src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256i          ewitab;
 113     __m128i          ewitab_lo,ewitab_hi;
 114     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 116     real             *ewtab;
 117     __m256           dummy_mask,cutoff_mask;
 118     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 119     __m256           one     = _mm256_set1_ps(1.0);
 120     __m256           two     = _mm256_set1_ps(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm256_set1_ps(fr->epsfac);
 133     charge           = mdatoms->chargeA;
 134     nvdwtype         = fr->ntype;
 135     vdwparam         = fr->nbfp;
 136     vdwtype          = mdatoms->typeA;
 137
 138     vftab            = kernel_data->table_vdw->data;
 139     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 140
 141     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 142     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 143     beta2            = _mm256_mul_ps(beta,beta);
 144     beta3            = _mm256_mul_ps(beta,beta2);
 145
 146     ewtab            = fr->ic->tabq_coul_FDV0;
 147     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 148     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 149
 150     /* Setup water-specific parameters */
 151     inr              = nlist->iinr[0];
 152     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 153     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 154     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 155     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163     j_coord_offsetE = 0;
 164     j_coord_offsetF = 0;
 165     j_coord_offsetG = 0;
 166     j_coord_offsetH = 0;
 167
 168     outeriter        = 0;
 169     inneriter        = 0;
 170
 171     for(iidx=0;iidx<4*DIM;iidx++)
 172     {
 173         scratch[iidx] = 0.0;
 174     }
 175
 176     /* Start outer loop over neighborlists */
 177     for(iidx=0; iidx<nri; iidx++)
 178     {
 179         /* Load shift vector for this list */
 180         i_shift_offset   = DIM*shiftidx[iidx];
 181
 182         /* Load limits for loop over neighbors */
 183         j_index_start    = jindex[iidx];
 184         j_index_end      = jindex[iidx+1];
 185
 186         /* Get outer coordinate index */
 187         inr              = iinr[iidx];
 188         i_coord_offset   = DIM*inr;
 189
 190         /* Load i particle coords and add shift vector */
 191         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 192                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 193
 194         fix0             = _mm256_setzero_ps();
 195         fiy0             = _mm256_setzero_ps();
 196         fiz0             = _mm256_setzero_ps();
 197         fix1             = _mm256_setzero_ps();
 198         fiy1             = _mm256_setzero_ps();
 199         fiz1             = _mm256_setzero_ps();
 200         fix2             = _mm256_setzero_ps();
 201         fiy2             = _mm256_setzero_ps();
 202         fiz2             = _mm256_setzero_ps();
 203
 204         /* Reset potential sums */
 205         velecsum         = _mm256_setzero_ps();
 206         vvdwsum          = _mm256_setzero_ps();
 207
 208         /* Start inner kernel loop */
 209         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 210         {
 211
 212             /* Get j neighbor index, and coordinate index */
 213             jnrA             = jjnr[jidx];
 214             jnrB             = jjnr[jidx+1];
 215             jnrC             = jjnr[jidx+2];
 216             jnrD             = jjnr[jidx+3];
 217             jnrE             = jjnr[jidx+4];
 218             jnrF             = jjnr[jidx+5];
 219             jnrG             = jjnr[jidx+6];
 220             jnrH             = jjnr[jidx+7];
 221             j_coord_offsetA  = DIM*jnrA;
 222             j_coord_offsetB  = DIM*jnrB;
 223             j_coord_offsetC  = DIM*jnrC;
 224             j_coord_offsetD  = DIM*jnrD;
 225             j_coord_offsetE  = DIM*jnrE;
 226             j_coord_offsetF  = DIM*jnrF;
 227             j_coord_offsetG  = DIM*jnrG;
 228             j_coord_offsetH  = DIM*jnrH;
 229
 230             /* load j atom coordinates */
 231             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 232                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 233                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 234                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 235                                                  &jx0,&jy0,&jz0);
 236
 237             /* Calculate displacement vector */
 238             dx00             = _mm256_sub_ps(ix0,jx0);
 239             dy00             = _mm256_sub_ps(iy0,jy0);
 240             dz00             = _mm256_sub_ps(iz0,jz0);
 241             dx10             = _mm256_sub_ps(ix1,jx0);
 242             dy10             = _mm256_sub_ps(iy1,jy0);
 243             dz10             = _mm256_sub_ps(iz1,jz0);
 244             dx20             = _mm256_sub_ps(ix2,jx0);
 245             dy20             = _mm256_sub_ps(iy2,jy0);
 246             dz20             = _mm256_sub_ps(iz2,jz0);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 250             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 251             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 252
 253             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 254             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 255             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 256
 257             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 258             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 259             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 260
 261             /* Load parameters for j particles */
 262             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 263                                                                  charge+jnrC+0,charge+jnrD+0,
 264                                                                  charge+jnrE+0,charge+jnrF+0,
 265                                                                  charge+jnrG+0,charge+jnrH+0);
 266             vdwjidx0A        = 2*vdwtype[jnrA+0];
 267             vdwjidx0B        = 2*vdwtype[jnrB+0];
 268             vdwjidx0C        = 2*vdwtype[jnrC+0];
 269             vdwjidx0D        = 2*vdwtype[jnrD+0];
 270             vdwjidx0E        = 2*vdwtype[jnrE+0];
 271             vdwjidx0F        = 2*vdwtype[jnrF+0];
 272             vdwjidx0G        = 2*vdwtype[jnrG+0];
 273             vdwjidx0H        = 2*vdwtype[jnrH+0];
 274
 275             fjx0             = _mm256_setzero_ps();
 276             fjy0             = _mm256_setzero_ps();
 277             fjz0             = _mm256_setzero_ps();
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r00              = _mm256_mul_ps(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq00             = _mm256_mul_ps(iq0,jq0);
 287             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 288                                             vdwioffsetptr0+vdwjidx0B,
 289                                             vdwioffsetptr0+vdwjidx0C,
 290                                             vdwioffsetptr0+vdwjidx0D,
 291                                             vdwioffsetptr0+vdwjidx0E,
 292                                             vdwioffsetptr0+vdwjidx0F,
 293                                             vdwioffsetptr0+vdwjidx0G,
 294                                             vdwioffsetptr0+vdwjidx0H,
 295                                             &c6_00,&c12_00);
 296
 297             /* Calculate table index by multiplying r with table scale and truncate to integer */
 298             rt               = _mm256_mul_ps(r00,vftabscale);
 299             vfitab           = _mm256_cvttps_epi32(rt);
 300             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 301             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 302             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 303             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 304             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 305             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Analytical PME correction */
