src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  98     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  99     __m256i          vfitab;
 100     __m128i          vfitab_lo,vfitab_hi;
 101     __m128i          ifour       = _mm_set1_epi32(4);
 102     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 103     real             *vftab;
 104     __m256i          ewitab;
 105     __m128i          ewitab_lo,ewitab_hi;
 106     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 108     real             *ewtab;
 109     __m256           dummy_mask,cutoff_mask;
 110     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 111     __m256           one     = _mm256_set1_ps(1.0);
 112     __m256           two     = _mm256_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 132
 133     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 134     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 135     beta2            = _mm256_mul_ps(beta,beta);
 136     beta3            = _mm256_mul_ps(beta,beta2);
 137
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148     j_coord_offsetE = 0;
 149     j_coord_offsetF = 0;
 150     j_coord_offsetG = 0;
 151     j_coord_offsetH = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     for(iidx=0;iidx<4*DIM;iidx++)
 157     {
 158         scratch[iidx] = 0.0;
 159     }
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181
 182         /* Load parameters for i particles */
 183         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 184         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm256_setzero_ps();
 188         vvdwsum          = _mm256_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             jnrE             = jjnr[jidx+4];
 200             jnrF             = jjnr[jidx+5];
 201             jnrG             = jjnr[jidx+6];
 202             jnrH             = jjnr[jidx+7];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207             j_coord_offsetE  = DIM*jnrE;
 208             j_coord_offsetF  = DIM*jnrF;
 209             j_coord_offsetG  = DIM*jnrG;
 210             j_coord_offsetH  = DIM*jnrH;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 216                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 217                                                  &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm256_sub_ps(ix0,jx0);
 221             dy00             = _mm256_sub_ps(iy0,jy0);
 222             dz00             = _mm256_sub_ps(iz0,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 226
 227             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 228
 229             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                                  charge+jnrC+0,charge+jnrD+0,
 234                                                                  charge+jnrE+0,charge+jnrF+0,
 235                                                                  charge+jnrG+0,charge+jnrH+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240             vdwjidx0E        = 2*vdwtype[jnrE+0];
 241             vdwjidx0F        = 2*vdwtype[jnrF+0];
 242             vdwjidx0G        = 2*vdwtype[jnrG+0];
 243             vdwjidx0H        = 2*vdwtype[jnrH+0];
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             r00              = _mm256_mul_ps(rsq00,rinv00);
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm256_mul_ps(iq0,jq0);
 253             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 254                                             vdwioffsetptr0+vdwjidx0B,
 255                                             vdwioffsetptr0+vdwjidx0C,
 256                                             vdwioffsetptr0+vdwjidx0D,
 257                                             vdwioffsetptr0+vdwjidx0E,
 258                                             vdwioffsetptr0+vdwjidx0F,
 259                                             vdwioffsetptr0+vdwjidx0G,
 260                                             vdwioffsetptr0+vdwjidx0H,
 261                                             &c6_00,&c12_00);
 262
 263             /* Calculate table index by multiplying r with table scale and truncate to integer */
 264             rt               = _mm256_mul_ps(r00,vftabscale);
 265             vfitab           = _mm256_cvttps_epi32(rt);
 266             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 267             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 268             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 269             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 270             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 271             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 272
 273             /* EWALD ELECTROSTATICS */
 274
 275             /* Analytical PME correction */
 276             zeta2            = _mm256_mul_ps(beta2,rsq00);
 277             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 278             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 279             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 280             felec            = _mm256_mul_ps(qq00,felec);
 281             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 282             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 283             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 284             velec            = _mm256_mul_ps(qq00,velec);
 285
 286             /* CUBIC SPLINE TABLE DISPERSION */
 287             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 288                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 289             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 291             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 293             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 295             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 296             Heps             = _mm256_mul_ps(vfeps,H);
 297             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 298             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 299             vvdw6            = _mm256_mul_ps(c6_00,VV);
 300             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 301             fvdw6            = _mm256_mul_ps(c6_00,FF);
 302
 303             /* CUBIC SPLINE TABLE REPULSION */
 304             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 305             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 306             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 308             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 310             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 312             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 314             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 315             Heps             = _mm256_mul_ps(vfeps,H);
