src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwLJ_GeomW3W3_avx_256_double.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_double 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_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3W3_VF_avx_256_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW3W3_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  95     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  97     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     int              nvdwtype;
 110     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 111     int              *vdwtype;
 112     real             *vdwparam;
 113     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 114     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 115     __m256d          dummy_mask,cutoff_mask;
 116     __m128           tmpmask0,tmpmask1;
 117     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 118     __m256d          one     = _mm256_set1_pd(1.0);
 119     __m256d          two     = _mm256_set1_pd(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm256_set1_pd(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133     nvdwtype         = fr->ntype;
 134     vdwparam         = fr->nbfp;
 135     vdwtype          = mdatoms->typeA;
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 140     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 141     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     jq0              = _mm256_set1_pd(charge[inr+0]);
 145     jq1              = _mm256_set1_pd(charge[inr+1]);
 146     jq2              = _mm256_set1_pd(charge[inr+2]);
 147     vdwjidx0A        = 2*vdwtype[inr+0];
 148     qq00             = _mm256_mul_pd(iq0,jq0);
 149     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 150     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 151     qq01             = _mm256_mul_pd(iq0,jq1);
 152     qq02             = _mm256_mul_pd(iq0,jq2);
 153     qq10             = _mm256_mul_pd(iq1,jq0);
 154     qq11             = _mm256_mul_pd(iq1,jq1);
 155     qq12             = _mm256_mul_pd(iq1,jq2);
 156     qq20             = _mm256_mul_pd(iq2,jq0);
 157     qq21             = _mm256_mul_pd(iq2,jq1);
 158     qq22             = _mm256_mul_pd(iq2,jq2);
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = jnrC = jnrD = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164     j_coord_offsetC = 0;
 165     j_coord_offsetD = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     for(iidx=0;iidx<4*DIM;iidx++)
 171     {
 172         scratch[iidx] = 0.0;
 173     }
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 191                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 192
 193         fix0             = _mm256_setzero_pd();
 194         fiy0             = _mm256_setzero_pd();
 195         fiz0             = _mm256_setzero_pd();
 196         fix1             = _mm256_setzero_pd();
 197         fiy1             = _mm256_setzero_pd();
 198         fiz1             = _mm256_setzero_pd();
 199         fix2             = _mm256_setzero_pd();
 200         fiy2             = _mm256_setzero_pd();
 201         fiz2             = _mm256_setzero_pd();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm256_setzero_pd();
 205         vvdwsum          = _mm256_setzero_pd();
 206
 207         /* Start inner kernel loop */
 208         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrA             = jjnr[jidx];
 213             jnrB             = jjnr[jidx+1];
 214             jnrC             = jjnr[jidx+2];
 215             jnrD             = jjnr[jidx+3];
 216             j_coord_offsetA  = DIM*jnrA;
 217             j_coord_offsetB  = DIM*jnrB;
 218             j_coord_offsetC  = DIM*jnrC;
 219             j_coord_offsetD  = DIM*jnrD;
 220
 221             /* load j atom coordinates */
 222             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 224                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_pd(ix0,jx0);
 228             dy00             = _mm256_sub_pd(iy0,jy0);
 229             dz00             = _mm256_sub_pd(iz0,jz0);
 230             dx01             = _mm256_sub_pd(ix0,jx1);
 231             dy01             = _mm256_sub_pd(iy0,jy1);
 232             dz01             = _mm256_sub_pd(iz0,jz1);
 233             dx02             = _mm256_sub_pd(ix0,jx2);
 234             dy02             = _mm256_sub_pd(iy0,jy2);
 235             dz02             = _mm256_sub_pd(iz0,jz2);
 236             dx10             = _mm256_sub_pd(ix1,jx0);
 237             dy10             = _mm256_sub_pd(iy1,jy0);
 238             dz10             = _mm256_sub_pd(iz1,jz0);
 239             dx11             = _mm256_sub_pd(ix1,jx1);
 240             dy11             = _mm256_sub_pd(iy1,jy1);
 241             dz11             = _mm256_sub_pd(iz1,jz1);
 242             dx12             = _mm256_sub_pd(ix1,jx2);
 243             dy12             = _mm256_sub_pd(iy1,jy2);
 244             dz12             = _mm256_sub_pd(iz1,jz2);
 245             dx20             = _mm256_sub_pd(ix2,jx0);
 246             dy20             = _mm256_sub_pd(iy2,jy0);
 247             dz20             = _mm256_sub_pd(iz2,jz0);
 248             dx21             = _mm256_sub_pd(ix2,jx1);
 249             dy21             = _mm256_sub_pd(iy2,jy1);
 250             dz21             = _mm256_sub_pd(iz2,jz1);
 251             dx22             = _mm256_sub_pd(ix2,jx2);
 252             dy22             = _mm256_sub_pd(iy2,jy2);
 253             dz22             = _mm256_sub_pd(iz2,jz2);
 254
 255             /* Calculate squared distance and things based on it */
 256             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 257             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 258             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 259             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 260             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 261             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 262             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 263             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 264             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 265
 266             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 267             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 268             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 269             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 270             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 271             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 272             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 273             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 274             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 275
 276             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 277             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 278             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 279             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 280             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 281             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 282             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 283             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 284             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 285
 286             fjx0             = _mm256_setzero_pd();
 287             fjy0             = _mm256_setzero_pd();
 288             fjz0             = _mm256_setzero_pd();
 289             fjx1             = _mm256_setzero_pd();
 290             fjy1             = _mm256_setzero_pd();
 291             fjz1             = _mm256_setzero_pd();
 292             fjx2             = _mm256_setzero_pd();
 293             fjy2             = _mm256_setzero_pd();
 294             fjz2             = _mm256_setzero_pd();
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             /* COULOMB ELECTROSTATICS */
 301             velec            = _mm256_mul_pd(qq00,rinv00);
 302             felec            = _mm256_mul_pd(velec,rinvsq00);
 303
 304             /* LENNARD-JONES DISPERSION/REPULSION */
 305
 306             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 307             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 308             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 309             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 310             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm256_add_pd(velecsum,velec);
 314             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 315
 316             fscal            = _mm256_add_pd(felec,fvdw);
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm256_mul_pd(fscal,dx00);
 320             ty               = _mm256_mul_pd(fscal,dy00);
 321             tz               = _mm256_mul_pd(fscal,dz00);
 322
 323             /* Update vectorial force */
