src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_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     __m256           dummy_mask,cutoff_mask;
 114     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 115     __m256           one     = _mm256_set1_ps(1.0);
 116     __m256           two     = _mm256_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     krf              = _mm256_set1_ps(fr->ic->k_rf);
 131     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 132     crf              = _mm256_set1_ps(fr->ic->c_rf);
 133     nvdwtype         = fr->ntype;
 134     vdwparam         = fr->nbfp;
 135     vdwtype          = mdatoms->typeA;
 136
 137     vftab            = kernel_data->table_vdw->data;
 138     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 143     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 144     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 145     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 146
 147     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 148     rcutoff_scalar   = fr->rcoulomb;
 149     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 150     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158     j_coord_offsetE = 0;
 159     j_coord_offsetF = 0;
 160     j_coord_offsetG = 0;
 161     j_coord_offsetH = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     for(iidx=0;iidx<4*DIM;iidx++)
 167     {
 168         scratch[iidx] = 0.0;
 169     }
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 187                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 188
 189         fix0             = _mm256_setzero_ps();
 190         fiy0             = _mm256_setzero_ps();
 191         fiz0             = _mm256_setzero_ps();
 192         fix1             = _mm256_setzero_ps();
 193         fiy1             = _mm256_setzero_ps();
 194         fiz1             = _mm256_setzero_ps();
 195         fix2             = _mm256_setzero_ps();
 196         fiy2             = _mm256_setzero_ps();
 197         fiz2             = _mm256_setzero_ps();
 198         fix3             = _mm256_setzero_ps();
 199         fiy3             = _mm256_setzero_ps();
 200         fiz3             = _mm256_setzero_ps();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm256_setzero_ps();
 204         vvdwsum          = _mm256_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             jnrE             = jjnr[jidx+4];
 216             jnrF             = jjnr[jidx+5];
 217             jnrG             = jjnr[jidx+6];
 218             jnrH             = jjnr[jidx+7];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223             j_coord_offsetE  = DIM*jnrE;
 224             j_coord_offsetF  = DIM*jnrF;
 225             j_coord_offsetG  = DIM*jnrG;
 226             j_coord_offsetH  = DIM*jnrH;
 227
 228             /* load j atom coordinates */
 229             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 230                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 231                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 232                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 233                                                  &jx0,&jy0,&jz0);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm256_sub_ps(ix0,jx0);
 237             dy00             = _mm256_sub_ps(iy0,jy0);
 238             dz00             = _mm256_sub_ps(iz0,jz0);
 239             dx10             = _mm256_sub_ps(ix1,jx0);
 240             dy10             = _mm256_sub_ps(iy1,jy0);
 241             dz10             = _mm256_sub_ps(iz1,jz0);
 242             dx20             = _mm256_sub_ps(ix2,jx0);
 243             dy20             = _mm256_sub_ps(iy2,jy0);
 244             dz20             = _mm256_sub_ps(iz2,jz0);
 245             dx30             = _mm256_sub_ps(ix3,jx0);
 246             dy30             = _mm256_sub_ps(iy3,jy0);
 247             dz30             = _mm256_sub_ps(iz3,jz0);
 248
 249             /* Calculate squared distance and things based on it */
 250             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 251             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 252             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 253             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 254
 255             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 256             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 257             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 258             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 259
 260             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 261             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 262             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 263
 264             /* Load parameters for j particles */
 265             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 266                                                                  charge+jnrC+0,charge+jnrD+0,
 267                                                                  charge+jnrE+0,charge+jnrF+0,
 268                                                                  charge+jnrG+0,charge+jnrH+0);
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270             vdwjidx0B        = 2*vdwtype[jnrB+0];
 271             vdwjidx0C        = 2*vdwtype[jnrC+0];
 272             vdwjidx0D        = 2*vdwtype[jnrD+0];
 273             vdwjidx0E        = 2*vdwtype[jnrE+0];
 274             vdwjidx0F        = 2*vdwtype[jnrF+0];
 275             vdwjidx0G        = 2*vdwtype[jnrG+0];
 276             vdwjidx0H        = 2*vdwtype[jnrH+0];
 277
 278             fjx0             = _mm256_setzero_ps();
 279             fjy0             = _mm256_setzero_ps();
 280             fjz0             = _mm256_setzero_ps();
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             r00              = _mm256_mul_ps(rsq00,rinv00);
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 290                                             vdwioffsetptr0+vdwjidx0B,
 291                                             vdwioffsetptr0+vdwjidx0C,
 292                                             vdwioffsetptr0+vdwjidx0D,
 293                                             vdwioffsetptr0+vdwjidx0E,
 294                                             vdwioffsetptr0+vdwjidx0F,
 295                                             vdwioffsetptr0+vdwjidx0G,
 296                                             vdwioffsetptr0+vdwjidx0H,
 297                                             &c6_00,&c12_00);
 298
 299             /* Calculate table index by multiplying r with table scale and truncate to integer */
 300             rt               = _mm256_mul_ps(r00,vftabscale);
 301             vfitab           = _mm256_cvttps_epi32(rt);
 302             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 303             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 304             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 305             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 306             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 307             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 308
 309             /* CUBIC SPLINE TABLE DISPERSION */
 310             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 312             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 314             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 316             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 317                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 318             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 319             Heps             = _mm256_mul_ps(vfeps,H);
