src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_VdwCSTab_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_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     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256i          vfitab;
 109     __m128i          vfitab_lo,vfitab_hi;
 110     __m128i          ifour       = _mm_set1_epi32(4);
 111     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 112     real             *vftab;
 113     __m256i          ewitab;
 114     __m128i          ewitab_lo,ewitab_hi;
 115     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 116     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 117     real             *ewtab;
 118     __m256           dummy_mask,cutoff_mask;
 119     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 120     __m256           one     = _mm256_set1_ps(1.0);
 121     __m256           two     = _mm256_set1_ps(2.0);
 122     x                = xx[0];
 123     f                = ff[0];
 124
 125     nri              = nlist->nri;
 126     iinr             = nlist->iinr;
 127     jindex           = nlist->jindex;
 128     jjnr             = nlist->jjnr;
 129     shiftidx         = nlist->shift;
 130     gid              = nlist->gid;
 131     shiftvec         = fr->shift_vec[0];
 132     fshift           = fr->fshift[0];
 133     facel            = _mm256_set1_ps(fr->epsfac);
 134     charge           = mdatoms->chargeA;
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138
 139     vftab            = kernel_data->table_vdw->data;
 140     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 141
 142     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 143     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 144     beta2            = _mm256_mul_ps(beta,beta);
 145     beta3            = _mm256_mul_ps(beta,beta2);
 146
 147     ewtab            = fr->ic->tabq_coul_FDV0;
 148     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 149     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 150
 151     /* Setup water-specific parameters */
 152     inr              = nlist->iinr[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     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 156     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 157
 158     /* Avoid stupid compiler warnings */
 159     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 160     j_coord_offsetA = 0;
 161     j_coord_offsetB = 0;
 162     j_coord_offsetC = 0;
 163     j_coord_offsetD = 0;
 164     j_coord_offsetE = 0;
 165     j_coord_offsetF = 0;
 166     j_coord_offsetG = 0;
 167     j_coord_offsetH = 0;
 168
 169     outeriter        = 0;
 170     inneriter        = 0;
 171
 172     for(iidx=0;iidx<4*DIM;iidx++)
 173     {
 174         scratch[iidx] = 0.0;
 175     }
 176
 177     /* Start outer loop over neighborlists */
 178     for(iidx=0; iidx<nri; iidx++)
 179     {
 180         /* Load shift vector for this list */
 181         i_shift_offset   = DIM*shiftidx[iidx];
 182
 183         /* Load limits for loop over neighbors */
 184         j_index_start    = jindex[iidx];
 185         j_index_end      = jindex[iidx+1];
 186
 187         /* Get outer coordinate index */
 188         inr              = iinr[iidx];
 189         i_coord_offset   = DIM*inr;
 190
 191         /* Load i particle coords and add shift vector */
 192         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 193                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 194
 195         fix0             = _mm256_setzero_ps();
 196         fiy0             = _mm256_setzero_ps();
 197         fiz0             = _mm256_setzero_ps();
 198         fix1             = _mm256_setzero_ps();
 199         fiy1             = _mm256_setzero_ps();
 200         fiz1             = _mm256_setzero_ps();
 201         fix2             = _mm256_setzero_ps();
 202         fiy2             = _mm256_setzero_ps();
 203         fiz2             = _mm256_setzero_ps();
 204         fix3             = _mm256_setzero_ps();
 205         fiy3             = _mm256_setzero_ps();
 206         fiz3             = _mm256_setzero_ps();
 207
 208         /* Reset potential sums */
 209         velecsum         = _mm256_setzero_ps();
 210         vvdwsum          = _mm256_setzero_ps();
 211
 212         /* Start inner kernel loop */
 213         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 214         {
 215
 216             /* Get j neighbor index, and coordinate index */
 217             jnrA             = jjnr[jidx];
 218             jnrB             = jjnr[jidx+1];
 219             jnrC             = jjnr[jidx+2];
 220             jnrD             = jjnr[jidx+3];
 221             jnrE             = jjnr[jidx+4];
 222             jnrF             = jjnr[jidx+5];
 223             jnrG             = jjnr[jidx+6];
 224             jnrH             = jjnr[jidx+7];
 225             j_coord_offsetA  = DIM*jnrA;
 226             j_coord_offsetB  = DIM*jnrB;
 227             j_coord_offsetC  = DIM*jnrC;
 228             j_coord_offsetD  = DIM*jnrD;
 229             j_coord_offsetE  = DIM*jnrE;
 230             j_coord_offsetF  = DIM*jnrF;
 231             j_coord_offsetG  = DIM*jnrG;
 232             j_coord_offsetH  = DIM*jnrH;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 237                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 238                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 239                                                  &jx0,&jy0,&jz0);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm256_sub_ps(ix0,jx0);
 243             dy00             = _mm256_sub_ps(iy0,jy0);
 244             dz00             = _mm256_sub_ps(iz0,jz0);
 245             dx10             = _mm256_sub_ps(ix1,jx0);
 246             dy10             = _mm256_sub_ps(iy1,jy0);
 247             dz10             = _mm256_sub_ps(iz1,jz0);
 248             dx20             = _mm256_sub_ps(ix2,jx0);
 249             dy20             = _mm256_sub_ps(iy2,jy0);
 250             dz20             = _mm256_sub_ps(iz2,jz0);
 251             dx30             = _mm256_sub_ps(ix3,jx0);
 252             dy30             = _mm256_sub_ps(iy3,jy0);
 253             dz30             = _mm256_sub_ps(iz3,jz0);
 254
 255             /* Calculate squared distance and things based on it */
 256             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 257             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 258             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 259             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 260
 261             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 262             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 263             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 264             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 265
 266             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 267             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 268             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 269
 270             /* Load parameters for j particles */
 271             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 272                                                                  charge+jnrC+0,charge+jnrD+0,
 273                                                                  charge+jnrE+0,charge+jnrF+0,
 274                                                                  charge+jnrG+0,charge+jnrH+0);
 275             vdwjidx0A        = 2*vdwtype[jnrA+0];
 276             vdwjidx0B        = 2*vdwtype[jnrB+0];
 277             vdwjidx0C        = 2*vdwtype[jnrC+0];
 278             vdwjidx0D        = 2*vdwtype[jnrD+0];
 279             vdwjidx0E        = 2*vdwtype[jnrE+0];
 280             vdwjidx0F        = 2*vdwtype[jnrF+0];
 281             vdwjidx0G        = 2*vdwtype[jnrG+0];
 282             vdwjidx0H        = 2*vdwtype[jnrH+0];
 283
 284             fjx0             = _mm256_setzero_ps();
 285             fjy0             = _mm256_setzero_ps();
 286             fjz0             = _mm256_setzero_ps();
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             r00              = _mm256_mul_ps(rsq00,rinv00);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 296                                             vdwioffsetptr0+vdwjidx0B,
 297                                             vdwioffsetptr0+vdwjidx0C,
 298                                             vdwioffsetptr0+vdwjidx0D,
 299                                             vdwioffsetptr0+vdwjidx0E,
 300                                             vdwioffsetptr0+vdwjidx0F,
 301                                             vdwioffsetptr0+vdwjidx0G,
 302                                             vdwioffsetptr0+vdwjidx0H,
 303                                             &c6_00,&c12_00);
 304
 305             /* Calculate table index by multiplying r with table scale and truncate to integer */
 306             rt               = _mm256_mul_ps(r00,vftabscale);
 307             vfitab           = _mm256_cvttps_epi32(rt);
 308             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 309             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 310             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 311             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 312             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 313             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 314
