src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          vfitab;
 108     __m128i          vfitab_lo,vfitab_hi;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->ic->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_elec->data;
 134     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 139     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 140     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 141     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149     j_coord_offsetE = 0;
 150     j_coord_offsetF = 0;
 151     j_coord_offsetG = 0;
 152     j_coord_offsetH = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183         fix1             = _mm256_setzero_ps();
 184         fiy1             = _mm256_setzero_ps();
 185         fiz1             = _mm256_setzero_ps();
 186         fix2             = _mm256_setzero_ps();
 187         fiy2             = _mm256_setzero_ps();
 188         fiz2             = _mm256_setzero_ps();
 189         fix3             = _mm256_setzero_ps();
 190         fiy3             = _mm256_setzero_ps();
 191         fiz3             = _mm256_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_ps();
 195         vvdwsum          = _mm256_setzero_ps();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             jnrE             = jjnr[jidx+4];
 207             jnrF             = jjnr[jidx+5];
 208             jnrG             = jjnr[jidx+6];
 209             jnrH             = jjnr[jidx+7];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214             j_coord_offsetE  = DIM*jnrE;
 215             j_coord_offsetF  = DIM*jnrF;
 216             j_coord_offsetG  = DIM*jnrG;
 217             j_coord_offsetH  = DIM*jnrH;
 218
 219             /* load j atom coordinates */
 220             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 221                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 222                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 223                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 224                                                  &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_ps(ix0,jx0);
 228             dy00             = _mm256_sub_ps(iy0,jy0);
 229             dz00             = _mm256_sub_ps(iz0,jz0);
 230             dx10             = _mm256_sub_ps(ix1,jx0);
 231             dy10             = _mm256_sub_ps(iy1,jy0);
 232             dz10             = _mm256_sub_ps(iz1,jz0);
 233             dx20             = _mm256_sub_ps(ix2,jx0);
 234             dy20             = _mm256_sub_ps(iy2,jy0);
 235             dz20             = _mm256_sub_ps(iz2,jz0);
 236             dx30             = _mm256_sub_ps(ix3,jx0);
 237             dy30             = _mm256_sub_ps(iy3,jy0);
 238             dz30             = _mm256_sub_ps(iz3,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 243             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 244             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 245
 246             rinv10           = avx256_invsqrt_f(rsq10);
 247             rinv20           = avx256_invsqrt_f(rsq20);
 248             rinv30           = avx256_invsqrt_f(rsq30);
 249
 250             rinvsq00         = avx256_inv_f(rsq00);
 251
 252             /* Load parameters for j particles */
 253             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 254                                                                  charge+jnrC+0,charge+jnrD+0,
 255                                                                  charge+jnrE+0,charge+jnrF+0,
 256                                                                  charge+jnrG+0,charge+jnrH+0);
 257             vdwjidx0A        = 2*vdwtype[jnrA+0];
 258             vdwjidx0B        = 2*vdwtype[jnrB+0];
 259             vdwjidx0C        = 2*vdwtype[jnrC+0];
 260             vdwjidx0D        = 2*vdwtype[jnrD+0];
 261             vdwjidx0E        = 2*vdwtype[jnrE+0];
 262             vdwjidx0F        = 2*vdwtype[jnrF+0];
 263             vdwjidx0G        = 2*vdwtype[jnrG+0];
 264             vdwjidx0H        = 2*vdwtype[jnrH+0];
 265
 266             fjx0             = _mm256_setzero_ps();
 267             fjy0             = _mm256_setzero_ps();
 268             fjz0             = _mm256_setzero_ps();
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 276                                             vdwioffsetptr0+vdwjidx0B,
 277                                             vdwioffsetptr0+vdwjidx0C,
 278                                             vdwioffsetptr0+vdwjidx0D,
 279                                             vdwioffsetptr0+vdwjidx0E,
 280                                             vdwioffsetptr0+vdwjidx0F,
 281                                             vdwioffsetptr0+vdwjidx0G,
 282                                             vdwioffsetptr0+vdwjidx0H,
 283                                             &c6_00,&c12_00);
 284
 285             /* LENNARD-JONES DISPERSION/REPULSION */
 286
 287             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 288             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 289             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 290             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 291             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 295
 296             fscal            = fvdw;
 297
 298             /* Calculate temporary vectorial force */
 299             tx               = _mm256_mul_ps(fscal,dx00);
 300             ty               = _mm256_mul_ps(fscal,dy00);
 301             tz               = _mm256_mul_ps(fscal,dz00);
 302
 303             /* Update vectorial force */
 304             fix0             = _mm256_add_ps(fix0,tx);
 305             fiy0             = _mm256_add_ps(fiy0,ty);
 306             fiz0             = _mm256_add_ps(fiz0,tz);
 307
 308             fjx0             = _mm256_add_ps(fjx0,tx);
 309             fjy0             = _mm256_add_ps(fjy0,ty);
 310             fjz0             = _mm256_add_ps(fjz0,tz);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             r10              = _mm256_mul_ps(rsq10,rinv10);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq10             = _mm256_mul_ps(iq1,jq0);
 320
 321             /* Calculate table index by multiplying r with table scale and truncate to integer */
 322             rt               = _mm256_mul_ps(r10,vftabscale);
 323             vfitab           = _mm256_cvttps_epi32(rt);
 324             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 325             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 326             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 327             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 328             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 329             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 330
 331             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 332             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 334             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 335                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 336             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 337                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 338             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 339                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 340             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 341             Heps             = _mm256_mul_ps(vfeps,H);