 316             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 317             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 318             vvdw12           = _mm256_mul_ps(c12_00,VV);
 319             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 320             fvdw12           = _mm256_mul_ps(c12_00,FF);
 321             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 322             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm256_add_ps(velecsum,velec);
 326             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 327
 328             fscal            = _mm256_add_ps(felec,fvdw);
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm256_mul_ps(fscal,dx00);
 332             ty               = _mm256_mul_ps(fscal,dy00);
 333             tz               = _mm256_mul_ps(fscal,dz00);
 334
 335             /* Update vectorial force */
 336             fix0             = _mm256_add_ps(fix0,tx);
 337             fiy0             = _mm256_add_ps(fiy0,ty);
 338             fiz0             = _mm256_add_ps(fiz0,tz);
 339
 340             fjptrA             = f+j_coord_offsetA;
 341             fjptrB             = f+j_coord_offsetB;
 342             fjptrC             = f+j_coord_offsetC;
 343             fjptrD             = f+j_coord_offsetD;
 344             fjptrE             = f+j_coord_offsetE;
 345             fjptrF             = f+j_coord_offsetF;
 346             fjptrG             = f+j_coord_offsetG;
 347             fjptrH             = f+j_coord_offsetH;
 348             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 349
 350             /* Inner loop uses 118 flops */
 351         }
 352
 353         if(jidx<j_index_end)
 354         {
 355
 356             /* Get j neighbor index, and coordinate index */
 357             jnrlistA         = jjnr[jidx];
 358             jnrlistB         = jjnr[jidx+1];
 359             jnrlistC         = jjnr[jidx+2];
 360             jnrlistD         = jjnr[jidx+3];
 361             jnrlistE         = jjnr[jidx+4];
 362             jnrlistF         = jjnr[jidx+5];
 363             jnrlistG         = jjnr[jidx+6];
 364             jnrlistH         = jjnr[jidx+7];
 365             /* Sign of each element will be negative for non-real atoms.
 366              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 367              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 368              */
 369             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 370                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 371
 372             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 373             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 374             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 375             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 376             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 377             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 378             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 379             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 380             j_coord_offsetA  = DIM*jnrA;
 381             j_coord_offsetB  = DIM*jnrB;
 382             j_coord_offsetC  = DIM*jnrC;
 383             j_coord_offsetD  = DIM*jnrD;
 384             j_coord_offsetE  = DIM*jnrE;
 385             j_coord_offsetF  = DIM*jnrF;
 386             j_coord_offsetG  = DIM*jnrG;
 387             j_coord_offsetH  = DIM*jnrH;
 388
 389             /* load j atom coordinates */
 390             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 391                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 392                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 393                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 394                                                  &jx0,&jy0,&jz0);
 395
 396             /* Calculate displacement vector */
 397             dx00             = _mm256_sub_ps(ix0,jx0);
 398             dy00             = _mm256_sub_ps(iy0,jy0);
 399             dz00             = _mm256_sub_ps(iz0,jz0);
 400
 401             /* Calculate squared distance and things based on it */
 402             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 403
 404             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 405
 406             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 407
 408             /* Load parameters for j particles */
 409             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 410                                                                  charge+jnrC+0,charge+jnrD+0,
 411                                                                  charge+jnrE+0,charge+jnrF+0,
 412                                                                  charge+jnrG+0,charge+jnrH+0);
 413             vdwjidx0A        = 2*vdwtype[jnrA+0];
 414             vdwjidx0B        = 2*vdwtype[jnrB+0];
 415             vdwjidx0C        = 2*vdwtype[jnrC+0];
 416             vdwjidx0D        = 2*vdwtype[jnrD+0];
 417             vdwjidx0E        = 2*vdwtype[jnrE+0];
 418             vdwjidx0F        = 2*vdwtype[jnrF+0];
 419             vdwjidx0G        = 2*vdwtype[jnrG+0];
 420             vdwjidx0H        = 2*vdwtype[jnrH+0];
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             r00              = _mm256_mul_ps(rsq00,rinv00);
 427             r00              = _mm256_andnot_ps(dummy_mask,r00);
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq00             = _mm256_mul_ps(iq0,jq0);
 431             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 432                                             vdwioffsetptr0+vdwjidx0B,
 433                                             vdwioffsetptr0+vdwjidx0C,
 434                                             vdwioffsetptr0+vdwjidx0D,
 435                                             vdwioffsetptr0+vdwjidx0E,
 436                                             vdwioffsetptr0+vdwjidx0F,
 437                                             vdwioffsetptr0+vdwjidx0G,
 438                                             vdwioffsetptr0+vdwjidx0H,
 439                                             &c6_00,&c12_00);
 440
 441             /* Calculate table index by multiplying r with table scale and truncate to integer */
 442             rt               = _mm256_mul_ps(r00,vftabscale);
 443             vfitab           = _mm256_cvttps_epi32(rt);
 444             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 445             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 446             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 447             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 448             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 449             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 450
