
1032         {
1033
1034             /* Get j neighbor index, and coordinate index */
1035             jnrlistA         = jjnr[jidx];
1036             jnrlistB         = jjnr[jidx+1];
1037             jnrlistC         = jjnr[jidx+2];
1038             jnrlistD         = jjnr[jidx+3];
1039             jnrlistE         = jjnr[jidx+4];
1040             jnrlistF         = jjnr[jidx+5];
1041             jnrlistG         = jjnr[jidx+6];
1042             jnrlistH         = jjnr[jidx+7];
1043             /* Sign of each element will be negative for non-real atoms.
1044              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1045              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1046              */
1047             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1048                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1049
1050             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1051             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1052             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1053             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1054             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1055             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1056             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1057             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1058             j_coord_offsetA  = DIM*jnrA;
1059             j_coord_offsetB  = DIM*jnrB;
1060             j_coord_offsetC  = DIM*jnrC;
1061             j_coord_offsetD  = DIM*jnrD;
1062             j_coord_offsetE  = DIM*jnrE;
1063             j_coord_offsetF  = DIM*jnrF;
1064             j_coord_offsetG  = DIM*jnrG;
1065             j_coord_offsetH  = DIM*jnrH;
1066
1067             /* load j atom coordinates */
1068             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1069                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1070                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1071                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1072                                                  &jx0,&jy0,&jz0);
1073
1074             /* Calculate displacement vector */
1075             dx00             = _mm256_sub_ps(ix0,jx0);
1076             dy00             = _mm256_sub_ps(iy0,jy0);
1077             dz00             = _mm256_sub_ps(iz0,jz0);
1078             dx10             = _mm256_sub_ps(ix1,jx0);
1079             dy10             = _mm256_sub_ps(iy1,jy0);
1080             dz10             = _mm256_sub_ps(iz1,jz0);
1081             dx20             = _mm256_sub_ps(ix2,jx0);
1082             dy20             = _mm256_sub_ps(iy2,jy0);
1083             dz20             = _mm256_sub_ps(iz2,jz0);
1084             dx30             = _mm256_sub_ps(ix3,jx0);
1085             dy30             = _mm256_sub_ps(iy3,jy0);
1086             dz30             = _mm256_sub_ps(iz3,jz0);
1087
1088             /* Calculate squared distance and things based on it */
1089             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1090             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1091             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1092             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1093
1094             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1095             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1096             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1097
1098             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1099             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1100             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1101             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1102
1103             /* Load parameters for j particles */
1104             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1105                                                                  charge+jnrC+0,charge+jnrD+0,
1106                                                                  charge+jnrE+0,charge+jnrF+0,
1107                                                                  charge+jnrG+0,charge+jnrH+0);
1108             vdwjidx0A        = 2*vdwtype[jnrA+0];
1109             vdwjidx0B        = 2*vdwtype[jnrB+0];
1110             vdwjidx0C        = 2*vdwtype[jnrC+0];
1111             vdwjidx0D        = 2*vdwtype[jnrD+0];
1112             vdwjidx0E        = 2*vdwtype[jnrE+0];
1113             vdwjidx0F        = 2*vdwtype[jnrF+0];
1114             vdwjidx0G        = 2*vdwtype[jnrG+0];
1115             vdwjidx0H        = 2*vdwtype[jnrH+0];
1116
1117             fjx0             = _mm256_setzero_ps();
1118             fjy0             = _mm256_setzero_ps();
1119             fjz0             = _mm256_setzero_ps();
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             /* Compute parameters for interactions between i and j atoms */
1126             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1127                                             vdwioffsetptr0+vdwjidx0B,
1128                                             vdwioffsetptr0+vdwjidx0C,
1129                                             vdwioffsetptr0+vdwjidx0D,
1130                                             vdwioffsetptr0+vdwjidx0E,
1131                                             vdwioffsetptr0+vdwjidx0F,
1132                                             vdwioffsetptr0+vdwjidx0G,
1133                                             vdwioffsetptr0+vdwjidx0H,
1134                                             &c6_00,&c12_00);
1135
1136             /* LENNARD-JONES DISPERSION/REPULSION */
