src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwLJ_GeomW3P1_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_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwLJ_GeomW3P1_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     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           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256           dummy_mask,cutoff_mask;
 108     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 109     __m256           one     = _mm256_set1_ps(1.0);
 110     __m256           two     = _mm256_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     krf              = _mm256_set1_ps(fr->ic->k_rf);
 125     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 126     crf              = _mm256_set1_ps(fr->ic->c_rf);
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 134     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 135     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144     j_coord_offsetE = 0;
 145     j_coord_offsetF = 0;
 146     j_coord_offsetG = 0;
 147     j_coord_offsetH = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 174
 175         fix0             = _mm256_setzero_ps();
 176         fiy0             = _mm256_setzero_ps();
 177         fiz0             = _mm256_setzero_ps();
 178         fix1             = _mm256_setzero_ps();
 179         fiy1             = _mm256_setzero_ps();
 180         fiz1             = _mm256_setzero_ps();
 181         fix2             = _mm256_setzero_ps();
 182         fiy2             = _mm256_setzero_ps();
 183         fiz2             = _mm256_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm256_setzero_ps();
 187         vvdwsum          = _mm256_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             jnrE             = jjnr[jidx+4];
 199             jnrF             = jjnr[jidx+5];
 200             jnrG             = jjnr[jidx+6];
 201             jnrH             = jjnr[jidx+7];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206             j_coord_offsetE  = DIM*jnrE;
 207             j_coord_offsetF  = DIM*jnrF;
 208             j_coord_offsetG  = DIM*jnrG;
 209             j_coord_offsetH  = DIM*jnrH;
 210
 211             /* load j atom coordinates */
 212             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 214                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 215                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 216                                                  &jx0,&jy0,&jz0);
 217
 218             /* Calculate displacement vector */
 219             dx00             = _mm256_sub_ps(ix0,jx0);
 220             dy00             = _mm256_sub_ps(iy0,jy0);
 221             dz00             = _mm256_sub_ps(iz0,jz0);
 222             dx10             = _mm256_sub_ps(ix1,jx0);
 223             dy10             = _mm256_sub_ps(iy1,jy0);
 224             dz10             = _mm256_sub_ps(iz1,jz0);
 225             dx20             = _mm256_sub_ps(ix2,jx0);
 226             dy20             = _mm256_sub_ps(iy2,jy0);
 227             dz20             = _mm256_sub_ps(iz2,jz0);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 231             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 232             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 233
 234             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 235             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 236             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 237
 238             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 239             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 240             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                                  charge+jnrC+0,charge+jnrD+0,
 245                                                                  charge+jnrE+0,charge+jnrF+0,
 246                                                                  charge+jnrG+0,charge+jnrH+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251             vdwjidx0E        = 2*vdwtype[jnrE+0];
 252             vdwjidx0F        = 2*vdwtype[jnrF+0];
 253             vdwjidx0G        = 2*vdwtype[jnrG+0];
 254             vdwjidx0H        = 2*vdwtype[jnrH+0];
 255
 256             fjx0             = _mm256_setzero_ps();
 257             fjy0             = _mm256_setzero_ps();
 258             fjz0             = _mm256_setzero_ps();
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             qq00             = _mm256_mul_ps(iq0,jq0);
 266             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 267                                             vdwioffsetptr0+vdwjidx0B,
 268                                             vdwioffsetptr0+vdwjidx0C,
 269                                             vdwioffsetptr0+vdwjidx0D,
 270                                             vdwioffsetptr0+vdwjidx0E,
 271                                             vdwioffsetptr0+vdwjidx0F,
 272                                             vdwioffsetptr0+vdwjidx0G,
 273                                             vdwioffsetptr0+vdwjidx0H,
 274                                             &c6_00,&c12_00);
 275
 276             /* REACTION-FIELD ELECTROSTATICS */
 277             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 278             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 279
 280             /* LENNARD-JONES DISPERSION/REPULSION */
 281
 282             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 283             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 284             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 285             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 286             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             velecsum         = _mm256_add_ps(velecsum,velec);
 290             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 291
 292             fscal            = _mm256_add_ps(felec,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             /* REACTION-FIELD ELECTROSTATICS */
 316             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 317             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 346             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 347             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 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             fjptrA             = f+j_coord_offsetA;
 369             fjptrB             = f+j_coord_offsetB;
 370             fjptrC             = f+j_coord_offsetC;
 371             fjptrD             = f+j_coord_offsetD;
 372             fjptrE             = f+j_coord_offsetE;
 373             fjptrF             = f+j_coord_offsetF;
 374             fjptrG             = f+j_coord_offsetG;
 375             fjptrH             = f+j_coord_offsetH;
 376
 377             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 378
 379             /* Inner loop uses 111 flops */
 380         }
 381
 382         if(jidx<j_index_end)
 383         {
 384
 385             /* Get j neighbor index, and coordinate index */
 386             jnrlistA         = jjnr[jidx];
 387             jnrlistB         = jjnr[jidx+1];
 388             jnrlistC         = jjnr[jidx+2];
 389             jnrlistD         = jjnr[jidx+3];
 390             jnrlistE         = jjnr[jidx+4];
 391             jnrlistF         = jjnr[jidx+5];
 392             jnrlistG         = jjnr[jidx+6];
 393             jnrlistH         = jjnr[jidx+7];
 394             /* Sign of each element will be negative for non-real atoms.