 315             /* CUBIC SPLINE TABLE DISPERSION */
 316             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 317                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 318             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 320             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 322             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 324             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 325             Heps             = _mm256_mul_ps(vfeps,H);
 326             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 327             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 328             vvdw6            = _mm256_mul_ps(c6_00,VV);
 329             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 330             fvdw6            = _mm256_mul_ps(c6_00,FF);
 331
 332             /* CUBIC SPLINE TABLE REPULSION */
 333             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 334             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 335             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 336                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 337             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 338                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 339             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 340                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 341             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 342                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 343             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 344             Heps             = _mm256_mul_ps(vfeps,H);
 345             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 346             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 347             vvdw12           = _mm256_mul_ps(c12_00,VV);
 348             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 349             fvdw12           = _mm256_mul_ps(c12_00,FF);
 350             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 351             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 355
 356             fscal            = fvdw;
 357
 358             /* Calculate temporary vectorial force */
 359             tx               = _mm256_mul_ps(fscal,dx00);
 360             ty               = _mm256_mul_ps(fscal,dy00);
 361             tz               = _mm256_mul_ps(fscal,dz00);
 362
 363             /* Update vectorial force */
 364             fix0             = _mm256_add_ps(fix0,tx);
 365             fiy0             = _mm256_add_ps(fiy0,ty);
 366             fiz0             = _mm256_add_ps(fiz0,tz);
 367
 368             fjx0             = _mm256_add_ps(fjx0,tx);
 369             fjy0             = _mm256_add_ps(fjy0,ty);
 370             fjz0             = _mm256_add_ps(fjz0,tz);
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             r10              = _mm256_mul_ps(rsq10,rinv10);
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq10             = _mm256_mul_ps(iq1,jq0);
 380
 381             /* EWALD ELECTROSTATICS */
 382
 383             /* Analytical PME correction */
 384             zeta2            = _mm256_mul_ps(beta2,rsq10);
 385             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 386             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 387             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 388             felec            = _mm256_mul_ps(qq10,felec);
 389             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 390             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 391             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 392             velec            = _mm256_mul_ps(qq10,velec);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velecsum         = _mm256_add_ps(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm256_mul_ps(fscal,dx10);
 401             ty               = _mm256_mul_ps(fscal,dy10);
 402             tz               = _mm256_mul_ps(fscal,dz10);
 403
 404             /* Update vectorial force */
 405             fix1             = _mm256_add_ps(fix1,tx);
 406             fiy1             = _mm256_add_ps(fiy1,ty);
 407             fiz1             = _mm256_add_ps(fiz1,tz);
 408
 409             fjx0             = _mm256_add_ps(fjx0,tx);
 410             fjy0             = _mm256_add_ps(fjy0,ty);
 411             fjz0             = _mm256_add_ps(fjz0,tz);
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             r20              = _mm256_mul_ps(rsq20,rinv20);
 418
 419             /* Compute parameters for interactions between i and j atoms */
 420             qq20             = _mm256_mul_ps(iq2,jq0);
 421
 422             /* EWALD ELECTROSTATICS */
 423
 424             /* Analytical PME correction */
 425             zeta2            = _mm256_mul_ps(beta2,rsq20);
 426             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 427             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 428             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 429             felec            = _mm256_mul_ps(qq20,felec);
 430             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 431             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 432             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 433             velec            = _mm256_mul_ps(qq20,velec);
 434
 435             /* Update potential sum for this i atom from the interaction with this j atom. */
 436             velecsum         = _mm256_add_ps(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             /* Calculate temporary vectorial force */
 441             tx               = _mm256_mul_ps(fscal,dx20);
 442             ty               = _mm256_mul_ps(fscal,dy20);
 443             tz               = _mm256_mul_ps(fscal,dz20);
 444
 445             /* Update vectorial force */
 446             fix2             = _mm256_add_ps(fix2,tx);
 447             fiy2             = _mm256_add_ps(fiy2,ty);
 448             fiz2             = _mm256_add_ps(fiz2,tz);
 449
 450             fjx0             = _mm256_add_ps(fjx0,tx);
 451             fjy0             = _mm256_add_ps(fjy0,ty);
 452             fjz0             = _mm256_add_ps(fjz0,tz);
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             r30              = _mm256_mul_ps(rsq30,rinv30);
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             qq30             = _mm256_mul_ps(iq3,jq0);
 462
 463             /* EWALD ELECTROSTATICS */
 464
 465             /* Analytical PME correction */
 466             zeta2            = _mm256_mul_ps(beta2,rsq30);
 467             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 468             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 469             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 470             felec            = _mm256_mul_ps(qq30,felec);
 471             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 472             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 473             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 474             velec            = _mm256_mul_ps(qq30,velec);
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velecsum         = _mm256_add_ps(velecsum,velec);
 478
 479             fscal            = felec;
 480
 481             /* Calculate temporary vectorial force */
 482             tx               = _mm256_mul_ps(fscal,dx30);
 483             ty               = _mm256_mul_ps(fscal,dy30);
 484             tz               = _mm256_mul_ps(fscal,dz30);
 485
 486             /* Update vectorial force */
 487             fix3             = _mm256_add_ps(fix3,tx);
 488             fiy3             = _mm256_add_ps(fiy3,ty);
 489             fiz3             = _mm256_add_ps(fiz3,tz);
 490
 491             fjx0             = _mm256_add_ps(fjx0,tx);
 492             fjy0             = _mm256_add_ps(fjy0,ty);
 493             fjz0             = _mm256_add_ps(fjz0,tz);
 494
 495             fjptrA             = f+j_coord_offsetA;
 496             fjptrB             = f+j_coord_offsetB;
 497             fjptrC             = f+j_coord_offsetC;
 498             fjptrD             = f+j_coord_offsetD;
 499             fjptrE             = f+j_coord_offsetE;
 500             fjptrF             = f+j_coord_offsetF;
 501             fjptrG             = f+j_coord_offsetG;
 502             fjptrH             = f+j_coord_offsetH;
 503
 504             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 505
 506             /* Inner loop uses 311 flops */
 507         }
 508
 509         if(jidx<j_index_end)
 510         {
 511
 512             /* Get j neighbor index, and coordinate index */
 513             jnrlistA         = jjnr[jidx];
 514             jnrlistB         = jjnr[jidx+1];
 515             jnrlistC         = jjnr[jidx+2];
 516             jnrlistD         = jjnr[jidx+3];
 517             jnrlistE         = jjnr[jidx+4];
 518             jnrlistF         = jjnr[jidx+5];
 519             jnrlistG         = jjnr[jidx+6];
 520             jnrlistH         = jjnr[jidx+7];
 521             /* Sign of each element will be negative for non-real atoms.