 310             zeta2            = _mm256_mul_ps(beta2,rsq00);
 311             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 312             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 313             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 314             felec            = _mm256_mul_ps(qq00,felec);
 315             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 316             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 317             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 318             velec            = _mm256_mul_ps(qq00,velec);
 319
 320             /* CUBIC SPLINE TABLE DISPERSION */
 321             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 323             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 325             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 327             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 329             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 330             Heps             = _mm256_mul_ps(vfeps,H);
 331             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 332             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 333             vvdw6            = _mm256_mul_ps(c6_00,VV);
 334             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 335             fvdw6            = _mm256_mul_ps(c6_00,FF);
 336
 337             /* CUBIC SPLINE TABLE REPULSION */
 338             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 339             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 340             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 341                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 342             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 343                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 344             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 345                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 346             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 347                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 348             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 349             Heps             = _mm256_mul_ps(vfeps,H);
 350             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 351             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 352             vvdw12           = _mm256_mul_ps(c12_00,VV);
 353             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 354             fvdw12           = _mm256_mul_ps(c12_00,FF);
 355             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 356             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm256_add_ps(velecsum,velec);
 360             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 361
 362             fscal            = _mm256_add_ps(felec,fvdw);
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm256_mul_ps(fscal,dx00);
 366             ty               = _mm256_mul_ps(fscal,dy00);
 367             tz               = _mm256_mul_ps(fscal,dz00);
 368
 369             /* Update vectorial force */
 370             fix0             = _mm256_add_ps(fix0,tx);
 371             fiy0             = _mm256_add_ps(fiy0,ty);
 372             fiz0             = _mm256_add_ps(fiz0,tz);
 373
 374             fjx0             = _mm256_add_ps(fjx0,tx);
 375             fjy0             = _mm256_add_ps(fjy0,ty);
 376             fjz0             = _mm256_add_ps(fjz0,tz);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             r10              = _mm256_mul_ps(rsq10,rinv10);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq10             = _mm256_mul_ps(iq1,jq0);
 386
 387             /* EWALD ELECTROSTATICS */
 388
 389             /* Analytical PME correction */
 390             zeta2            = _mm256_mul_ps(beta2,rsq10);
 391             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 392             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 393             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 394             felec            = _mm256_mul_ps(qq10,felec);
 395             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 396             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 397             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 398             velec            = _mm256_mul_ps(qq10,velec);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velecsum         = _mm256_add_ps(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_ps(fscal,dx10);
 407             ty               = _mm256_mul_ps(fscal,dy10);
 408             tz               = _mm256_mul_ps(fscal,dz10);
 409
 410             /* Update vectorial force */
 411             fix1             = _mm256_add_ps(fix1,tx);
 412             fiy1             = _mm256_add_ps(fiy1,ty);
 413             fiz1             = _mm256_add_ps(fiz1,tz);
 414
 415             fjx0             = _mm256_add_ps(fjx0,tx);
 416             fjy0             = _mm256_add_ps(fjy0,ty);
 417             fjz0             = _mm256_add_ps(fjz0,tz);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             r20              = _mm256_mul_ps(rsq20,rinv20);
 424
 425             /* Compute parameters for interactions between i and j atoms */
 426             qq20             = _mm256_mul_ps(iq2,jq0);
 427
 428             /* EWALD ELECTROSTATICS */
 429
 430             /* Analytical PME correction */
 431             zeta2            = _mm256_mul_ps(beta2,rsq20);
 432             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 433             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 434             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 435             felec            = _mm256_mul_ps(qq20,felec);
 436             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 437             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 438             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 439             velec            = _mm256_mul_ps(qq20,velec);
 440
 441             /* Update potential sum for this i atom from the interaction with this j atom. */
 442             velecsum         = _mm256_add_ps(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm256_mul_ps(fscal,dx20);
 448             ty               = _mm256_mul_ps(fscal,dy20);
 449             tz               = _mm256_mul_ps(fscal,dz20);
 450
 451             /* Update vectorial force */
 452             fix2             = _mm256_add_ps(fix2,tx);
 453             fiy2             = _mm256_add_ps(fiy2,ty);
 454             fiz2             = _mm256_add_ps(fiz2,tz);
 455
 456             fjx0             = _mm256_add_ps(fjx0,tx);
 457             fjy0             = _mm256_add_ps(fjy0,ty);
 458             fjz0             = _mm256_add_ps(fjz0,tz);
 459
 460             fjptrA             = f+j_coord_offsetA;
 461             fjptrB             = f+j_coord_offsetB;
 462             fjptrC             = f+j_coord_offsetC;
 463             fjptrD             = f+j_coord_offsetD;
 464             fjptrE             = f+j_coord_offsetE;
 465             fjptrF             = f+j_coord_offsetF;
 466             fjptrG             = f+j_coord_offsetG;
 467             fjptrH             = f+j_coord_offsetH;
 468
 469             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 470
 471             /* Inner loop uses 289 flops */
 472         }
 473
 474         if(jidx<j_index_end)
 475         {
 476
 477             /* Get j neighbor index, and coordinate index */
 478             jnrlistA         = jjnr[jidx];
 479             jnrlistB         = jjnr[jidx+1];
 480             jnrlistC         = jjnr[jidx+2];
 481             jnrlistD         = jjnr[jidx+3];
 482             jnrlistE         = jjnr[jidx+4];
 483             jnrlistF         = jjnr[jidx+5];
 484             jnrlistG         = jjnr[jidx+6];
 485             jnrlistH         = jjnr[jidx+7];
 486             /* Sign of each element will be negative for non-real atoms.