 324             fix0             = _mm256_add_pd(fix0,tx);
 325             fiy0             = _mm256_add_pd(fiy0,ty);
 326             fiz0             = _mm256_add_pd(fiz0,tz);
 327
 328             fjx0             = _mm256_add_pd(fjx0,tx);
 329             fjy0             = _mm256_add_pd(fjy0,ty);
 330             fjz0             = _mm256_add_pd(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             /* COULOMB ELECTROSTATICS */
 337             velec            = _mm256_mul_pd(qq01,rinv01);
 338             felec            = _mm256_mul_pd(velec,rinvsq01);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm256_add_pd(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm256_mul_pd(fscal,dx01);
 347             ty               = _mm256_mul_pd(fscal,dy01);
 348             tz               = _mm256_mul_pd(fscal,dz01);
 349
 350             /* Update vectorial force */
 351             fix0             = _mm256_add_pd(fix0,tx);
 352             fiy0             = _mm256_add_pd(fiy0,ty);
 353             fiz0             = _mm256_add_pd(fiz0,tz);
 354
 355             fjx1             = _mm256_add_pd(fjx1,tx);
 356             fjy1             = _mm256_add_pd(fjy1,ty);
 357             fjz1             = _mm256_add_pd(fjz1,tz);
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             /* COULOMB ELECTROSTATICS */
 364             velec            = _mm256_mul_pd(qq02,rinv02);
 365             felec            = _mm256_mul_pd(velec,rinvsq02);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velecsum         = _mm256_add_pd(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm256_mul_pd(fscal,dx02);
 374             ty               = _mm256_mul_pd(fscal,dy02);
 375             tz               = _mm256_mul_pd(fscal,dz02);
 376
 377             /* Update vectorial force */
 378             fix0             = _mm256_add_pd(fix0,tx);
 379             fiy0             = _mm256_add_pd(fiy0,ty);
 380             fiz0             = _mm256_add_pd(fiz0,tz);
 381
 382             fjx2             = _mm256_add_pd(fjx2,tx);
 383             fjy2             = _mm256_add_pd(fjy2,ty);
 384             fjz2             = _mm256_add_pd(fjz2,tz);
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             /* COULOMB ELECTROSTATICS */
 391             velec            = _mm256_mul_pd(qq10,rinv10);
 392             felec            = _mm256_mul_pd(velec,rinvsq10);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velecsum         = _mm256_add_pd(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm256_mul_pd(fscal,dx10);
 401             ty               = _mm256_mul_pd(fscal,dy10);
 402             tz               = _mm256_mul_pd(fscal,dz10);
 403
 404             /* Update vectorial force */
 405             fix1             = _mm256_add_pd(fix1,tx);
 406             fiy1             = _mm256_add_pd(fiy1,ty);
 407             fiz1             = _mm256_add_pd(fiz1,tz);
 408
 409             fjx0             = _mm256_add_pd(fjx0,tx);
 410             fjy0             = _mm256_add_pd(fjy0,ty);
 411             fjz0             = _mm256_add_pd(fjz0,tz);
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             /* COULOMB ELECTROSTATICS */
 418             velec            = _mm256_mul_pd(qq11,rinv11);
 419             felec            = _mm256_mul_pd(velec,rinvsq11);
 420
 421             /* Update potential sum for this i atom from the interaction with this j atom. */
 422             velecsum         = _mm256_add_pd(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm256_mul_pd(fscal,dx11);
 428             ty               = _mm256_mul_pd(fscal,dy11);
 429             tz               = _mm256_mul_pd(fscal,dz11);
 430
 431             /* Update vectorial force */
 432             fix1             = _mm256_add_pd(fix1,tx);
 433             fiy1             = _mm256_add_pd(fiy1,ty);
 434             fiz1             = _mm256_add_pd(fiz1,tz);
 435
 436             fjx1             = _mm256_add_pd(fjx1,tx);
 437             fjy1             = _mm256_add_pd(fjy1,ty);
 438             fjz1             = _mm256_add_pd(fjz1,tz);
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             /* COULOMB ELECTROSTATICS */
 445             velec            = _mm256_mul_pd(qq12,rinv12);
 446             felec            = _mm256_mul_pd(velec,rinvsq12);
 447
 448             /* Update potential sum for this i atom from the interaction with this j atom. */
 449             velecsum         = _mm256_add_pd(velecsum,velec);
 450
 451             fscal            = felec;
 452
 453             /* Calculate temporary vectorial force */
 454             tx               = _mm256_mul_pd(fscal,dx12);
 455             ty               = _mm256_mul_pd(fscal,dy12);
 456             tz               = _mm256_mul_pd(fscal,dz12);
 457
 458             /* Update vectorial force */
 459             fix1             = _mm256_add_pd(fix1,tx);
 460             fiy1             = _mm256_add_pd(fiy1,ty);
 461             fiz1             = _mm256_add_pd(fiz1,tz);
 462
 463             fjx2             = _mm256_add_pd(fjx2,tx);
 464             fjy2             = _mm256_add_pd(fjy2,ty);
 465             fjz2             = _mm256_add_pd(fjz2,tz);
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             /* COULOMB ELECTROSTATICS */
 472             velec            = _mm256_mul_pd(qq20,rinv20);
 473             felec            = _mm256_mul_pd(velec,rinvsq20);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             velecsum         = _mm256_add_pd(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm256_mul_pd(fscal,dx20);
 482             ty               = _mm256_mul_pd(fscal,dy20);
 483             tz               = _mm256_mul_pd(fscal,dz20);
 484
 485             /* Update vectorial force */
 486             fix2             = _mm256_add_pd(fix2,tx);
 487             fiy2             = _mm256_add_pd(fiy2,ty);
 488             fiz2             = _mm256_add_pd(fiz2,tz);
 489
 490             fjx0             = _mm256_add_pd(fjx0,tx);
 491             fjy0             = _mm256_add_pd(fjy0,ty);
 492             fjz0             = _mm256_add_pd(fjz0,tz);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             /* COULOMB ELECTROSTATICS */
 499             velec            = _mm256_mul_pd(qq21,rinv21);
 500             felec            = _mm256_mul_pd(velec,rinvsq21);
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velecsum         = _mm256_add_pd(velecsum,velec);
 504
 505             fscal            = felec;
 506
 507             /* Calculate temporary vectorial force */
 508             tx               = _mm256_mul_pd(fscal,dx21);
 509             ty               = _mm256_mul_pd(fscal,dy21);
 510             tz               = _mm256_mul_pd(fscal,dz21);
 511
 512             /* Update vectorial force */
 513             fix2             = _mm256_add_pd(fix2,tx);
 514             fiy2             = _mm256_add_pd(fiy2,ty);
 515             fiz2             = _mm256_add_pd(fiz2,tz);
 516
 517             fjx1             = _mm256_add_pd(fjx1,tx);
 518             fjy1             = _mm256_add_pd(fjy1,ty);
 519             fjz1             = _mm256_add_pd(fjz1,tz);
 520
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             /* COULOMB ELECTROSTATICS */
 526             velec            = _mm256_mul_pd(qq22,rinv22);
 527             felec            = _mm256_mul_pd(velec,rinvsq22);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velecsum         = _mm256_add_pd(velecsum,velec);
 531
 532             fscal            = felec;
 533
 534             /* Calculate temporary vectorial force */
 535             tx               = _mm256_mul_pd(fscal,dx22);
 536             ty               = _mm256_mul_pd(fscal,dy22);
 537             tz               = _mm256_mul_pd(fscal,dz22);
 538
 539             /* Update vectorial force */
 540             fix2             = _mm256_add_pd(fix2,tx);
 541             fiy2             = _mm256_add_pd(fiy2,ty);
 542             fiz2             = _mm256_add_pd(fiz2,tz);
 543
 544             fjx2             = _mm256_add_pd(fjx2,tx);
 545             fjy2             = _mm256_add_pd(fjy2,ty);
 546             fjz2             = _mm256_add_pd(fjz2,tz);
 547
 548             fjptrA             = f+j_coord_offsetA;
 549             fjptrB             = f+j_coord_offsetB;
 550             fjptrC             = f+j_coord_offsetC;
 551             fjptrD             = f+j_coord_offsetD;
 552
 553             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 554                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 555
 556             /* Inner loop uses 255 flops */
 557         }
 558
 559         if(jidx<j_index_end)
 560         {
 561
 562             /* Get j neighbor index, and coordinate index */
 563             jnrlistA         = jjnr[jidx];
 564             jnrlistB         = jjnr[jidx+1];
 565             jnrlistC         = jjnr[jidx+2];
 566             jnrlistD         = jjnr[jidx+3];
 567             /* Sign of each element will be negative for non-real atoms.