 320             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 321             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 322             vvdw6            = _mm256_mul_ps(c6_00,VV);
 323             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 324             fvdw6            = _mm256_mul_ps(c6_00,FF);
 325
 326             /* CUBIC SPLINE TABLE REPULSION */
 327             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 328             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 329             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 331             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 332                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 333             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 334                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 335             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 336                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 337             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 338             Heps             = _mm256_mul_ps(vfeps,H);
 339             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 340             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 341             vvdw12           = _mm256_mul_ps(c12_00,VV);
 342             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 343             fvdw12           = _mm256_mul_ps(c12_00,FF);
 344             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 345             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 349
 350             fscal            = fvdw;
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_ps(fscal,dx00);
 354             ty               = _mm256_mul_ps(fscal,dy00);
 355             tz               = _mm256_mul_ps(fscal,dz00);
 356
 357             /* Update vectorial force */
 358             fix0             = _mm256_add_ps(fix0,tx);
 359             fiy0             = _mm256_add_ps(fiy0,ty);
 360             fiz0             = _mm256_add_ps(fiz0,tz);
 361
 362             fjx0             = _mm256_add_ps(fjx0,tx);
 363             fjy0             = _mm256_add_ps(fjy0,ty);
 364             fjz0             = _mm256_add_ps(fjz0,tz);
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 371             {
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq10             = _mm256_mul_ps(iq1,jq0);
 375
 376             /* REACTION-FIELD ELECTROSTATICS */
 377             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 378             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 379
 380             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velec            = _mm256_and_ps(velec,cutoff_mask);
 384             velecsum         = _mm256_add_ps(velecsum,velec);
 385
 386             fscal            = felec;
 387
 388             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm256_mul_ps(fscal,dx10);
 392             ty               = _mm256_mul_ps(fscal,dy10);
 393             tz               = _mm256_mul_ps(fscal,dz10);
 394
 395             /* Update vectorial force */
 396             fix1             = _mm256_add_ps(fix1,tx);
 397             fiy1             = _mm256_add_ps(fiy1,ty);
 398             fiz1             = _mm256_add_ps(fiz1,tz);
 399
 400             fjx0             = _mm256_add_ps(fjx0,tx);
 401             fjy0             = _mm256_add_ps(fjy0,ty);
 402             fjz0             = _mm256_add_ps(fjz0,tz);
 403
 404             }
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 411             {
 412
 413             /* Compute parameters for interactions between i and j atoms */
 414             qq20             = _mm256_mul_ps(iq2,jq0);
 415
 416             /* REACTION-FIELD ELECTROSTATICS */
 417             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 418             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 419
 420             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velec            = _mm256_and_ps(velec,cutoff_mask);
 424             velecsum         = _mm256_add_ps(velecsum,velec);
 425
 426             fscal            = felec;
 427
 428             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 429
 430             /* Calculate temporary vectorial force */
 431             tx               = _mm256_mul_ps(fscal,dx20);
 432             ty               = _mm256_mul_ps(fscal,dy20);
 433             tz               = _mm256_mul_ps(fscal,dz20);
 434
 435             /* Update vectorial force */
 436             fix2             = _mm256_add_ps(fix2,tx);
 437             fiy2             = _mm256_add_ps(fiy2,ty);
 438             fiz2             = _mm256_add_ps(fiz2,tz);
 439
 440             fjx0             = _mm256_add_ps(fjx0,tx);
 441             fjy0             = _mm256_add_ps(fjy0,ty);
 442             fjz0             = _mm256_add_ps(fjz0,tz);
 443
 444             }
 445
 446             /**************************
 447              * CALCULATE INTERACTIONS *
 448              **************************/
 449
 450             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 451             {
 452
 453             /* Compute parameters for interactions between i and j atoms */
 454             qq30             = _mm256_mul_ps(iq3,jq0);
 455
 456             /* REACTION-FIELD ELECTROSTATICS */
 457             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 458             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 459
 460             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 461
 462             /* Update potential sum for this i atom from the interaction with this j atom. */
 463             velec            = _mm256_and_ps(velec,cutoff_mask);
 464             velecsum         = _mm256_add_ps(velecsum,velec);
 465
 466             fscal            = felec;
 467
 468             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 469
 470             /* Calculate temporary vectorial force */
 471             tx               = _mm256_mul_ps(fscal,dx30);
 472             ty               = _mm256_mul_ps(fscal,dy30);
 473             tz               = _mm256_mul_ps(fscal,dz30);
 474
 475             /* Update vectorial force */
 476             fix3             = _mm256_add_ps(fix3,tx);
 477             fiy3             = _mm256_add_ps(fiy3,ty);
 478             fiz3             = _mm256_add_ps(fiz3,tz);
 479
 480             fjx0             = _mm256_add_ps(fjx0,tx);
 481             fjy0             = _mm256_add_ps(fjy0,ty);
 482             fjz0             = _mm256_add_ps(fjz0,tz);
 483
 484             }
 485
 486             fjptrA             = f+j_coord_offsetA;
 487             fjptrB             = f+j_coord_offsetB;
 488             fjptrC             = f+j_coord_offsetC;
 489             fjptrD             = f+j_coord_offsetD;
 490             fjptrE             = f+j_coord_offsetE;
 491             fjptrF             = f+j_coord_offsetF;
 492             fjptrG             = f+j_coord_offsetG;
 493             fjptrH             = f+j_coord_offsetH;
 494
 495             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 496
 497             /* Inner loop uses 167 flops */
 498         }
 499
 500         if(jidx<j_index_end)
 501         {
 502
 503             /* Get j neighbor index, and coordinate index */
 504             jnrlistA         = jjnr[jidx];
 505             jnrlistB         = jjnr[jidx+1];
 506             jnrlistC         = jjnr[jidx+2];
 507             jnrlistD         = jjnr[jidx+3];
 508             jnrlistE         = jjnr[jidx+4];
 509             jnrlistF         = jjnr[jidx+5];
 510             jnrlistG         = jjnr[jidx+6];
 511             jnrlistH         = jjnr[jidx+7];
 512             /* Sign of each element will be negative for non-real atoms.
