src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 104     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 105     __m256i          ewitab;
 106     __m128i          ewitab_lo,ewitab_hi;
 107     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 109     real             *ewtab;
 110     __m256           dummy_mask,cutoff_mask;
 111     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 112     __m256           one     = _mm256_set1_ps(1.0);
 113     __m256           two     = _mm256_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 132     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 133     beta2            = _mm256_mul_ps(beta,beta);
 134     beta3            = _mm256_mul_ps(beta,beta2);
 135
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 143     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 144     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 145     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153     j_coord_offsetE = 0;
 154     j_coord_offsetF = 0;
 155     j_coord_offsetG = 0;
 156     j_coord_offsetH = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm256_setzero_ps();
 185         fiy0             = _mm256_setzero_ps();
 186         fiz0             = _mm256_setzero_ps();
 187         fix1             = _mm256_setzero_ps();
 188         fiy1             = _mm256_setzero_ps();
 189         fiz1             = _mm256_setzero_ps();
 190         fix2             = _mm256_setzero_ps();
 191         fiy2             = _mm256_setzero_ps();
 192         fiz2             = _mm256_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_ps();
 196         vvdwsum          = _mm256_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             jnrE             = jjnr[jidx+4];
 208             jnrF             = jjnr[jidx+5];
 209             jnrG             = jjnr[jidx+6];
 210             jnrH             = jjnr[jidx+7];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215             j_coord_offsetE  = DIM*jnrE;
 216             j_coord_offsetF  = DIM*jnrF;
 217             j_coord_offsetG  = DIM*jnrG;
 218             j_coord_offsetH  = DIM*jnrH;
 219
 220             /* load j atom coordinates */
 221             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 223                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 224                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 225                                                  &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm256_sub_ps(ix0,jx0);
 229             dy00             = _mm256_sub_ps(iy0,jy0);
 230             dz00             = _mm256_sub_ps(iz0,jz0);
 231             dx10             = _mm256_sub_ps(ix1,jx0);
 232             dy10             = _mm256_sub_ps(iy1,jy0);
 233             dz10             = _mm256_sub_ps(iz1,jz0);
 234             dx20             = _mm256_sub_ps(ix2,jx0);
 235             dy20             = _mm256_sub_ps(iy2,jy0);
 236             dz20             = _mm256_sub_ps(iz2,jz0);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 240             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 241             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 242
 243             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 244             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 245             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 246
 247             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 248             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 249             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 250
 251             /* Load parameters for j particles */
 252             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 253                                                                  charge+jnrC+0,charge+jnrD+0,
 254                                                                  charge+jnrE+0,charge+jnrF+0,
 255                                                                  charge+jnrG+0,charge+jnrH+0);
 256             vdwjidx0A        = 2*vdwtype[jnrA+0];
 257             vdwjidx0B        = 2*vdwtype[jnrB+0];
 258             vdwjidx0C        = 2*vdwtype[jnrC+0];
 259             vdwjidx0D        = 2*vdwtype[jnrD+0];
 260             vdwjidx0E        = 2*vdwtype[jnrE+0];
 261             vdwjidx0F        = 2*vdwtype[jnrF+0];
 262             vdwjidx0G        = 2*vdwtype[jnrG+0];
 263             vdwjidx0H        = 2*vdwtype[jnrH+0];
 264
 265             fjx0             = _mm256_setzero_ps();
 266             fjy0             = _mm256_setzero_ps();
 267             fjz0             = _mm256_setzero_ps();
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             r00              = _mm256_mul_ps(rsq00,rinv00);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             qq00             = _mm256_mul_ps(iq0,jq0);
 277             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 278                                             vdwioffsetptr0+vdwjidx0B,
 279                                             vdwioffsetptr0+vdwjidx0C,
 280                                             vdwioffsetptr0+vdwjidx0D,
 281                                             vdwioffsetptr0+vdwjidx0E,
 282                                             vdwioffsetptr0+vdwjidx0F,
 283                                             vdwioffsetptr0+vdwjidx0G,
 284                                             vdwioffsetptr0+vdwjidx0H,
 285                                             &c6_00,&c12_00);
 286
 287             /* EWALD ELECTROSTATICS */
 288
 289             /* Analytical PME correction */
 290             zeta2            = _mm256_mul_ps(beta2,rsq00);
 291             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 292             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 293             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 294             felec            = _mm256_mul_ps(qq00,felec);