 451             /* EWALD ELECTROSTATICS */
 452
 453             /* Analytical PME correction */
 454             zeta2            = _mm256_mul_ps(beta2,rsq00);
 455             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 456             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 457             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 458             felec            = _mm256_mul_ps(qq00,felec);
 459             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 460             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 461             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 462             velec            = _mm256_mul_ps(qq00,velec);
 463
 464             /* CUBIC SPLINE TABLE DISPERSION */
 465             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 466                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 467             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 468                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 469             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 470                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 471             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 472                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 473             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 474             Heps             = _mm256_mul_ps(vfeps,H);
 475             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 476             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 477             vvdw6            = _mm256_mul_ps(c6_00,VV);
 478             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 479             fvdw6            = _mm256_mul_ps(c6_00,FF);
 480
 481             /* CUBIC SPLINE TABLE REPULSION */
 482             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 483             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 484             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 485                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 486             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 487                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 488             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 489                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 490             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 491                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 492             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 493             Heps             = _mm256_mul_ps(vfeps,H);
 494             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 495             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 496             vvdw12           = _mm256_mul_ps(c12_00,VV);
 497             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 498             fvdw12           = _mm256_mul_ps(c12_00,FF);
 499             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 500             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velec            = _mm256_andnot_ps(dummy_mask,velec);
 504             velecsum         = _mm256_add_ps(velecsum,velec);
 505             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 506             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 507
 508             fscal            = _mm256_add_ps(felec,fvdw);
 509
 510             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 511
 512             /* Calculate temporary vectorial force */
 513             tx               = _mm256_mul_ps(fscal,dx00);
 514             ty               = _mm256_mul_ps(fscal,dy00);
 515             tz               = _mm256_mul_ps(fscal,dz00);
 516
 517             /* Update vectorial force */
 518             fix0             = _mm256_add_ps(fix0,tx);
 519             fiy0             = _mm256_add_ps(fiy0,ty);
 520             fiz0             = _mm256_add_ps(fiz0,tz);
 521
 522             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 523             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 524             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 525             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 526             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 527             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 528             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 529             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 530             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 531
 532             /* Inner loop uses 119 flops */
 533         }
 534
 535         /* End of innermost loop */
 536
 537         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 538                                                  f+i_coord_offset,fshift+i_shift_offset);
 539
 540         ggid                        = gid[iidx];
 541         /* Update potential energies */
 542         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 543         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 544
 545         /* Increment number of inner iterations */
 546         inneriter                  += j_index_end - j_index_start;
 547
 548         /* Outer loop uses 9 flops */
 549     }
 550
 551     /* Increment number of outer iterations */
 552     outeriter        += nri;
 553
 554     /* Update outer/inner flops */
 555
 556     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*119);
 557 }
 558 /*
 559  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_256_single
 560  * Electrostatics interaction: Ewald
 561  * VdW interaction:            CubicSplineTable
 562  * Geometry:                   Particle-Particle
 563  * Calculate force/pot:        Force
 564  */
 565 void
 566 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_256_single
 567                     (t_nblist                    * gmx_restrict       nlist,
 568                      rvec                        * gmx_restrict          xx,
 569                      rvec                        * gmx_restrict          ff,
 570                      t_forcerec                  * gmx_restrict          fr,
 571                      t_mdatoms                   * gmx_restrict     mdatoms,
 572                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 573                      t_nrnb                      * gmx_restrict        nrnb)
 574 {
 575     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 576      * just 0 for non-waters.