 568              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 569              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 570              */
 571             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 572
 573             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 574             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 575             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 576
 577             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 578             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 579             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 580             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 581             j_coord_offsetA  = DIM*jnrA;
 582             j_coord_offsetB  = DIM*jnrB;
 583             j_coord_offsetC  = DIM*jnrC;
 584             j_coord_offsetD  = DIM*jnrD;
 585
 586             /* load j atom coordinates */
 587             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 588                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 589                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 590
 591             /* Calculate displacement vector */
 592             dx00             = _mm256_sub_pd(ix0,jx0);
 593             dy00             = _mm256_sub_pd(iy0,jy0);
 594             dz00             = _mm256_sub_pd(iz0,jz0);
 595             dx01             = _mm256_sub_pd(ix0,jx1);
 596             dy01             = _mm256_sub_pd(iy0,jy1);
 597             dz01             = _mm256_sub_pd(iz0,jz1);
 598             dx02             = _mm256_sub_pd(ix0,jx2);
 599             dy02             = _mm256_sub_pd(iy0,jy2);
 600             dz02             = _mm256_sub_pd(iz0,jz2);
 601             dx10             = _mm256_sub_pd(ix1,jx0);
 602             dy10             = _mm256_sub_pd(iy1,jy0);
 603             dz10             = _mm256_sub_pd(iz1,jz0);
 604             dx11             = _mm256_sub_pd(ix1,jx1);
 605             dy11             = _mm256_sub_pd(iy1,jy1);
 606             dz11             = _mm256_sub_pd(iz1,jz1);
 607             dx12             = _mm256_sub_pd(ix1,jx2);
 608             dy12             = _mm256_sub_pd(iy1,jy2);
 609             dz12             = _mm256_sub_pd(iz1,jz2);
 610             dx20             = _mm256_sub_pd(ix2,jx0);
 611             dy20             = _mm256_sub_pd(iy2,jy0);
 612             dz20             = _mm256_sub_pd(iz2,jz0);
 613             dx21             = _mm256_sub_pd(ix2,jx1);
 614             dy21             = _mm256_sub_pd(iy2,jy1);
 615             dz21             = _mm256_sub_pd(iz2,jz1);
 616             dx22             = _mm256_sub_pd(ix2,jx2);
 617             dy22             = _mm256_sub_pd(iy2,jy2);
 618             dz22             = _mm256_sub_pd(iz2,jz2);
 619
 620             /* Calculate squared distance and things based on it */
 621             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 622             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 623             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 624             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 625             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 626             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 627             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 628             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 629             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 630
 631             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 632             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 633             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 634             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 635             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 636             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 637             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 638             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 639             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 640
 641             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 642             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
 643             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
 644             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 645             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 646             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 647             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 648             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 649             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 650
 651             fjx0             = _mm256_setzero_pd();
 652             fjy0             = _mm256_setzero_pd();
 653             fjz0             = _mm256_setzero_pd();
 654             fjx1             = _mm256_setzero_pd();
 655             fjy1             = _mm256_setzero_pd();
 656             fjz1             = _mm256_setzero_pd();
 657             fjx2             = _mm256_setzero_pd();
 658             fjy2             = _mm256_setzero_pd();
 659             fjz2             = _mm256_setzero_pd();
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             /* COULOMB ELECTROSTATICS */
 666             velec            = _mm256_mul_pd(qq00,rinv00);
 667             felec            = _mm256_mul_pd(velec,rinvsq00);
 668
 669             /* LENNARD-JONES DISPERSION/REPULSION */
 670
 671             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 672             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 673             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 674             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 675             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 676
 677             /* Update potential sum for this i atom from the interaction with this j atom. */
 678             velec            = _mm256_andnot_pd(dummy_mask,velec);
 679             velecsum         = _mm256_add_pd(velecsum,velec);
 680             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 681             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 682
 683             fscal            = _mm256_add_pd(felec,fvdw);
 684
 685             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 686
 687             /* Calculate temporary vectorial force */
 688             tx               = _mm256_mul_pd(fscal,dx00);
 689             ty               = _mm256_mul_pd(fscal,dy00);
 690             tz               = _mm256_mul_pd(fscal,dz00);
 691
 692             /* Update vectorial force */
 693             fix0             = _mm256_add_pd(fix0,tx);
 694             fiy0             = _mm256_add_pd(fiy0,ty);
 695             fiz0             = _mm256_add_pd(fiz0,tz);
 696
 697             fjx0             = _mm256_add_pd(fjx0,tx);
 698             fjy0             = _mm256_add_pd(fjy0,ty);
 699             fjz0             = _mm256_add_pd(fjz0,tz);
 700
 701             /**************************
 702              * CALCULATE INTERACTIONS *
 703              **************************/
 704
 705             /* COULOMB ELECTROSTATICS */
 706             velec            = _mm256_mul_pd(qq01,rinv01);
 707             felec            = _mm256_mul_pd(velec,rinvsq01);
 708
 709             /* Update potential sum for this i atom from the interaction with this j atom. */
 710             velec            = _mm256_andnot_pd(dummy_mask,velec);
 711             velecsum         = _mm256_add_pd(velecsum,velec);
 712
 713             fscal            = felec;
 714
 715             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm256_mul_pd(fscal,dx01);
 719             ty               = _mm256_mul_pd(fscal,dy01);
 720             tz               = _mm256_mul_pd(fscal,dz01);
 721
 722             /* Update vectorial force */
 723             fix0             = _mm256_add_pd(fix0,tx);
 724             fiy0             = _mm256_add_pd(fiy0,ty);
 725             fiz0             = _mm256_add_pd(fiz0,tz);
 726
 727             fjx1             = _mm256_add_pd(fjx1,tx);
 728             fjy1             = _mm256_add_pd(fjy1,ty);
 729             fjz1             = _mm256_add_pd(fjz1,tz);
 730
 731             /**************************
 732              * CALCULATE INTERACTIONS *
 733              **************************/
 734
 735             /* COULOMB ELECTROSTATICS */
 736             velec            = _mm256_mul_pd(qq02,rinv02);
 737             felec            = _mm256_mul_pd(velec,rinvsq02);
 738
 739             /* Update potential sum for this i atom from the interaction with this j atom. */
 740             velec            = _mm256_andnot_pd(dummy_mask,velec);
 741             velecsum         = _mm256_add_pd(velecsum,velec);
 742
 743             fscal            = felec;
 744
 745             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 746
 747             /* Calculate temporary vectorial force */
 748             tx               = _mm256_mul_pd(fscal,dx02);
 749             ty               = _mm256_mul_pd(fscal,dy02);
 750             tz               = _mm256_mul_pd(fscal,dz02);
 751
 752             /* Update vectorial force */
 753             fix0             = _mm256_add_pd(fix0,tx);
 754             fiy0             = _mm256_add_pd(fiy0,ty);
 755             fiz0             = _mm256_add_pd(fiz0,tz);
 756
 757             fjx2             = _mm256_add_pd(fjx2,tx);
 758             fjy2             = _mm256_add_pd(fjy2,ty);
 759             fjz2             = _mm256_add_pd(fjz2,tz);
 760
 761             /**************************
 762              * CALCULATE INTERACTIONS *
 763              **************************/