 342             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 343             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 344             velec            = _mm256_mul_ps(qq10,VV);
 345             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 346             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm256_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_ps(fscal,dx10);
 355             ty               = _mm256_mul_ps(fscal,dy10);
 356             tz               = _mm256_mul_ps(fscal,dz10);
 357
 358             /* Update vectorial force */
 359             fix1             = _mm256_add_ps(fix1,tx);
 360             fiy1             = _mm256_add_ps(fiy1,ty);
 361             fiz1             = _mm256_add_ps(fiz1,tz);
 362
 363             fjx0             = _mm256_add_ps(fjx0,tx);
 364             fjy0             = _mm256_add_ps(fjy0,ty);
 365             fjz0             = _mm256_add_ps(fjz0,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r20              = _mm256_mul_ps(rsq20,rinv20);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq20             = _mm256_mul_ps(iq2,jq0);
 375
 376             /* Calculate table index by multiplying r with table scale and truncate to integer */
 377             rt               = _mm256_mul_ps(r20,vftabscale);
 378             vfitab           = _mm256_cvttps_epi32(rt);
 379             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 380             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 381             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 382             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 383             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 384             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 385
 386             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 387             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 388                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 389             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 390                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 391             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 392                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 393             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 394                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 395             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 396             Heps             = _mm256_mul_ps(vfeps,H);
 397             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 398             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 399             velec            = _mm256_mul_ps(qq20,VV);
 400             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 401             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velecsum         = _mm256_add_ps(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_ps(fscal,dx20);
 410             ty               = _mm256_mul_ps(fscal,dy20);
 411             tz               = _mm256_mul_ps(fscal,dz20);
 412
 413             /* Update vectorial force */
 414             fix2             = _mm256_add_ps(fix2,tx);
 415             fiy2             = _mm256_add_ps(fiy2,ty);
 416             fiz2             = _mm256_add_ps(fiz2,tz);
 417
 418             fjx0             = _mm256_add_ps(fjx0,tx);
 419             fjy0             = _mm256_add_ps(fjy0,ty);
 420             fjz0             = _mm256_add_ps(fjz0,tz);
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             r30              = _mm256_mul_ps(rsq30,rinv30);
 427
 428             /* Compute parameters for interactions between i and j atoms */
 429             qq30             = _mm256_mul_ps(iq3,jq0);
 430
 431             /* Calculate table index by multiplying r with table scale and truncate to integer */
 432             rt               = _mm256_mul_ps(r30,vftabscale);
 433             vfitab           = _mm256_cvttps_epi32(rt);
 434             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 435             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 436             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 437             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 438             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 439             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 440
 441             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 442             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 443                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 444             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 445                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 446             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 447                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 448             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 449                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 450             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 451             Heps             = _mm256_mul_ps(vfeps,H);
 452             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 453             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 454             velec            = _mm256_mul_ps(qq30,VV);
 455             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 456             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velecsum         = _mm256_add_ps(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm256_mul_ps(fscal,dx30);
 465             ty               = _mm256_mul_ps(fscal,dy30);
 466             tz               = _mm256_mul_ps(fscal,dz30);
 467
 468             /* Update vectorial force */
 469             fix3             = _mm256_add_ps(fix3,tx);
 470             fiy3             = _mm256_add_ps(fiy3,ty);
 471             fiz3             = _mm256_add_ps(fiz3,tz);
 472
 473             fjx0             = _mm256_add_ps(fjx0,tx);
 474             fjy0             = _mm256_add_ps(fjy0,ty);
 475             fjz0             = _mm256_add_ps(fjz0,tz);
 476
 477             fjptrA             = f+j_coord_offsetA;
 478             fjptrB             = f+j_coord_offsetB;
 479             fjptrC             = f+j_coord_offsetC;
 480             fjptrD             = f+j_coord_offsetD;
 481             fjptrE             = f+j_coord_offsetE;
 482             fjptrF             = f+j_coord_offsetF;
 483             fjptrG             = f+j_coord_offsetG;
 484             fjptrH             = f+j_coord_offsetH;
 485
 486             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 487
 488             /* Inner loop uses 164 flops */
 489         }
 490
 491         if(jidx<j_index_end)
 492         {
 493
 494             /* Get j neighbor index, and coordinate index */
 495             jnrlistA         = jjnr[jidx];
 496             jnrlistB         = jjnr[jidx+1];
 497             jnrlistC         = jjnr[jidx+2];
 498             jnrlistD         = jjnr[jidx+3];
 499             jnrlistE         = jjnr[jidx+4];
 500             jnrlistF         = jjnr[jidx+5];
 501             jnrlistG         = jjnr[jidx+6];
 502             jnrlistH         = jjnr[jidx+7];
 503             /* Sign of each element will be negative for non-real atoms.