 577      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 578      * jnr indices corresponding to data put in the four positions in the SIMD register.
 579      */
 580     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 581     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 582     int              jnrA,jnrB,jnrC,jnrD;
 583     int              jnrE,jnrF,jnrG,jnrH;
 584     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 585     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 586     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 587     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 588     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 589     real             rcutoff_scalar;
 590     real             *shiftvec,*fshift,*x,*f;
 591     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 592     real             scratch[4*DIM];
 593     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 594     real *           vdwioffsetptr0;
 595     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 596     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 597     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 598     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 599     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 600     real             *charge;
 601     int              nvdwtype;
 602     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 603     int              *vdwtype;
 604     real             *vdwparam;
 605     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 606     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 607     __m256i          vfitab;
 608     __m128i          vfitab_lo,vfitab_hi;
 609     __m128i          ifour       = _mm_set1_epi32(4);
 610     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 611     real             *vftab;
 612     __m256i          ewitab;
 613     __m128i          ewitab_lo,ewitab_hi;
 614     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 615     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 616     real             *ewtab;
 617     __m256           dummy_mask,cutoff_mask;
 618     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 619     __m256           one     = _mm256_set1_ps(1.0);
 620     __m256           two     = _mm256_set1_ps(2.0);
 621     x                = xx[0];
 622     f                = ff[0];
 623
 624     nri              = nlist->nri;
 625     iinr             = nlist->iinr;
 626     jindex           = nlist->jindex;
 627     jjnr             = nlist->jjnr;
 628     shiftidx         = nlist->shift;
 629     gid              = nlist->gid;
 630     shiftvec         = fr->shift_vec[0];
 631     fshift           = fr->fshift[0];
 632     facel            = _mm256_set1_ps(fr->epsfac);
 633     charge           = mdatoms->chargeA;
 634     nvdwtype         = fr->ntype;
 635     vdwparam         = fr->nbfp;
 636     vdwtype          = mdatoms->typeA;
 637
 638     vftab            = kernel_data->table_vdw->data;
 639     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 640
 641     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 642     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 643     beta2            = _mm256_mul_ps(beta,beta);
 644     beta3            = _mm256_mul_ps(beta,beta2);
 645
 646     ewtab            = fr->ic->tabq_coul_F;
 647     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 648     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 649
 650     /* Avoid stupid compiler warnings */
 651     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 652     j_coord_offsetA = 0;
 653     j_coord_offsetB = 0;
 654     j_coord_offsetC = 0;
 655     j_coord_offsetD = 0;
 656     j_coord_offsetE = 0;
 657     j_coord_offsetF = 0;
 658     j_coord_offsetG = 0;
 659     j_coord_offsetH = 0;
 660
 661     outeriter        = 0;
 662     inneriter        = 0;
 663
 664     for(iidx=0;iidx<4*DIM;iidx++)
 665     {
 666         scratch[iidx] = 0.0;
 667     }
 668
 669     /* Start outer loop over neighborlists */
 670     for(iidx=0; iidx<nri; iidx++)
 671     {
 672         /* Load shift vector for this list */
 673         i_shift_offset   = DIM*shiftidx[iidx];
 674
 675         /* Load limits for loop over neighbors */
 676         j_index_start    = jindex[iidx];
 677         j_index_end      = jindex[iidx+1];
 678
 679         /* Get outer coordinate index */
 680         inr              = iinr[iidx];
 681         i_coord_offset   = DIM*inr;
 682
 683         /* Load i particle coords and add shift vector */