 513              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 514              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 515              */
 516             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 517                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 518
 519             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 520             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 521             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 522             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 523             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 524             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 525             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 526             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 527             j_coord_offsetA  = DIM*jnrA;
 528             j_coord_offsetB  = DIM*jnrB;
 529             j_coord_offsetC  = DIM*jnrC;
 530             j_coord_offsetD  = DIM*jnrD;
 531             j_coord_offsetE  = DIM*jnrE;
 532             j_coord_offsetF  = DIM*jnrF;
 533             j_coord_offsetG  = DIM*jnrG;
 534             j_coord_offsetH  = DIM*jnrH;
 535
 536             /* load j atom coordinates */
 537             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 538                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 539                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 540                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 541                                                  &jx0,&jy0,&jz0);
 542
 543             /* Calculate displacement vector */
 544             dx00             = _mm256_sub_ps(ix0,jx0);
 545             dy00             = _mm256_sub_ps(iy0,jy0);
 546             dz00             = _mm256_sub_ps(iz0,jz0);
 547             dx10             = _mm256_sub_ps(ix1,jx0);
 548             dy10             = _mm256_sub_ps(iy1,jy0);
 549             dz10             = _mm256_sub_ps(iz1,jz0);
 550             dx20             = _mm256_sub_ps(ix2,jx0);
 551             dy20             = _mm256_sub_ps(iy2,jy0);
 552             dz20             = _mm256_sub_ps(iz2,jz0);
 553             dx30             = _mm256_sub_ps(ix3,jx0);
 554             dy30             = _mm256_sub_ps(iy3,jy0);
 555             dz30             = _mm256_sub_ps(iz3,jz0);
 556
 557             /* Calculate squared distance and things based on it */
 558             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 559             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 560             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 561             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 562
 563             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 564             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 565             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 566             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 567
 568             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 569             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 570             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 571
 572             /* Load parameters for j particles */
 573             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 574                                                                  charge+jnrC+0,charge+jnrD+0,
 575                                                                  charge+jnrE+0,charge+jnrF+0,
 576                                                                  charge+jnrG+0,charge+jnrH+0);
 577             vdwjidx0A        = 2*vdwtype[jnrA+0];
 578             vdwjidx0B        = 2*vdwtype[jnrB+0];
 579             vdwjidx0C        = 2*vdwtype[jnrC+0];
 580             vdwjidx0D        = 2*vdwtype[jnrD+0];
 581             vdwjidx0E        = 2*vdwtype[jnrE+0];
 582             vdwjidx0F        = 2*vdwtype[jnrF+0];
 583             vdwjidx0G        = 2*vdwtype[jnrG+0];
 584             vdwjidx0H        = 2*vdwtype[jnrH+0];
 585
 586             fjx0             = _mm256_setzero_ps();
 587             fjy0             = _mm256_setzero_ps();
 588             fjz0             = _mm256_setzero_ps();
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             r00              = _mm256_mul_ps(rsq00,rinv00);
 595             r00              = _mm256_andnot_ps(dummy_mask,r00);
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 599                                             vdwioffsetptr0+vdwjidx0B,
 600                                             vdwioffsetptr0+vdwjidx0C,
 601                                             vdwioffsetptr0+vdwjidx0D,
 602                                             vdwioffsetptr0+vdwjidx0E,
 603                                             vdwioffsetptr0+vdwjidx0F,
 604                                             vdwioffsetptr0+vdwjidx0G,
 605                                             vdwioffsetptr0+vdwjidx0H,
 606                                             &c6_00,&c12_00);
 607
 608             /* Calculate table index by multiplying r with table scale and truncate to integer */
 609             rt               = _mm256_mul_ps(r00,vftabscale);
 610             vfitab           = _mm256_cvttps_epi32(rt);
 611             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 612             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 613             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 614             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 615             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 616             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 617
 618             /* CUBIC SPLINE TABLE DISPERSION */
 619             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 620                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 621             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 622                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 623             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 625             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 627             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 628             Heps             = _mm256_mul_ps(vfeps,H);
 629             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 630             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 631             vvdw6            = _mm256_mul_ps(c6_00,VV);
 632             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 633             fvdw6            = _mm256_mul_ps(c6_00,FF);
 634
 635             /* CUBIC SPLINE TABLE REPULSION */
 636             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 637             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 638             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 639                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 640             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 641                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 642             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 643                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 644             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 645                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 646             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 647             Heps             = _mm256_mul_ps(vfeps,H);
 648             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 649             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 650             vvdw12           = _mm256_mul_ps(c12_00,VV);
 651             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 652             fvdw12           = _mm256_mul_ps(c12_00,FF);
 653             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 654             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 658             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 659
 660             fscal            = fvdw;
 661
 662             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm256_mul_ps(fscal,dx00);
 666             ty               = _mm256_mul_ps(fscal,dy00);