 504              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 505              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 506              */
 507             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 508                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 509
 510             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 511             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 512             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 513             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 514             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 515             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 516             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 517             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 518             j_coord_offsetA  = DIM*jnrA;
 519             j_coord_offsetB  = DIM*jnrB;
 520             j_coord_offsetC  = DIM*jnrC;
 521             j_coord_offsetD  = DIM*jnrD;
 522             j_coord_offsetE  = DIM*jnrE;
 523             j_coord_offsetF  = DIM*jnrF;
 524             j_coord_offsetG  = DIM*jnrG;
 525             j_coord_offsetH  = DIM*jnrH;
 526
 527             /* load j atom coordinates */
 528             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 529                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 530                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 531                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 532                                                  &jx0,&jy0,&jz0);
 533
 534             /* Calculate displacement vector */
 535             dx00             = _mm256_sub_ps(ix0,jx0);
 536             dy00             = _mm256_sub_ps(iy0,jy0);
 537             dz00             = _mm256_sub_ps(iz0,jz0);
 538             dx10             = _mm256_sub_ps(ix1,jx0);
 539             dy10             = _mm256_sub_ps(iy1,jy0);
 540             dz10             = _mm256_sub_ps(iz1,jz0);
 541             dx20             = _mm256_sub_ps(ix2,jx0);
 542             dy20             = _mm256_sub_ps(iy2,jy0);
 543             dz20             = _mm256_sub_ps(iz2,jz0);
 544             dx30             = _mm256_sub_ps(ix3,jx0);
 545             dy30             = _mm256_sub_ps(iy3,jy0);
 546             dz30             = _mm256_sub_ps(iz3,jz0);
 547
 548             /* Calculate squared distance and things based on it */
 549             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 550             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 551             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 552             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 553
 554             rinv10           = avx256_invsqrt_f(rsq10);
 555             rinv20           = avx256_invsqrt_f(rsq20);
 556             rinv30           = avx256_invsqrt_f(rsq30);
 557
 558             rinvsq00         = avx256_inv_f(rsq00);
 559
 560             /* Load parameters for j particles */
 561             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 562                                                                  charge+jnrC+0,charge+jnrD+0,
 563                                                                  charge+jnrE+0,charge+jnrF+0,
 564                                                                  charge+jnrG+0,charge+jnrH+0);
 565             vdwjidx0A        = 2*vdwtype[jnrA+0];
 566             vdwjidx0B        = 2*vdwtype[jnrB+0];
 567             vdwjidx0C        = 2*vdwtype[jnrC+0];
 568             vdwjidx0D        = 2*vdwtype[jnrD+0];
 569             vdwjidx0E        = 2*vdwtype[jnrE+0];
 570             vdwjidx0F        = 2*vdwtype[jnrF+0];
 571             vdwjidx0G        = 2*vdwtype[jnrG+0];
 572             vdwjidx0H        = 2*vdwtype[jnrH+0];
 573
 574             fjx0             = _mm256_setzero_ps();
 575             fjy0             = _mm256_setzero_ps();
 576             fjz0             = _mm256_setzero_ps();
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 584                                             vdwioffsetptr0+vdwjidx0B,
 585                                             vdwioffsetptr0+vdwjidx0C,
 586                                             vdwioffsetptr0+vdwjidx0D,
 587                                             vdwioffsetptr0+vdwjidx0E,
 588                                             vdwioffsetptr0+vdwjidx0F,
 589                                             vdwioffsetptr0+vdwjidx0G,
 590                                             vdwioffsetptr0+vdwjidx0H,
 591                                             &c6_00,&c12_00);
 592
 593             /* LENNARD-JONES DISPERSION/REPULSION */
 594
 595             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 596             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 597             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 598             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 599             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 600
 601             /* Update potential sum for this i atom from the interaction with this j atom. */
 602             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 603             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 604
 605             fscal            = fvdw;
 606
 607             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm256_mul_ps(fscal,dx00);
 611             ty               = _mm256_mul_ps(fscal,dy00);
 612             tz               = _mm256_mul_ps(fscal,dz00);
 613
 614             /* Update vectorial force */
 615             fix0             = _mm256_add_ps(fix0,tx);
 616             fiy0             = _mm256_add_ps(fiy0,ty);
 617             fiz0             = _mm256_add_ps(fiz0,tz);
 618
 619             fjx0             = _mm256_add_ps(fjx0,tx);
 620             fjy0             = _mm256_add_ps(fjy0,ty);
 621             fjz0             = _mm256_add_ps(fjz0,tz);
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             r10              = _mm256_mul_ps(rsq10,rinv10);
 628             r10              = _mm256_andnot_ps(dummy_mask,r10);
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq10             = _mm256_mul_ps(iq1,jq0);
 632
 633             /* Calculate table index by multiplying r with table scale and truncate to integer */
 634             rt               = _mm256_mul_ps(r10,vftabscale);
 635             vfitab           = _mm256_cvttps_epi32(rt);
 636             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 637             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 638             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 639             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 640             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 641             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 642
 643             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 644             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 645                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 646             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 647                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 648             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 649                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 650             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 651                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 652             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 653             Heps             = _mm256_mul_ps(vfeps,H);
 654             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 655             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 656             velec            = _mm256_mul_ps(qq10,VV);
 657             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 658             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm256_andnot_ps(dummy_mask,velec);
 662             velecsum         = _mm256_add_ps(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm256_mul_ps(fscal,dx10);
 670             ty               = _mm256_mul_ps(fscal,dy10);
 671             tz               = _mm256_mul_ps(fscal,dz10);
 672
 673             /* Update vectorial force */
 674             fix1             = _mm256_add_ps(fix1,tx);
 675             fiy1             = _mm256_add_ps(fiy1,ty);
 676             fiz1             = _mm256_add_ps(fiz1,tz);
 677
 678             fjx0             = _mm256_add_ps(fjx0,tx);
 679             fjy0             = _mm256_add_ps(fjy0,ty);
 680             fjz0             = _mm256_add_ps(fjz0,tz);
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             r20              = _mm256_mul_ps(rsq20,rinv20);