 684         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 685
 686         fix0             = _mm256_setzero_ps();
 687         fiy0             = _mm256_setzero_ps();
 688         fiz0             = _mm256_setzero_ps();
 689
 690         /* Load parameters for i particles */
 691         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 692         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 693
 694         /* Start inner kernel loop */
 695         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 696         {
 697
 698             /* Get j neighbor index, and coordinate index */
 699             jnrA             = jjnr[jidx];
 700             jnrB             = jjnr[jidx+1];
 701             jnrC             = jjnr[jidx+2];
 702             jnrD             = jjnr[jidx+3];
 703             jnrE             = jjnr[jidx+4];
 704             jnrF             = jjnr[jidx+5];
 705             jnrG             = jjnr[jidx+6];
 706             jnrH             = jjnr[jidx+7];
 707             j_coord_offsetA  = DIM*jnrA;
 708             j_coord_offsetB  = DIM*jnrB;
 709             j_coord_offsetC  = DIM*jnrC;
 710             j_coord_offsetD  = DIM*jnrD;
 711             j_coord_offsetE  = DIM*jnrE;
 712             j_coord_offsetF  = DIM*jnrF;
 713             j_coord_offsetG  = DIM*jnrG;
 714             j_coord_offsetH  = DIM*jnrH;
 715
 716             /* load j atom coordinates */
 717             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 718                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 719                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 720                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 721                                                  &jx0,&jy0,&jz0);
 722
 723             /* Calculate displacement vector */
 724             dx00             = _mm256_sub_ps(ix0,jx0);
 725             dy00             = _mm256_sub_ps(iy0,jy0);
 726             dz00             = _mm256_sub_ps(iz0,jz0);
 727
 728             /* Calculate squared distance and things based on it */
 729             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 730
 731             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 732
 733             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 734
 735             /* Load parameters for j particles */
 736             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 737                                                                  charge+jnrC+0,charge+jnrD+0,
 738                                                                  charge+jnrE+0,charge+jnrF+0,
 739                                                                  charge+jnrG+0,charge+jnrH+0);
 740             vdwjidx0A        = 2*vdwtype[jnrA+0];
 741             vdwjidx0B        = 2*vdwtype[jnrB+0];
 742             vdwjidx0C        = 2*vdwtype[jnrC+0];
 743             vdwjidx0D        = 2*vdwtype[jnrD+0];
 744             vdwjidx0E        = 2*vdwtype[jnrE+0];
 745             vdwjidx0F        = 2*vdwtype[jnrF+0];
 746             vdwjidx0G        = 2*vdwtype[jnrG+0];
 747             vdwjidx0H        = 2*vdwtype[jnrH+0];
 748
 749             /**************************
 750              * CALCULATE INTERACTIONS *
 751              **************************/
 752
 753             r00              = _mm256_mul_ps(rsq00,rinv00);
 754
 755             /* Compute parameters for interactions between i and j atoms */
 756             qq00             = _mm256_mul_ps(iq0,jq0);
 757             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 758                                             vdwioffsetptr0+vdwjidx0B,
 759                                             vdwioffsetptr0+vdwjidx0C,
 760                                             vdwioffsetptr0+vdwjidx0D,
 761                                             vdwioffsetptr0+vdwjidx0E,
 762                                             vdwioffsetptr0+vdwjidx0F,
 763                                             vdwioffsetptr0+vdwjidx0G,
 764                                             vdwioffsetptr0+vdwjidx0H,
 765                                             &c6_00,&c12_00);
 766
 767             /* Calculate table index by multiplying r with table scale and truncate to integer */
 768             rt               = _mm256_mul_ps(r00,vftabscale);
 769             vfitab           = _mm256_cvttps_epi32(rt);