 295             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 296             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 297             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 298             velec            = _mm256_mul_ps(qq00,velec);
 299
 300             /* LENNARD-JONES DISPERSION/REPULSION */
 301
 302             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 303             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 304             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 305             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 306             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velecsum         = _mm256_add_ps(velecsum,velec);
 310             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 311
 312             fscal            = _mm256_add_ps(felec,fvdw);
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm256_mul_ps(fscal,dx00);
 316             ty               = _mm256_mul_ps(fscal,dy00);
 317             tz               = _mm256_mul_ps(fscal,dz00);
 318
 319             /* Update vectorial force */
 320             fix0             = _mm256_add_ps(fix0,tx);
 321             fiy0             = _mm256_add_ps(fiy0,ty);
 322             fiz0             = _mm256_add_ps(fiz0,tz);
 323
 324             fjx0             = _mm256_add_ps(fjx0,tx);
 325             fjy0             = _mm256_add_ps(fjy0,ty);
 326             fjz0             = _mm256_add_ps(fjz0,tz);
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             r10              = _mm256_mul_ps(rsq10,rinv10);
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq10             = _mm256_mul_ps(iq1,jq0);
 336
 337             /* EWALD ELECTROSTATICS */
 338
 339             /* Analytical PME correction */
 340             zeta2            = _mm256_mul_ps(beta2,rsq10);
 341             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 342             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 343             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 344             felec            = _mm256_mul_ps(qq10,felec);
 345             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 346             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 347             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 348             velec            = _mm256_mul_ps(qq10,velec);
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velecsum         = _mm256_add_ps(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm256_mul_ps(fscal,dx10);
 357             ty               = _mm256_mul_ps(fscal,dy10);
 358             tz               = _mm256_mul_ps(fscal,dz10);
 359
 360             /* Update vectorial force */
 361             fix1             = _mm256_add_ps(fix1,tx);
 362             fiy1             = _mm256_add_ps(fiy1,ty);
 363             fiz1             = _mm256_add_ps(fiz1,tz);
 364
 365             fjx0             = _mm256_add_ps(fjx0,tx);
 366             fjy0             = _mm256_add_ps(fjy0,ty);
 367             fjz0             = _mm256_add_ps(fjz0,tz);
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             r20              = _mm256_mul_ps(rsq20,rinv20);
 374
 375             /* Compute parameters for interactions between i and j atoms */
 376             qq20             = _mm256_mul_ps(iq2,jq0);
 377
 378             /* EWALD ELECTROSTATICS */
 379
 380             /* Analytical PME correction */
 381             zeta2            = _mm256_mul_ps(beta2,rsq20);
 382             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 383             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 384             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 385             felec            = _mm256_mul_ps(qq20,felec);
 386             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 387             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 388             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 389             velec            = _mm256_mul_ps(qq20,velec);
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm256_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm256_mul_ps(fscal,dx20);
 398             ty               = _mm256_mul_ps(fscal,dy20);
 399             tz               = _mm256_mul_ps(fscal,dz20);
 400
 401             /* Update vectorial force */
 402             fix2             = _mm256_add_ps(fix2,tx);
 403             fiy2             = _mm256_add_ps(fiy2,ty);
 404             fiz2             = _mm256_add_ps(fiz2,tz);
 405
 406             fjx0             = _mm256_add_ps(fjx0,tx);
 407             fjy0             = _mm256_add_ps(fjy0,ty);
 408             fjz0             = _mm256_add_ps(fjz0,tz);
 409
 410             fjptrA             = f+j_coord_offsetA;
 411             fjptrB             = f+j_coord_offsetB;
 412             fjptrC             = f+j_coord_offsetC;
 413             fjptrD             = f+j_coord_offsetD;
 414             fjptrE             = f+j_coord_offsetE;
 415             fjptrF             = f+j_coord_offsetF;
 416             fjptrG             = f+j_coord_offsetG;
 417             fjptrH             = f+j_coord_offsetH;
 418
 419             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 420
 421             /* Inner loop uses 267 flops */
 422         }
 423
 424         if(jidx<j_index_end)
 425         {
 426
 427             /* Get j neighbor index, and coordinate index */
 428             jnrlistA         = jjnr[jidx];
 429             jnrlistB         = jjnr[jidx+1];
 430             jnrlistC         = jjnr[jidx+2];
 431             jnrlistD         = jjnr[jidx+3];
 432             jnrlistE         = jjnr[jidx+4];
 433             jnrlistF         = jjnr[jidx+5];
 434             jnrlistG         = jjnr[jidx+6];
 435             jnrlistH         = jjnr[jidx+7];
 436             /* Sign of each element will be negative for non-real atoms.