 395              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 396              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 397              */
 398             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 399                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 400
 401             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 402             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 403             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 404             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 405             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 406             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 407             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 408             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 409             j_coord_offsetA  = DIM*jnrA;
 410             j_coord_offsetB  = DIM*jnrB;
 411             j_coord_offsetC  = DIM*jnrC;
 412             j_coord_offsetD  = DIM*jnrD;
 413             j_coord_offsetE  = DIM*jnrE;
 414             j_coord_offsetF  = DIM*jnrF;
 415             j_coord_offsetG  = DIM*jnrG;
 416             j_coord_offsetH  = DIM*jnrH;
 417
 418             /* load j atom coordinates */
 419             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 420                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 421                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 422                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 423                                                  &jx0,&jy0,&jz0);
 424
 425             /* Calculate displacement vector */
 426             dx00             = _mm256_sub_ps(ix0,jx0);
 427             dy00             = _mm256_sub_ps(iy0,jy0);
 428             dz00             = _mm256_sub_ps(iz0,jz0);
 429             dx10             = _mm256_sub_ps(ix1,jx0);
 430             dy10             = _mm256_sub_ps(iy1,jy0);
 431             dz10             = _mm256_sub_ps(iz1,jz0);
 432             dx20             = _mm256_sub_ps(ix2,jx0);
 433             dy20             = _mm256_sub_ps(iy2,jy0);
 434             dz20             = _mm256_sub_ps(iz2,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 438             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 439             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 440
 441             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 442             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 443             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 444
 445             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 446             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 447             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 448
 449             /* Load parameters for j particles */
 450             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 451                                                                  charge+jnrC+0,charge+jnrD+0,
 452                                                                  charge+jnrE+0,charge+jnrF+0,
 453                                                                  charge+jnrG+0,charge+jnrH+0);
 454             vdwjidx0A        = 2*vdwtype[jnrA+0];
 455             vdwjidx0B        = 2*vdwtype[jnrB+0];
 456             vdwjidx0C        = 2*vdwtype[jnrC+0];
 457             vdwjidx0D        = 2*vdwtype[jnrD+0];
 458             vdwjidx0E        = 2*vdwtype[jnrE+0];
 459             vdwjidx0F        = 2*vdwtype[jnrF+0];
 460             vdwjidx0G        = 2*vdwtype[jnrG+0];
 461             vdwjidx0H        = 2*vdwtype[jnrH+0];
 462
 463             fjx0             = _mm256_setzero_ps();
 464             fjy0             = _mm256_setzero_ps();
 465             fjz0             = _mm256_setzero_ps();
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq00             = _mm256_mul_ps(iq0,jq0);
 473             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 474                                             vdwioffsetptr0+vdwjidx0B,
 475                                             vdwioffsetptr0+vdwjidx0C,
 476                                             vdwioffsetptr0+vdwjidx0D,
 477                                             vdwioffsetptr0+vdwjidx0E,
 478                                             vdwioffsetptr0+vdwjidx0F,
 479                                             vdwioffsetptr0+vdwjidx0G,
 480                                             vdwioffsetptr0+vdwjidx0H,
 481                                             &c6_00,&c12_00);
 482
 483             /* REACTION-FIELD ELECTROSTATICS */
 484             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 485             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 486
 487             /* LENNARD-JONES DISPERSION/REPULSION */
 488
 489             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 490             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 491             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 492             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 493             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 494