 487              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 488              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 489              */
 490             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 491                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 492
 493             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 494             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 495             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 496             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 497             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 498             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 499             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 500             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 501             j_coord_offsetA  = DIM*jnrA;
 502             j_coord_offsetB  = DIM*jnrB;
 503             j_coord_offsetC  = DIM*jnrC;
 504             j_coord_offsetD  = DIM*jnrD;
 505             j_coord_offsetE  = DIM*jnrE;
 506             j_coord_offsetF  = DIM*jnrF;
 507             j_coord_offsetG  = DIM*jnrG;
 508             j_coord_offsetH  = DIM*jnrH;
 509
 510             /* load j atom coordinates */
 511             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 512                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 513                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 514                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 515                                                  &jx0,&jy0,&jz0);
 516
 517             /* Calculate displacement vector */
 518             dx00             = _mm256_sub_ps(ix0,jx0);
 519             dy00             = _mm256_sub_ps(iy0,jy0);
 520             dz00             = _mm256_sub_ps(iz0,jz0);
 521             dx10             = _mm256_sub_ps(ix1,jx0);
 522             dy10             = _mm256_sub_ps(iy1,jy0);
 523             dz10             = _mm256_sub_ps(iz1,jz0);
 524             dx20             = _mm256_sub_ps(ix2,jx0);
 525             dy20             = _mm256_sub_ps(iy2,jy0);
 526             dz20             = _mm256_sub_ps(iz2,jz0);
 527
 528             /* Calculate squared distance and things based on it */
 529             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 530             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 531             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 532
 533             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 534             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 535             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 536
 537             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 538             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 539             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 540
 541             /* Load parameters for j particles */
 542             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 543                                                                  charge+jnrC+0,charge+jnrD+0,
 544                                                                  charge+jnrE+0,charge+jnrF+0,
 545                                                                  charge+jnrG+0,charge+jnrH+0);
 546             vdwjidx0A        = 2*vdwtype[jnrA+0];
 547             vdwjidx0B        = 2*vdwtype[jnrB+0];
 548             vdwjidx0C        = 2*vdwtype[jnrC+0];
 549             vdwjidx0D        = 2*vdwtype[jnrD+0];
 550             vdwjidx0E        = 2*vdwtype[jnrE+0];
 551             vdwjidx0F        = 2*vdwtype[jnrF+0];
 552             vdwjidx0G        = 2*vdwtype[jnrG+0];
 553             vdwjidx0H        = 2*vdwtype[jnrH+0];
 554
 555             fjx0             = _mm256_setzero_ps();
 556             fjy0             = _mm256_setzero_ps();
 557             fjz0             = _mm256_setzero_ps();
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r00              = _mm256_mul_ps(rsq00,rinv00);
 564             r00              = _mm256_andnot_ps(dummy_mask,r00);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq00             = _mm256_mul_ps(iq0,jq0);
 568             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 569                                             vdwioffsetptr0+vdwjidx0B,
 570                                             vdwioffsetptr0+vdwjidx0C,
 571                                             vdwioffsetptr0+vdwjidx0D,
 572                                             vdwioffsetptr0+vdwjidx0E,
 573                                             vdwioffsetptr0+vdwjidx0F,
 574                                             vdwioffsetptr0+vdwjidx0G,
 575                                             vdwioffsetptr0+vdwjidx0H,
 576                                             &c6_00,&c12_00);
 577
 578             /* Calculate table index by multiplying r with table scale and truncate to integer */
 579             rt               = _mm256_mul_ps(r00,vftabscale);
 580             vfitab           = _mm256_cvttps_epi32(rt);
 581             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 582             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 583             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 584             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 585             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 586             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 587
 588             /* EWALD ELECTROSTATICS */
 589
 590             /* Analytical PME correction */
 591             zeta2            = _mm256_mul_ps(beta2,rsq00);
 592             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 593             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 594             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 595             felec            = _mm256_mul_ps(qq00,felec);
 596             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 597             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 598             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 599             velec            = _mm256_mul_ps(qq00,velec);
 600
 601             /* CUBIC SPLINE TABLE DISPERSION */
 602             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 603                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 604             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 605                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 606             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 607                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 608             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 609                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 610             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 611             Heps             = _mm256_mul_ps(vfeps,H);
 612             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 613             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 614             vvdw6            = _mm256_mul_ps(c6_00,VV);
 615             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 616             fvdw6            = _mm256_mul_ps(c6_00,FF);
 617
 618             /* CUBIC SPLINE TABLE REPULSION */
 619             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 620             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 621             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 622                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 623             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 625             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 627             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 628                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 629             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 630             Heps             = _mm256_mul_ps(vfeps,H);