 437              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 438              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 439              */
 440             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 441                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 442
 443             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 444             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 445             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 446             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 447             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 448             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 449             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 450             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 451             j_coord_offsetA  = DIM*jnrA;
 452             j_coord_offsetB  = DIM*jnrB;
 453             j_coord_offsetC  = DIM*jnrC;
 454             j_coord_offsetD  = DIM*jnrD;
 455             j_coord_offsetE  = DIM*jnrE;
 456             j_coord_offsetF  = DIM*jnrF;
 457             j_coord_offsetG  = DIM*jnrG;
 458             j_coord_offsetH  = DIM*jnrH;
 459
 460             /* load j atom coordinates */
 461             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 463                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 464                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 465                                                  &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 468             dx00             = _mm256_sub_ps(ix0,jx0);
 469             dy00             = _mm256_sub_ps(iy0,jy0);
 470             dz00             = _mm256_sub_ps(iz0,jz0);
 471             dx10             = _mm256_sub_ps(ix1,jx0);
 472             dy10             = _mm256_sub_ps(iy1,jy0);
 473             dz10             = _mm256_sub_ps(iz1,jz0);
 474             dx20             = _mm256_sub_ps(ix2,jx0);
 475             dy20             = _mm256_sub_ps(iy2,jy0);
 476             dz20             = _mm256_sub_ps(iz2,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 480             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 481             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 482
 483             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 484             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 485             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 486
 487             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 488             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 489             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 490
 491             /* Load parameters for j particles */
 492             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 493                                                                  charge+jnrC+0,charge+jnrD+0,
 494                                                                  charge+jnrE+0,charge+jnrF+0,
 495                                                                  charge+jnrG+0,charge+jnrH+0);
 496             vdwjidx0A        = 2*vdwtype[jnrA+0];
 497             vdwjidx0B        = 2*vdwtype[jnrB+0];
 498             vdwjidx0C        = 2*vdwtype[jnrC+0];
 499             vdwjidx0D        = 2*vdwtype[jnrD+0];
 500             vdwjidx0E        = 2*vdwtype[jnrE+0];
 501             vdwjidx0F        = 2*vdwtype[jnrF+0];
 502             vdwjidx0G        = 2*vdwtype[jnrG+0];
 503             vdwjidx0H        = 2*vdwtype[jnrH+0];
 504
 505             fjx0             = _mm256_setzero_ps();
 506             fjy0             = _mm256_setzero_ps();
 507             fjz0             = _mm256_setzero_ps();
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             r00              = _mm256_mul_ps(rsq00,rinv00);
 514             r00              = _mm256_andnot_ps(dummy_mask,r00);
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             qq00             = _mm256_mul_ps(iq0,jq0);
 518             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 519                                             vdwioffsetptr0+vdwjidx0B,
 520                                             vdwioffsetptr0+vdwjidx0C,
 521                                             vdwioffsetptr0+vdwjidx0D,
 522                                             vdwioffsetptr0+vdwjidx0E,
 523                                             vdwioffsetptr0+vdwjidx0F,
 524                                             vdwioffsetptr0+vdwjidx0G,
 525                                             vdwioffsetptr0+vdwjidx0H,
 526                                             &c6_00,&c12_00);
 527
 528             /* EWALD ELECTROSTATICS */
 529
 530             /* Analytical PME correction */
 531             zeta2            = _mm256_mul_ps(beta2,rsq00);
 532             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 533             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 534             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 535             felec            = _mm256_mul_ps(qq00,felec);
 536             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 537             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 538             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 539             velec            = _mm256_mul_ps(qq00,velec);
 540
 541             /* LENNARD-JONES DISPERSION/REPULSION */
 542
 543             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 544             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 545             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 546             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 547             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm256_andnot_ps(dummy_mask,velec);
 551             velecsum         = _mm256_add_ps(velecsum,velec);
 552             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 553             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 554
 555             fscal            = _mm256_add_ps(felec,fvdw);
 556
 557             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm256_mul_ps(fscal,dx00);
 561             ty               = _mm256_mul_ps(fscal,dy00);
 562             tz               = _mm256_mul_ps(fscal,dz00);
 563
 564             /* Update vectorial force */
 565             fix0             = _mm256_add_ps(fix0,tx);
 566             fiy0             = _mm256_add_ps(fiy0,ty);