 495             /* Update potential sum for this i atom from the interaction with this j atom. */
 496             velec            = _mm256_andnot_ps(dummy_mask,velec);
 497             velecsum         = _mm256_add_ps(velecsum,velec);
 498             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 499             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 500
 501             fscal            = _mm256_add_ps(felec,fvdw);
 502
 503             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm256_mul_ps(fscal,dx00);
 507             ty               = _mm256_mul_ps(fscal,dy00);
 508             tz               = _mm256_mul_ps(fscal,dz00);
 509
 510             /* Update vectorial force */
 511             fix0             = _mm256_add_ps(fix0,tx);
 512             fiy0             = _mm256_add_ps(fiy0,ty);
 513             fiz0             = _mm256_add_ps(fiz0,tz);
 514
 515             fjx0             = _mm256_add_ps(fjx0,tx);
 516             fjy0             = _mm256_add_ps(fjy0,ty);
 517             fjz0             = _mm256_add_ps(fjz0,tz);
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq10             = _mm256_mul_ps(iq1,jq0);
 525
 526             /* REACTION-FIELD ELECTROSTATICS */
 527             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 528             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm256_andnot_ps(dummy_mask,velec);
 532             velecsum         = _mm256_add_ps(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm256_mul_ps(fscal,dx10);
 540             ty               = _mm256_mul_ps(fscal,dy10);
 541             tz               = _mm256_mul_ps(fscal,dz10);
 542
 543             /* Update vectorial force */
 544             fix1             = _mm256_add_ps(fix1,tx);
 545             fiy1             = _mm256_add_ps(fiy1,ty);
 546             fiz1             = _mm256_add_ps(fiz1,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             qq20             = _mm256_mul_ps(iq2,jq0);
 558
 559             /* REACTION-FIELD ELECTROSTATICS */
 560             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 561             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),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,dx20);
 573             ty               = _mm256_mul_ps(fscal,dy20);
 574             tz               = _mm256_mul_ps(fscal,dz20);
 575
 576             /* Update vectorial force */
 577             fix2             = _mm256_add_ps(fix2,tx);
 578             fiy2             = _mm256_add_ps(fiy2,ty);
 579             fiz2             = _mm256_add_ps(fiz2,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 586             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 587             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 588             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 589             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 590             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 591             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 592             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 593
 594             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 595
 596             /* Inner loop uses 111 flops */
 597         }
 598
 599         /* End of innermost loop */
 600
 601         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 602                                                  f+i_coord_offset,fshift+i_shift_offset);
 603
 604         ggid                        = gid[iidx];
 605         /* Update potential energies */
 606         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 607         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 608
 609         /* Increment number of inner iterations */
 610         inneriter                  += j_index_end - j_index_start;
 611
 612         /* Outer loop uses 20 flops */
 613     }
 614
 615     /* Increment number of outer iterations */
 616     outeriter        += nri;
 617
 618     /* Update outer/inner flops */
 619
 620     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*111);
 621 }
 622 /*
 623  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_single
 624  * Electrostatics interaction: ReactionField
 625  * VdW interaction:            LennardJones
 626  * Geometry:                   Water3-Particle
 627  * Calculate force/pot:        Force
 628  */
 629 void
 630 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_single
 631                     (t_nblist                    * gmx_restrict       nlist,
 632                      rvec                        * gmx_restrict          xx,
 633                      rvec                        * gmx_restrict          ff,
 634                      t_forcerec                  * gmx_restrict          fr,
 635                      t_mdatoms                   * gmx_restrict     mdatoms,
 636                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 637                      t_nrnb                      * gmx_restrict        nrnb)
 638 {
 639     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 640      * just 0 for non-waters.
 641      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 642      * jnr indices corresponding to data put in the four positions in the SIMD register.