 770             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 771             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 772             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 773             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 774             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 775             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 776
 777             /* EWALD ELECTROSTATICS */
 778
 779             /* Analytical PME correction */
 780             zeta2            = _mm256_mul_ps(beta2,rsq00);
 781             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 782             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 783             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 784             felec            = _mm256_mul_ps(qq00,felec);
 785
 786             /* CUBIC SPLINE TABLE DISPERSION */
 787             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 788                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 789             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 790                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 791             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 792                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 793             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 794                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 795             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 796             Heps             = _mm256_mul_ps(vfeps,H);
 797             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 798             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 799             fvdw6            = _mm256_mul_ps(c6_00,FF);
 800
 801             /* CUBIC SPLINE TABLE REPULSION */
 802             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 803             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 804             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 805                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 806             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 807                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 808             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 809                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 810             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 811                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 812             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 813             Heps             = _mm256_mul_ps(vfeps,H);
 814             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 815             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 816             fvdw12           = _mm256_mul_ps(c12_00,FF);
 817             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 818
 819             fscal            = _mm256_add_ps(felec,fvdw);
 820
 821             /* Calculate temporary vectorial force */
 822             tx               = _mm256_mul_ps(fscal,dx00);
 823             ty               = _mm256_mul_ps(fscal,dy00);
 824             tz               = _mm256_mul_ps(fscal,dz00);
 825
 826             /* Update vectorial force */
 827             fix0             = _mm256_add_ps(fix0,tx);
 828             fiy0             = _mm256_add_ps(fiy0,ty);
 829             fiz0             = _mm256_add_ps(fiz0,tz);
 830
 831             fjptrA             = f+j_coord_offsetA;
 832             fjptrB             = f+j_coord_offsetB;
 833             fjptrC             = f+j_coord_offsetC;
 834             fjptrD             = f+j_coord_offsetD;
 835             fjptrE             = f+j_coord_offsetE;
 836             fjptrF             = f+j_coord_offsetF;
 837             fjptrG             = f+j_coord_offsetG;
 838             fjptrH             = f+j_coord_offsetH;
 839             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 840
 841             /* Inner loop uses 82 flops */
 842         }
 843
 844         if(jidx<j_index_end)
 845         {
 846
 847             /* Get j neighbor index, and coordinate index */
 848             jnrlistA         = jjnr[jidx];
 849             jnrlistB         = jjnr[jidx+1];
 850             jnrlistC         = jjnr[jidx+2];
 851             jnrlistD         = jjnr[jidx+3];
 852             jnrlistE         = jjnr[jidx+4];
 853             jnrlistF         = jjnr[jidx+5];
 854             jnrlistG         = jjnr[jidx+6];
 855             jnrlistH         = jjnr[jidx+7];
 856             /* Sign of each element will be negative for non-real atoms.