1137
1138             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1139             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1140
1141             fscal            = fvdw;
1142
1143             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1144
1145             /* Calculate temporary vectorial force */
1146             tx               = _mm256_mul_ps(fscal,dx00);
1147             ty               = _mm256_mul_ps(fscal,dy00);
1148             tz               = _mm256_mul_ps(fscal,dz00);
1149
1150             /* Update vectorial force */
1151             fix0             = _mm256_add_ps(fix0,tx);
1152             fiy0             = _mm256_add_ps(fiy0,ty);
1153             fiz0             = _mm256_add_ps(fiz0,tz);
1154
1155             fjx0             = _mm256_add_ps(fjx0,tx);
1156             fjy0             = _mm256_add_ps(fjy0,ty);
1157             fjz0             = _mm256_add_ps(fjz0,tz);
1158
1159             /**************************
1160              * CALCULATE INTERACTIONS *
1161              **************************/
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq10             = _mm256_mul_ps(iq1,jq0);
1165
1166             /* REACTION-FIELD ELECTROSTATICS */
1167             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1168
1169             fscal            = felec;
1170
1171             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1172
1173             /* Calculate temporary vectorial force */
1174             tx               = _mm256_mul_ps(fscal,dx10);
1175             ty               = _mm256_mul_ps(fscal,dy10);
1176             tz               = _mm256_mul_ps(fscal,dz10);
1177
1178             /* Update vectorial force */
1179             fix1             = _mm256_add_ps(fix1,tx);
1180             fiy1             = _mm256_add_ps(fiy1,ty);
1181             fiz1             = _mm256_add_ps(fiz1,tz);
1182
1183             fjx0             = _mm256_add_ps(fjx0,tx);
1184             fjy0             = _mm256_add_ps(fjy0,ty);
1185             fjz0             = _mm256_add_ps(fjz0,tz);
1186
1187             /**************************
1188              * CALCULATE INTERACTIONS *
1189              **************************/
1190
1191             /* Compute parameters for interactions between i and j atoms */
1192             qq20             = _mm256_mul_ps(iq2,jq0);
1193
1194             /* REACTION-FIELD ELECTROSTATICS */
1195             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1200
1201             /* Calculate temporary vectorial force */
1202             tx               = _mm256_mul_ps(fscal,dx20);
1203             ty               = _mm256_mul_ps(fscal,dy20);
1204             tz               = _mm256_mul_ps(fscal,dz20);
1205
1206             /* Update vectorial force */
1207             fix2             = _mm256_add_ps(fix2,tx);
1208             fiy2             = _mm256_add_ps(fiy2,ty);
1209             fiz2             = _mm256_add_ps(fiz2,tz);
1210
1211             fjx0             = _mm256_add_ps(fjx0,tx);
1212             fjy0             = _mm256_add_ps(fjy0,ty);
1213             fjz0             = _mm256_add_ps(fjz0,tz);
1214
1215             /**************************
1216              * CALCULATE INTERACTIONS *
1217              **************************/
1218
1219             /* Compute parameters for interactions between i and j atoms */
1220             qq30             = _mm256_mul_ps(iq3,jq0);
1221
1222             /* REACTION-FIELD ELECTROSTATICS */
1223             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1224
1225             fscal            = felec;
1226
1227             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1228
1229             /* Calculate temporary vectorial force */
1230             tx               = _mm256_mul_ps(fscal,dx30);
1231             ty               = _mm256_mul_ps(fscal,dy30);
1232             tz               = _mm256_mul_ps(fscal,dz30);
1233
1234             /* Update vectorial force */
1235             fix3             = _mm256_add_ps(fix3,tx);
1236             fiy3             = _mm256_add_ps(fiy3,ty);
1237             fiz3             = _mm256_add_ps(fiz3,tz);
1238
1239             fjx0             = _mm256_add_ps(fjx0,tx);
1240             fjy0             = _mm256_add_ps(fjy0,ty);
1241             fjz0             = _mm256_add_ps(fjz0,tz);
1242
1243             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1244             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1245             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1246             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1247             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1248             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1249             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1250             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1251
1252             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1253
1254             /* Inner loop uses 111 flops */
1255         }
1256
1257         /* End of innermost loop */
1258
1259         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1260                                                  f+i_coord_offset,fshift+i_shift_offset);
1261
1262         /* Increment number of inner iterations */
1263         inneriter                  += j_index_end - j_index_start;
1264
1265         /* Outer loop uses 24 flops */
1266     }
1267
1268     /* Increment number of outer iterations */
1269     outeriter        += nri;
1270
1271     /* Update outer/inner flops */
1272
1273     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*111);
1274 }