 643      */
 644     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 645     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 646     int              jnrA,jnrB,jnrC,jnrD;
 647     int              jnrE,jnrF,jnrG,jnrH;
 648     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 649     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 650     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 651     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 652     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 653     real             rcutoff_scalar;
 654     real             *shiftvec,*fshift,*x,*f;
 655     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 656     real             scratch[4*DIM];
 657     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 658     real *           vdwioffsetptr0;
 659     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 660     real *           vdwioffsetptr1;
 661     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 662     real *           vdwioffsetptr2;
 663     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 664     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 665     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 666     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 667     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 668     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 669     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 670     real             *charge;
 671     int              nvdwtype;
 672     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 673     int              *vdwtype;
 674     real             *vdwparam;
 675     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 676     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 677     __m256           dummy_mask,cutoff_mask;
 678     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 679     __m256           one     = _mm256_set1_ps(1.0);
 680     __m256           two     = _mm256_set1_ps(2.0);
 681     x                = xx[0];
 682     f                = ff[0];
 683
 684     nri              = nlist->nri;
 685     iinr             = nlist->iinr;
 686     jindex           = nlist->jindex;
 687     jjnr             = nlist->jjnr;
 688     shiftidx         = nlist->shift;
 689     gid              = nlist->gid;
 690     shiftvec         = fr->shift_vec[0];
 691     fshift           = fr->fshift[0];
 692     facel            = _mm256_set1_ps(fr->epsfac);
 693     charge           = mdatoms->chargeA;
 694     krf              = _mm256_set1_ps(fr->ic->k_rf);
 695     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 696     crf              = _mm256_set1_ps(fr->ic->c_rf);
 697     nvdwtype         = fr->ntype;
 698     vdwparam         = fr->nbfp;
 699     vdwtype          = mdatoms->typeA;
 700
 701     /* Setup water-specific parameters */
 702     inr              = nlist->iinr[0];
 703     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 704     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 705     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 706     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 707
 708     /* Avoid stupid compiler warnings */
 709     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 710     j_coord_offsetA = 0;
 711     j_coord_offsetB = 0;
 712     j_coord_offsetC = 0;
 713     j_coord_offsetD = 0;
 714     j_coord_offsetE = 0;
 715     j_coord_offsetF = 0;
 716     j_coord_offsetG = 0;
 717     j_coord_offsetH = 0;
 718
 719     outeriter        = 0;
 720     inneriter        = 0;
 721
 722     for(iidx=0;iidx<4*DIM;iidx++)
 723     {
 724         scratch[iidx] = 0.0;
 725     }
 726
 727     /* Start outer loop over neighborlists */
 728     for(iidx=0; iidx<nri; iidx++)
 729     {
 730         /* Load shift vector for this list */
 731         i_shift_offset   = DIM*shiftidx[iidx];
 732
 733         /* Load limits for loop over neighbors */
 734         j_index_start    = jindex[iidx];
 735         j_index_end      = jindex[iidx+1];
 736
 737         /* Get outer coordinate index */
 738         inr              = iinr[iidx];
 739         i_coord_offset   = DIM*inr;
 740
 741         /* Load i particle coords and add shift vector */
 742         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 743                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 744
 745         fix0             = _mm256_setzero_ps();
 746         fiy0             = _mm256_setzero_ps();
 747         fiz0             = _mm256_setzero_ps();
 748         fix1             = _mm256_setzero_ps();
 749         fiy1             = _mm256_setzero_ps();
 750         fiz1             = _mm256_setzero_ps();
 751         fix2             = _mm256_setzero_ps();
 752         fiy2             = _mm256_setzero_ps();
 753         fiz2             = _mm256_setzero_ps();
 754
 755         /* Start inner kernel loop */
 756         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 757         {
 758
 759             /* Get j neighbor index, and coordinate index */
 760             jnrA             = jjnr[jidx];
 761             jnrB             = jjnr[jidx+1];
 762             jnrC             = jjnr[jidx+2];
 763             jnrD             = jjnr[jidx+3];
 764             jnrE             = jjnr[jidx+4];
 765             jnrF             = jjnr[jidx+5];
 766             jnrG             = jjnr[jidx+6];
 767             jnrH             = jjnr[jidx+7];
 768             j_coord_offsetA  = DIM*jnrA;
 769             j_coord_offsetB  = DIM*jnrB;
 770             j_coord_offsetC  = DIM*jnrC;
 771             j_coord_offsetD  = DIM*jnrD;
 772             j_coord_offsetE  = DIM*jnrE;
 773             j_coord_offsetF  = DIM*jnrF;