 567             fiz0             = _mm256_add_ps(fiz0,tz);
 568
 569             fjx0             = _mm256_add_ps(fjx0,tx);
 570             fjy0             = _mm256_add_ps(fjy0,ty);
 571             fjz0             = _mm256_add_ps(fjz0,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r10              = _mm256_mul_ps(rsq10,rinv10);
 578             r10              = _mm256_andnot_ps(dummy_mask,r10);
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq10             = _mm256_mul_ps(iq1,jq0);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Analytical PME correction */
 586             zeta2            = _mm256_mul_ps(beta2,rsq10);
 587             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 588             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 589             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 590             felec            = _mm256_mul_ps(qq10,felec);
 591             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 592             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 593             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 594             velec            = _mm256_mul_ps(qq10,velec);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm256_andnot_ps(dummy_mask,velec);
 598             velecsum         = _mm256_add_ps(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm256_mul_ps(fscal,dx10);
 606             ty               = _mm256_mul_ps(fscal,dy10);
 607             tz               = _mm256_mul_ps(fscal,dz10);
 608
 609             /* Update vectorial force */
 610             fix1             = _mm256_add_ps(fix1,tx);
 611             fiy1             = _mm256_add_ps(fiy1,ty);
 612             fiz1             = _mm256_add_ps(fiz1,tz);
 613
 614             fjx0             = _mm256_add_ps(fjx0,tx);
 615             fjy0             = _mm256_add_ps(fjy0,ty);
 616             fjz0             = _mm256_add_ps(fjz0,tz);
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             r20              = _mm256_mul_ps(rsq20,rinv20);
 623             r20              = _mm256_andnot_ps(dummy_mask,r20);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq20             = _mm256_mul_ps(iq2,jq0);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Analytical PME correction */
 631             zeta2            = _mm256_mul_ps(beta2,rsq20);
 632             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 633             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 634             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 635             felec            = _mm256_mul_ps(qq20,felec);
 636             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 637             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 638             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 639             velec            = _mm256_mul_ps(qq20,velec);
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm256_andnot_ps(dummy_mask,velec);
 643             velecsum         = _mm256_add_ps(velecsum,velec);
 644
 645             fscal            = felec;
 646
 647             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 648
 649             /* Calculate temporary vectorial force */
 650             tx               = _mm256_mul_ps(fscal,dx20);
 651             ty               = _mm256_mul_ps(fscal,dy20);
 652             tz               = _mm256_mul_ps(fscal,dz20);
 653
 654             /* Update vectorial force */
 655             fix2             = _mm256_add_ps(fix2,tx);
 656             fiy2             = _mm256_add_ps(fiy2,ty);
 657             fiz2             = _mm256_add_ps(fiz2,tz);
 658
 659             fjx0             = _mm256_add_ps(fjx0,tx);
 660             fjy0             = _mm256_add_ps(fjy0,ty);
 661             fjz0             = _mm256_add_ps(fjz0,tz);
 662
 663             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 664             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 665             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 666             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 667             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 668             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 669             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 670             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 671
 672             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 673
 674             /* Inner loop uses 270 flops */
 675         }
 676
 677         /* End of innermost loop */
 678
 679         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 680                                                  f+i_coord_offset,fshift+i_shift_offset);
 681
 682         ggid                        = gid[iidx];
 683         /* Update potential energies */
 684         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 685         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 686
 687         /* Increment number of inner iterations */
 688         inneriter                  += j_index_end - j_index_start;
 689
 690         /* Outer loop uses 20 flops */
 691     }
 692
 693     /* Increment number of outer iterations */
 694     outeriter        += nri;
 695
 696     /* Update outer/inner flops */
 697
 698     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*270);
 699 }
 700 /*
 701  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_single
 702  * Electrostatics interaction: Ewald
 703  * VdW interaction:            LennardJones
 704  * Geometry:                   Water3-Particle
 705  * Calculate force/pot:        Force
 706  */
 707 void
 708 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_single
 709                     (t_nblist                    * gmx_restrict       nlist,
 710                      rvec                        * gmx_restrict          xx,
 711                      rvec                        * gmx_restrict          ff,
 712                      t_forcerec                  * gmx_restrict          fr,
 713                      t_mdatoms                   * gmx_restrict     mdatoms,
 714                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 715                      t_nrnb                      * gmx_restrict        nrnb)
 716 {
 717     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 718      * just 0 for non-waters.
 719      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 720      * jnr indices corresponding to data put in the four positions in the SIMD register.