 764
 765             /* COULOMB ELECTROSTATICS */
 766             velec            = _mm256_mul_pd(qq10,rinv10);
 767             felec            = _mm256_mul_pd(velec,rinvsq10);
 768
 769             /* Update potential sum for this i atom from the interaction with this j atom. */
 770             velec            = _mm256_andnot_pd(dummy_mask,velec);
 771             velecsum         = _mm256_add_pd(velecsum,velec);
 772
 773             fscal            = felec;
 774
 775             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 776
 777             /* Calculate temporary vectorial force */
 778             tx               = _mm256_mul_pd(fscal,dx10);
 779             ty               = _mm256_mul_pd(fscal,dy10);
 780             tz               = _mm256_mul_pd(fscal,dz10);
 781
 782             /* Update vectorial force */
 783             fix1             = _mm256_add_pd(fix1,tx);
 784             fiy1             = _mm256_add_pd(fiy1,ty);
 785             fiz1             = _mm256_add_pd(fiz1,tz);
 786
 787             fjx0             = _mm256_add_pd(fjx0,tx);
 788             fjy0             = _mm256_add_pd(fjy0,ty);
 789             fjz0             = _mm256_add_pd(fjz0,tz);
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             /* COULOMB ELECTROSTATICS */
 796             velec            = _mm256_mul_pd(qq11,rinv11);
 797             felec            = _mm256_mul_pd(velec,rinvsq11);
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm256_andnot_pd(dummy_mask,velec);
 801             velecsum         = _mm256_add_pd(velecsum,velec);
 802
 803             fscal            = felec;
 804
 805             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 806
 807             /* Calculate temporary vectorial force */
 808             tx               = _mm256_mul_pd(fscal,dx11);
 809             ty               = _mm256_mul_pd(fscal,dy11);
 810             tz               = _mm256_mul_pd(fscal,dz11);
 811
 812             /* Update vectorial force */
 813             fix1             = _mm256_add_pd(fix1,tx);
 814             fiy1             = _mm256_add_pd(fiy1,ty);
 815             fiz1             = _mm256_add_pd(fiz1,tz);
 816
 817             fjx1             = _mm256_add_pd(fjx1,tx);
 818             fjy1             = _mm256_add_pd(fjy1,ty);
 819             fjz1             = _mm256_add_pd(fjz1,tz);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             /* COULOMB ELECTROSTATICS */
 826             velec            = _mm256_mul_pd(qq12,rinv12);
 827             felec            = _mm256_mul_pd(velec,rinvsq12);
 828
 829             /* Update potential sum for this i atom from the interaction with this j atom. */
 830             velec            = _mm256_andnot_pd(dummy_mask,velec);
 831             velecsum         = _mm256_add_pd(velecsum,velec);
 832
 833             fscal            = felec;
 834
 835             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 836
 837             /* Calculate temporary vectorial force */
 838             tx               = _mm256_mul_pd(fscal,dx12);
 839             ty               = _mm256_mul_pd(fscal,dy12);
 840             tz               = _mm256_mul_pd(fscal,dz12);
 841
 842             /* Update vectorial force */
 843             fix1             = _mm256_add_pd(fix1,tx);
 844             fiy1             = _mm256_add_pd(fiy1,ty);
 845             fiz1             = _mm256_add_pd(fiz1,tz);
 846
 847             fjx2             = _mm256_add_pd(fjx2,tx);
 848             fjy2             = _mm256_add_pd(fjy2,ty);
 849             fjz2             = _mm256_add_pd(fjz2,tz);
 850
 851             /**************************
 852              * CALCULATE INTERACTIONS *
 853              **************************/
 854
 855             /* COULOMB ELECTROSTATICS */
 856             velec            = _mm256_mul_pd(qq20,rinv20);
 857             felec            = _mm256_mul_pd(velec,rinvsq20);
 858
 859             /* Update potential sum for this i atom from the interaction with this j atom. */
 860             velec            = _mm256_andnot_pd(dummy_mask,velec);
 861             velecsum         = _mm256_add_pd(velecsum,velec);
 862
 863             fscal            = felec;
 864
 865             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 866
 867             /* Calculate temporary vectorial force */
 868             tx               = _mm256_mul_pd(fscal,dx20);
 869             ty               = _mm256_mul_pd(fscal,dy20);
 870             tz               = _mm256_mul_pd(fscal,dz20);
 871
 872             /* Update vectorial force */
 873             fix2             = _mm256_add_pd(fix2,tx);
 874             fiy2             = _mm256_add_pd(fiy2,ty);
 875             fiz2             = _mm256_add_pd(fiz2,tz);
 876
 877             fjx0             = _mm256_add_pd(fjx0,tx);
 878             fjy0             = _mm256_add_pd(fjy0,ty);
 879             fjz0             = _mm256_add_pd(fjz0,tz);
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             /* COULOMB ELECTROSTATICS */
 886             velec            = _mm256_mul_pd(qq21,rinv21);
 887             felec            = _mm256_mul_pd(velec,rinvsq21);
 888
 889             /* Update potential sum for this i atom from the interaction with this j atom. */
 890             velec            = _mm256_andnot_pd(dummy_mask,velec);
 891             velecsum         = _mm256_add_pd(velecsum,velec);
 892
 893             fscal            = felec;
 894
 895             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm256_mul_pd(fscal,dx21);
 899             ty               = _mm256_mul_pd(fscal,dy21);
 900             tz               = _mm256_mul_pd(fscal,dz21);
 901
 902             /* Update vectorial force */
 903             fix2             = _mm256_add_pd(fix2,tx);
 904             fiy2             = _mm256_add_pd(fiy2,ty);
 905             fiz2             = _mm256_add_pd(fiz2,tz);
 906
 907             fjx1             = _mm256_add_pd(fjx1,tx);
 908             fjy1             = _mm256_add_pd(fjy1,ty);
 909             fjz1             = _mm256_add_pd(fjz1,tz);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             /* COULOMB ELECTROSTATICS */
 916             velec            = _mm256_mul_pd(qq22,rinv22);
 917             felec            = _mm256_mul_pd(velec,rinvsq22);
 918
 919             /* Update potential sum for this i atom from the interaction with this j atom. */
 920             velec            = _mm256_andnot_pd(dummy_mask,velec);
 921             velecsum         = _mm256_add_pd(velecsum,velec);
 922
 923             fscal            = felec;
 924
 925             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 926
 927             /* Calculate temporary vectorial force */
 928             tx               = _mm256_mul_pd(fscal,dx22);
 929             ty               = _mm256_mul_pd(fscal,dy22);
 930             tz               = _mm256_mul_pd(fscal,dz22);
 931
 932             /* Update vectorial force */
 933             fix2             = _mm256_add_pd(fix2,tx);
 934             fiy2             = _mm256_add_pd(fiy2,ty);
 935             fiz2             = _mm256_add_pd(fiz2,tz);
 936
 937             fjx2             = _mm256_add_pd(fjx2,tx);
 938             fjy2             = _mm256_add_pd(fjy2,ty);
 939             fjz2             = _mm256_add_pd(fjz2,tz);
 940
 941             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 942             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 943             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 944             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 945
 946             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 947                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 948
 949             /* Inner loop uses 255 flops */
 950         }
 951
 952         /* End of innermost loop */
 953
 954         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 955                                                  f+i_coord_offset,fshift+i_shift_offset);
 956
 957         ggid                        = gid[iidx];
 958         /* Update potential energies */
 959         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 960         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 961
 962         /* Increment number of inner iterations */
 963         inneriter                  += j_index_end - j_index_start;
 964
 965         /* Outer loop uses 20 flops */
 966     }
 967
 968     /* Increment number of outer iterations */
 969     outeriter        += nri;
 970
 971     /* Update outer/inner flops */
 972
 973     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*255);
 974 }
 975 /*
 976  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3W3_F_avx_256_double
 977  * Electrostatics interaction: Coulomb
 978  * VdW interaction:            LennardJones
 979  * Geometry:                   Water3-Water3
 980  * Calculate force/pot:        Force
 981  */
 982 void
 983 nb_kernel_ElecCoul_VdwLJ_GeomW3W3_F_avx_256_double
 984                     (t_nblist                    * gmx_restrict       nlist,
 985                      rvec                        * gmx_restrict          xx,
 986                      rvec                        * gmx_restrict          ff,
 987                      t_forcerec                  * gmx_restrict          fr,
 988                      t_mdatoms                   * gmx_restrict     mdatoms,
 989                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 990                      t_nrnb                      * gmx_restrict        nrnb)
 991 {
 992     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 993      * just 0 for non-waters.
 994      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 995      * jnr indices corresponding to data put in the four positions in the SIMD register.