 667             tz               = _mm256_mul_ps(fscal,dz00);
 668
 669             /* Update vectorial force */
 670             fix0             = _mm256_add_ps(fix0,tx);
 671             fiy0             = _mm256_add_ps(fiy0,ty);
 672             fiz0             = _mm256_add_ps(fiz0,tz);
 673
 674             fjx0             = _mm256_add_ps(fjx0,tx);
 675             fjy0             = _mm256_add_ps(fjy0,ty);
 676             fjz0             = _mm256_add_ps(fjz0,tz);
 677
 678             /**************************
 679              * CALCULATE INTERACTIONS *
 680              **************************/
 681
 682             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 683             {
 684
 685             /* Compute parameters for interactions between i and j atoms */
 686             qq10             = _mm256_mul_ps(iq1,jq0);
 687
 688             /* REACTION-FIELD ELECTROSTATICS */
 689             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 690             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 691
 692             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 693
 694             /* Update potential sum for this i atom from the interaction with this j atom. */
 695             velec            = _mm256_and_ps(velec,cutoff_mask);
 696             velec            = _mm256_andnot_ps(dummy_mask,velec);
 697             velecsum         = _mm256_add_ps(velecsum,velec);
 698
 699             fscal            = felec;
 700
 701             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 702
 703             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm256_mul_ps(fscal,dx10);
 707             ty               = _mm256_mul_ps(fscal,dy10);
 708             tz               = _mm256_mul_ps(fscal,dz10);
 709
 710             /* Update vectorial force */
 711             fix1             = _mm256_add_ps(fix1,tx);
 712             fiy1             = _mm256_add_ps(fiy1,ty);
 713             fiz1             = _mm256_add_ps(fiz1,tz);
 714
 715             fjx0             = _mm256_add_ps(fjx0,tx);
 716             fjy0             = _mm256_add_ps(fjy0,ty);
 717             fjz0             = _mm256_add_ps(fjz0,tz);
 718
 719             }
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 726             {
 727
 728             /* Compute parameters for interactions between i and j atoms */
 729             qq20             = _mm256_mul_ps(iq2,jq0);
 730
 731             /* REACTION-FIELD ELECTROSTATICS */
 732             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 733             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 734
 735             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 736
 737             /* Update potential sum for this i atom from the interaction with this j atom. */
 738             velec            = _mm256_and_ps(velec,cutoff_mask);
 739             velec            = _mm256_andnot_ps(dummy_mask,velec);
 740             velecsum         = _mm256_add_ps(velecsum,velec);
 741
 742             fscal            = felec;
 743
 744             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 745
 746             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 747
 748             /* Calculate temporary vectorial force */
 749             tx               = _mm256_mul_ps(fscal,dx20);
 750             ty               = _mm256_mul_ps(fscal,dy20);
 751             tz               = _mm256_mul_ps(fscal,dz20);
 752
 753             /* Update vectorial force */
 754             fix2             = _mm256_add_ps(fix2,tx);
 755             fiy2             = _mm256_add_ps(fiy2,ty);
 756             fiz2             = _mm256_add_ps(fiz2,tz);
 757
 758             fjx0             = _mm256_add_ps(fjx0,tx);
 759             fjy0             = _mm256_add_ps(fjy0,ty);
 760             fjz0             = _mm256_add_ps(fjz0,tz);
 761
 762             }
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 769             {
 770
 771             /* Compute parameters for interactions between i and j atoms */
 772             qq30             = _mm256_mul_ps(iq3,jq0);
 773
 774             /* REACTION-FIELD ELECTROSTATICS */
 775             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 776             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 777
 778             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 779
 780             /* Update potential sum for this i atom from the interaction with this j atom. */
 781             velec            = _mm256_and_ps(velec,cutoff_mask);
 782             velec            = _mm256_andnot_ps(dummy_mask,velec);
 783             velecsum         = _mm256_add_ps(velecsum,velec);
 784
 785             fscal            = felec;
 786
 787             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 788
 789             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 790
 791             /* Calculate temporary vectorial force */
 792             tx               = _mm256_mul_ps(fscal,dx30);
 793             ty               = _mm256_mul_ps(fscal,dy30);
 794             tz               = _mm256_mul_ps(fscal,dz30);
 795
 796             /* Update vectorial force */
 797             fix3             = _mm256_add_ps(fix3,tx);
 798             fiy3             = _mm256_add_ps(fiy3,ty);
 799             fiz3             = _mm256_add_ps(fiz3,tz);
 800
 801             fjx0             = _mm256_add_ps(fjx0,tx);
 802             fjy0             = _mm256_add_ps(fjy0,ty);
 803             fjz0             = _mm256_add_ps(fjz0,tz);
 804
 805             }
 806
 807             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 808             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 809             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 810             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 811             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 812             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 813             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 814             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 815
 816             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 817
 818             /* Inner loop uses 168 flops */
 819         }
 820
 821         /* End of innermost loop */
 822
 823         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 824                                                  f+i_coord_offset,fshift+i_shift_offset);
 825
 826         ggid                        = gid[iidx];
 827         /* Update potential energies */
 828         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 829         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 830
 831         /* Increment number of inner iterations */
 832         inneriter                  += j_index_end - j_index_start;
 833
 834         /* Outer loop uses 26 flops */
 835     }
 836
 837     /* Increment number of outer iterations */
 838     outeriter        += nri;
 839
 840     /* Update outer/inner flops */
 841
 842     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
 843 }
 844 /*
 845  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_single
 846  * Electrostatics interaction: ReactionField
 847  * VdW interaction:            CubicSplineTable
 848  * Geometry:                   Water4-Particle
 849  * Calculate force/pot:        Force
 850  */
 851 void
 852 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_single
 853                     (t_nblist                    * gmx_restrict       nlist,
 854                      rvec                        * gmx_restrict          xx,
 855                      rvec                        * gmx_restrict          ff,
 856                      t_forcerec                  * gmx_restrict          fr,
 857                      t_mdatoms                   * gmx_restrict     mdatoms,
 858                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 859                      t_nrnb                      * gmx_restrict        nrnb)
 860 {
 861     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 862      * just 0 for non-waters.
 863      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 864      * jnr indices corresponding to data put in the four positions in the SIMD register.