 687             r20              = _mm256_andnot_ps(dummy_mask,r20);
 688
 689             /* Compute parameters for interactions between i and j atoms */
 690             qq20             = _mm256_mul_ps(iq2,jq0);
 691
 692             /* Calculate table index by multiplying r with table scale and truncate to integer */
 693             rt               = _mm256_mul_ps(r20,vftabscale);
 694             vfitab           = _mm256_cvttps_epi32(rt);
 695             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 696             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 697             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 698             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 699             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 700             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 701
 702             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 703             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 704                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 705             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 706                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 707             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 708                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 709             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 710                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 711             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 712             Heps             = _mm256_mul_ps(vfeps,H);
 713             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 714             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 715             velec            = _mm256_mul_ps(qq20,VV);
 716             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 717             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 718
 719             /* Update potential sum for this i atom from the interaction with this j atom. */
 720             velec            = _mm256_andnot_ps(dummy_mask,velec);
 721             velecsum         = _mm256_add_ps(velecsum,velec);
 722
 723             fscal            = felec;
 724
 725             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 726
 727             /* Calculate temporary vectorial force */
 728             tx               = _mm256_mul_ps(fscal,dx20);
 729             ty               = _mm256_mul_ps(fscal,dy20);
 730             tz               = _mm256_mul_ps(fscal,dz20);
 731
 732             /* Update vectorial force */
 733             fix2             = _mm256_add_ps(fix2,tx);
 734             fiy2             = _mm256_add_ps(fiy2,ty);
 735             fiz2             = _mm256_add_ps(fiz2,tz);
 736
 737             fjx0             = _mm256_add_ps(fjx0,tx);
 738             fjy0             = _mm256_add_ps(fjy0,ty);
 739             fjz0             = _mm256_add_ps(fjz0,tz);
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             r30              = _mm256_mul_ps(rsq30,rinv30);
 746             r30              = _mm256_andnot_ps(dummy_mask,r30);
 747
 748             /* Compute parameters for interactions between i and j atoms */
 749             qq30             = _mm256_mul_ps(iq3,jq0);
 750
 751             /* Calculate table index by multiplying r with table scale and truncate to integer */
 752             rt               = _mm256_mul_ps(r30,vftabscale);
 753             vfitab           = _mm256_cvttps_epi32(rt);
 754             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 755             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 756             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 757             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 758             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 759             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 760
 761             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 762             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 763                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 764             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 765                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 766             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 767                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 768             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 769                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 770             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 771             Heps             = _mm256_mul_ps(vfeps,H);
 772             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 773             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 774             velec            = _mm256_mul_ps(qq30,VV);
 775             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 776             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
 777
 778             /* Update potential sum for this i atom from the interaction with this j atom. */
 779             velec            = _mm256_andnot_ps(dummy_mask,velec);
 780             velecsum         = _mm256_add_ps(velecsum,velec);
 781
 782             fscal            = felec;
 783
 784             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 785
 786             /* Calculate temporary vectorial force */
 787             tx               = _mm256_mul_ps(fscal,dx30);
 788             ty               = _mm256_mul_ps(fscal,dy30);
 789             tz               = _mm256_mul_ps(fscal,dz30);
 790
 791             /* Update vectorial force */
 792             fix3             = _mm256_add_ps(fix3,tx);
 793             fiy3             = _mm256_add_ps(fiy3,ty);
 794             fiz3             = _mm256_add_ps(fiz3,tz);
 795
 796             fjx0             = _mm256_add_ps(fjx0,tx);
 797             fjy0             = _mm256_add_ps(fjy0,ty);
 798             fjz0             = _mm256_add_ps(fjz0,tz);
 799
 800             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 801             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 802             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 803             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 804             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 805             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 806             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 807             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 808
 809             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 810
 811             /* Inner loop uses 167 flops */
 812         }
 813
 814         /* End of innermost loop */
 815
 816         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 817                                                  f+i_coord_offset,fshift+i_shift_offset);
 818
 819         ggid                        = gid[iidx];
 820         /* Update potential energies */
 821         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 822         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 823
 824         /* Increment number of inner iterations */
 825         inneriter                  += j_index_end - j_index_start;
 826
 827         /* Outer loop uses 26 flops */
 828     }
 829
 830     /* Increment number of outer iterations */
 831     outeriter        += nri;
 832
 833     /* Update outer/inner flops */
 834
 835     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 836 }
 837 /*
 838  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_single
 839  * Electrostatics interaction: CubicSplineTable
 840  * VdW interaction:            LennardJones
 841  * Geometry:                   Water4-Particle
 842  * Calculate force/pot:        Force
 843  */
 844 void
 845 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_single
 846                     (t_nblist                    * gmx_restrict       nlist,
 847                      rvec                        * gmx_restrict          xx,
 848                      rvec                        * gmx_restrict          ff,
 849                      struct t_forcerec           * gmx_restrict          fr,
 850                      t_mdatoms                   * gmx_restrict     mdatoms,
 851                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 852                      t_nrnb                      * gmx_restrict        nrnb)
 853 {
 854     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 855      * just 0 for non-waters.
 856      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 857      * jnr indices corresponding to data put in the four positions in the SIMD register.