 996      */
 997     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 998     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 999     int              jnrA,jnrB,jnrC,jnrD;
1000     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1001     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1002     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1003     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1004     real             rcutoff_scalar;
1005     real             *shiftvec,*fshift,*x,*f;
1006     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1007     real             scratch[4*DIM];
1008     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1009     real *           vdwioffsetptr0;
1010     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1011     real *           vdwioffsetptr1;
1012     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1013     real *           vdwioffsetptr2;
1014     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1015     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1016     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1017     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1018     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1019     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1020     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1021     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1022     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1023     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1024     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1025     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1026     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1027     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1028     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1029     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1030     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1031     real             *charge;
1032     int              nvdwtype;
1033     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1034     int              *vdwtype;
1035     real             *vdwparam;
1036     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1037     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1038     __m256d          dummy_mask,cutoff_mask;
1039     __m128           tmpmask0,tmpmask1;
1040     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1041     __m256d          one     = _mm256_set1_pd(1.0);
1042     __m256d          two     = _mm256_set1_pd(2.0);
1043     x                = xx[0];
1044     f                = ff[0];
1045
1046     nri              = nlist->nri;
1047     iinr             = nlist->iinr;
1048     jindex           = nlist->jindex;
1049     jjnr             = nlist->jjnr;
1050     shiftidx         = nlist->shift;
1051     gid              = nlist->gid;
1052     shiftvec         = fr->shift_vec[0];
1053     fshift           = fr->fshift[0];
1054     facel            = _mm256_set1_pd(fr->epsfac);
1055     charge           = mdatoms->chargeA;
1056     nvdwtype         = fr->ntype;
1057     vdwparam         = fr->nbfp;
1058     vdwtype          = mdatoms->typeA;
1059
1060     /* Setup water-specific parameters */
1061     inr              = nlist->iinr[0];
1062     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1063     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1064     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1065     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1066
1067     jq0              = _mm256_set1_pd(charge[inr+0]);
1068     jq1              = _mm256_set1_pd(charge[inr+1]);
1069     jq2              = _mm256_set1_pd(charge[inr+2]);
1070     vdwjidx0A        = 2*vdwtype[inr+0];
1071     qq00             = _mm256_mul_pd(iq0,jq0);
1072     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1073     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1074     qq01             = _mm256_mul_pd(iq0,jq1);
1075     qq02             = _mm256_mul_pd(iq0,jq2);
1076     qq10             = _mm256_mul_pd(iq1,jq0);
1077     qq11             = _mm256_mul_pd(iq1,jq1);
1078     qq12             = _mm256_mul_pd(iq1,jq2);
1079     qq20             = _mm256_mul_pd(iq2,jq0);
1080     qq21             = _mm256_mul_pd(iq2,jq1);
1081     qq22             = _mm256_mul_pd(iq2,jq2);
1082
1083     /* Avoid stupid compiler warnings */
1084     jnrA = jnrB = jnrC = jnrD = 0;
1085     j_coord_offsetA = 0;
1086     j_coord_offsetB = 0;
1087     j_coord_offsetC = 0;
1088     j_coord_offsetD = 0;
1089
1090     outeriter        = 0;
1091     inneriter        = 0;
1092
1093     for(iidx=0;iidx<4*DIM;iidx++)
1094     {
1095         scratch[iidx] = 0.0;
1096     }
1097
1098     /* Start outer loop over neighborlists */
1099     for(iidx=0; iidx<nri; iidx++)
1100     {
1101         /* Load shift vector for this list */
1102         i_shift_offset   = DIM*shiftidx[iidx];
1103
1104         /* Load limits for loop over neighbors */
1105         j_index_start    = jindex[iidx];
1106         j_index_end      = jindex[iidx+1];
1107
1108         /* Get outer coordinate index */
1109         inr              = iinr[iidx];
1110         i_coord_offset   = DIM*inr;
1111
1112         /* Load i particle coords and add shift vector */
1113         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1114                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1115
1116         fix0             = _mm256_setzero_pd();
1117         fiy0             = _mm256_setzero_pd();
1118         fiz0             = _mm256_setzero_pd();
1119         fix1             = _mm256_setzero_pd();
1120         fiy1             = _mm256_setzero_pd();
1121         fiz1             = _mm256_setzero_pd();
1122         fix2             = _mm256_setzero_pd();
1123         fiy2             = _mm256_setzero_pd();
1124         fiz2             = _mm256_setzero_pd();
1125
1126         /* Start inner kernel loop */
1127         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1128         {
1129
1130             /* Get j neighbor index, and coordinate index */
1131             jnrA             = jjnr[jidx];
1132             jnrB             = jjnr[jidx+1];
1133             jnrC             = jjnr[jidx+2];
1134             jnrD             = jjnr[jidx+3];
1135             j_coord_offsetA  = DIM*jnrA;
1136             j_coord_offsetB  = DIM*jnrB;
1137             j_coord_offsetC  = DIM*jnrC;
1138             j_coord_offsetD  = DIM*jnrD;
1139
1140             /* load j atom coordinates */
1141             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1142                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1143                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1144
1145             /* Calculate displacement vector */
1146             dx00             = _mm256_sub_pd(ix0,jx0);
1147             dy00             = _mm256_sub_pd(iy0,jy0);
1148             dz00             = _mm256_sub_pd(iz0,jz0);
1149             dx01             = _mm256_sub_pd(ix0,jx1);
1150             dy01             = _mm256_sub_pd(iy0,jy1);
1151             dz01             = _mm256_sub_pd(iz0,jz1);
1152             dx02             = _mm256_sub_pd(ix0,jx2);
1153             dy02             = _mm256_sub_pd(iy0,jy2);
1154             dz02             = _mm256_sub_pd(iz0,jz2);
1155             dx10             = _mm256_sub_pd(ix1,jx0);
1156             dy10             = _mm256_sub_pd(iy1,jy0);
1157             dz10             = _mm256_sub_pd(iz1,jz0);
1158             dx11             = _mm256_sub_pd(ix1,jx1);
1159             dy11             = _mm256_sub_pd(iy1,jy1);
1160             dz11             = _mm256_sub_pd(iz1,jz1);
1161             dx12             = _mm256_sub_pd(ix1,jx2);
1162             dy12             = _mm256_sub_pd(iy1,jy2);
1163             dz12             = _mm256_sub_pd(iz1,jz2);
1164             dx20             = _mm256_sub_pd(ix2,jx0);
1165             dy20             = _mm256_sub_pd(iy2,jy0);
1166             dz20             = _mm256_sub_pd(iz2,jz0);
1167             dx21             = _mm256_sub_pd(ix2,jx1);
1168             dy21             = _mm256_sub_pd(iy2,jy1);
1169             dz21             = _mm256_sub_pd(iz2,jz1);
1170             dx22             = _mm256_sub_pd(ix2,jx2);
1171             dy22             = _mm256_sub_pd(iy2,jy2);
1172             dz22             = _mm256_sub_pd(iz2,jz2);
1173
1174             /* Calculate squared distance and things based on it */
1175             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1176             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1177             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1178             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1179             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1180             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1181             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1182             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1183             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1184
1185             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1186             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1187             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1188             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1189             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1190             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1191             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1192             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1193             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1194
1195             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1196             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1197             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1198             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1199             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1200             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1201             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1202             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1203             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1204
1205             fjx0             = _mm256_setzero_pd();
1206             fjy0             = _mm256_setzero_pd();
1207             fjz0             = _mm256_setzero_pd();
1208             fjx1             = _mm256_setzero_pd();