 865      */
 866     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 867     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 868     int              jnrA,jnrB,jnrC,jnrD;
 869     int              jnrE,jnrF,jnrG,jnrH;
 870     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 871     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 872     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 873     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 874     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 875     real             rcutoff_scalar;
 876     real             *shiftvec,*fshift,*x,*f;
 877     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 878     real             scratch[4*DIM];
 879     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 880     real *           vdwioffsetptr0;
 881     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 882     real *           vdwioffsetptr1;
 883     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 884     real *           vdwioffsetptr2;
 885     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 886     real *           vdwioffsetptr3;
 887     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 888     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 889     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 890     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 891     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 892     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 893     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 894     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 895     real             *charge;
 896     int              nvdwtype;
 897     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 898     int              *vdwtype;
 899     real             *vdwparam;
 900     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 901     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 902     __m256i          vfitab;
 903     __m128i          vfitab_lo,vfitab_hi;
 904     __m128i          ifour       = _mm_set1_epi32(4);
 905     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 906     real             *vftab;
 907     __m256           dummy_mask,cutoff_mask;
 908     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 909     __m256           one     = _mm256_set1_ps(1.0);
 910     __m256           two     = _mm256_set1_ps(2.0);
 911     x                = xx[0];
 912     f                = ff[0];
 913
 914     nri              = nlist->nri;
 915     iinr             = nlist->iinr;
 916     jindex           = nlist->jindex;
 917     jjnr             = nlist->jjnr;
 918     shiftidx         = nlist->shift;
 919     gid              = nlist->gid;
 920     shiftvec         = fr->shift_vec[0];
 921     fshift           = fr->fshift[0];
 922     facel            = _mm256_set1_ps(fr->epsfac);
 923     charge           = mdatoms->chargeA;
 924     krf              = _mm256_set1_ps(fr->ic->k_rf);
 925     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 926     crf              = _mm256_set1_ps(fr->ic->c_rf);
 927     nvdwtype         = fr->ntype;
 928     vdwparam         = fr->nbfp;
 929     vdwtype          = mdatoms->typeA;
 930
 931     vftab            = kernel_data->table_vdw->data;
 932     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 933
 934     /* Setup water-specific parameters */
 935     inr              = nlist->iinr[0];
 936     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 937     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 938     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 939     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 940
 941     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 942     rcutoff_scalar   = fr->rcoulomb;
 943     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 944     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 945
 946     /* Avoid stupid compiler warnings */
 947     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 948     j_coord_offsetA = 0;
 949     j_coord_offsetB = 0;
 950     j_coord_offsetC = 0;
 951     j_coord_offsetD = 0;
 952     j_coord_offsetE = 0;
 953     j_coord_offsetF = 0;
 954     j_coord_offsetG = 0;
 955     j_coord_offsetH = 0;
 956
 957     outeriter        = 0;
 958     inneriter        = 0;
 959
 960     for(iidx=0;iidx<4*DIM;iidx++)
 961     {
 962         scratch[iidx] = 0.0;
 963     }
 964
 965     /* Start outer loop over neighborlists */
 966     for(iidx=0; iidx<nri; iidx++)
 967     {
 968         /* Load shift vector for this list */
 969         i_shift_offset   = DIM*shiftidx[iidx];
 970
 971         /* Load limits for loop over neighbors */
 972         j_index_start    = jindex[iidx];
 973         j_index_end      = jindex[iidx+1];
 974
 975         /* Get outer coordinate index */
 976         inr              = iinr[iidx];
 977         i_coord_offset   = DIM*inr;
 978
 979         /* Load i particle coords and add shift vector */
 980         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 981                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 982
 983         fix0             = _mm256_setzero_ps();
 984         fiy0             = _mm256_setzero_ps();
 985         fiz0             = _mm256_setzero_ps();
 986         fix1             = _mm256_setzero_ps();
 987         fiy1             = _mm256_setzero_ps();
 988         fiz1             = _mm256_setzero_ps();
 989         fix2             = _mm256_setzero_ps();
 990         fiy2             = _mm256_setzero_ps();
 991         fiz2             = _mm256_setzero_ps();
 992         fix3             = _mm256_setzero_ps();
 993         fiy3             = _mm256_setzero_ps();
 994         fiz3             = _mm256_setzero_ps();
 995
 996         /* Start inner kernel loop */
 997         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 998         {
 999
1000             /* Get j neighbor index, and coordinate index */
1001             jnrA             = jjnr[jidx];
1002             jnrB             = jjnr[jidx+1];
1003             jnrC             = jjnr[jidx+2];
1004             jnrD             = jjnr[jidx+3];
1005             jnrE             = jjnr[jidx+4];
1006             jnrF             = jjnr[jidx+5];
1007             jnrG             = jjnr[jidx+6];
1008             jnrH             = jjnr[jidx+7];
1009             j_coord_offsetA  = DIM*jnrA;
1010             j_coord_offsetB  = DIM*jnrB;
1011             j_coord_offsetC  = DIM*jnrC;
1012             j_coord_offsetD  = DIM*jnrD;
1013             j_coord_offsetE  = DIM*jnrE;
1014             j_coord_offsetF  = DIM*jnrF;
1015             j_coord_offsetG  = DIM*jnrG;
1016             j_coord_offsetH  = DIM*jnrH;
1017
1018             /* load j atom coordinates */
1019             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1020                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1021                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1022                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1023                                                  &jx0,&jy0,&jz0);
1024
1025             /* Calculate displacement vector */
1026             dx00             = _mm256_sub_ps(ix0,jx0);
1027             dy00             = _mm256_sub_ps(iy0,jy0);
1028             dz00             = _mm256_sub_ps(iz0,jz0);
1029             dx10             = _mm256_sub_ps(ix1,jx0);
1030             dy10             = _mm256_sub_ps(iy1,jy0);
1031             dz10             = _mm256_sub_ps(iz1,jz0);
1032             dx20             = _mm256_sub_ps(ix2,jx0);
1033             dy20             = _mm256_sub_ps(iy2,jy0);
1034             dz20             = _mm256_sub_ps(iz2,jz0);
1035             dx30             = _mm256_sub_ps(ix3,jx0);
1036             dy30             = _mm256_sub_ps(iy3,jy0);
1037             dz30             = _mm256_sub_ps(iz3,jz0);
1038
1039             /* Calculate squared distance and things based on it */
1040             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1041             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1042             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1043             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1044