 774             j_coord_offsetG  = DIM*jnrG;
 775             j_coord_offsetH  = DIM*jnrH;
 776
 777             /* load j atom coordinates */
 778             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 779                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 780                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 781                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 782                                                  &jx0,&jy0,&jz0);
 783
 784             /* Calculate displacement vector */
 785             dx00             = _mm256_sub_ps(ix0,jx0);
 786             dy00             = _mm256_sub_ps(iy0,jy0);
 787             dz00             = _mm256_sub_ps(iz0,jz0);
 788             dx10             = _mm256_sub_ps(ix1,jx0);
 789             dy10             = _mm256_sub_ps(iy1,jy0);
 790             dz10             = _mm256_sub_ps(iz1,jz0);
 791             dx20             = _mm256_sub_ps(ix2,jx0);
 792             dy20             = _mm256_sub_ps(iy2,jy0);
 793             dz20             = _mm256_sub_ps(iz2,jz0);
 794
 795             /* Calculate squared distance and things based on it */
 796             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 797             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 798             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 799
 800             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 801             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 802             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 803
 804             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 805             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 806             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 807
 808             /* Load parameters for j particles */
 809             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 810                                                                  charge+jnrC+0,charge+jnrD+0,
 811                                                                  charge+jnrE+0,charge+jnrF+0,
 812                                                                  charge+jnrG+0,charge+jnrH+0);
 813             vdwjidx0A        = 2*vdwtype[jnrA+0];
 814             vdwjidx0B        = 2*vdwtype[jnrB+0];
 815             vdwjidx0C        = 2*vdwtype[jnrC+0];
 816             vdwjidx0D        = 2*vdwtype[jnrD+0];
 817             vdwjidx0E        = 2*vdwtype[jnrE+0];
 818             vdwjidx0F        = 2*vdwtype[jnrF+0];
 819             vdwjidx0G        = 2*vdwtype[jnrG+0];
 820             vdwjidx0H        = 2*vdwtype[jnrH+0];
 821
 822             fjx0             = _mm256_setzero_ps();
 823             fjy0             = _mm256_setzero_ps();
 824             fjz0             = _mm256_setzero_ps();
 825
 826             /**************************
 827              * CALCULATE INTERACTIONS *
 828              **************************/
 829
 830             /* Compute parameters for interactions between i and j atoms */
 831             qq00             = _mm256_mul_ps(iq0,jq0);
 832             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 833                                             vdwioffsetptr0+vdwjidx0B,
 834                                             vdwioffsetptr0+vdwjidx0C,
 835                                             vdwioffsetptr0+vdwjidx0D,
 836                                             vdwioffsetptr0+vdwjidx0E,
 837                                             vdwioffsetptr0+vdwjidx0F,
 838                                             vdwioffsetptr0+vdwjidx0G,
 839                                             vdwioffsetptr0+vdwjidx0H,
 840                                             &c6_00,&c12_00);
 841
 842             /* REACTION-FIELD ELECTROSTATICS */
 843             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 844
 845             /* LENNARD-JONES DISPERSION/REPULSION */
 846
 847             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 848             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 849
 850             fscal            = _mm256_add_ps(felec,fvdw);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm256_mul_ps(fscal,dx00);
 854             ty               = _mm256_mul_ps(fscal,dy00);
 855             tz               = _mm256_mul_ps(fscal,dz00);
 856
 857             /* Update vectorial force */
 858             fix0             = _mm256_add_ps(fix0,tx);
 859             fiy0             = _mm256_add_ps(fiy0,ty);
 860             fiz0             = _mm256_add_ps(fiz0,tz);
 861
 862             fjx0             = _mm256_add_ps(fjx0,tx);
 863             fjy0             = _mm256_add_ps(fjy0,ty);
 864             fjz0             = _mm256_add_ps(fjz0,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq10             = _mm256_mul_ps(iq1,jq0);
 872
 873             /* REACTION-FIELD ELECTROSTATICS */
 874             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 875
 876             fscal            = felec;
 877
 878             /* Calculate temporary vectorial force */
 879             tx               = _mm256_mul_ps(fscal,dx10);
 880             ty               = _mm256_mul_ps(fscal,dy10);
 881             tz               = _mm256_mul_ps(fscal,dz10);
 882
 883             /* Update vectorial force */
 884             fix1             = _mm256_add_ps(fix1,tx);
 885             fiy1             = _mm256_add_ps(fiy1,ty);
 886             fiz1             = _mm256_add_ps(fiz1,tz);
 887
 888             fjx0             = _mm256_add_ps(fjx0,tx);
 889             fjy0             = _mm256_add_ps(fjy0,ty);
 890             fjz0             = _mm256_add_ps(fjz0,tz);