 631             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 632             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 633             vvdw12           = _mm256_mul_ps(c12_00,VV);
 634             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 635             fvdw12           = _mm256_mul_ps(c12_00,FF);
 636             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 637             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm256_andnot_ps(dummy_mask,velec);
 641             velecsum         = _mm256_add_ps(velecsum,velec);
 642             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 643             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 644
 645             fscal            = _mm256_add_ps(felec,fvdw);
 646
 647             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 648
 649             /* Calculate temporary vectorial force */
 650             tx               = _mm256_mul_ps(fscal,dx00);
 651             ty               = _mm256_mul_ps(fscal,dy00);
 652             tz               = _mm256_mul_ps(fscal,dz00);
 653
 654             /* Update vectorial force */
 655             fix0             = _mm256_add_ps(fix0,tx);
 656             fiy0             = _mm256_add_ps(fiy0,ty);
 657             fiz0             = _mm256_add_ps(fiz0,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             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             r10              = _mm256_mul_ps(rsq10,rinv10);
 668             r10              = _mm256_andnot_ps(dummy_mask,r10);
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq10             = _mm256_mul_ps(iq1,jq0);
 672
 673             /* EWALD ELECTROSTATICS */
 674
 675             /* Analytical PME correction */
 676             zeta2            = _mm256_mul_ps(beta2,rsq10);
 677             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 678             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 679             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 680             felec            = _mm256_mul_ps(qq10,felec);
 681             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 682             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 683             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 684             velec            = _mm256_mul_ps(qq10,velec);
 685
 686             /* Update potential sum for this i atom from the interaction with this j atom. */
 687             velec            = _mm256_andnot_ps(dummy_mask,velec);
 688             velecsum         = _mm256_add_ps(velecsum,velec);
 689
 690             fscal            = felec;
 691
 692             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm256_mul_ps(fscal,dx10);
 696             ty               = _mm256_mul_ps(fscal,dy10);
 697             tz               = _mm256_mul_ps(fscal,dz10);
 698
 699             /* Update vectorial force */
 700             fix1             = _mm256_add_ps(fix1,tx);
 701             fiy1             = _mm256_add_ps(fiy1,ty);
 702             fiz1             = _mm256_add_ps(fiz1,tz);
 703
 704             fjx0             = _mm256_add_ps(fjx0,tx);
 705             fjy0             = _mm256_add_ps(fjy0,ty);
 706             fjz0             = _mm256_add_ps(fjz0,tz);
 707
 708             /**************************
 709              * CALCULATE INTERACTIONS *
 710              **************************/
 711
 712             r20              = _mm256_mul_ps(rsq20,rinv20);
 713             r20              = _mm256_andnot_ps(dummy_mask,r20);
 714
 715             /* Compute parameters for interactions between i and j atoms */
 716             qq20             = _mm256_mul_ps(iq2,jq0);
 717
 718             /* EWALD ELECTROSTATICS */
 719
 720             /* Analytical PME correction */
 721             zeta2            = _mm256_mul_ps(beta2,rsq20);
 722             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 723             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 724             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 725             felec            = _mm256_mul_ps(qq20,felec);
 726             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 727             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 728             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 729             velec            = _mm256_mul_ps(qq20,velec);
 730
 731             /* Update potential sum for this i atom from the interaction with this j atom. */
 732             velec            = _mm256_andnot_ps(dummy_mask,velec);
 733             velecsum         = _mm256_add_ps(velecsum,velec);
 734
 735             fscal            = felec;
 736
 737             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 738
 739             /* Calculate temporary vectorial force */
 740             tx               = _mm256_mul_ps(fscal,dx20);
 741             ty               = _mm256_mul_ps(fscal,dy20);
 742             tz               = _mm256_mul_ps(fscal,dz20);
 743
 744             /* Update vectorial force */
 745             fix2             = _mm256_add_ps(fix2,tx);
 746             fiy2             = _mm256_add_ps(fiy2,ty);
 747             fiz2             = _mm256_add_ps(fiz2,tz);
 748
 749             fjx0             = _mm256_add_ps(fjx0,tx);
 750             fjy0             = _mm256_add_ps(fjy0,ty);
 751             fjz0             = _mm256_add_ps(fjz0,tz);
 752
 753             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 754             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 755             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 756             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 757             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 758             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 759             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 760             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 761
 762             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 763
 764             /* Inner loop uses 292 flops */
 765         }
 766
 767         /* End of innermost loop */
 768
 769         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 770                                                  f+i_coord_offset,fshift+i_shift_offset);
 771
 772         ggid                        = gid[iidx];
 773         /* Update potential energies */
 774         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 775         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 776
 777         /* Increment number of inner iterations */
 778         inneriter                  += j_index_end - j_index_start;
 779
 780         /* Outer loop uses 20 flops */
 781     }
 782
 783     /* Increment number of outer iterations */
 784     outeriter        += nri;
 785
 786     /* Update outer/inner flops */
 787
 788     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*292);
 789 }
 790 /*
 791  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_single
 792  * Electrostatics interaction: Ewald
 793  * VdW interaction:            CubicSplineTable
 794  * Geometry:                   Water3-Particle
 795  * Calculate force/pot:        Force
 796  */
 797 void
 798 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_single
 799                     (t_nblist                    * gmx_restrict       nlist,
 800                      rvec                        * gmx_restrict          xx,
 801                      rvec                        * gmx_restrict          ff,
 802                      t_forcerec                  * gmx_restrict          fr,
 803                      t_mdatoms                   * gmx_restrict     mdatoms,
 804                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 805                      t_nrnb                      * gmx_restrict        nrnb)
 806 {
 807     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 808      * just 0 for non-waters.
 809      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 810      * jnr indices corresponding to data put in the four positions in the SIMD register.