 721      */
 722     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 723     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 724     int              jnrA,jnrB,jnrC,jnrD;
 725     int              jnrE,jnrF,jnrG,jnrH;
 726     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 727     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 728     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 729     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 730     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 731     real             rcutoff_scalar;
 732     real             *shiftvec,*fshift,*x,*f;
 733     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 734     real             scratch[4*DIM];
 735     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 736     real *           vdwioffsetptr0;
 737     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 738     real *           vdwioffsetptr1;
 739     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 740     real *           vdwioffsetptr2;
 741     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 742     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 743     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 744     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 745     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 746     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 747     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 748     real             *charge;
 749     int              nvdwtype;
 750     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 751     int              *vdwtype;
 752     real             *vdwparam;
 753     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 754     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 755     __m256i          ewitab;
 756     __m128i          ewitab_lo,ewitab_hi;
 757     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 758     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 759     real             *ewtab;
 760     __m256           dummy_mask,cutoff_mask;
 761     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 762     __m256           one     = _mm256_set1_ps(1.0);
 763     __m256           two     = _mm256_set1_ps(2.0);
 764     x                = xx[0];
 765     f                = ff[0];
 766
 767     nri              = nlist->nri;
 768     iinr             = nlist->iinr;
 769     jindex           = nlist->jindex;
 770     jjnr             = nlist->jjnr;
 771     shiftidx         = nlist->shift;
 772     gid              = nlist->gid;
 773     shiftvec         = fr->shift_vec[0];
 774     fshift           = fr->fshift[0];
 775     facel            = _mm256_set1_ps(fr->epsfac);
 776     charge           = mdatoms->chargeA;
 777     nvdwtype         = fr->ntype;
 778     vdwparam         = fr->nbfp;
 779     vdwtype          = mdatoms->typeA;
 780
 781     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 782     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 783     beta2            = _mm256_mul_ps(beta,beta);
 784     beta3            = _mm256_mul_ps(beta,beta2);
 785
 786     ewtab            = fr->ic->tabq_coul_F;
 787     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 788     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 789
 790     /* Setup water-specific parameters */
 791     inr              = nlist->iinr[0];
 792     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 793     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 794     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 795     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 796
 797     /* Avoid stupid compiler warnings */
 798     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 799     j_coord_offsetA = 0;
 800     j_coord_offsetB = 0;
 801     j_coord_offsetC = 0;
 802     j_coord_offsetD = 0;
 803     j_coord_offsetE = 0;
 804     j_coord_offsetF = 0;
 805     j_coord_offsetG = 0;
 806     j_coord_offsetH = 0;
 807
 808     outeriter        = 0;
 809     inneriter        = 0;
 810
 811     for(iidx=0;iidx<4*DIM;iidx++)
 812     {
 813         scratch[iidx] = 0.0;
 814     }
 815
 816     /* Start outer loop over neighborlists */
 817     for(iidx=0; iidx<nri; iidx++)
 818     {
 819         /* Load shift vector for this list */
 820         i_shift_offset   = DIM*shiftidx[iidx];
 821
 822         /* Load limits for loop over neighbors */
 823         j_index_start    = jindex[iidx];
 824         j_index_end      = jindex[iidx+1];
 825
 826         /* Get outer coordinate index */
 827         inr              = iinr[iidx];
 828         i_coord_offset   = DIM*inr;
 829
 830         /* Load i particle coords and add shift vector */
 831         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 832                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 833
 834         fix0             = _mm256_setzero_ps();
 835         fiy0             = _mm256_setzero_ps();
 836         fiz0             = _mm256_setzero_ps();
 837         fix1             = _mm256_setzero_ps();
 838         fiy1             = _mm256_setzero_ps();
 839         fiz1             = _mm256_setzero_ps();
 840         fix2             = _mm256_setzero_ps();
 841         fiy2             = _mm256_setzero_ps();
 842         fiz2             = _mm256_setzero_ps();
 843
 844         /* Start inner kernel loop */
 845         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 846         {
 847
 848             /* Get j neighbor index, and coordinate index */
 849             jnrA             = jjnr[jidx];
 850             jnrB             = jjnr[jidx+1];
 851             jnrC             = jjnr[jidx+2];
 852             jnrD             = jjnr[jidx+3];
 853             jnrE             = jjnr[jidx+4];
 854             jnrF             = jjnr[jidx+5];
 855             jnrG             = jjnr[jidx+6];
 856             jnrH             = jjnr[jidx+7];
 857             j_coord_offsetA  = DIM*jnrA;
 858             j_coord_offsetB  = DIM*jnrB;
 859             j_coord_offsetC  = DIM*jnrC;
 860             j_coord_offsetD  = DIM*jnrD;
 861             j_coord_offsetE  = DIM*jnrE;
 862             j_coord_offsetF  = DIM*jnrF;
 863             j_coord_offsetG  = DIM*jnrG;
 864             j_coord_offsetH  = DIM*jnrH;
 865
 866             /* load j atom coordinates */