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             /* Compute parameters for interactions between i and j atoms */
 897             qq20             = _mm256_mul_ps(iq2,jq0);
 898
 899             /* REACTION-FIELD ELECTROSTATICS */
 900             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 901
 902             fscal            = felec;
 903
 904             /* Calculate temporary vectorial force */
 905             tx               = _mm256_mul_ps(fscal,dx20);
 906             ty               = _mm256_mul_ps(fscal,dy20);
 907             tz               = _mm256_mul_ps(fscal,dz20);
 908
 909             /* Update vectorial force */
 910             fix2             = _mm256_add_ps(fix2,tx);
 911             fiy2             = _mm256_add_ps(fiy2,ty);
 912             fiz2             = _mm256_add_ps(fiz2,tz);
 913
 914             fjx0             = _mm256_add_ps(fjx0,tx);
 915             fjy0             = _mm256_add_ps(fjy0,ty);
 916             fjz0             = _mm256_add_ps(fjz0,tz);
 917
 918             fjptrA             = f+j_coord_offsetA;
 919             fjptrB             = f+j_coord_offsetB;
 920             fjptrC             = f+j_coord_offsetC;
 921             fjptrD             = f+j_coord_offsetD;
 922             fjptrE             = f+j_coord_offsetE;
 923             fjptrF             = f+j_coord_offsetF;
 924             fjptrG             = f+j_coord_offsetG;
 925             fjptrH             = f+j_coord_offsetH;
 926
 927             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 928
 929             /* Inner loop uses 91 flops */
 930         }
 931
 932         if(jidx<j_index_end)
 933         {
 934
 935             /* Get j neighbor index, and coordinate index */
 936             jnrlistA         = jjnr[jidx];
 937             jnrlistB         = jjnr[jidx+1];
 938             jnrlistC         = jjnr[jidx+2];
 939             jnrlistD         = jjnr[jidx+3];
 940             jnrlistE         = jjnr[jidx+4];
 941             jnrlistF         = jjnr[jidx+5];
 942             jnrlistG         = jjnr[jidx+6];
 943             jnrlistH         = jjnr[jidx+7];
 944             /* Sign of each element will be negative for non-real atoms.
 945              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 946              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 947              */
 948             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 949                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 950
 951             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 952             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 953             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 954             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 955             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 956             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 957             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 958             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 959             j_coord_offsetA  = DIM*jnrA;
 960             j_coord_offsetB  = DIM*jnrB;
 961             j_coord_offsetC  = DIM*jnrC;
 962             j_coord_offsetD  = DIM*jnrD;
 963             j_coord_offsetE  = DIM*jnrE;
 964             j_coord_offsetF  = DIM*jnrF;
 965             j_coord_offsetG  = DIM*jnrG;
 966             j_coord_offsetH  = DIM*jnrH;
 967
 968             /* load j atom coordinates */
 969             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 970                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 971                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 972                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 973                                                  &jx0,&jy0,&jz0);
 974
 975             /* Calculate displacement vector */
 976             dx00             = _mm256_sub_ps(ix0,jx0);
 977             dy00             = _mm256_sub_ps(iy0,jy0);
 978             dz00             = _mm256_sub_ps(iz0,jz0);
 979             dx10             = _mm256_sub_ps(ix1,jx0);
 980             dy10             = _mm256_sub_ps(iy1,jy0);
 981             dz10             = _mm256_sub_ps(iz1,jz0);
 982             dx20             = _mm256_sub_ps(ix2,jx0);
 983             dy20             = _mm256_sub_ps(iy2,jy0);
 984             dz20             = _mm256_sub_ps(iz2,jz0);
 985
 986             /* Calculate squared distance and things based on it */
 987             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 988             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 989             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 990
 991             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 992             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 993             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 994
 995             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 996             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 997             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 998
 999             /* Load parameters for j particles */
1000             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1001                                                                  charge+jnrC+0,charge+jnrD+0,
1002                                                                  charge+jnrE+0,charge+jnrF+0,
1003                                                                  charge+jnrG+0,charge+jnrH+0);
1004             vdwjidx0A        = 2*vdwtype[jnrA+0];
1005             vdwjidx0B        = 2*vdwtype[jnrB+0];
1006             vdwjidx0C        = 2*vdwtype[jnrC+0];