 811      */
 812     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 813     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 814     int              jnrA,jnrB,jnrC,jnrD;
 815     int              jnrE,jnrF,jnrG,jnrH;
 816     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 817     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 818     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 819     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 820     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 821     real             rcutoff_scalar;
 822     real             *shiftvec,*fshift,*x,*f;
 823     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 824     real             scratch[4*DIM];
 825     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 826     real *           vdwioffsetptr0;
 827     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 828     real *           vdwioffsetptr1;
 829     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 830     real *           vdwioffsetptr2;
 831     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 832     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 833     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 834     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 835     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 836     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 837     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 838     real             *charge;
 839     int              nvdwtype;
 840     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 841     int              *vdwtype;
 842     real             *vdwparam;
 843     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 844     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 845     __m256i          vfitab;
 846     __m128i          vfitab_lo,vfitab_hi;
 847     __m128i          ifour       = _mm_set1_epi32(4);
 848     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 849     real             *vftab;
 850     __m256i          ewitab;
 851     __m128i          ewitab_lo,ewitab_hi;
 852     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 853     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 854     real             *ewtab;
 855     __m256           dummy_mask,cutoff_mask;
 856     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 857     __m256           one     = _mm256_set1_ps(1.0);
 858     __m256           two     = _mm256_set1_ps(2.0);
 859     x                = xx[0];
 860     f                = ff[0];
 861
 862     nri              = nlist->nri;
 863     iinr             = nlist->iinr;
 864     jindex           = nlist->jindex;
 865     jjnr             = nlist->jjnr;
 866     shiftidx         = nlist->shift;
 867     gid              = nlist->gid;
 868     shiftvec         = fr->shift_vec[0];
 869     fshift           = fr->fshift[0];
 870     facel            = _mm256_set1_ps(fr->epsfac);
 871     charge           = mdatoms->chargeA;
 872     nvdwtype         = fr->ntype;
 873     vdwparam         = fr->nbfp;
 874     vdwtype          = mdatoms->typeA;
 875
 876     vftab            = kernel_data->table_vdw->data;
 877     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 878
 879     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 880     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 881     beta2            = _mm256_mul_ps(beta,beta);
 882     beta3            = _mm256_mul_ps(beta,beta2);
 883
 884     ewtab            = fr->ic->tabq_coul_F;
 885     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 886     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 887
 888     /* Setup water-specific parameters */
 889     inr              = nlist->iinr[0];
 890     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 891     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 892     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 893     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 894
 895     /* Avoid stupid compiler warnings */
 896     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 897     j_coord_offsetA = 0;
 898     j_coord_offsetB = 0;
 899     j_coord_offsetC = 0;
 900     j_coord_offsetD = 0;
 901     j_coord_offsetE = 0;
 902     j_coord_offsetF = 0;
 903     j_coord_offsetG = 0;
 904     j_coord_offsetH = 0;
 905
 906     outeriter        = 0;
 907     inneriter        = 0;
 908
 909     for(iidx=0;iidx<4*DIM;iidx++)
 910     {
 911         scratch[iidx] = 0.0;
 912     }
 913
 914     /* Start outer loop over neighborlists */
 915     for(iidx=0; iidx<nri; iidx++)
 916     {
 917         /* Load shift vector for this list */
 918         i_shift_offset   = DIM*shiftidx[iidx];
 919
 920         /* Load limits for loop over neighbors */
 921         j_index_start    = jindex[iidx];
 922         j_index_end      = jindex[iidx+1];
 923
 924         /* Get outer coordinate index */
 925         inr              = iinr[iidx];
 926         i_coord_offset   = DIM*inr;
 927
 928         /* Load i particle coords and add shift vector */
 929         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 930                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 931
 932         fix0             = _mm256_setzero_ps();
 933         fiy0             = _mm256_setzero_ps();
 934         fiz0             = _mm256_setzero_ps();
 935         fix1             = _mm256_setzero_ps();
 936         fiy1             = _mm256_setzero_ps();
 937         fiz1             = _mm256_setzero_ps();
 938         fix2             = _mm256_setzero_ps();
 939         fiy2             = _mm256_setzero_ps();
 940         fiz2             = _mm256_setzero_ps();
 941
 942         /* Start inner kernel loop */
 943         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 944         {
 945
 946             /* Get j neighbor index, and coordinate index */
 947             jnrA             = jjnr[jidx];
 948             jnrB             = jjnr[jidx+1];
 949             jnrC             = jjnr[jidx+2];
 950             jnrD             = jjnr[jidx+3];
 951             jnrE             = jjnr[jidx+4];
 952             jnrF             = jjnr[jidx+5];
 953             jnrG             = jjnr[jidx+6];
 954             jnrH             = jjnr[jidx+7];
 955             j_coord_offsetA  = DIM*jnrA;
 956             j_coord_offsetB  = DIM*jnrB;
 957             j_coord_offsetC  = DIM*jnrC;
 958             j_coord_offsetD  = DIM*jnrD;
 959             j_coord_offsetE  = DIM*jnrE;
 960             j_coord_offsetF  = DIM*jnrF;
 961             j_coord_offsetG  = DIM*jnrG;
 962             j_coord_offsetH  = DIM*jnrH;
 963
 964             /* load j atom coordinates */
 965             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 966                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 967                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 968                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 969                                                  &jx0,&jy0,&jz0);
 970
 971             /* Calculate displacement vector */
 972             dx00             = _mm256_sub_ps(ix0,jx0);
 973             dy00             = _mm256_sub_ps(iy0,jy0);
 974             dz00             = _mm256_sub_ps(iz0,jz0);
 975             dx10             = _mm256_sub_ps(ix1,jx0);
 976             dy10             = _mm256_sub_ps(iy1,jy0);
 977             dz10             = _mm256_sub_ps(iz1,jz0);
 978             dx20             = _mm256_sub_ps(ix2,jx0);
 979             dy20             = _mm256_sub_ps(iy2,jy0);
 980             dz20             = _mm256_sub_ps(iz2,jz0);
 981
 982             /* Calculate squared distance and things based on it */