1209             fjy1             = _mm256_setzero_pd();
1210             fjz1             = _mm256_setzero_pd();
1211             fjx2             = _mm256_setzero_pd();
1212             fjy2             = _mm256_setzero_pd();
1213             fjz2             = _mm256_setzero_pd();
1214
1215             /**************************
1216              * CALCULATE INTERACTIONS *
1217              **************************/
1218
1219             /* COULOMB ELECTROSTATICS */
1220             velec            = _mm256_mul_pd(qq00,rinv00);
1221             felec            = _mm256_mul_pd(velec,rinvsq00);
1222
1223             /* LENNARD-JONES DISPERSION/REPULSION */
1224
1225             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1226             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1227
1228             fscal            = _mm256_add_pd(felec,fvdw);
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm256_mul_pd(fscal,dx00);
1232             ty               = _mm256_mul_pd(fscal,dy00);
1233             tz               = _mm256_mul_pd(fscal,dz00);
1234
1235             /* Update vectorial force */
1236             fix0             = _mm256_add_pd(fix0,tx);
1237             fiy0             = _mm256_add_pd(fiy0,ty);
1238             fiz0             = _mm256_add_pd(fiz0,tz);
1239
1240             fjx0             = _mm256_add_pd(fjx0,tx);
1241             fjy0             = _mm256_add_pd(fjy0,ty);
1242             fjz0             = _mm256_add_pd(fjz0,tz);
1243
1244             /**************************
1245              * CALCULATE INTERACTIONS *
1246              **************************/
1247
1248             /* COULOMB ELECTROSTATICS */
1249             velec            = _mm256_mul_pd(qq01,rinv01);
1250             felec            = _mm256_mul_pd(velec,rinvsq01);
1251
1252             fscal            = felec;
1253
1254             /* Calculate temporary vectorial force */
1255             tx               = _mm256_mul_pd(fscal,dx01);
1256             ty               = _mm256_mul_pd(fscal,dy01);
1257             tz               = _mm256_mul_pd(fscal,dz01);
1258
1259             /* Update vectorial force */
1260             fix0             = _mm256_add_pd(fix0,tx);
1261             fiy0             = _mm256_add_pd(fiy0,ty);
1262             fiz0             = _mm256_add_pd(fiz0,tz);
1263
1264             fjx1             = _mm256_add_pd(fjx1,tx);
1265             fjy1             = _mm256_add_pd(fjy1,ty);
1266             fjz1             = _mm256_add_pd(fjz1,tz);
1267
1268             /**************************
1269              * CALCULATE INTERACTIONS *
1270              **************************/
1271
1272             /* COULOMB ELECTROSTATICS */
1273             velec            = _mm256_mul_pd(qq02,rinv02);
1274             felec            = _mm256_mul_pd(velec,rinvsq02);
1275
1276             fscal            = felec;
1277
1278             /* Calculate temporary vectorial force */
1279             tx               = _mm256_mul_pd(fscal,dx02);
1280             ty               = _mm256_mul_pd(fscal,dy02);
1281             tz               = _mm256_mul_pd(fscal,dz02);
1282
1283             /* Update vectorial force */
1284             fix0             = _mm256_add_pd(fix0,tx);
1285             fiy0             = _mm256_add_pd(fiy0,ty);
1286             fiz0             = _mm256_add_pd(fiz0,tz);
1287
1288             fjx2             = _mm256_add_pd(fjx2,tx);
1289             fjy2             = _mm256_add_pd(fjy2,ty);
1290             fjz2             = _mm256_add_pd(fjz2,tz);
1291
1292             /**************************
1293              * CALCULATE INTERACTIONS *
1294              **************************/
1295
1296             /* COULOMB ELECTROSTATICS */
1297             velec            = _mm256_mul_pd(qq10,rinv10);
1298             felec            = _mm256_mul_pd(velec,rinvsq10);
1299
1300             fscal            = felec;
1301
1302             /* Calculate temporary vectorial force */
1303             tx               = _mm256_mul_pd(fscal,dx10);
1304             ty               = _mm256_mul_pd(fscal,dy10);
1305             tz               = _mm256_mul_pd(fscal,dz10);
1306
1307             /* Update vectorial force */
1308             fix1             = _mm256_add_pd(fix1,tx);
1309             fiy1             = _mm256_add_pd(fiy1,ty);
1310             fiz1             = _mm256_add_pd(fiz1,tz);
1311
1312             fjx0             = _mm256_add_pd(fjx0,tx);
1313             fjy0             = _mm256_add_pd(fjy0,ty);
1314             fjz0             = _mm256_add_pd(fjz0,tz);
1315
1316             /**************************
1317              * CALCULATE INTERACTIONS *
1318              **************************/
1319
1320             /* COULOMB ELECTROSTATICS */
1321             velec            = _mm256_mul_pd(qq11,rinv11);
1322             felec            = _mm256_mul_pd(velec,rinvsq11);
1323
1324             fscal            = felec;
1325
1326             /* Calculate temporary vectorial force */
1327             tx               = _mm256_mul_pd(fscal,dx11);
1328             ty               = _mm256_mul_pd(fscal,dy11);
1329             tz               = _mm256_mul_pd(fscal,dz11);
1330
1331             /* Update vectorial force */
1332             fix1             = _mm256_add_pd(fix1,tx);
1333             fiy1             = _mm256_add_pd(fiy1,ty);
1334             fiz1             = _mm256_add_pd(fiz1,tz);
1335
1336             fjx1             = _mm256_add_pd(fjx1,tx);
1337             fjy1             = _mm256_add_pd(fjy1,ty);
1338             fjz1             = _mm256_add_pd(fjz1,tz);
1339
1340             /**************************
1341              * CALCULATE INTERACTIONS *
1342              **************************/
1343
1344             /* COULOMB ELECTROSTATICS */
1345             velec            = _mm256_mul_pd(qq12,rinv12);
1346             felec            = _mm256_mul_pd(velec,rinvsq12);
1347
1348             fscal            = felec;
1349
1350             /* Calculate temporary vectorial force */
1351             tx               = _mm256_mul_pd(fscal,dx12);
1352             ty               = _mm256_mul_pd(fscal,dy12);
1353             tz               = _mm256_mul_pd(fscal,dz12);
1354
1355             /* Update vectorial force */
1356             fix1             = _mm256_add_pd(fix1,tx);
1357             fiy1             = _mm256_add_pd(fiy1,ty);
1358             fiz1             = _mm256_add_pd(fiz1,tz);
1359
1360             fjx2             = _mm256_add_pd(fjx2,tx);
1361             fjy2             = _mm256_add_pd(fjy2,ty);
1362             fjz2             = _mm256_add_pd(fjz2,tz);
1363
1364             /**************************
1365              * CALCULATE INTERACTIONS *
1366              **************************/
1367
1368             /* COULOMB ELECTROSTATICS */
1369             velec            = _mm256_mul_pd(qq20,rinv20);
1370             felec            = _mm256_mul_pd(velec,rinvsq20);
1371
1372             fscal            = felec;
1373
1374             /* Calculate temporary vectorial force */
1375             tx               = _mm256_mul_pd(fscal,dx20);
1376             ty               = _mm256_mul_pd(fscal,dy20);
1377             tz               = _mm256_mul_pd(fscal,dz20);
1378
1379             /* Update vectorial force */
1380             fix2             = _mm256_add_pd(fix2,tx);
1381             fiy2             = _mm256_add_pd(fiy2,ty);
1382             fiz2             = _mm256_add_pd(fiz2,tz);
1383
1384             fjx0             = _mm256_add_pd(fjx0,tx);
1385             fjy0             = _mm256_add_pd(fjy0,ty);
1386             fjz0             = _mm256_add_pd(fjz0,tz);
1387
1388             /**************************
1389              * CALCULATE INTERACTIONS *
1390              **************************/
1391
1392             /* COULOMB ELECTROSTATICS */
1393             velec            = _mm256_mul_pd(qq21,rinv21);
1394             felec            = _mm256_mul_pd(velec,rinvsq21);
1395
1396             fscal            = felec;
1397
1398             /* Calculate temporary vectorial force */
1399             tx               = _mm256_mul_pd(fscal,dx21);
1400             ty               = _mm256_mul_pd(fscal,dy21);
1401             tz               = _mm256_mul_pd(fscal,dz21);
1402
1403             /* Update vectorial force */
1404             fix2             = _mm256_add_pd(fix2,tx);
1405             fiy2             = _mm256_add_pd(fiy2,ty);
1406             fiz2             = _mm256_add_pd(fiz2,tz);
1407
1408             fjx1             = _mm256_add_pd(fjx1,tx);
1409             fjy1             = _mm256_add_pd(fjy1,ty);
1410             fjz1             = _mm256_add_pd(fjz1,tz);
1411
1412             /**************************
1413              * CALCULATE INTERACTIONS *
1414              **************************/
1415
1416             /* COULOMB ELECTROSTATICS */
1417             velec            = _mm256_mul_pd(qq22,rinv22);
1418             felec            = _mm256_mul_pd(velec,rinvsq22);
1419
1420             fscal            = felec;
1421
1422             /* Calculate temporary vectorial force */
1423             tx               = _mm256_mul_pd(fscal,dx22);
1424             ty               = _mm256_mul_pd(fscal,dy22);
1425             tz               = _mm256_mul_pd(fscal,dz22);
1426
1427             /* Update vectorial force */
1428             fix2             = _mm256_add_pd(fix2,tx);
1429             fiy2             = _mm256_add_pd(fiy2,ty);
1430             fiz2             = _mm256_add_pd(fiz2,tz);
1431
1432             fjx2             = _mm256_add_pd(fjx2,tx);
1433             fjy2             = _mm256_add_pd(fjy2,ty);
1434             fjz2             = _mm256_add_pd(fjz2,tz);
1435
1436             fjptrA             = f+j_coord_offsetA;
1437             fjptrB             = f+j_coord_offsetB;
1438             fjptrC             = f+j_coord_offsetC;
1439             fjptrD             = f+j_coord_offsetD;
1440
1441             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1442                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1443
1444             /* Inner loop uses 241 flops */
1445         }
1446
1447         if(jidx<j_index_end)
1448         {
1449
1450             /* Get j neighbor index, and coordinate index */
1451             jnrlistA         = jjnr[jidx];
1452             jnrlistB         = jjnr[jidx+1];
1453             jnrlistC         = jjnr[jidx+2];
1454             jnrlistD         = jjnr[jidx+3];
1455             /* Sign of each element will be negative for non-real atoms.