 867             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 868                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 869                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 870                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 871                                                  &jx0,&jy0,&jz0);
 872
 873             /* Calculate displacement vector */
 874             dx00             = _mm256_sub_ps(ix0,jx0);
 875             dy00             = _mm256_sub_ps(iy0,jy0);
 876             dz00             = _mm256_sub_ps(iz0,jz0);
 877             dx10             = _mm256_sub_ps(ix1,jx0);
 878             dy10             = _mm256_sub_ps(iy1,jy0);
 879             dz10             = _mm256_sub_ps(iz1,jz0);
 880             dx20             = _mm256_sub_ps(ix2,jx0);
 881             dy20             = _mm256_sub_ps(iy2,jy0);
 882             dz20             = _mm256_sub_ps(iz2,jz0);
 883
 884             /* Calculate squared distance and things based on it */
 885             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 886             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 887             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 888
 889             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 890             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 891             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 892
 893             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 894             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 895             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 896
 897             /* Load parameters for j particles */
 898             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 899                                                                  charge+jnrC+0,charge+jnrD+0,
 900                                                                  charge+jnrE+0,charge+jnrF+0,
 901                                                                  charge+jnrG+0,charge+jnrH+0);
 902             vdwjidx0A        = 2*vdwtype[jnrA+0];
 903             vdwjidx0B        = 2*vdwtype[jnrB+0];
 904             vdwjidx0C        = 2*vdwtype[jnrC+0];
 905             vdwjidx0D        = 2*vdwtype[jnrD+0];
 906             vdwjidx0E        = 2*vdwtype[jnrE+0];
 907             vdwjidx0F        = 2*vdwtype[jnrF+0];
 908             vdwjidx0G        = 2*vdwtype[jnrG+0];
 909             vdwjidx0H        = 2*vdwtype[jnrH+0];
 910
 911             fjx0             = _mm256_setzero_ps();
 912             fjy0             = _mm256_setzero_ps();
 913             fjz0             = _mm256_setzero_ps();
 914
 915             /**************************
 916              * CALCULATE INTERACTIONS *
 917              **************************/
 918
 919             r00              = _mm256_mul_ps(rsq00,rinv00);
 920
 921             /* Compute parameters for interactions between i and j atoms */
 922             qq00             = _mm256_mul_ps(iq0,jq0);
 923             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 924                                             vdwioffsetptr0+vdwjidx0B,
 925                                             vdwioffsetptr0+vdwjidx0C,
 926                                             vdwioffsetptr0+vdwjidx0D,
 927                                             vdwioffsetptr0+vdwjidx0E,
 928                                             vdwioffsetptr0+vdwjidx0F,
 929                                             vdwioffsetptr0+vdwjidx0G,
 930                                             vdwioffsetptr0+vdwjidx0H,
 931                                             &c6_00,&c12_00);
 932
 933             /* EWALD ELECTROSTATICS */
 934
 935             /* Analytical PME correction */
 936             zeta2            = _mm256_mul_ps(beta2,rsq00);
 937             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 938             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 939             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 940             felec            = _mm256_mul_ps(qq00,felec);
 941
 942             /* LENNARD-JONES DISPERSION/REPULSION */
 943
 944             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 945             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 946
 947             fscal            = _mm256_add_ps(felec,fvdw);
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm256_mul_ps(fscal,dx00);
 951             ty               = _mm256_mul_ps(fscal,dy00);
 952             tz               = _mm256_mul_ps(fscal,dz00);
 953
 954             /* Update vectorial force */
 955             fix0             = _mm256_add_ps(fix0,tx);
 956             fiy0             = _mm256_add_ps(fiy0,ty);
 957             fiz0             = _mm256_add_ps(fiz0,tz);
 958
 959             fjx0             = _mm256_add_ps(fjx0,tx);
 960             fjy0             = _mm256_add_ps(fjy0,ty);
 961             fjz0             = _mm256_add_ps(fjz0,tz);
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             r10              = _mm256_mul_ps(rsq10,rinv10);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq10             = _mm256_mul_ps(iq1,jq0);
 971
 972             /* EWALD ELECTROSTATICS */
 973
 974             /* Analytical PME correction */
 975             zeta2            = _mm256_mul_ps(beta2,rsq10);
 976             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 977             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 978             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 979             felec            = _mm256_mul_ps(qq10,felec);
 980
 981             fscal            = felec;
 982
 983             /* Calculate temporary vectorial force */
 984             tx               = _mm256_mul_ps(fscal,dx10);
 985             ty               = _mm256_mul_ps(fscal,dy10);
 986             tz               = _mm256_mul_ps(fscal,dz10);
 987
 988             /* Update vectorial force */
 989             fix1             = _mm256_add_ps(fix1,tx);
 990             fiy1             = _mm256_add_ps(fiy1,ty);
 991             fiz1             = _mm256_add_ps(fiz1,tz);
 992
 993             fjx0             = _mm256_add_ps(fjx0,tx);
 994             fjy0             = _mm256_add_ps(fjy0,ty);
 995             fjz0             = _mm256_add_ps(fjz0,tz);
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             r20              = _mm256_mul_ps(rsq20,rinv20);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq20             = _mm256_mul_ps(iq2,jq0);
1005
1006             /* EWALD ELECTROSTATICS */
1007