 983             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 984             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 985             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 986
 987             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 988             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 989             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 990
 991             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 992             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 993             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 994
 995             /* Load parameters for j particles */
 996             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 997                                                                  charge+jnrC+0,charge+jnrD+0,
 998                                                                  charge+jnrE+0,charge+jnrF+0,
 999                                                                  charge+jnrG+0,charge+jnrH+0);
1000             vdwjidx0A        = 2*vdwtype[jnrA+0];
1001             vdwjidx0B        = 2*vdwtype[jnrB+0];
1002             vdwjidx0C        = 2*vdwtype[jnrC+0];
1003             vdwjidx0D        = 2*vdwtype[jnrD+0];
1004             vdwjidx0E        = 2*vdwtype[jnrE+0];
1005             vdwjidx0F        = 2*vdwtype[jnrF+0];
1006             vdwjidx0G        = 2*vdwtype[jnrG+0];
1007             vdwjidx0H        = 2*vdwtype[jnrH+0];
1008
1009             fjx0             = _mm256_setzero_ps();
1010             fjy0             = _mm256_setzero_ps();
1011             fjz0             = _mm256_setzero_ps();
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r00              = _mm256_mul_ps(rsq00,rinv00);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq00             = _mm256_mul_ps(iq0,jq0);
1021             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1022                                             vdwioffsetptr0+vdwjidx0B,
1023                                             vdwioffsetptr0+vdwjidx0C,
1024                                             vdwioffsetptr0+vdwjidx0D,
1025                                             vdwioffsetptr0+vdwjidx0E,
1026                                             vdwioffsetptr0+vdwjidx0F,
1027                                             vdwioffsetptr0+vdwjidx0G,
1028                                             vdwioffsetptr0+vdwjidx0H,
1029                                             &c6_00,&c12_00);
1030
1031             /* Calculate table index by multiplying r with table scale and truncate to integer */
1032             rt               = _mm256_mul_ps(r00,vftabscale);
1033             vfitab           = _mm256_cvttps_epi32(rt);
1034             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1035             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1036             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1037             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1038             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1039             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1040
1041             /* EWALD ELECTROSTATICS */
1042
1043             /* Analytical PME correction */
1044             zeta2            = _mm256_mul_ps(beta2,rsq00);
1045             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1046             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1047             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1048             felec            = _mm256_mul_ps(qq00,felec);
1049
1050             /* CUBIC SPLINE TABLE DISPERSION */
1051             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1052                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1053             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1054                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1055             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1056                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1057             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1058                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1059             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1060             Heps             = _mm256_mul_ps(vfeps,H);
1061             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1062             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1063             fvdw6            = _mm256_mul_ps(c6_00,FF);
1064
1065             /* CUBIC SPLINE TABLE REPULSION */
1066             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1067             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1068             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1069                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1070             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1071                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1072             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1073                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1074             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1075                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1076             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1077             Heps             = _mm256_mul_ps(vfeps,H);
1078             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1079             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1080             fvdw12           = _mm256_mul_ps(c12_00,FF);
1081             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1082
1083             fscal            = _mm256_add_ps(felec,fvdw);
1084
1085             /* Calculate temporary vectorial force */
1086             tx               = _mm256_mul_ps(fscal,dx00);
1087             ty               = _mm256_mul_ps(fscal,dy00);
1088             tz               = _mm256_mul_ps(fscal,dz00);
1089
1090             /* Update vectorial force */
1091             fix0             = _mm256_add_ps(fix0,tx);
1092             fiy0             = _mm256_add_ps(fiy0,ty);
1093             fiz0             = _mm256_add_ps(fiz0,tz);
1094
1095             fjx0             = _mm256_add_ps(fjx0,tx);
1096             fjy0             = _mm256_add_ps(fjy0,ty);
1097             fjz0             = _mm256_add_ps(fjz0,tz);
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             r10              = _mm256_mul_ps(rsq10,rinv10);
1104
1105             /* Compute parameters for interactions between i and j atoms */
1106             qq10             = _mm256_mul_ps(iq1,jq0);
1107
1108             /* EWALD ELECTROSTATICS */
1109
1110             /* Analytical PME correction */
1111             zeta2            = _mm256_mul_ps(beta2,rsq10);
1112             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1113             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1114             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1115             felec            = _mm256_mul_ps(qq10,felec);
1116
1117             fscal            = felec;
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm256_mul_ps(fscal,dx10);
1121             ty               = _mm256_mul_ps(fscal,dy10);
1122             tz               = _mm256_mul_ps(fscal,dz10);
1123
1124             /* Update vectorial force */
1125             fix1             = _mm256_add_ps(fix1,tx);
1126             fiy1             = _mm256_add_ps(fiy1,ty);
1127             fiz1             = _mm256_add_ps(fiz1,tz);
1128
1129             fjx0             = _mm256_add_ps(fjx0,tx);
1130             fjy0             = _mm256_add_ps(fjy0,ty);
1131             fjz0             = _mm256_add_ps(fjz0,tz);
1132
1133             /**************************
1134              * CALCULATE INTERACTIONS *
1135              **************************/