 522              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 523              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 524              */
 525             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 526                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 527
 528             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 529             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 530             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 531             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 532             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 533             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 534             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 535             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 536             j_coord_offsetA  = DIM*jnrA;
 537             j_coord_offsetB  = DIM*jnrB;
 538             j_coord_offsetC  = DIM*jnrC;
 539             j_coord_offsetD  = DIM*jnrD;
 540             j_coord_offsetE  = DIM*jnrE;
 541             j_coord_offsetF  = DIM*jnrF;
 542             j_coord_offsetG  = DIM*jnrG;
 543             j_coord_offsetH  = DIM*jnrH;
 544
 545             /* load j atom coordinates */
 546             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 547                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 548                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 549                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 550                                                  &jx0,&jy0,&jz0);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _mm256_sub_ps(ix0,jx0);
 554             dy00             = _mm256_sub_ps(iy0,jy0);
 555             dz00             = _mm256_sub_ps(iz0,jz0);
 556             dx10             = _mm256_sub_ps(ix1,jx0);
 557             dy10             = _mm256_sub_ps(iy1,jy0);
 558             dz10             = _mm256_sub_ps(iz1,jz0);
 559             dx20             = _mm256_sub_ps(ix2,jx0);
 560             dy20             = _mm256_sub_ps(iy2,jy0);
 561             dz20             = _mm256_sub_ps(iz2,jz0);
 562             dx30             = _mm256_sub_ps(ix3,jx0);
 563             dy30             = _mm256_sub_ps(iy3,jy0);
 564             dz30             = _mm256_sub_ps(iz3,jz0);
 565
 566             /* Calculate squared distance and things based on it */
 567             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 568             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 569             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 570             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 571
 572             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 573             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 574             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 575             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 576
 577             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 578             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 579             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 580
 581             /* Load parameters for j particles */
 582             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 583                                                                  charge+jnrC+0,charge+jnrD+0,
 584                                                                  charge+jnrE+0,charge+jnrF+0,
 585                                                                  charge+jnrG+0,charge+jnrH+0);
 586             vdwjidx0A        = 2*vdwtype[jnrA+0];
 587             vdwjidx0B        = 2*vdwtype[jnrB+0];
 588             vdwjidx0C        = 2*vdwtype[jnrC+0];
 589             vdwjidx0D        = 2*vdwtype[jnrD+0];
 590             vdwjidx0E        = 2*vdwtype[jnrE+0];
 591             vdwjidx0F        = 2*vdwtype[jnrF+0];
 592             vdwjidx0G        = 2*vdwtype[jnrG+0];
 593             vdwjidx0H        = 2*vdwtype[jnrH+0];
 594
 595             fjx0             = _mm256_setzero_ps();
 596             fjy0             = _mm256_setzero_ps();
 597             fjz0             = _mm256_setzero_ps();
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             r00              = _mm256_mul_ps(rsq00,rinv00);
 604             r00              = _mm256_andnot_ps(dummy_mask,r00);
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 608                                             vdwioffsetptr0+vdwjidx0B,
 609                                             vdwioffsetptr0+vdwjidx0C,
 610                                             vdwioffsetptr0+vdwjidx0D,
 611                                             vdwioffsetptr0+vdwjidx0E,
 612                                             vdwioffsetptr0+vdwjidx0F,
 613                                             vdwioffsetptr0+vdwjidx0G,
 614                                             vdwioffsetptr0+vdwjidx0H,
 615                                             &c6_00,&c12_00);
 616
 617             /* Calculate table index by multiplying r with table scale and truncate to integer */
 618             rt               = _mm256_mul_ps(r00,vftabscale);
 619             vfitab           = _mm256_cvttps_epi32(rt);
 620             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 621             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 622             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 623             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 624             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 625             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 626
 627             /* CUBIC SPLINE TABLE DISPERSION */
 628             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 629                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 630             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 631                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 632             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 633                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 634             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 635                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 636             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 637             Heps             = _mm256_mul_ps(vfeps,H);
 638             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 639             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 640             vvdw6            = _mm256_mul_ps(c6_00,VV);
 641             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 642             fvdw6            = _mm256_mul_ps(c6_00,FF);
 643
 644             /* CUBIC SPLINE TABLE REPULSION */
 645             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 646             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 647             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 648                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 649             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 650                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 651             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 652                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 653             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 654                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 655             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 656             Heps             = _mm256_mul_ps(vfeps,H);
 657             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 658             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 659             vvdw12           = _mm256_mul_ps(c12_00,VV);
 660             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 661             fvdw12           = _mm256_mul_ps(c12_00,FF);
 662             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 663             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 664
 665             /* Update potential sum for this i atom from the interaction with this j atom. */
 666             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 667             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 668
 669             fscal            = fvdw;
 670
 671             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 672
 673             /* Calculate temporary vectorial force */
 674             tx               = _mm256_mul_ps(fscal,dx00);
 675             ty               = _mm256_mul_ps(fscal,dy00);
 676             tz               = _mm256_mul_ps(fscal,dz00);
 677
 678             /* Update vectorial force */
 679             fix0             = _mm256_add_ps(fix0,tx);
 680             fiy0             = _mm256_add_ps(fiy0,ty);
 681             fiz0             = _mm256_add_ps(fiz0,tz);
 682
 683             fjx0             = _mm256_add_ps(fjx0,tx);
 684             fjy0             = _mm256_add_ps(fjy0,ty);