 858      */
 859     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 860     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 861     int              jnrA,jnrB,jnrC,jnrD;
 862     int              jnrE,jnrF,jnrG,jnrH;
 863     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 864     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 865     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 866     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 867     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 868     real             rcutoff_scalar;
 869     real             *shiftvec,*fshift,*x,*f;
 870     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 871     real             scratch[4*DIM];
 872     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 873     real *           vdwioffsetptr0;
 874     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 875     real *           vdwioffsetptr1;
 876     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 877     real *           vdwioffsetptr2;
 878     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 879     real *           vdwioffsetptr3;
 880     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 881     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 882     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 883     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 884     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 885     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 886     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 887     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 888     real             *charge;
 889     int              nvdwtype;
 890     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 891     int              *vdwtype;
 892     real             *vdwparam;
 893     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 894     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 895     __m256i          vfitab;
 896     __m128i          vfitab_lo,vfitab_hi;
 897     __m128i          ifour       = _mm_set1_epi32(4);
 898     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 899     real             *vftab;
 900     __m256           dummy_mask,cutoff_mask;
 901     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 902     __m256           one     = _mm256_set1_ps(1.0);
 903     __m256           two     = _mm256_set1_ps(2.0);
 904     x                = xx[0];
 905     f                = ff[0];
 906
 907     nri              = nlist->nri;
 908     iinr             = nlist->iinr;
 909     jindex           = nlist->jindex;
 910     jjnr             = nlist->jjnr;
 911     shiftidx         = nlist->shift;
 912     gid              = nlist->gid;
 913     shiftvec         = fr->shift_vec[0];
 914     fshift           = fr->fshift[0];
 915     facel            = _mm256_set1_ps(fr->ic->epsfac);
 916     charge           = mdatoms->chargeA;
 917     nvdwtype         = fr->ntype;
 918     vdwparam         = fr->nbfp;
 919     vdwtype          = mdatoms->typeA;
 920
 921     vftab            = kernel_data->table_elec->data;
 922     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 923
 924     /* Setup water-specific parameters */
 925     inr              = nlist->iinr[0];
 926     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 927     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 928     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 929     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 930
 931     /* Avoid stupid compiler warnings */
 932     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 933     j_coord_offsetA = 0;
 934     j_coord_offsetB = 0;
 935     j_coord_offsetC = 0;
 936     j_coord_offsetD = 0;
 937     j_coord_offsetE = 0;
 938     j_coord_offsetF = 0;
 939     j_coord_offsetG = 0;
 940     j_coord_offsetH = 0;
 941
 942     outeriter        = 0;
 943     inneriter        = 0;
 944
 945     for(iidx=0;iidx<4*DIM;iidx++)
 946     {
 947         scratch[iidx] = 0.0;
 948     }
 949
 950     /* Start outer loop over neighborlists */
 951     for(iidx=0; iidx<nri; iidx++)
 952     {
 953         /* Load shift vector for this list */
 954         i_shift_offset   = DIM*shiftidx[iidx];
 955
 956         /* Load limits for loop over neighbors */
 957         j_index_start    = jindex[iidx];
 958         j_index_end      = jindex[iidx+1];
 959
 960         /* Get outer coordinate index */
 961         inr              = iinr[iidx];
 962         i_coord_offset   = DIM*inr;
 963
 964         /* Load i particle coords and add shift vector */
 965         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 966                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 967
 968         fix0             = _mm256_setzero_ps();
 969         fiy0             = _mm256_setzero_ps();
 970         fiz0             = _mm256_setzero_ps();
 971         fix1             = _mm256_setzero_ps();
 972         fiy1             = _mm256_setzero_ps();
 973         fiz1             = _mm256_setzero_ps();
 974         fix2             = _mm256_setzero_ps();
 975         fiy2             = _mm256_setzero_ps();
 976         fiz2             = _mm256_setzero_ps();
 977         fix3             = _mm256_setzero_ps();
 978         fiy3             = _mm256_setzero_ps();
 979         fiz3             = _mm256_setzero_ps();
 980
 981         /* Start inner kernel loop */
 982         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 983         {
 984
 985             /* Get j neighbor index, and coordinate index */
 986             jnrA             = jjnr[jidx];
 987             jnrB             = jjnr[jidx+1];
 988             jnrC             = jjnr[jidx+2];
 989             jnrD             = jjnr[jidx+3];
 990             jnrE             = jjnr[jidx+4];
 991             jnrF             = jjnr[jidx+5];
 992             jnrG             = jjnr[jidx+6];
 993             jnrH             = jjnr[jidx+7];
 994             j_coord_offsetA  = DIM*jnrA;
 995             j_coord_offsetB  = DIM*jnrB;
 996             j_coord_offsetC  = DIM*jnrC;
 997             j_coord_offsetD  = DIM*jnrD;
 998             j_coord_offsetE  = DIM*jnrE;
 999             j_coord_offsetF  = DIM*jnrF;
1000             j_coord_offsetG  = DIM*jnrG;
1001             j_coord_offsetH  = DIM*jnrH;
1002
1003             /* load j atom coordinates */
1004             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1005                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1006                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1007                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1008                                                  &jx0,&jy0,&jz0);
1009
1010             /* Calculate displacement vector */
1011             dx00             = _mm256_sub_ps(ix0,jx0);
1012             dy00             = _mm256_sub_ps(iy0,jy0);
1013             dz00             = _mm256_sub_ps(iz0,jz0);
1014             dx10             = _mm256_sub_ps(ix1,jx0);
1015             dy10             = _mm256_sub_ps(iy1,jy0);
1016             dz10             = _mm256_sub_ps(iz1,jz0);
1017             dx20             = _mm256_sub_ps(ix2,jx0);
1018             dy20             = _mm256_sub_ps(iy2,jy0);
1019             dz20             = _mm256_sub_ps(iz2,jz0);
1020             dx30             = _mm256_sub_ps(ix3,jx0);
1021             dy30             = _mm256_sub_ps(iy3,jy0);
1022             dz30             = _mm256_sub_ps(iz3,jz0);
1023
1024             /* Calculate squared distance and things based on it */
1025             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1026             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1027             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1028             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1029
1030             rinv10           = avx256_invsqrt_f(rsq10);
1031             rinv20           = avx256_invsqrt_f(rsq20);
1032             rinv30           = avx256_invsqrt_f(rsq30);
1033
1034             rinvsq00         = avx256_inv_f(rsq00);
1035
1036             /* Load parameters for j particles */
1037             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1038                                                                  charge+jnrC+0,charge+jnrD+0,
1039                                                                  charge+jnrE+0,charge+jnrF+0,
1040                                                                  charge+jnrG+0,charge+jnrH+0);
1041             vdwjidx0A        = 2*vdwtype[jnrA+0];
1042             vdwjidx0B        = 2*vdwtype[jnrB+0];
1043             vdwjidx0C        = 2*vdwtype[jnrC+0];
1044             vdwjidx0D        = 2*vdwtype[jnrD+0];
1045             vdwjidx0E        = 2*vdwtype[jnrE+0];
1046             vdwjidx0F        = 2*vdwtype[jnrF+0];
1047             vdwjidx0G        = 2*vdwtype[jnrG+0];
1048             vdwjidx0H        = 2*vdwtype[jnrH+0];
1049
1050             fjx0             = _mm256_setzero_ps();
1051             fjy0             = _mm256_setzero_ps();
1052             fjz0             = _mm256_setzero_ps();
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1060                                             vdwioffsetptr0+vdwjidx0B,
1061                                             vdwioffsetptr0+vdwjidx0C,
1062                                             vdwioffsetptr0+vdwjidx0D,
1063                                             vdwioffsetptr0+vdwjidx0E,
1064                                             vdwioffsetptr0+vdwjidx0F,
1065                                             vdwioffsetptr0+vdwjidx0G,
1066                                             vdwioffsetptr0+vdwjidx0H,
1067                                             &c6_00,&c12_00);