1136
1137             r20              = _mm256_mul_ps(rsq20,rinv20);
1138
1139             /* Compute parameters for interactions between i and j atoms */
1140             qq20             = _mm256_mul_ps(iq2,jq0);
1141
1142             /* EWALD ELECTROSTATICS */
1143
1144             /* Analytical PME correction */
1145             zeta2            = _mm256_mul_ps(beta2,rsq20);
1146             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1147             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1148             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1149             felec            = _mm256_mul_ps(qq20,felec);
1150
1151             fscal            = felec;
1152
1153             /* Calculate temporary vectorial force */
1154             tx               = _mm256_mul_ps(fscal,dx20);
1155             ty               = _mm256_mul_ps(fscal,dy20);
1156             tz               = _mm256_mul_ps(fscal,dz20);
1157
1158             /* Update vectorial force */
1159             fix2             = _mm256_add_ps(fix2,tx);
1160             fiy2             = _mm256_add_ps(fiy2,ty);
1161             fiz2             = _mm256_add_ps(fiz2,tz);
1162
1163             fjx0             = _mm256_add_ps(fjx0,tx);
1164             fjy0             = _mm256_add_ps(fjy0,ty);
1165             fjz0             = _mm256_add_ps(fjz0,tz);
1166
1167             fjptrA             = f+j_coord_offsetA;
1168             fjptrB             = f+j_coord_offsetB;
1169             fjptrC             = f+j_coord_offsetC;
1170             fjptrD             = f+j_coord_offsetD;
1171             fjptrE             = f+j_coord_offsetE;
1172             fjptrF             = f+j_coord_offsetF;
1173             fjptrG             = f+j_coord_offsetG;
1174             fjptrH             = f+j_coord_offsetH;
1175
1176             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1177
1178             /* Inner loop uses 197 flops */
1179         }
1180
1181         if(jidx<j_index_end)
1182         {
1183
1184             /* Get j neighbor index, and coordinate index */
1185             jnrlistA         = jjnr[jidx];
1186             jnrlistB         = jjnr[jidx+1];
1187             jnrlistC         = jjnr[jidx+2];
1188             jnrlistD         = jjnr[jidx+3];
1189             jnrlistE         = jjnr[jidx+4];
1190             jnrlistF         = jjnr[jidx+5];
1191             jnrlistG         = jjnr[jidx+6];
1192             jnrlistH         = jjnr[jidx+7];
1193             /* Sign of each element will be negative for non-real atoms.
1194              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1195              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1196              */
1197             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1198                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1199
1200             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1201             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1202             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1203             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1204             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1205             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1206             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1207             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1208             j_coord_offsetA  = DIM*jnrA;
1209             j_coord_offsetB  = DIM*jnrB;
1210             j_coord_offsetC  = DIM*jnrC;
1211             j_coord_offsetD  = DIM*jnrD;
1212             j_coord_offsetE  = DIM*jnrE;
1213             j_coord_offsetF  = DIM*jnrF;
1214             j_coord_offsetG  = DIM*jnrG;
1215             j_coord_offsetH  = DIM*jnrH;
1216
1217             /* load j atom coordinates */
1218             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1219                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1220                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1221                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1222                                                  &jx0,&jy0,&jz0);
1223
1224             /* Calculate displacement vector */
1225             dx00             = _mm256_sub_ps(ix0,jx0);
1226             dy00             = _mm256_sub_ps(iy0,jy0);
1227             dz00             = _mm256_sub_ps(iz0,jz0);
1228             dx10             = _mm256_sub_ps(ix1,jx0);
1229             dy10             = _mm256_sub_ps(iy1,jy0);
1230             dz10             = _mm256_sub_ps(iz1,jz0);
1231             dx20             = _mm256_sub_ps(ix2,jx0);
1232             dy20             = _mm256_sub_ps(iy2,jy0);
1233             dz20             = _mm256_sub_ps(iz2,jz0);
1234
1235             /* Calculate squared distance and things based on it */
1236             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1237             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1238             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1239
1240             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1241             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1242             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1243
1244             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1245             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1246             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1247
1248             /* Load parameters for j particles */
1249             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1250                                                                  charge+jnrC+0,charge+jnrD+0,
1251                                                                  charge+jnrE+0,charge+jnrF+0,
1252                                                                  charge+jnrG+0,charge+jnrH+0);
1253             vdwjidx0A        = 2*vdwtype[jnrA+0];
1254             vdwjidx0B        = 2*vdwtype[jnrB+0];
1255             vdwjidx0C        = 2*vdwtype[jnrC+0];
1256             vdwjidx0D        = 2*vdwtype[jnrD+0];
1257             vdwjidx0E        = 2*vdwtype[jnrE+0];
1258             vdwjidx0F        = 2*vdwtype[jnrF+0];
1259             vdwjidx0G        = 2*vdwtype[jnrG+0];
1260             vdwjidx0H        = 2*vdwtype[jnrH+0];
1261
1262             fjx0             = _mm256_setzero_ps();
1263             fjy0             = _mm256_setzero_ps();
1264             fjz0             = _mm256_setzero_ps();
1265
1266             /**************************
1267              * CALCULATE INTERACTIONS *
1268              **************************/
1269
1270             r00              = _mm256_mul_ps(rsq00,rinv00);
1271             r00              = _mm256_andnot_ps(dummy_mask,r00);
1272
1273             /* Compute parameters for interactions between i and j atoms */
1274             qq00             = _mm256_mul_ps(iq0,jq0);
1275             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1276                                             vdwioffsetptr0+vdwjidx0B,
1277                                             vdwioffsetptr0+vdwjidx0C,
1278                                             vdwioffsetptr0+vdwjidx0D,
1279                                             vdwioffsetptr0+vdwjidx0E,
1280                                             vdwioffsetptr0+vdwjidx0F,
1281                                             vdwioffsetptr0+vdwjidx0G,
1282                                             vdwioffsetptr0+vdwjidx0H,
1283                                             &c6_00,&c12_00);
1284
1285             /* Calculate table index by multiplying r with table scale and truncate to integer */
1286             rt               = _mm256_mul_ps(r00,vftabscale);
1287             vfitab           = _mm256_cvttps_epi32(rt);