1045             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1046             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1047             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1048             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1049
1050             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1051             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1052             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1053
1054             /* Load parameters for j particles */
1055             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1056                                                                  charge+jnrC+0,charge+jnrD+0,
1057                                                                  charge+jnrE+0,charge+jnrF+0,
1058                                                                  charge+jnrG+0,charge+jnrH+0);
1059             vdwjidx0A        = 2*vdwtype[jnrA+0];
1060             vdwjidx0B        = 2*vdwtype[jnrB+0];
1061             vdwjidx0C        = 2*vdwtype[jnrC+0];
1062             vdwjidx0D        = 2*vdwtype[jnrD+0];
1063             vdwjidx0E        = 2*vdwtype[jnrE+0];
1064             vdwjidx0F        = 2*vdwtype[jnrF+0];
1065             vdwjidx0G        = 2*vdwtype[jnrG+0];
1066             vdwjidx0H        = 2*vdwtype[jnrH+0];
1067
1068             fjx0             = _mm256_setzero_ps();
1069             fjy0             = _mm256_setzero_ps();
1070             fjz0             = _mm256_setzero_ps();
1071
1072             /**************************
1073              * CALCULATE INTERACTIONS *
1074              **************************/
1075
1076             r00              = _mm256_mul_ps(rsq00,rinv00);
1077
1078             /* Compute parameters for interactions between i and j atoms */
1079             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1080                                             vdwioffsetptr0+vdwjidx0B,
1081                                             vdwioffsetptr0+vdwjidx0C,
1082                                             vdwioffsetptr0+vdwjidx0D,
1083                                             vdwioffsetptr0+vdwjidx0E,
1084                                             vdwioffsetptr0+vdwjidx0F,
1085                                             vdwioffsetptr0+vdwjidx0G,
1086                                             vdwioffsetptr0+vdwjidx0H,
1087                                             &c6_00,&c12_00);
1088
1089             /* Calculate table index by multiplying r with table scale and truncate to integer */
1090             rt               = _mm256_mul_ps(r00,vftabscale);
1091             vfitab           = _mm256_cvttps_epi32(rt);
1092             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1093             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1094             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1095             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1096             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1097             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1098
1099             /* CUBIC SPLINE TABLE DISPERSION */
1100             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1101                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1102             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1103                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1104             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1105                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1106             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1107                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1108             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1109             Heps             = _mm256_mul_ps(vfeps,H);
1110             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1111             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1112             fvdw6            = _mm256_mul_ps(c6_00,FF);
1113
1114             /* CUBIC SPLINE TABLE REPULSION */
1115             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1116             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1117             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1118                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1119             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1120                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1121             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1122                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1123             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1124                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1125             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1126             Heps             = _mm256_mul_ps(vfeps,H);
1127             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1128             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1129             fvdw12           = _mm256_mul_ps(c12_00,FF);
1130             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1131
1132             fscal            = fvdw;
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm256_mul_ps(fscal,dx00);
1136             ty               = _mm256_mul_ps(fscal,dy00);
1137             tz               = _mm256_mul_ps(fscal,dz00);
1138
1139             /* Update vectorial force */
1140             fix0             = _mm256_add_ps(fix0,tx);
1141             fiy0             = _mm256_add_ps(fiy0,ty);
1142             fiz0             = _mm256_add_ps(fiz0,tz);
1143
1144             fjx0             = _mm256_add_ps(fjx0,tx);
1145             fjy0             = _mm256_add_ps(fjy0,ty);
1146             fjz0             = _mm256_add_ps(fjz0,tz);
1147
1148             /**************************
1149              * CALCULATE INTERACTIONS *
1150              **************************/
1151
1152             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1153             {
1154
1155             /* Compute parameters for interactions between i and j atoms */
1156             qq10             = _mm256_mul_ps(iq1,jq0);
1157
1158             /* REACTION-FIELD ELECTROSTATICS */
1159             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1160
1161             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm256_mul_ps(fscal,dx10);
1169             ty               = _mm256_mul_ps(fscal,dy10);
1170             tz               = _mm256_mul_ps(fscal,dz10);
1171
1172             /* Update vectorial force */
1173             fix1             = _mm256_add_ps(fix1,tx);
1174             fiy1             = _mm256_add_ps(fiy1,ty);
1175             fiz1             = _mm256_add_ps(fiz1,tz);
1176
1177             fjx0             = _mm256_add_ps(fjx0,tx);
1178             fjy0             = _mm256_add_ps(fjy0,ty);
1179             fjz0             = _mm256_add_ps(fjz0,tz);
1180
1181             }
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1188             {
1189
1190             /* Compute parameters for interactions between i and j atoms */
1191             qq20             = _mm256_mul_ps(iq2,jq0);
1192
1193             /* REACTION-FIELD ELECTROSTATICS */
1194             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1195
1196             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm256_mul_ps(fscal,dx20);
1204             ty               = _mm256_mul_ps(fscal,dy20);
1205             tz               = _mm256_mul_ps(fscal,dz20);
1206
1207             /* Update vectorial force */
1208             fix2             = _mm256_add_ps(fix2,tx);
1209             fiy2             = _mm256_add_ps(fiy2,ty);
1210             fiz2             = _mm256_add_ps(fiz2,tz);
1211
1212             fjx0             = _mm256_add_ps(fjx0,tx);
1213             fjy0             = _mm256_add_ps(fjy0,ty);
1214             fjz0             = _mm256_add_ps(fjz0,tz);
1215
1216             }
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1223             {
1224
1225             /* Compute parameters for interactions between i and j atoms */