1068
1069             /* LENNARD-JONES DISPERSION/REPULSION */
1070
1071             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1072             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1073
1074             fscal            = fvdw;
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm256_mul_ps(fscal,dx00);
1078             ty               = _mm256_mul_ps(fscal,dy00);
1079             tz               = _mm256_mul_ps(fscal,dz00);
1080
1081             /* Update vectorial force */
1082             fix0             = _mm256_add_ps(fix0,tx);
1083             fiy0             = _mm256_add_ps(fiy0,ty);
1084             fiz0             = _mm256_add_ps(fiz0,tz);
1085
1086             fjx0             = _mm256_add_ps(fjx0,tx);
1087             fjy0             = _mm256_add_ps(fjy0,ty);
1088             fjz0             = _mm256_add_ps(fjz0,tz);
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
1092              **************************/
1093
1094             r10              = _mm256_mul_ps(rsq10,rinv10);
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             qq10             = _mm256_mul_ps(iq1,jq0);
1098
1099             /* Calculate table index by multiplying r with table scale and truncate to integer */
1100             rt               = _mm256_mul_ps(r10,vftabscale);
1101             vfitab           = _mm256_cvttps_epi32(rt);
1102             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1103             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1104             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1105             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1106             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1107             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1108
1109             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1110             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1111                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1112             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1113                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1114             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1115                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1116             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1117                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1118             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1119             Heps             = _mm256_mul_ps(vfeps,H);
1120             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1121             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1122             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1123
1124             fscal            = felec;
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm256_mul_ps(fscal,dx10);
1128             ty               = _mm256_mul_ps(fscal,dy10);
1129             tz               = _mm256_mul_ps(fscal,dz10);
1130
1131             /* Update vectorial force */
1132             fix1             = _mm256_add_ps(fix1,tx);
1133             fiy1             = _mm256_add_ps(fiy1,ty);
1134             fiz1             = _mm256_add_ps(fiz1,tz);
1135
1136             fjx0             = _mm256_add_ps(fjx0,tx);
1137             fjy0             = _mm256_add_ps(fjy0,ty);
1138             fjz0             = _mm256_add_ps(fjz0,tz);
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             r20              = _mm256_mul_ps(rsq20,rinv20);
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             qq20             = _mm256_mul_ps(iq2,jq0);
1148
1149             /* Calculate table index by multiplying r with table scale and truncate to integer */
1150             rt               = _mm256_mul_ps(r20,vftabscale);
1151             vfitab           = _mm256_cvttps_epi32(rt);
1152             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1153             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1154             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1155             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1156             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1157             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1158
1159             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1160             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1161                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1162             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1163                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1164             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1165                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1166             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1167                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1168             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1169             Heps             = _mm256_mul_ps(vfeps,H);
1170             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1171             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1172             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1173
1174             fscal            = felec;
1175
1176             /* Calculate temporary vectorial force */
1177             tx               = _mm256_mul_ps(fscal,dx20);
1178             ty               = _mm256_mul_ps(fscal,dy20);
1179             tz               = _mm256_mul_ps(fscal,dz20);
1180
1181             /* Update vectorial force */
1182             fix2             = _mm256_add_ps(fix2,tx);
1183             fiy2             = _mm256_add_ps(fiy2,ty);
1184             fiz2             = _mm256_add_ps(fiz2,tz);
1185
1186             fjx0             = _mm256_add_ps(fjx0,tx);
1187             fjy0             = _mm256_add_ps(fjy0,ty);
1188             fjz0             = _mm256_add_ps(fjz0,tz);
1189
1190             /**************************
1191              * CALCULATE INTERACTIONS *
1192              **************************/
1193
1194             r30              = _mm256_mul_ps(rsq30,rinv30);
1195
1196             /* Compute parameters for interactions between i and j atoms */
1197             qq30             = _mm256_mul_ps(iq3,jq0);
1198
1199             /* Calculate table index by multiplying r with table scale and truncate to integer */
1200             rt               = _mm256_mul_ps(r30,vftabscale);
1201             vfitab           = _mm256_cvttps_epi32(rt);
1202             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1203             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1204             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1205             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1206             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1207             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1208
1209             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1210             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1212             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1214             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1215                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1216             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1217                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1218             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1219             Heps             = _mm256_mul_ps(vfeps,H);
1220             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1221             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1222             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1223
1224             fscal            = felec;
1225
1226             /* Calculate temporary vectorial force */
1227             tx               = _mm256_mul_ps(fscal,dx30);
1228             ty               = _mm256_mul_ps(fscal,dy30);
1229             tz               = _mm256_mul_ps(fscal,dz30);
1230
1231             /* Update vectorial force */
1232             fix3             = _mm256_add_ps(fix3,tx);
1233             fiy3             = _mm256_add_ps(fiy3,ty);
1234             fiz3             = _mm256_add_ps(fiz3,tz);
1235
1236             fjx0             = _mm256_add_ps(fjx0,tx);
1237             fjy0             = _mm256_add_ps(fjy0,ty);
1238             fjz0             = _mm256_add_ps(fjz0,tz);
1239
1240             fjptrA             = f+j_coord_offsetA;
1241             fjptrB             = f+j_coord_offsetB;
1242             fjptrC             = f+j_coord_offsetC;
1243             fjptrD             = f+j_coord_offsetD;
1244             fjptrE             = f+j_coord_offsetE;
1245             fjptrF             = f+j_coord_offsetF;
1246             fjptrG             = f+j_coord_offsetG;
1247             fjptrH             = f+j_coord_offsetH;
1248
1249             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1250
1251             /* Inner loop uses 147 flops */
1252         }
1253
1254         if(jidx<j_index_end)
1255         {
1256
1257             /* Get j neighbor index, and coordinate index */
1258             jnrlistA         = jjnr[jidx];
1259             jnrlistB         = jjnr[jidx+1];
1260             jnrlistC         = jjnr[jidx+2];
1261             jnrlistD         = jjnr[jidx+3];
1262             jnrlistE         = jjnr[jidx+4];
1263             jnrlistF         = jjnr[jidx+5];
1264             jnrlistG         = jjnr[jidx+6];
1265             jnrlistH         = jjnr[jidx+7];
1266             /* Sign of each element will be negative for non-real atoms.