1007             vdwjidx0D        = 2*vdwtype[jnrD+0];
1008             vdwjidx0E        = 2*vdwtype[jnrE+0];
1009             vdwjidx0F        = 2*vdwtype[jnrF+0];
1010             vdwjidx0G        = 2*vdwtype[jnrG+0];
1011             vdwjidx0H        = 2*vdwtype[jnrH+0];
1012
1013             fjx0             = _mm256_setzero_ps();
1014             fjy0             = _mm256_setzero_ps();
1015             fjz0             = _mm256_setzero_ps();
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq00             = _mm256_mul_ps(iq0,jq0);
1023             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1024                                             vdwioffsetptr0+vdwjidx0B,
1025                                             vdwioffsetptr0+vdwjidx0C,
1026                                             vdwioffsetptr0+vdwjidx0D,
1027                                             vdwioffsetptr0+vdwjidx0E,
1028                                             vdwioffsetptr0+vdwjidx0F,
1029                                             vdwioffsetptr0+vdwjidx0G,
1030                                             vdwioffsetptr0+vdwjidx0H,
1031                                             &c6_00,&c12_00);
1032
1033             /* REACTION-FIELD ELECTROSTATICS */
1034             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1035
1036             /* LENNARD-JONES DISPERSION/REPULSION */
1037
1038             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1039             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1040
1041             fscal            = _mm256_add_ps(felec,fvdw);
1042
1043             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm256_mul_ps(fscal,dx00);
1047             ty               = _mm256_mul_ps(fscal,dy00);
1048             tz               = _mm256_mul_ps(fscal,dz00);
1049
1050             /* Update vectorial force */
1051             fix0             = _mm256_add_ps(fix0,tx);
1052             fiy0             = _mm256_add_ps(fiy0,ty);
1053             fiz0             = _mm256_add_ps(fiz0,tz);
1054
1055             fjx0             = _mm256_add_ps(fjx0,tx);
1056             fjy0             = _mm256_add_ps(fjy0,ty);
1057             fjz0             = _mm256_add_ps(fjz0,tz);
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             qq10             = _mm256_mul_ps(iq1,jq0);
1065
1066             /* REACTION-FIELD ELECTROSTATICS */
1067             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1068
1069             fscal            = felec;
1070
1071             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1072
1073             /* Calculate temporary vectorial force */
1074             tx               = _mm256_mul_ps(fscal,dx10);
1075             ty               = _mm256_mul_ps(fscal,dy10);
1076             tz               = _mm256_mul_ps(fscal,dz10);
1077
1078             /* Update vectorial force */
1079             fix1             = _mm256_add_ps(fix1,tx);
1080             fiy1             = _mm256_add_ps(fiy1,ty);
1081             fiz1             = _mm256_add_ps(fiz1,tz);
1082
1083             fjx0             = _mm256_add_ps(fjx0,tx);
1084             fjy0             = _mm256_add_ps(fjy0,ty);
1085             fjz0             = _mm256_add_ps(fjz0,tz);
1086
1087             /**************************
1088              * CALCULATE INTERACTIONS *
1089              **************************/
1090
1091             /* Compute parameters for interactions between i and j atoms */
1092             qq20             = _mm256_mul_ps(iq2,jq0);
1093
1094             /* REACTION-FIELD ELECTROSTATICS */
1095             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1096
1097             fscal            = felec;
1098
1099             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1100
1101             /* Calculate temporary vectorial force */
1102             tx               = _mm256_mul_ps(fscal,dx20);
1103             ty               = _mm256_mul_ps(fscal,dy20);
1104             tz               = _mm256_mul_ps(fscal,dz20);
1105
1106             /* Update vectorial force */
1107             fix2             = _mm256_add_ps(fix2,tx);
1108             fiy2             = _mm256_add_ps(fiy2,ty);
1109             fiz2             = _mm256_add_ps(fiz2,tz);
1110
1111             fjx0             = _mm256_add_ps(fjx0,tx);
1112             fjy0             = _mm256_add_ps(fjy0,ty);
1113             fjz0             = _mm256_add_ps(fjz0,tz);
1114
1115             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1116             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1117             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1118             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1119             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1120             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1121             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1122             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1123
1124             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1125
1126             /* Inner loop uses 91 flops */
1127         }
1128
1129         /* End of innermost loop */
1130
1131         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1132                                                  f+i_coord_offset,fshift+i_shift_offset);
1133
1134         /* Increment number of inner iterations */
1135         inneriter                  += j_index_end - j_index_start;
1136
1137         /* Outer loop uses 18 flops */
1138     }
1139
1140     /* Increment number of outer iterations */
1141     outeriter        += nri;
1142
1143     /* Update outer/inner flops */
1144
1145     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);
1146 }