 685             fjz0             = _mm256_add_ps(fjz0,tz);
 686
 687             /**************************
 688              * CALCULATE INTERACTIONS *
 689              **************************/
 690
 691             r10              = _mm256_mul_ps(rsq10,rinv10);
 692             r10              = _mm256_andnot_ps(dummy_mask,r10);
 693
 694             /* Compute parameters for interactions between i and j atoms */
 695             qq10             = _mm256_mul_ps(iq1,jq0);
 696
 697             /* EWALD ELECTROSTATICS */
 698
 699             /* Analytical PME correction */
 700             zeta2            = _mm256_mul_ps(beta2,rsq10);
 701             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 702             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 703             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 704             felec            = _mm256_mul_ps(qq10,felec);
 705             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 706             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 707             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 708             velec            = _mm256_mul_ps(qq10,velec);
 709
 710             /* Update potential sum for this i atom from the interaction with this j atom. */
 711             velec            = _mm256_andnot_ps(dummy_mask,velec);
 712             velecsum         = _mm256_add_ps(velecsum,velec);
 713
 714             fscal            = felec;
 715
 716             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 717
 718             /* Calculate temporary vectorial force */
 719             tx               = _mm256_mul_ps(fscal,dx10);
 720             ty               = _mm256_mul_ps(fscal,dy10);
 721             tz               = _mm256_mul_ps(fscal,dz10);
 722
 723             /* Update vectorial force */
 724             fix1             = _mm256_add_ps(fix1,tx);
 725             fiy1             = _mm256_add_ps(fiy1,ty);
 726             fiz1             = _mm256_add_ps(fiz1,tz);
 727
 728             fjx0             = _mm256_add_ps(fjx0,tx);
 729             fjy0             = _mm256_add_ps(fjy0,ty);
 730             fjz0             = _mm256_add_ps(fjz0,tz);
 731
 732             /**************************
 733              * CALCULATE INTERACTIONS *
 734              **************************/
 735
 736             r20              = _mm256_mul_ps(rsq20,rinv20);
 737             r20              = _mm256_andnot_ps(dummy_mask,r20);
 738
 739             /* Compute parameters for interactions between i and j atoms */
 740             qq20             = _mm256_mul_ps(iq2,jq0);
 741
 742             /* EWALD ELECTROSTATICS */
 743
 744             /* Analytical PME correction */
 745             zeta2            = _mm256_mul_ps(beta2,rsq20);
 746             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 747             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 748             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 749             felec            = _mm256_mul_ps(qq20,felec);
 750             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 751             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 752             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 753             velec            = _mm256_mul_ps(qq20,velec);
 754
 755             /* Update potential sum for this i atom from the interaction with this j atom. */
 756             velec            = _mm256_andnot_ps(dummy_mask,velec);
 757             velecsum         = _mm256_add_ps(velecsum,velec);
 758
 759             fscal            = felec;
 760
 761             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 762
 763             /* Calculate temporary vectorial force */
 764             tx               = _mm256_mul_ps(fscal,dx20);
 765             ty               = _mm256_mul_ps(fscal,dy20);
 766             tz               = _mm256_mul_ps(fscal,dz20);
 767
 768             /* Update vectorial force */
 769             fix2             = _mm256_add_ps(fix2,tx);
 770             fiy2             = _mm256_add_ps(fiy2,ty);
 771             fiz2             = _mm256_add_ps(fiz2,tz);
 772
 773             fjx0             = _mm256_add_ps(fjx0,tx);
 774             fjy0             = _mm256_add_ps(fjy0,ty);
 775             fjz0             = _mm256_add_ps(fjz0,tz);
 776
 777             /**************************
 778              * CALCULATE INTERACTIONS *
 779              **************************/
 780
 781             r30              = _mm256_mul_ps(rsq30,rinv30);
 782             r30              = _mm256_andnot_ps(dummy_mask,r30);
 783
 784             /* Compute parameters for interactions between i and j atoms */
 785             qq30             = _mm256_mul_ps(iq3,jq0);
 786
 787             /* EWALD ELECTROSTATICS */
 788
 789             /* Analytical PME correction */
 790             zeta2            = _mm256_mul_ps(beta2,rsq30);
 791             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 792             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 793             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 794             felec            = _mm256_mul_ps(qq30,felec);
 795             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 796             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 797             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 798             velec            = _mm256_mul_ps(qq30,velec);
 799
 800             /* Update potential sum for this i atom from the interaction with this j atom. */
 801             velec            = _mm256_andnot_ps(dummy_mask,velec);
 802             velecsum         = _mm256_add_ps(velecsum,velec);
 803
 804             fscal            = felec;
 805
 806             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 807
 808             /* Calculate temporary vectorial force */
 809             tx               = _mm256_mul_ps(fscal,dx30);
 810             ty               = _mm256_mul_ps(fscal,dy30);
 811             tz               = _mm256_mul_ps(fscal,dz30);
 812
 813             /* Update vectorial force */
 814             fix3             = _mm256_add_ps(fix3,tx);
 815             fiy3             = _mm256_add_ps(fiy3,ty);
 816             fiz3             = _mm256_add_ps(fiz3,tz);
 817
 818             fjx0             = _mm256_add_ps(fjx0,tx);
 819             fjy0             = _mm256_add_ps(fjy0,ty);
 820             fjz0             = _mm256_add_ps(fjz0,tz);
 821
 822             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 823             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 824             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 825             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 826             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 827             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 828             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 829             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 830
 831             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 832
 833             /* Inner loop uses 315 flops */
 834         }
 835
 836         /* End of innermost loop */
 837
 838         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 839                                                  f+i_coord_offset,fshift+i_shift_offset);
 840
 841         ggid                        = gid[iidx];
 842         /* Update potential energies */
 843         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 844         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 845
 846         /* Increment number of inner iterations */
 847         inneriter                  += j_index_end - j_index_start;
 848
 849         /* Outer loop uses 26 flops */
 850     }
 851
 852     /* Increment number of outer iterations */
 853     outeriter        += nri;
 854
 855     /* Update outer/inner flops */
 856
 857     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*315);
 858 }
 859 /*
 860  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_single
 861  * Electrostatics interaction: Ewald
 862  * VdW interaction:            CubicSplineTable
 863  * Geometry:                   Water4-Particle
 864  * Calculate force/pot:        Force
 865  */
 866 void
 867 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_single
 868                     (t_nblist                    * gmx_restrict       nlist,
 869                      rvec                        * gmx_restrict          xx,
 870                      rvec                        * gmx_restrict          ff,
 871                      t_forcerec                  * gmx_restrict          fr,
 872                      t_mdatoms                   * gmx_restrict     mdatoms,
 873                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 874                      t_nrnb                      * gmx_restrict        nrnb)
 875 {
 876     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 877      * just 0 for non-waters.