1008             /* Analytical PME correction */
1009             zeta2            = _mm256_mul_ps(beta2,rsq20);
1010             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1011             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1012             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1013             felec            = _mm256_mul_ps(qq20,felec);
1014
1015             fscal            = felec;
1016
1017             /* Calculate temporary vectorial force */
1018             tx               = _mm256_mul_ps(fscal,dx20);
1019             ty               = _mm256_mul_ps(fscal,dy20);
1020             tz               = _mm256_mul_ps(fscal,dz20);
1021
1022             /* Update vectorial force */
1023             fix2             = _mm256_add_ps(fix2,tx);
1024             fiy2             = _mm256_add_ps(fiy2,ty);
1025             fiz2             = _mm256_add_ps(fiz2,tz);
1026
1027             fjx0             = _mm256_add_ps(fjx0,tx);
1028             fjy0             = _mm256_add_ps(fjy0,ty);
1029             fjz0             = _mm256_add_ps(fjz0,tz);
1030
1031             fjptrA             = f+j_coord_offsetA;
1032             fjptrB             = f+j_coord_offsetB;
1033             fjptrC             = f+j_coord_offsetC;
1034             fjptrD             = f+j_coord_offsetD;
1035             fjptrE             = f+j_coord_offsetE;
1036             fjptrF             = f+j_coord_offsetF;
1037             fjptrG             = f+j_coord_offsetG;
1038             fjptrH             = f+j_coord_offsetH;
1039
1040             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1041
1042             /* Inner loop uses 178 flops */
1043         }
1044
1045         if(jidx<j_index_end)
1046         {
1047
1048             /* Get j neighbor index, and coordinate index */
1049             jnrlistA         = jjnr[jidx];
1050             jnrlistB         = jjnr[jidx+1];
1051             jnrlistC         = jjnr[jidx+2];
1052             jnrlistD         = jjnr[jidx+3];
1053             jnrlistE         = jjnr[jidx+4];
1054             jnrlistF         = jjnr[jidx+5];
1055             jnrlistG         = jjnr[jidx+6];
1056             jnrlistH         = jjnr[jidx+7];
1057             /* Sign of each element will be negative for non-real atoms.
1058              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1059              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1060              */
1061             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1062                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1063
1064             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1065             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1066             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1067             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1068             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1069             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1070             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1071             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1072             j_coord_offsetA  = DIM*jnrA;
1073             j_coord_offsetB  = DIM*jnrB;
1074             j_coord_offsetC  = DIM*jnrC;
1075             j_coord_offsetD  = DIM*jnrD;
1076             j_coord_offsetE  = DIM*jnrE;
1077             j_coord_offsetF  = DIM*jnrF;
1078             j_coord_offsetG  = DIM*jnrG;
1079             j_coord_offsetH  = DIM*jnrH;
1080
1081             /* load j atom coordinates */
1082             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1083                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1084                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1085                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1086                                                  &jx0,&jy0,&jz0);
1087
1088             /* Calculate displacement vector */
1089             dx00             = _mm256_sub_ps(ix0,jx0);
1090             dy00             = _mm256_sub_ps(iy0,jy0);
1091             dz00             = _mm256_sub_ps(iz0,jz0);
1092             dx10             = _mm256_sub_ps(ix1,jx0);
1093             dy10             = _mm256_sub_ps(iy1,jy0);
1094             dz10             = _mm256_sub_ps(iz1,jz0);
1095             dx20             = _mm256_sub_ps(ix2,jx0);
1096             dy20             = _mm256_sub_ps(iy2,jy0);
1097             dz20             = _mm256_sub_ps(iz2,jz0);
1098
1099             /* Calculate squared distance and things based on it */
1100             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1101             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1102             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1103
1104             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1105             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1106             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1107
1108             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1109             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1110             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1111
1112             /* Load parameters for j particles */
1113             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1114                                                                  charge+jnrC+0,charge+jnrD+0,
1115                                                                  charge+jnrE+0,charge+jnrF+0,
1116                                                                  charge+jnrG+0,charge+jnrH+0);
1117             vdwjidx0A        = 2*vdwtype[jnrA+0];
1118             vdwjidx0B        = 2*vdwtype[jnrB+0];
1119             vdwjidx0C        = 2*vdwtype[jnrC+0];
1120             vdwjidx0D        = 2*vdwtype[jnrD+0];
1121             vdwjidx0E        = 2*vdwtype[jnrE+0];
1122             vdwjidx0F        = 2*vdwtype[jnrF+0];
1123             vdwjidx0G        = 2*vdwtype[jnrG+0];
1124             vdwjidx0H        = 2*vdwtype[jnrH+0];
1125
1126             fjx0             = _mm256_setzero_ps();
1127             fjy0             = _mm256_setzero_ps();
1128             fjz0             = _mm256_setzero_ps();
1129
1130             /**************************
1131              * CALCULATE INTERACTIONS *
1132              **************************/
1133
1134             r00              = _mm256_mul_ps(rsq00,rinv00);