1267              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1268              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1269              */
1270             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1271                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1272
1273             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1274             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1275             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1276             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1277             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1278             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1279             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1280             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1281             j_coord_offsetA  = DIM*jnrA;
1282             j_coord_offsetB  = DIM*jnrB;
1283             j_coord_offsetC  = DIM*jnrC;
1284             j_coord_offsetD  = DIM*jnrD;
1285             j_coord_offsetE  = DIM*jnrE;
1286             j_coord_offsetF  = DIM*jnrF;
1287             j_coord_offsetG  = DIM*jnrG;
1288             j_coord_offsetH  = DIM*jnrH;
1289
1290             /* load j atom coordinates */
1291             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1292                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1293                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1294                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1295                                                  &jx0,&jy0,&jz0);
1296
1297             /* Calculate displacement vector */
1298             dx00             = _mm256_sub_ps(ix0,jx0);
1299             dy00             = _mm256_sub_ps(iy0,jy0);
1300             dz00             = _mm256_sub_ps(iz0,jz0);
1301             dx10             = _mm256_sub_ps(ix1,jx0);
1302             dy10             = _mm256_sub_ps(iy1,jy0);
1303             dz10             = _mm256_sub_ps(iz1,jz0);
1304             dx20             = _mm256_sub_ps(ix2,jx0);
1305             dy20             = _mm256_sub_ps(iy2,jy0);
1306             dz20             = _mm256_sub_ps(iz2,jz0);
1307             dx30             = _mm256_sub_ps(ix3,jx0);
1308             dy30             = _mm256_sub_ps(iy3,jy0);
1309             dz30             = _mm256_sub_ps(iz3,jz0);
1310
1311             /* Calculate squared distance and things based on it */
1312             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1313             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1314             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1315             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1316
1317             rinv10           = avx256_invsqrt_f(rsq10);
1318             rinv20           = avx256_invsqrt_f(rsq20);
1319             rinv30           = avx256_invsqrt_f(rsq30);
1320
1321             rinvsq00         = avx256_inv_f(rsq00);
1322
1323             /* Load parameters for j particles */
1324             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1325                                                                  charge+jnrC+0,charge+jnrD+0,
1326                                                                  charge+jnrE+0,charge+jnrF+0,
1327                                                                  charge+jnrG+0,charge+jnrH+0);
1328             vdwjidx0A        = 2*vdwtype[jnrA+0];
1329             vdwjidx0B        = 2*vdwtype[jnrB+0];
1330             vdwjidx0C        = 2*vdwtype[jnrC+0];
1331             vdwjidx0D        = 2*vdwtype[jnrD+0];
1332             vdwjidx0E        = 2*vdwtype[jnrE+0];
1333             vdwjidx0F        = 2*vdwtype[jnrF+0];
1334             vdwjidx0G        = 2*vdwtype[jnrG+0];
1335             vdwjidx0H        = 2*vdwtype[jnrH+0];
1336
1337             fjx0             = _mm256_setzero_ps();
1338             fjy0             = _mm256_setzero_ps();
1339             fjz0             = _mm256_setzero_ps();
1340
1341             /**************************
1342              * CALCULATE INTERACTIONS *
1343              **************************/
1344
1345             /* Compute parameters for interactions between i and j atoms */
1346             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1347                                             vdwioffsetptr0+vdwjidx0B,
1348                                             vdwioffsetptr0+vdwjidx0C,
1349                                             vdwioffsetptr0+vdwjidx0D,
1350                                             vdwioffsetptr0+vdwjidx0E,
1351                                             vdwioffsetptr0+vdwjidx0F,
1352                                             vdwioffsetptr0+vdwjidx0G,
1353                                             vdwioffsetptr0+vdwjidx0H,
1354                                             &c6_00,&c12_00);
1355
1356             /* LENNARD-JONES DISPERSION/REPULSION */
1357
1358             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1359             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1360
1361             fscal            = fvdw;
1362
1363             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1364
1365             /* Calculate temporary vectorial force */
1366             tx               = _mm256_mul_ps(fscal,dx00);
1367             ty               = _mm256_mul_ps(fscal,dy00);
1368             tz               = _mm256_mul_ps(fscal,dz00);
1369
1370             /* Update vectorial force */
1371             fix0             = _mm256_add_ps(fix0,tx);
1372             fiy0             = _mm256_add_ps(fiy0,ty);
1373             fiz0             = _mm256_add_ps(fiz0,tz);
1374
1375             fjx0             = _mm256_add_ps(fjx0,tx);
1376             fjy0             = _mm256_add_ps(fjy0,ty);
1377             fjz0             = _mm256_add_ps(fjz0,tz);
1378
1379             /**************************
1380              * CALCULATE INTERACTIONS *
1381              **************************/
1382
1383             r10              = _mm256_mul_ps(rsq10,rinv10);
1384             r10              = _mm256_andnot_ps(dummy_mask,r10);
1385
1386             /* Compute parameters for interactions between i and j atoms */
1387             qq10             = _mm256_mul_ps(iq1,jq0);
1388
1389             /* Calculate table index by multiplying r with table scale and truncate to integer */
1390             rt               = _mm256_mul_ps(r10,vftabscale);
1391             vfitab           = _mm256_cvttps_epi32(rt);
1392             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1393             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1394             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1395             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1396             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1397             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1398
1399             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1400             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1401                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1402             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1403                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1404             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1405                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1406             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1407                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1408             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1409             Heps             = _mm256_mul_ps(vfeps,H);
1410             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1411             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1412             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1413