 878      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 879      * jnr indices corresponding to data put in the four positions in the SIMD register.
 880      */
 881     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 882     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 883     int              jnrA,jnrB,jnrC,jnrD;
 884     int              jnrE,jnrF,jnrG,jnrH;
 885     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 886     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 887     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 888     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 889     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 890     real             rcutoff_scalar;
 891     real             *shiftvec,*fshift,*x,*f;
 892     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 893     real             scratch[4*DIM];
 894     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 895     real *           vdwioffsetptr0;
 896     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 897     real *           vdwioffsetptr1;
 898     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 899     real *           vdwioffsetptr2;
 900     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 901     real *           vdwioffsetptr3;
 902     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 903     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 904     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 905     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 906     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 907     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 908     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 909     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 910     real             *charge;
 911     int              nvdwtype;
 912     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 913     int              *vdwtype;
 914     real             *vdwparam;
 915     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 916     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 917     __m256i          vfitab;
 918     __m128i          vfitab_lo,vfitab_hi;
 919     __m128i          ifour       = _mm_set1_epi32(4);
 920     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 921     real             *vftab;
 922     __m256i          ewitab;
 923     __m128i          ewitab_lo,ewitab_hi;
 924     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 925     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 926     real             *ewtab;
 927     __m256           dummy_mask,cutoff_mask;
 928     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 929     __m256           one     = _mm256_set1_ps(1.0);
 930     __m256           two     = _mm256_set1_ps(2.0);
 931     x                = xx[0];
 932     f                = ff[0];
 933
 934     nri              = nlist->nri;
 935     iinr             = nlist->iinr;
 936     jindex           = nlist->jindex;
 937     jjnr             = nlist->jjnr;
 938     shiftidx         = nlist->shift;
 939     gid              = nlist->gid;
 940     shiftvec         = fr->shift_vec[0];
 941     fshift           = fr->fshift[0];
 942     facel            = _mm256_set1_ps(fr->epsfac);
 943     charge           = mdatoms->chargeA;
 944     nvdwtype         = fr->ntype;
 945     vdwparam         = fr->nbfp;
 946     vdwtype          = mdatoms->typeA;
 947
 948     vftab            = kernel_data->table_vdw->data;
 949     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 950
 951     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 952     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 953     beta2            = _mm256_mul_ps(beta,beta);
 954     beta3            = _mm256_mul_ps(beta,beta2);
 955
 956     ewtab            = fr->ic->tabq_coul_F;
 957     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 958     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 959
 960     /* Setup water-specific parameters */
 961     inr              = nlist->iinr[0];
 962     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 963     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 964     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 965     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 966
 967     /* Avoid stupid compiler warnings */
 968     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 969     j_coord_offsetA = 0;
 970     j_coord_offsetB = 0;
 971     j_coord_offsetC = 0;
 972     j_coord_offsetD = 0;
 973     j_coord_offsetE = 0;
 974     j_coord_offsetF = 0;
 975     j_coord_offsetG = 0;
 976     j_coord_offsetH = 0;
 977
 978     outeriter        = 0;
 979     inneriter        = 0;
 980
 981     for(iidx=0;iidx<4*DIM;iidx++)
 982     {
 983         scratch[iidx] = 0.0;
 984     }
 985
 986     /* Start outer loop over neighborlists */
 987     for(iidx=0; iidx<nri; iidx++)
 988     {
 989         /* Load shift vector for this list */
 990         i_shift_offset   = DIM*shiftidx[iidx];
 991
 992         /* Load limits for loop over neighbors */
 993         j_index_start    = jindex[iidx];
 994         j_index_end      = jindex[iidx+1];
 995
 996         /* Get outer coordinate index */
 997         inr              = iinr[iidx];
 998         i_coord_offset   = DIM*inr;
 999
1000         /* Load i particle coords and add shift vector */
1001         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1002                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1003
1004         fix0             = _mm256_setzero_ps();
1005         fiy0             = _mm256_setzero_ps();
1006         fiz0             = _mm256_setzero_ps();
1007         fix1             = _mm256_setzero_ps();
1008         fiy1             = _mm256_setzero_ps();
1009         fiz1             = _mm256_setzero_ps();
1010         fix2             = _mm256_setzero_ps();
1011         fiy2             = _mm256_setzero_ps();
1012         fiz2             = _mm256_setzero_ps();
1013         fix3             = _mm256_setzero_ps();
1014         fiy3             = _mm256_setzero_ps();
1015         fiz3             = _mm256_setzero_ps();
1016
1017         /* Start inner kernel loop */
1018         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1019         {
1020
1021             /* Get j neighbor index, and coordinate index */
1022             jnrA             = jjnr[jidx];
1023             jnrB             = jjnr[jidx+1];
1024             jnrC             = jjnr[jidx+2];
1025             jnrD             = jjnr[jidx+3];
1026             jnrE             = jjnr[jidx+4];
1027             jnrF             = jjnr[jidx+5];
1028             jnrG             = jjnr[jidx+6];
1029             jnrH             = jjnr[jidx+7];
1030             j_coord_offsetA  = DIM*jnrA;
1031             j_coord_offsetB  = DIM*jnrB;
1032             j_coord_offsetC  = DIM*jnrC;
1033             j_coord_offsetD  = DIM*jnrD;
1034             j_coord_offsetE  = DIM*jnrE;
1035             j_coord_offsetF  = DIM*jnrF;
1036             j_coord_offsetG  = DIM*jnrG;
1037             j_coord_offsetH  = DIM*jnrH;
1038
1039             /* load j atom coordinates */
1040             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1041                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1042                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1043                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1044                                                  &jx0,&jy0,&jz0);
1045
1046             /* Calculate displacement vector */
1047             dx00             = _mm256_sub_ps(ix0,jx0);
1048             dy00             = _mm256_sub_ps(iy0,jy0);
1049             dz00             = _mm256_sub_ps(iz0,jz0);
1050             dx10             = _mm256_sub_ps(ix1,jx0);
1051             dy10             = _mm256_sub_ps(iy1,jy0);
1052             dz10             = _mm256_sub_ps(iz1,jz0);
1053             dx20             = _mm256_sub_ps(ix2,jx0);
1054             dy20             = _mm256_sub_ps(iy2,jy0);
1055             dz20             = _mm256_sub_ps(iz2,jz0);
1056             dx30             = _mm256_sub_ps(ix3,jx0);
1057             dy30             = _mm256_sub_ps(iy3,jy0);
1058             dz30             = _mm256_sub_ps(iz3,jz0);
1059
1060             /* Calculate squared distance and things based on it */
1061             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1062             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1063             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1064             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1065