1456              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1457              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1458              */
1459             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1460
1461             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1462             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1463             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1464
1465             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1466             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1467             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1468             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1469             j_coord_offsetA  = DIM*jnrA;
1470             j_coord_offsetB  = DIM*jnrB;
1471             j_coord_offsetC  = DIM*jnrC;
1472             j_coord_offsetD  = DIM*jnrD;
1473
1474             /* load j atom coordinates */
1475             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1476                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1477                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1478
1479             /* Calculate displacement vector */
1480             dx00             = _mm256_sub_pd(ix0,jx0);
1481             dy00             = _mm256_sub_pd(iy0,jy0);
1482             dz00             = _mm256_sub_pd(iz0,jz0);
1483             dx01             = _mm256_sub_pd(ix0,jx1);
1484             dy01             = _mm256_sub_pd(iy0,jy1);
1485             dz01             = _mm256_sub_pd(iz0,jz1);
1486             dx02             = _mm256_sub_pd(ix0,jx2);
1487             dy02             = _mm256_sub_pd(iy0,jy2);
1488             dz02             = _mm256_sub_pd(iz0,jz2);
1489             dx10             = _mm256_sub_pd(ix1,jx0);
1490             dy10             = _mm256_sub_pd(iy1,jy0);
1491             dz10             = _mm256_sub_pd(iz1,jz0);
1492             dx11             = _mm256_sub_pd(ix1,jx1);
1493             dy11             = _mm256_sub_pd(iy1,jy1);
1494             dz11             = _mm256_sub_pd(iz1,jz1);
1495             dx12             = _mm256_sub_pd(ix1,jx2);
1496             dy12             = _mm256_sub_pd(iy1,jy2);
1497             dz12             = _mm256_sub_pd(iz1,jz2);
1498             dx20             = _mm256_sub_pd(ix2,jx0);
1499             dy20             = _mm256_sub_pd(iy2,jy0);
1500             dz20             = _mm256_sub_pd(iz2,jz0);
1501             dx21             = _mm256_sub_pd(ix2,jx1);
1502             dy21             = _mm256_sub_pd(iy2,jy1);
1503             dz21             = _mm256_sub_pd(iz2,jz1);
1504             dx22             = _mm256_sub_pd(ix2,jx2);
1505             dy22             = _mm256_sub_pd(iy2,jy2);
1506             dz22             = _mm256_sub_pd(iz2,jz2);
1507
1508             /* Calculate squared distance and things based on it */
1509             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1510             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1511             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1512             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1513             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1514             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1515             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1516             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1517             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1518
1519             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1520             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1521             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1522             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1523             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1524             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1525             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1526             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1527             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1528
1529             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1530             rinvsq01         = _mm256_mul_pd(rinv01,rinv01);
1531             rinvsq02         = _mm256_mul_pd(rinv02,rinv02);
1532             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1533             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1534             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1535             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1536             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1537             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1538
1539             fjx0             = _mm256_setzero_pd();
1540             fjy0             = _mm256_setzero_pd();
1541             fjz0             = _mm256_setzero_pd();
1542             fjx1             = _mm256_setzero_pd();
1543             fjy1             = _mm256_setzero_pd();
1544             fjz1             = _mm256_setzero_pd();
1545             fjx2             = _mm256_setzero_pd();
1546             fjy2             = _mm256_setzero_pd();
1547             fjz2             = _mm256_setzero_pd();
1548
1549             /**************************
1550              * CALCULATE INTERACTIONS *
1551              **************************/
1552
1553             /* COULOMB ELECTROSTATICS */
1554             velec            = _mm256_mul_pd(qq00,rinv00);
1555             felec            = _mm256_mul_pd(velec,rinvsq00);
1556
1557             /* LENNARD-JONES DISPERSION/REPULSION */
1558
1559             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1560             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1561
1562             fscal            = _mm256_add_pd(felec,fvdw);
1563
1564             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1565
1566             /* Calculate temporary vectorial force */
1567             tx               = _mm256_mul_pd(fscal,dx00);
1568             ty               = _mm256_mul_pd(fscal,dy00);
1569             tz               = _mm256_mul_pd(fscal,dz00);
1570
1571             /* Update vectorial force */
1572             fix0             = _mm256_add_pd(fix0,tx);
1573             fiy0             = _mm256_add_pd(fiy0,ty);
1574             fiz0             = _mm256_add_pd(fiz0,tz);
1575
1576             fjx0             = _mm256_add_pd(fjx0,tx);
1577             fjy0             = _mm256_add_pd(fjy0,ty);
1578             fjz0             = _mm256_add_pd(fjz0,tz);
1579
1580             /**************************
1581              * CALCULATE INTERACTIONS *
1582              **************************/
1583
1584             /* COULOMB ELECTROSTATICS */
1585             velec            = _mm256_mul_pd(qq01,rinv01);
1586             felec            = _mm256_mul_pd(velec,rinvsq01);
1587
1588             fscal            = felec;
1589
1590             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1591
1592             /* Calculate temporary vectorial force */
1593             tx               = _mm256_mul_pd(fscal,dx01);
1594             ty               = _mm256_mul_pd(fscal,dy01);
1595             tz               = _mm256_mul_pd(fscal,dz01);
1596
1597             /* Update vectorial force */
1598             fix0             = _mm256_add_pd(fix0,tx);
1599             fiy0             = _mm256_add_pd(fiy0,ty);
1600             fiz0             = _mm256_add_pd(fiz0,tz);
1601
1602             fjx1             = _mm256_add_pd(fjx1,tx);
1603             fjy1             = _mm256_add_pd(fjy1,ty);
1604             fjz1             = _mm256_add_pd(fjz1,tz);
1605
1606             /**************************
1607              * CALCULATE INTERACTIONS *
1608              **************************/
1609
1610             /* COULOMB ELECTROSTATICS */
1611             velec            = _mm256_mul_pd(qq02,rinv02);
1612             felec            = _mm256_mul_pd(velec,rinvsq02);
1613
1614             fscal            = felec;
1615
1616             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1617
1618             /* Calculate temporary vectorial force */
1619             tx               = _mm256_mul_pd(fscal,dx02);