1226             qq30             = _mm256_mul_ps(iq3,jq0);
1227
1228             /* REACTION-FIELD ELECTROSTATICS */
1229             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1230
1231             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1232
1233             fscal            = felec;
1234
1235             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1236
1237             /* Calculate temporary vectorial force */
1238             tx               = _mm256_mul_ps(fscal,dx30);
1239             ty               = _mm256_mul_ps(fscal,dy30);
1240             tz               = _mm256_mul_ps(fscal,dz30);
1241
1242             /* Update vectorial force */
1243             fix3             = _mm256_add_ps(fix3,tx);
1244             fiy3             = _mm256_add_ps(fiy3,ty);
1245             fiz3             = _mm256_add_ps(fiz3,tz);
1246
1247             fjx0             = _mm256_add_ps(fjx0,tx);
1248             fjy0             = _mm256_add_ps(fjy0,ty);
1249             fjz0             = _mm256_add_ps(fjz0,tz);
1250
1251             }
1252
1253             fjptrA             = f+j_coord_offsetA;
1254             fjptrB             = f+j_coord_offsetB;
1255             fjptrC             = f+j_coord_offsetC;
1256             fjptrD             = f+j_coord_offsetD;
1257             fjptrE             = f+j_coord_offsetE;
1258             fjptrF             = f+j_coord_offsetF;
1259             fjptrG             = f+j_coord_offsetG;
1260             fjptrH             = f+j_coord_offsetH;
1261
1262             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1263
1264             /* Inner loop uses 141 flops */
1265         }
1266
1267         if(jidx<j_index_end)
1268         {
1269
1270             /* Get j neighbor index, and coordinate index */
1271             jnrlistA         = jjnr[jidx];
1272             jnrlistB         = jjnr[jidx+1];
1273             jnrlistC         = jjnr[jidx+2];
1274             jnrlistD         = jjnr[jidx+3];
1275             jnrlistE         = jjnr[jidx+4];
1276             jnrlistF         = jjnr[jidx+5];
1277             jnrlistG         = jjnr[jidx+6];
1278             jnrlistH         = jjnr[jidx+7];
1279             /* Sign of each element will be negative for non-real atoms.
1280              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1281              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1282              */
1283             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1284                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1285
1286             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1287             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1288             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1289             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1290             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1291             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1292             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1293             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1294             j_coord_offsetA  = DIM*jnrA;
1295             j_coord_offsetB  = DIM*jnrB;
1296             j_coord_offsetC  = DIM*jnrC;
1297             j_coord_offsetD  = DIM*jnrD;
1298             j_coord_offsetE  = DIM*jnrE;
1299             j_coord_offsetF  = DIM*jnrF;
1300             j_coord_offsetG  = DIM*jnrG;
1301             j_coord_offsetH  = DIM*jnrH;
1302
1303             /* load j atom coordinates */
1304             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1305                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1306                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1307                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1308                                                  &jx0,&jy0,&jz0);
1309
1310             /* Calculate displacement vector */
1311             dx00             = _mm256_sub_ps(ix0,jx0);
1312             dy00             = _mm256_sub_ps(iy0,jy0);
1313             dz00             = _mm256_sub_ps(iz0,jz0);
1314             dx10             = _mm256_sub_ps(ix1,jx0);
1315             dy10             = _mm256_sub_ps(iy1,jy0);
1316             dz10             = _mm256_sub_ps(iz1,jz0);
1317             dx20             = _mm256_sub_ps(ix2,jx0);
1318             dy20             = _mm256_sub_ps(iy2,jy0);
1319             dz20             = _mm256_sub_ps(iz2,jz0);
1320             dx30             = _mm256_sub_ps(ix3,jx0);
1321             dy30             = _mm256_sub_ps(iy3,jy0);
1322             dz30             = _mm256_sub_ps(iz3,jz0);
1323
1324             /* Calculate squared distance and things based on it */
1325             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1326             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1327             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1328             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1329
1330             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1331             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1332             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1333             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1334
1335             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1336             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1337             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1338
1339             /* Load parameters for j particles */
1340             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1341                                                                  charge+jnrC+0,charge+jnrD+0,
1342                                                                  charge+jnrE+0,charge+jnrF+0,
1343                                                                  charge+jnrG+0,charge+jnrH+0);
1344             vdwjidx0A        = 2*vdwtype[jnrA+0];
1345             vdwjidx0B        = 2*vdwtype[jnrB+0];
1346             vdwjidx0C        = 2*vdwtype[jnrC+0];
1347             vdwjidx0D        = 2*vdwtype[jnrD+0];
1348             vdwjidx0E        = 2*vdwtype[jnrE+0];
1349             vdwjidx0F        = 2*vdwtype[jnrF+0];
1350             vdwjidx0G        = 2*vdwtype[jnrG+0];
1351             vdwjidx0H        = 2*vdwtype[jnrH+0];
1352
1353             fjx0             = _mm256_setzero_ps();
1354             fjy0             = _mm256_setzero_ps();
1355             fjz0             = _mm256_setzero_ps();
1356
1357             /**************************
1358              * CALCULATE INTERACTIONS *
1359              **************************/
1360
1361             r00              = _mm256_mul_ps(rsq00,rinv00);
1362             r00              = _mm256_andnot_ps(dummy_mask,r00);
1363
1364             /* Compute parameters for interactions between i and j atoms */
1365             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1366                                             vdwioffsetptr0+vdwjidx0B,
1367                                             vdwioffsetptr0+vdwjidx0C,
1368                                             vdwioffsetptr0+vdwjidx0D,
1369                                             vdwioffsetptr0+vdwjidx0E,
1370                                             vdwioffsetptr0+vdwjidx0F,
1371                                             vdwioffsetptr0+vdwjidx0G,
1372                                             vdwioffsetptr0+vdwjidx0H,
1373                                             &c6_00,&c12_00);
1374
1375             /* Calculate table index by multiplying r with table scale and truncate to integer */
1376             rt               = _mm256_mul_ps(r00,vftabscale);
1377             vfitab           = _mm256_cvttps_epi32(rt);
1378             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1379             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1380             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1381             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1382             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1383             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1384
1385             /* CUBIC SPLINE TABLE DISPERSION */