1066             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1067             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1068             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1069             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1070
1071             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1072             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1073             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1074
1075             /* Load parameters for j particles */
1076             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1077                                                                  charge+jnrC+0,charge+jnrD+0,
1078                                                                  charge+jnrE+0,charge+jnrF+0,
1079                                                                  charge+jnrG+0,charge+jnrH+0);
1080             vdwjidx0A        = 2*vdwtype[jnrA+0];
1081             vdwjidx0B        = 2*vdwtype[jnrB+0];
1082             vdwjidx0C        = 2*vdwtype[jnrC+0];
1083             vdwjidx0D        = 2*vdwtype[jnrD+0];
1084             vdwjidx0E        = 2*vdwtype[jnrE+0];
1085             vdwjidx0F        = 2*vdwtype[jnrF+0];
1086             vdwjidx0G        = 2*vdwtype[jnrG+0];
1087             vdwjidx0H        = 2*vdwtype[jnrH+0];
1088
1089             fjx0             = _mm256_setzero_ps();
1090             fjy0             = _mm256_setzero_ps();
1091             fjz0             = _mm256_setzero_ps();
1092
1093             /**************************
1094              * CALCULATE INTERACTIONS *
1095              **************************/
1096
1097             r00              = _mm256_mul_ps(rsq00,rinv00);
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1101                                             vdwioffsetptr0+vdwjidx0B,
1102                                             vdwioffsetptr0+vdwjidx0C,
1103                                             vdwioffsetptr0+vdwjidx0D,
1104                                             vdwioffsetptr0+vdwjidx0E,
1105                                             vdwioffsetptr0+vdwjidx0F,
1106                                             vdwioffsetptr0+vdwjidx0G,
1107                                             vdwioffsetptr0+vdwjidx0H,
1108                                             &c6_00,&c12_00);
1109
1110             /* Calculate table index by multiplying r with table scale and truncate to integer */
1111             rt               = _mm256_mul_ps(r00,vftabscale);
1112             vfitab           = _mm256_cvttps_epi32(rt);
1113             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1114             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1115             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1116             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1117             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1118             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1119
1120             /* CUBIC SPLINE TABLE DISPERSION */
1121             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1122                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1123             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1124                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1125             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1126                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1127             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1128                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1129             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1130             Heps             = _mm256_mul_ps(vfeps,H);
1131             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1132             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1133             fvdw6            = _mm256_mul_ps(c6_00,FF);
1134
1135             /* CUBIC SPLINE TABLE REPULSION */
1136             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1137             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1138             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1139                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1140             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1141                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1142             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1143                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1144             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1145                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1146             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1147             Heps             = _mm256_mul_ps(vfeps,H);
1148             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1149             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1150             fvdw12           = _mm256_mul_ps(c12_00,FF);
1151             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1152
1153             fscal            = fvdw;
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm256_mul_ps(fscal,dx00);
1157             ty               = _mm256_mul_ps(fscal,dy00);
1158             tz               = _mm256_mul_ps(fscal,dz00);
1159
1160             /* Update vectorial force */
1161             fix0             = _mm256_add_ps(fix0,tx);
1162             fiy0             = _mm256_add_ps(fiy0,ty);
1163             fiz0             = _mm256_add_ps(fiz0,tz);
1164
1165             fjx0             = _mm256_add_ps(fjx0,tx);
1166             fjy0             = _mm256_add_ps(fjy0,ty);
1167             fjz0             = _mm256_add_ps(fjz0,tz);
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             r10              = _mm256_mul_ps(rsq10,rinv10);
1174
1175             /* Compute parameters for interactions between i and j atoms */
1176             qq10             = _mm256_mul_ps(iq1,jq0);
1177
1178             /* EWALD ELECTROSTATICS */
1179
1180             /* Analytical PME correction */
1181             zeta2            = _mm256_mul_ps(beta2,rsq10);
1182             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1183             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1184             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1185             felec            = _mm256_mul_ps(qq10,felec);
1186
1187             fscal            = felec;
1188
1189             /* Calculate temporary vectorial force */
1190             tx               = _mm256_mul_ps(fscal,dx10);
1191             ty               = _mm256_mul_ps(fscal,dy10);
1192             tz               = _mm256_mul_ps(fscal,dz10);
1193
1194             /* Update vectorial force */
1195             fix1             = _mm256_add_ps(fix1,tx);
1196             fiy1             = _mm256_add_ps(fiy1,ty);
1197             fiz1             = _mm256_add_ps(fiz1,tz);
1198
1199             fjx0             = _mm256_add_ps(fjx0,tx);
1200             fjy0             = _mm256_add_ps(fjy0,ty);
1201             fjz0             = _mm256_add_ps(fjz0,tz);
1202
1203             /**************************
1204              * CALCULATE INTERACTIONS *
1205              **************************/
1206
1207             r20              = _mm256_mul_ps(rsq20,rinv20);
1208
1209             /* Compute parameters for interactions between i and j atoms */
1210             qq20             = _mm256_mul_ps(iq2,jq0);
1211
1212             /* EWALD ELECTROSTATICS */
1213
1214             /* Analytical PME correction */
1215             zeta2            = _mm256_mul_ps(beta2,rsq20);
1216             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1217             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1218             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1219             felec            = _mm256_mul_ps(qq20,felec);
1220
1221             fscal            = felec;
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm256_mul_ps(fscal,dx20);
1225             ty               = _mm256_mul_ps(fscal,dy20);
1226             tz               = _mm256_mul_ps(fscal,dz20);
1227
1228             /* Update vectorial force */
1229             fix2             = _mm256_add_ps(fix2,tx);
1230             fiy2             = _mm256_add_ps(fiy2,ty);
1231             fiz2             = _mm256_add_ps(fiz2,tz);
1232
1233             fjx0             = _mm256_add_ps(fjx0,tx);
1234             fjy0             = _mm256_add_ps(fjy0,ty);
1235             fjz0             = _mm256_add_ps(fjz0,tz);
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             r30              = _mm256_mul_ps(rsq30,rinv30);
1242
1243             /* Compute parameters for interactions between i and j atoms */