1135             r00              = _mm256_andnot_ps(dummy_mask,r00);
1136
1137             /* Compute parameters for interactions between i and j atoms */
1138             qq00             = _mm256_mul_ps(iq0,jq0);
1139             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1140                                             vdwioffsetptr0+vdwjidx0B,
1141                                             vdwioffsetptr0+vdwjidx0C,
1142                                             vdwioffsetptr0+vdwjidx0D,
1143                                             vdwioffsetptr0+vdwjidx0E,
1144                                             vdwioffsetptr0+vdwjidx0F,
1145                                             vdwioffsetptr0+vdwjidx0G,
1146                                             vdwioffsetptr0+vdwjidx0H,
1147                                             &c6_00,&c12_00);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Analytical PME correction */
1152             zeta2            = _mm256_mul_ps(beta2,rsq00);
1153             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1154             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1155             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1156             felec            = _mm256_mul_ps(qq00,felec);
1157
1158             /* LENNARD-JONES DISPERSION/REPULSION */
1159
1160             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1161             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1162
1163             fscal            = _mm256_add_ps(felec,fvdw);
1164
1165             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm256_mul_ps(fscal,dx00);
1169             ty               = _mm256_mul_ps(fscal,dy00);
1170             tz               = _mm256_mul_ps(fscal,dz00);
1171
1172             /* Update vectorial force */
1173             fix0             = _mm256_add_ps(fix0,tx);
1174             fiy0             = _mm256_add_ps(fiy0,ty);
1175             fiz0             = _mm256_add_ps(fiz0,tz);
1176
1177             fjx0             = _mm256_add_ps(fjx0,tx);
1178             fjy0             = _mm256_add_ps(fjy0,ty);
1179             fjz0             = _mm256_add_ps(fjz0,tz);
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             r10              = _mm256_mul_ps(rsq10,rinv10);
1186             r10              = _mm256_andnot_ps(dummy_mask,r10);
1187
1188             /* Compute parameters for interactions between i and j atoms */
1189             qq10             = _mm256_mul_ps(iq1,jq0);
1190
1191             /* EWALD ELECTROSTATICS */
1192
1193             /* Analytical PME correction */
1194             zeta2            = _mm256_mul_ps(beta2,rsq10);
1195             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1196             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1197             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1198             felec            = _mm256_mul_ps(qq10,felec);
1199
1200             fscal            = felec;
1201
1202             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm256_mul_ps(fscal,dx10);
1206             ty               = _mm256_mul_ps(fscal,dy10);
1207             tz               = _mm256_mul_ps(fscal,dz10);
1208
1209             /* Update vectorial force */
1210             fix1             = _mm256_add_ps(fix1,tx);
1211             fiy1             = _mm256_add_ps(fiy1,ty);
1212             fiz1             = _mm256_add_ps(fiz1,tz);
1213
1214             fjx0             = _mm256_add_ps(fjx0,tx);
1215             fjy0             = _mm256_add_ps(fjy0,ty);
1216             fjz0             = _mm256_add_ps(fjz0,tz);
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             r20              = _mm256_mul_ps(rsq20,rinv20);
1223             r20              = _mm256_andnot_ps(dummy_mask,r20);
1224
1225             /* Compute parameters for interactions between i and j atoms */
1226             qq20             = _mm256_mul_ps(iq2,jq0);
1227
1228             /* EWALD ELECTROSTATICS */
1229
1230             /* Analytical PME correction */
1231             zeta2            = _mm256_mul_ps(beta2,rsq20);
1232             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1233             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1234             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1235             felec            = _mm256_mul_ps(qq20,felec);
1236
1237             fscal            = felec;
1238
1239             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1240
1241             /* Calculate temporary vectorial force */
1242             tx               = _mm256_mul_ps(fscal,dx20);
1243             ty               = _mm256_mul_ps(fscal,dy20);
1244             tz               = _mm256_mul_ps(fscal,dz20);
1245
1246             /* Update vectorial force */
1247             fix2             = _mm256_add_ps(fix2,tx);
1248             fiy2             = _mm256_add_ps(fiy2,ty);
1249             fiz2             = _mm256_add_ps(fiz2,tz);
1250
1251             fjx0             = _mm256_add_ps(fjx0,tx);
1252             fjy0             = _mm256_add_ps(fjy0,ty);
1253             fjz0             = _mm256_add_ps(fjz0,tz);
1254
1255             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1256             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1257             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1258             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1259             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1260             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1261             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1262             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1263
1264             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1265
1266             /* Inner loop uses 181 flops */
1267         }
1268
1269         /* End of innermost loop */
1270
1271         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1272                                                  f+i_coord_offset,fshift+i_shift_offset);
1273
1274         /* Increment number of inner iterations */
1275         inneriter                  += j_index_end - j_index_start;
1276
1277         /* Outer loop uses 18 flops */
1278     }
1279
1280     /* Increment number of outer iterations */
1281     outeriter        += nri;
1282
1283     /* Update outer/inner flops */
1284
1285     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*181);
1286 }