1386             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1388             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1390             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1392             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1394             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1395             Heps             = _mm256_mul_ps(vfeps,H);
1396             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1397             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1398             fvdw6            = _mm256_mul_ps(c6_00,FF);
1399
1400             /* CUBIC SPLINE TABLE REPULSION */
1401             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1402             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1403             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1404                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1405             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1406                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1407             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1408                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1409             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1410                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1411             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1412             Heps             = _mm256_mul_ps(vfeps,H);
1413             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1414             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1415             fvdw12           = _mm256_mul_ps(c12_00,FF);
1416             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1417
1418             fscal            = fvdw;
1419
1420             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1421
1422             /* Calculate temporary vectorial force */
1423             tx               = _mm256_mul_ps(fscal,dx00);
1424             ty               = _mm256_mul_ps(fscal,dy00);
1425             tz               = _mm256_mul_ps(fscal,dz00);
1426
1427             /* Update vectorial force */
1428             fix0             = _mm256_add_ps(fix0,tx);
1429             fiy0             = _mm256_add_ps(fiy0,ty);
1430             fiz0             = _mm256_add_ps(fiz0,tz);
1431
1432             fjx0             = _mm256_add_ps(fjx0,tx);
1433             fjy0             = _mm256_add_ps(fjy0,ty);
1434             fjz0             = _mm256_add_ps(fjz0,tz);
1435
1436             /**************************
1437              * CALCULATE INTERACTIONS *
1438              **************************/
1439
1440             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1441             {
1442
1443             /* Compute parameters for interactions between i and j atoms */
1444             qq10             = _mm256_mul_ps(iq1,jq0);
1445
1446             /* REACTION-FIELD ELECTROSTATICS */
1447             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1448
1449             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1450
1451             fscal            = felec;
1452
1453             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1454
1455             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1456
1457             /* Calculate temporary vectorial force */
1458             tx               = _mm256_mul_ps(fscal,dx10);
1459             ty               = _mm256_mul_ps(fscal,dy10);
1460             tz               = _mm256_mul_ps(fscal,dz10);
1461
1462             /* Update vectorial force */
1463             fix1             = _mm256_add_ps(fix1,tx);
1464             fiy1             = _mm256_add_ps(fiy1,ty);
1465             fiz1             = _mm256_add_ps(fiz1,tz);
1466
1467             fjx0             = _mm256_add_ps(fjx0,tx);
1468             fjy0             = _mm256_add_ps(fjy0,ty);
1469             fjz0             = _mm256_add_ps(fjz0,tz);
1470
1471             }
1472
1473             /**************************
1474              * CALCULATE INTERACTIONS *
1475              **************************/
1476
1477             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1478             {
1479
1480             /* Compute parameters for interactions between i and j atoms */
1481             qq20             = _mm256_mul_ps(iq2,jq0);
1482
1483             /* REACTION-FIELD ELECTROSTATICS */
1484             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1485
1486             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1487
1488             fscal            = felec;
1489
1490             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1491
1492             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1493
1494             /* Calculate temporary vectorial force */
1495             tx               = _mm256_mul_ps(fscal,dx20);
1496             ty               = _mm256_mul_ps(fscal,dy20);
1497             tz               = _mm256_mul_ps(fscal,dz20);
1498
1499             /* Update vectorial force */
1500             fix2             = _mm256_add_ps(fix2,tx);
1501             fiy2             = _mm256_add_ps(fiy2,ty);
1502             fiz2             = _mm256_add_ps(fiz2,tz);
1503
1504             fjx0             = _mm256_add_ps(fjx0,tx);
1505             fjy0             = _mm256_add_ps(fjy0,ty);
1506             fjz0             = _mm256_add_ps(fjz0,tz);
1507
1508             }
1509
1510             /**************************
1511              * CALCULATE INTERACTIONS *
1512              **************************/
1513
1514             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1515             {
1516
1517             /* Compute parameters for interactions between i and j atoms */
1518             qq30             = _mm256_mul_ps(iq3,jq0);
1519
1520             /* REACTION-FIELD ELECTROSTATICS */
1521             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1522
1523             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1524
1525             fscal            = felec;
1526
1527             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1528
1529             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1530
1531             /* Calculate temporary vectorial force */
1532             tx               = _mm256_mul_ps(fscal,dx30);
1533             ty               = _mm256_mul_ps(fscal,dy30);
1534             tz               = _mm256_mul_ps(fscal,dz30);
1535
1536             /* Update vectorial force */
1537             fix3             = _mm256_add_ps(fix3,tx);
1538             fiy3             = _mm256_add_ps(fiy3,ty);
1539             fiz3             = _mm256_add_ps(fiz3,tz);
1540
1541             fjx0             = _mm256_add_ps(fjx0,tx);
1542             fjy0             = _mm256_add_ps(fjy0,ty);
1543             fjz0             = _mm256_add_ps(fjz0,tz);
1544
1545             }
1546
1547             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1548             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1549             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1550             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1551             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1552             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1553             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1554             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1555
1556             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1557
1558             /* Inner loop uses 142 flops */
1559         }
1560
1561         /* End of innermost loop */
1562
1563         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1564                                                  f+i_coord_offset,fshift+i_shift_offset);
1565
1566         /* Increment number of inner iterations */
1567         inneriter                  += j_index_end - j_index_start;
1568
1569         /* Outer loop uses 24 flops */
1570     }
1571
1572     /* Increment number of outer iterations */
1573     outeriter        += nri;
1574
1575     /* Update outer/inner flops */
1576
1577     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);
1578 }