1620             ty               = _mm256_mul_pd(fscal,dy02);
1621             tz               = _mm256_mul_pd(fscal,dz02);
1622
1623             /* Update vectorial force */
1624             fix0             = _mm256_add_pd(fix0,tx);
1625             fiy0             = _mm256_add_pd(fiy0,ty);
1626             fiz0             = _mm256_add_pd(fiz0,tz);
1627
1628             fjx2             = _mm256_add_pd(fjx2,tx);
1629             fjy2             = _mm256_add_pd(fjy2,ty);
1630             fjz2             = _mm256_add_pd(fjz2,tz);
1631
1632             /**************************
1633              * CALCULATE INTERACTIONS *
1634              **************************/
1635
1636             /* COULOMB ELECTROSTATICS */
1637             velec            = _mm256_mul_pd(qq10,rinv10);
1638             felec            = _mm256_mul_pd(velec,rinvsq10);
1639
1640             fscal            = felec;
1641
1642             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1643
1644             /* Calculate temporary vectorial force */
1645             tx               = _mm256_mul_pd(fscal,dx10);
1646             ty               = _mm256_mul_pd(fscal,dy10);
1647             tz               = _mm256_mul_pd(fscal,dz10);
1648
1649             /* Update vectorial force */
1650             fix1             = _mm256_add_pd(fix1,tx);
1651             fiy1             = _mm256_add_pd(fiy1,ty);
1652             fiz1             = _mm256_add_pd(fiz1,tz);
1653
1654             fjx0             = _mm256_add_pd(fjx0,tx);
1655             fjy0             = _mm256_add_pd(fjy0,ty);
1656             fjz0             = _mm256_add_pd(fjz0,tz);
1657
1658             /**************************
1659              * CALCULATE INTERACTIONS *
1660              **************************/
1661
1662             /* COULOMB ELECTROSTATICS */
1663             velec            = _mm256_mul_pd(qq11,rinv11);
1664             felec            = _mm256_mul_pd(velec,rinvsq11);
1665
1666             fscal            = felec;
1667
1668             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1669
1670             /* Calculate temporary vectorial force */
1671             tx               = _mm256_mul_pd(fscal,dx11);
1672             ty               = _mm256_mul_pd(fscal,dy11);
1673             tz               = _mm256_mul_pd(fscal,dz11);
1674
1675             /* Update vectorial force */
1676             fix1             = _mm256_add_pd(fix1,tx);
1677             fiy1             = _mm256_add_pd(fiy1,ty);
1678             fiz1             = _mm256_add_pd(fiz1,tz);
1679
1680             fjx1             = _mm256_add_pd(fjx1,tx);
1681             fjy1             = _mm256_add_pd(fjy1,ty);
1682             fjz1             = _mm256_add_pd(fjz1,tz);
1683
1684             /**************************
1685              * CALCULATE INTERACTIONS *
1686              **************************/
1687
1688             /* COULOMB ELECTROSTATICS */
1689             velec            = _mm256_mul_pd(qq12,rinv12);
1690             felec            = _mm256_mul_pd(velec,rinvsq12);
1691
1692             fscal            = felec;
1693
1694             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1695
1696             /* Calculate temporary vectorial force */
1697             tx               = _mm256_mul_pd(fscal,dx12);
1698             ty               = _mm256_mul_pd(fscal,dy12);
1699             tz               = _mm256_mul_pd(fscal,dz12);
1700
1701             /* Update vectorial force */
1702             fix1             = _mm256_add_pd(fix1,tx);
1703             fiy1             = _mm256_add_pd(fiy1,ty);
1704             fiz1             = _mm256_add_pd(fiz1,tz);
1705
1706             fjx2             = _mm256_add_pd(fjx2,tx);
1707             fjy2             = _mm256_add_pd(fjy2,ty);
1708             fjz2             = _mm256_add_pd(fjz2,tz);
1709
1710             /**************************
1711              * CALCULATE INTERACTIONS *
1712              **************************/
1713
1714             /* COULOMB ELECTROSTATICS */
1715             velec            = _mm256_mul_pd(qq20,rinv20);
1716             felec            = _mm256_mul_pd(velec,rinvsq20);
1717
1718             fscal            = felec;
1719
1720             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1721
1722             /* Calculate temporary vectorial force */
1723             tx               = _mm256_mul_pd(fscal,dx20);
1724             ty               = _mm256_mul_pd(fscal,dy20);
1725             tz               = _mm256_mul_pd(fscal,dz20);
1726
1727             /* Update vectorial force */
1728             fix2             = _mm256_add_pd(fix2,tx);
1729             fiy2             = _mm256_add_pd(fiy2,ty);
1730             fiz2             = _mm256_add_pd(fiz2,tz);
1731
1732             fjx0             = _mm256_add_pd(fjx0,tx);
1733             fjy0             = _mm256_add_pd(fjy0,ty);
1734             fjz0             = _mm256_add_pd(fjz0,tz);
1735
1736             /**************************
1737              * CALCULATE INTERACTIONS *
1738              **************************/
1739
1740             /* COULOMB ELECTROSTATICS */
1741             velec            = _mm256_mul_pd(qq21,rinv21);
1742             felec            = _mm256_mul_pd(velec,rinvsq21);
1743
1744             fscal            = felec;
1745
1746             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1747
1748             /* Calculate temporary vectorial force */
1749             tx               = _mm256_mul_pd(fscal,dx21);
1750             ty               = _mm256_mul_pd(fscal,dy21);
1751             tz               = _mm256_mul_pd(fscal,dz21);
1752
1753             /* Update vectorial force */
1754             fix2             = _mm256_add_pd(fix2,tx);
1755             fiy2             = _mm256_add_pd(fiy2,ty);
1756             fiz2             = _mm256_add_pd(fiz2,tz);
1757
1758             fjx1             = _mm256_add_pd(fjx1,tx);
1759             fjy1             = _mm256_add_pd(fjy1,ty);
1760             fjz1             = _mm256_add_pd(fjz1,tz);
1761
1762             /**************************
1763              * CALCULATE INTERACTIONS *
1764              **************************/
1765
1766             /* COULOMB ELECTROSTATICS */
1767             velec            = _mm256_mul_pd(qq22,rinv22);
1768             felec            = _mm256_mul_pd(velec,rinvsq22);
1769
1770             fscal            = felec;
1771
1772             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1773
1774             /* Calculate temporary vectorial force */
1775             tx               = _mm256_mul_pd(fscal,dx22);
1776             ty               = _mm256_mul_pd(fscal,dy22);
1777             tz               = _mm256_mul_pd(fscal,dz22);
1778
1779             /* Update vectorial force */
1780             fix2             = _mm256_add_pd(fix2,tx);
1781             fiy2             = _mm256_add_pd(fiy2,ty);
1782             fiz2             = _mm256_add_pd(fiz2,tz);
1783
1784             fjx2             = _mm256_add_pd(fjx2,tx);
1785             fjy2             = _mm256_add_pd(fjy2,ty);
1786             fjz2             = _mm256_add_pd(fjz2,tz);
1787
1788             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1789             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1790             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1791             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1792
1793             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1794                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1795
1796             /* Inner loop uses 241 flops */
1797         }
1798
1799         /* End of innermost loop */
1800
1801         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1802                                                  f+i_coord_offset,fshift+i_shift_offset);
1803
1804         /* Increment number of inner iterations */
1805         inneriter                  += j_index_end - j_index_start;
1806
1807         /* Outer loop uses 18 flops */
1808     }
1809
1810     /* Increment number of outer iterations */
1811     outeriter        += nri;
1812
1813     /* Update outer/inner flops */
1814
1815     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*241);
1816 }