1414             fscal            = felec;
1415
1416             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1417
1418             /* Calculate temporary vectorial force */
1419             tx               = _mm256_mul_ps(fscal,dx10);
1420             ty               = _mm256_mul_ps(fscal,dy10);
1421             tz               = _mm256_mul_ps(fscal,dz10);
1422
1423             /* Update vectorial force */
1424             fix1             = _mm256_add_ps(fix1,tx);
1425             fiy1             = _mm256_add_ps(fiy1,ty);
1426             fiz1             = _mm256_add_ps(fiz1,tz);
1427
1428             fjx0             = _mm256_add_ps(fjx0,tx);
1429             fjy0             = _mm256_add_ps(fjy0,ty);
1430             fjz0             = _mm256_add_ps(fjz0,tz);
1431
1432             /**************************
1433              * CALCULATE INTERACTIONS *
1434              **************************/
1435
1436             r20              = _mm256_mul_ps(rsq20,rinv20);
1437             r20              = _mm256_andnot_ps(dummy_mask,r20);
1438
1439             /* Compute parameters for interactions between i and j atoms */
1440             qq20             = _mm256_mul_ps(iq2,jq0);
1441
1442             /* Calculate table index by multiplying r with table scale and truncate to integer */
1443             rt               = _mm256_mul_ps(r20,vftabscale);
1444             vfitab           = _mm256_cvttps_epi32(rt);
1445             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1446             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1447             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1448             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1449             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1450             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1451
1452             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1453             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1454                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1455             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1456                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1457             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1458                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1459             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1460                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1461             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1462             Heps             = _mm256_mul_ps(vfeps,H);
1463             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1464             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1465             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1466
1467             fscal            = felec;
1468
1469             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1470
1471             /* Calculate temporary vectorial force */
1472             tx               = _mm256_mul_ps(fscal,dx20);
1473             ty               = _mm256_mul_ps(fscal,dy20);
1474             tz               = _mm256_mul_ps(fscal,dz20);
1475
1476             /* Update vectorial force */
1477             fix2             = _mm256_add_ps(fix2,tx);
1478             fiy2             = _mm256_add_ps(fiy2,ty);
1479             fiz2             = _mm256_add_ps(fiz2,tz);
1480
1481             fjx0             = _mm256_add_ps(fjx0,tx);
1482             fjy0             = _mm256_add_ps(fjy0,ty);
1483             fjz0             = _mm256_add_ps(fjz0,tz);
1484
1485             /**************************
1486              * CALCULATE INTERACTIONS *
1487              **************************/
1488
1489             r30              = _mm256_mul_ps(rsq30,rinv30);
1490             r30              = _mm256_andnot_ps(dummy_mask,r30);
1491
1492             /* Compute parameters for interactions between i and j atoms */
1493             qq30             = _mm256_mul_ps(iq3,jq0);
1494
1495             /* Calculate table index by multiplying r with table scale and truncate to integer */
1496             rt               = _mm256_mul_ps(r30,vftabscale);
1497             vfitab           = _mm256_cvttps_epi32(rt);
1498             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1499             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1500             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1501             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1502             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1503             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1504
1505             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1506             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1507                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1508             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1509                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1510             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1511                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1512             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1513                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1514             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1515             Heps             = _mm256_mul_ps(vfeps,H);
1516             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1517             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1518             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq30,FF),_mm256_mul_ps(vftabscale,rinv30)));
1519
1520             fscal            = felec;
1521
1522             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1523
1524             /* Calculate temporary vectorial force */
1525             tx               = _mm256_mul_ps(fscal,dx30);
1526             ty               = _mm256_mul_ps(fscal,dy30);
1527             tz               = _mm256_mul_ps(fscal,dz30);
1528
1529             /* Update vectorial force */
1530             fix3             = _mm256_add_ps(fix3,tx);
1531             fiy3             = _mm256_add_ps(fiy3,ty);
1532             fiz3             = _mm256_add_ps(fiz3,tz);
1533
1534             fjx0             = _mm256_add_ps(fjx0,tx);
1535             fjy0             = _mm256_add_ps(fjy0,ty);
1536             fjz0             = _mm256_add_ps(fjz0,tz);
1537
1538             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1539             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1540             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1541             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1542             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1543             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1544             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1545             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1546
1547             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1548
1549             /* Inner loop uses 150 flops */
1550         }
1551
1552         /* End of innermost loop */
1553
1554         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1555                                                  f+i_coord_offset,fshift+i_shift_offset);
1556
1557         /* Increment number of inner iterations */
1558         inneriter                  += j_index_end - j_index_start;
1559
1560         /* Outer loop uses 24 flops */
1561     }
1562
1563     /* Increment number of outer iterations */
1564     outeriter        += nri;
1565
1566     /* Update outer/inner flops */
1567
1568     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1569 }