1244             qq30             = _mm256_mul_ps(iq3,jq0);
1245
1246             /* EWALD ELECTROSTATICS */
1247
1248             /* Analytical PME correction */
1249             zeta2            = _mm256_mul_ps(beta2,rsq30);
1250             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1251             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1252             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1253             felec            = _mm256_mul_ps(qq30,felec);
1254
1255             fscal            = felec;
1256
1257             /* Calculate temporary vectorial force */
1258             tx               = _mm256_mul_ps(fscal,dx30);
1259             ty               = _mm256_mul_ps(fscal,dy30);
1260             tz               = _mm256_mul_ps(fscal,dz30);
1261
1262             /* Update vectorial force */
1263             fix3             = _mm256_add_ps(fix3,tx);
1264             fiy3             = _mm256_add_ps(fiy3,ty);
1265             fiz3             = _mm256_add_ps(fiz3,tz);
1266
1267             fjx0             = _mm256_add_ps(fjx0,tx);
1268             fjy0             = _mm256_add_ps(fjy0,ty);
1269             fjz0             = _mm256_add_ps(fjz0,tz);
1270
1271             fjptrA             = f+j_coord_offsetA;
1272             fjptrB             = f+j_coord_offsetB;
1273             fjptrC             = f+j_coord_offsetC;
1274             fjptrD             = f+j_coord_offsetD;
1275             fjptrE             = f+j_coord_offsetE;
1276             fjptrF             = f+j_coord_offsetF;
1277             fjptrG             = f+j_coord_offsetG;
1278             fjptrH             = f+j_coord_offsetH;
1279
1280             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1281
1282             /* Inner loop uses 219 flops */
1283         }
1284
1285         if(jidx<j_index_end)
1286         {
1287
1288             /* Get j neighbor index, and coordinate index */
1289             jnrlistA         = jjnr[jidx];
1290             jnrlistB         = jjnr[jidx+1];
1291             jnrlistC         = jjnr[jidx+2];
1292             jnrlistD         = jjnr[jidx+3];
1293             jnrlistE         = jjnr[jidx+4];
1294             jnrlistF         = jjnr[jidx+5];
1295             jnrlistG         = jjnr[jidx+6];
1296             jnrlistH         = jjnr[jidx+7];
1297             /* Sign of each element will be negative for non-real atoms.
1298              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1299              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1300              */
1301             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1302                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1303
1304             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1305             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1306             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1307             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1308             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1309             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1310             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1311             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1312             j_coord_offsetA  = DIM*jnrA;
1313             j_coord_offsetB  = DIM*jnrB;
1314             j_coord_offsetC  = DIM*jnrC;
1315             j_coord_offsetD  = DIM*jnrD;
1316             j_coord_offsetE  = DIM*jnrE;
1317             j_coord_offsetF  = DIM*jnrF;
1318             j_coord_offsetG  = DIM*jnrG;
1319             j_coord_offsetH  = DIM*jnrH;
1320
1321             /* load j atom coordinates */
1322             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1323                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1324                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1325                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1326                                                  &jx0,&jy0,&jz0);
1327
1328             /* Calculate displacement vector */
1329             dx00             = _mm256_sub_ps(ix0,jx0);
1330             dy00             = _mm256_sub_ps(iy0,jy0);
1331             dz00             = _mm256_sub_ps(iz0,jz0);
1332             dx10             = _mm256_sub_ps(ix1,jx0);
1333             dy10             = _mm256_sub_ps(iy1,jy0);
1334             dz10             = _mm256_sub_ps(iz1,jz0);
1335             dx20             = _mm256_sub_ps(ix2,jx0);
1336             dy20             = _mm256_sub_ps(iy2,jy0);
1337             dz20             = _mm256_sub_ps(iz2,jz0);
1338             dx30             = _mm256_sub_ps(ix3,jx0);
1339             dy30             = _mm256_sub_ps(iy3,jy0);
1340             dz30             = _mm256_sub_ps(iz3,jz0);
1341
1342             /* Calculate squared distance and things based on it */
1343             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1344             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1345             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1346             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1347
1348             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1349             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1350             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1351             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1352
1353             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1354             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1355             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1356
1357             /* Load parameters for j particles */
1358             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1359                                                                  charge+jnrC+0,charge+jnrD+0,
1360                                                                  charge+jnrE+0,charge+jnrF+0,
1361                                                                  charge+jnrG+0,charge+jnrH+0);
1362             vdwjidx0A        = 2*vdwtype[jnrA+0];
1363             vdwjidx0B        = 2*vdwtype[jnrB+0];
1364             vdwjidx0C        = 2*vdwtype[jnrC+0];
1365             vdwjidx0D        = 2*vdwtype[jnrD+0];
1366             vdwjidx0E        = 2*vdwtype[jnrE+0];
1367             vdwjidx0F        = 2*vdwtype[jnrF+0];
1368             vdwjidx0G        = 2*vdwtype[jnrG+0];
1369             vdwjidx0H        = 2*vdwtype[jnrH+0];
1370
1371             fjx0             = _mm256_setzero_ps();
1372             fjy0             = _mm256_setzero_ps();
1373             fjz0             = _mm256_setzero_ps();
1374
1375             /**************************
1376              * CALCULATE INTERACTIONS *
1377              **************************/
1378
1379             r00              = _mm256_mul_ps(rsq00,rinv00);
1380             r00              = _mm256_andnot_ps(dummy_mask,r00);
1381
1382             /* Compute parameters for interactions between i and j atoms */
1383             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1384                                             vdwioffsetptr0+vdwjidx0B,
1385                                             vdwioffsetptr0+vdwjidx0C,
1386                                             vdwioffsetptr0+vdwjidx0D,
1387                                             vdwioffsetptr0+vdwjidx0E,
1388                                             vdwioffsetptr0+vdwjidx0F,
1389                                             vdwioffsetptr0+vdwjidx0G,
1390                                             vdwioffsetptr0+vdwjidx0H,
1391                                             &c6_00,&c12_00);
1392
1393             /* Calculate table index by multiplying r with table scale and truncate to integer */
1394             rt               = _mm256_mul_ps(r00,vftabscale);
1395             vfitab           = _mm256_cvttps_epi32(rt);
1396             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1397             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1398             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1399             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1400             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1401             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1402
1403             /* CUBIC SPLINE TABLE DISPERSION */