1288             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1289             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1290             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1291             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1292             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1293             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1294
1295             /* EWALD ELECTROSTATICS */
1296
1297             /* Analytical PME correction */
1298             zeta2            = _mm256_mul_ps(beta2,rsq00);
1299             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1300             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1301             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1302             felec            = _mm256_mul_ps(qq00,felec);
1303
1304             /* CUBIC SPLINE TABLE DISPERSION */
1305             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1307             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1309             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1311             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1313             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1314             Heps             = _mm256_mul_ps(vfeps,H);
1315             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1316             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1317             fvdw6            = _mm256_mul_ps(c6_00,FF);
1318
1319             /* CUBIC SPLINE TABLE REPULSION */
1320             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1321             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1322             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1324             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1326             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1328             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1330             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1331             Heps             = _mm256_mul_ps(vfeps,H);
1332             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1333             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1334             fvdw12           = _mm256_mul_ps(c12_00,FF);
1335             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1336
1337             fscal            = _mm256_add_ps(felec,fvdw);
1338
1339             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1340
1341             /* Calculate temporary vectorial force */
1342             tx               = _mm256_mul_ps(fscal,dx00);
1343             ty               = _mm256_mul_ps(fscal,dy00);
1344             tz               = _mm256_mul_ps(fscal,dz00);
1345
1346             /* Update vectorial force */
1347             fix0             = _mm256_add_ps(fix0,tx);
1348             fiy0             = _mm256_add_ps(fiy0,ty);
1349             fiz0             = _mm256_add_ps(fiz0,tz);
1350
1351             fjx0             = _mm256_add_ps(fjx0,tx);
1352             fjy0             = _mm256_add_ps(fjy0,ty);
1353             fjz0             = _mm256_add_ps(fjz0,tz);
1354
1355             /**************************
1356              * CALCULATE INTERACTIONS *
1357              **************************/
1358
1359             r10              = _mm256_mul_ps(rsq10,rinv10);
1360             r10              = _mm256_andnot_ps(dummy_mask,r10);
1361
1362             /* Compute parameters for interactions between i and j atoms */
1363             qq10             = _mm256_mul_ps(iq1,jq0);
1364
1365             /* EWALD ELECTROSTATICS */
1366
1367             /* Analytical PME correction */
1368             zeta2            = _mm256_mul_ps(beta2,rsq10);
1369             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1370             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1371             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1372             felec            = _mm256_mul_ps(qq10,felec);
1373
1374             fscal            = felec;
1375
1376             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1377
1378             /* Calculate temporary vectorial force */
1379             tx               = _mm256_mul_ps(fscal,dx10);
1380             ty               = _mm256_mul_ps(fscal,dy10);
1381             tz               = _mm256_mul_ps(fscal,dz10);
1382
1383             /* Update vectorial force */
1384             fix1             = _mm256_add_ps(fix1,tx);
1385             fiy1             = _mm256_add_ps(fiy1,ty);
1386             fiz1             = _mm256_add_ps(fiz1,tz);
1387
1388             fjx0             = _mm256_add_ps(fjx0,tx);
1389             fjy0             = _mm256_add_ps(fjy0,ty);
1390             fjz0             = _mm256_add_ps(fjz0,tz);
1391
1392             /**************************
1393              * CALCULATE INTERACTIONS *
1394              **************************/
1395
1396             r20              = _mm256_mul_ps(rsq20,rinv20);
1397             r20              = _mm256_andnot_ps(dummy_mask,r20);
1398
1399             /* Compute parameters for interactions between i and j atoms */
1400             qq20             = _mm256_mul_ps(iq2,jq0);
1401
1402             /* EWALD ELECTROSTATICS */
1403
1404             /* Analytical PME correction */
1405             zeta2            = _mm256_mul_ps(beta2,rsq20);
1406             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1407             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1408             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1409             felec            = _mm256_mul_ps(qq20,felec);
1410
1411             fscal            = felec;
1412
1413             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1414
1415             /* Calculate temporary vectorial force */
1416             tx               = _mm256_mul_ps(fscal,dx20);
1417             ty               = _mm256_mul_ps(fscal,dy20);
1418             tz               = _mm256_mul_ps(fscal,dz20);
1419
1420             /* Update vectorial force */
1421             fix2             = _mm256_add_ps(fix2,tx);
1422             fiy2             = _mm256_add_ps(fiy2,ty);
1423             fiz2             = _mm256_add_ps(fiz2,tz);
1424
1425             fjx0             = _mm256_add_ps(fjx0,tx);
1426             fjy0             = _mm256_add_ps(fjy0,ty);
1427             fjz0             = _mm256_add_ps(fjz0,tz);
1428
1429             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1430             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1431             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1432             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1433             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1434             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1435             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1436             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1437
1438             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1439
1440             /* Inner loop uses 200 flops */
1441         }
1442
1443         /* End of innermost loop */
1444
1445         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1446                                                  f+i_coord_offset,fshift+i_shift_offset);
1447
1448         /* Increment number of inner iterations */
1449         inneriter                  += j_index_end - j_index_start;
1450
1451         /* Outer loop uses 18 flops */
1452     }
1453
1454     /* Increment number of outer iterations */
1455     outeriter        += nri;
1456
1457     /* Update outer/inner flops */
1458
1459     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*200);
1460 }