1404             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1405                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1406             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1407                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1408             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1409                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1410             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1411                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1412             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1413             Heps             = _mm256_mul_ps(vfeps,H);
1414             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1415             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1416             fvdw6            = _mm256_mul_ps(c6_00,FF);
1417
1418             /* CUBIC SPLINE TABLE REPULSION */
1419             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1420             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1421             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1422                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1423             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1424                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1425             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1426                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1427             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1428                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1429             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1430             Heps             = _mm256_mul_ps(vfeps,H);
1431             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1432             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1433             fvdw12           = _mm256_mul_ps(c12_00,FF);
1434             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1435
1436             fscal            = fvdw;
1437
1438             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1439
1440             /* Calculate temporary vectorial force */
1441             tx               = _mm256_mul_ps(fscal,dx00);
1442             ty               = _mm256_mul_ps(fscal,dy00);
1443             tz               = _mm256_mul_ps(fscal,dz00);
1444
1445             /* Update vectorial force */
1446             fix0             = _mm256_add_ps(fix0,tx);
1447             fiy0             = _mm256_add_ps(fiy0,ty);
1448             fiz0             = _mm256_add_ps(fiz0,tz);
1449
1450             fjx0             = _mm256_add_ps(fjx0,tx);
1451             fjy0             = _mm256_add_ps(fjy0,ty);
1452             fjz0             = _mm256_add_ps(fjz0,tz);
1453
1454             /**************************
1455              * CALCULATE INTERACTIONS *
1456              **************************/
1457
1458             r10              = _mm256_mul_ps(rsq10,rinv10);
1459             r10              = _mm256_andnot_ps(dummy_mask,r10);
1460
1461             /* Compute parameters for interactions between i and j atoms */
1462             qq10             = _mm256_mul_ps(iq1,jq0);
1463
1464             /* EWALD ELECTROSTATICS */
1465
1466             /* Analytical PME correction */
1467             zeta2            = _mm256_mul_ps(beta2,rsq10);
1468             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1469             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1470             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1471             felec            = _mm256_mul_ps(qq10,felec);
1472
1473             fscal            = felec;
1474
1475             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1476
1477             /* Calculate temporary vectorial force */
1478             tx               = _mm256_mul_ps(fscal,dx10);
1479             ty               = _mm256_mul_ps(fscal,dy10);
1480             tz               = _mm256_mul_ps(fscal,dz10);
1481
1482             /* Update vectorial force */
1483             fix1             = _mm256_add_ps(fix1,tx);
1484             fiy1             = _mm256_add_ps(fiy1,ty);
1485             fiz1             = _mm256_add_ps(fiz1,tz);
1486
1487             fjx0             = _mm256_add_ps(fjx0,tx);
1488             fjy0             = _mm256_add_ps(fjy0,ty);
1489             fjz0             = _mm256_add_ps(fjz0,tz);
1490
1491             /**************************
1492              * CALCULATE INTERACTIONS *
1493              **************************/
1494
1495             r20              = _mm256_mul_ps(rsq20,rinv20);
1496             r20              = _mm256_andnot_ps(dummy_mask,r20);
1497
1498             /* Compute parameters for interactions between i and j atoms */
1499             qq20             = _mm256_mul_ps(iq2,jq0);
1500
1501             /* EWALD ELECTROSTATICS */
1502
1503             /* Analytical PME correction */
1504             zeta2            = _mm256_mul_ps(beta2,rsq20);
1505             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1506             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1507             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1508             felec            = _mm256_mul_ps(qq20,felec);
1509
1510             fscal            = felec;
1511
1512             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1513
1514             /* Calculate temporary vectorial force */
1515             tx               = _mm256_mul_ps(fscal,dx20);
1516             ty               = _mm256_mul_ps(fscal,dy20);
1517             tz               = _mm256_mul_ps(fscal,dz20);
1518
1519             /* Update vectorial force */
1520             fix2             = _mm256_add_ps(fix2,tx);
1521             fiy2             = _mm256_add_ps(fiy2,ty);
1522             fiz2             = _mm256_add_ps(fiz2,tz);
1523
1524             fjx0             = _mm256_add_ps(fjx0,tx);
1525             fjy0             = _mm256_add_ps(fjy0,ty);
1526             fjz0             = _mm256_add_ps(fjz0,tz);
1527
1528             /**************************
1529              * CALCULATE INTERACTIONS *
1530              **************************/
1531
1532             r30              = _mm256_mul_ps(rsq30,rinv30);
1533             r30              = _mm256_andnot_ps(dummy_mask,r30);
1534
1535             /* Compute parameters for interactions between i and j atoms */
1536             qq30             = _mm256_mul_ps(iq3,jq0);
1537
1538             /* EWALD ELECTROSTATICS */
1539
1540             /* Analytical PME correction */
1541             zeta2            = _mm256_mul_ps(beta2,rsq30);
1542             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1543             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1544             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1545             felec            = _mm256_mul_ps(qq30,felec);
1546
1547             fscal            = felec;
1548
1549             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1550
1551             /* Calculate temporary vectorial force */
1552             tx               = _mm256_mul_ps(fscal,dx30);
1553             ty               = _mm256_mul_ps(fscal,dy30);
1554             tz               = _mm256_mul_ps(fscal,dz30);
1555
1556             /* Update vectorial force */
1557             fix3             = _mm256_add_ps(fix3,tx);
1558             fiy3             = _mm256_add_ps(fiy3,ty);
1559             fiz3             = _mm256_add_ps(fiz3,tz);
1560
1561             fjx0             = _mm256_add_ps(fjx0,tx);
1562             fjy0             = _mm256_add_ps(fjy0,ty);
1563             fjz0             = _mm256_add_ps(fjz0,tz);
1564
1565             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1566             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1567             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1568             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1569             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1570             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1571             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1572             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1573
1574             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1575
1576             /* Inner loop uses 223 flops */
1577         }
1578
1579         /* End of innermost loop */
1580
1581         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1582                                                  f+i_coord_offset,fshift+i_shift_offset);
1583
1584         /* Increment number of inner iterations */
1585         inneriter                  += j_index_end - j_index_start;
1586
1587         /* Outer loop uses 24 flops */
1588     }
1589
1590     /* Increment number of outer iterations */
1591     outeriter        += nri;
1592
1593     /* Update outer/inner flops */
1594
1595     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*223);
1596 }