 857              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 858              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 859              */
 860             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 861                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 862
 863             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 864             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 865             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 866             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 867             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 868             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 869             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 870             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 871             j_coord_offsetA  = DIM*jnrA;
 872             j_coord_offsetB  = DIM*jnrB;
 873             j_coord_offsetC  = DIM*jnrC;
 874             j_coord_offsetD  = DIM*jnrD;
 875             j_coord_offsetE  = DIM*jnrE;
 876             j_coord_offsetF  = DIM*jnrF;
 877             j_coord_offsetG  = DIM*jnrG;
 878             j_coord_offsetH  = DIM*jnrH;
 879
 880             /* load j atom coordinates */
 881             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 882                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 883                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 884                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 885                                                  &jx0,&jy0,&jz0);
 886
 887             /* Calculate displacement vector */
 888             dx00             = _mm256_sub_ps(ix0,jx0);
 889             dy00             = _mm256_sub_ps(iy0,jy0);
 890             dz00             = _mm256_sub_ps(iz0,jz0);
 891
 892             /* Calculate squared distance and things based on it */
 893             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 894
 895             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 896
 897             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 898
 899             /* Load parameters for j particles */
 900             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 901                                                                  charge+jnrC+0,charge+jnrD+0,
 902                                                                  charge+jnrE+0,charge+jnrF+0,
 903                                                                  charge+jnrG+0,charge+jnrH+0);
 904             vdwjidx0A        = 2*vdwtype[jnrA+0];
 905             vdwjidx0B        = 2*vdwtype[jnrB+0];
 906             vdwjidx0C        = 2*vdwtype[jnrC+0];
 907             vdwjidx0D        = 2*vdwtype[jnrD+0];
 908             vdwjidx0E        = 2*vdwtype[jnrE+0];
 909             vdwjidx0F        = 2*vdwtype[jnrF+0];
 910             vdwjidx0G        = 2*vdwtype[jnrG+0];
 911             vdwjidx0H        = 2*vdwtype[jnrH+0];
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r00              = _mm256_mul_ps(rsq00,rinv00);
 918             r00              = _mm256_andnot_ps(dummy_mask,r00);
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             qq00             = _mm256_mul_ps(iq0,jq0);
 922             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 923                                             vdwioffsetptr0+vdwjidx0B,
 924                                             vdwioffsetptr0+vdwjidx0C,
 925                                             vdwioffsetptr0+vdwjidx0D,
 926                                             vdwioffsetptr0+vdwjidx0E,
 927                                             vdwioffsetptr0+vdwjidx0F,
 928                                             vdwioffsetptr0+vdwjidx0G,
 929                                             vdwioffsetptr0+vdwjidx0H,
 930                                             &c6_00,&c12_00);
 931
 932             /* Calculate table index by multiplying r with table scale and truncate to integer */
 933             rt               = _mm256_mul_ps(r00,vftabscale);
 934             vfitab           = _mm256_cvttps_epi32(rt);
 935             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 936             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 937             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 938             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 939             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 940             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 941
 942             /* EWALD ELECTROSTATICS */
 943
 944             /* Analytical PME correction */
 945             zeta2            = _mm256_mul_ps(beta2,rsq00);
 946             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 947             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 948             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 949             felec            = _mm256_mul_ps(qq00,felec);
 950
 951             /* CUBIC SPLINE TABLE DISPERSION */
 952             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 953                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 954             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 955                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 956             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 957                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 958             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 959                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 960             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 961             Heps             = _mm256_mul_ps(vfeps,H);
 962             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 963             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 964             fvdw6            = _mm256_mul_ps(c6_00,FF);
 965
 966             /* CUBIC SPLINE TABLE REPULSION */
 967             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 968             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 969             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 970                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 971             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 972                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 973             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 974                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 975             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 976                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 977             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 978             Heps             = _mm256_mul_ps(vfeps,H);
 979             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 980             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 981             fvdw12           = _mm256_mul_ps(c12_00,FF);
 982             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 983
 984             fscal            = _mm256_add_ps(felec,fvdw);
 985
 986             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 987
 988             /* Calculate temporary vectorial force */
 989             tx               = _mm256_mul_ps(fscal,dx00);
 990             ty               = _mm256_mul_ps(fscal,dy00);
 991             tz               = _mm256_mul_ps(fscal,dz00);
 992
 993             /* Update vectorial force */
 994             fix0             = _mm256_add_ps(fix0,tx);
 995             fiy0             = _mm256_add_ps(fiy0,ty);
 996             fiz0             = _mm256_add_ps(fiz0,tz);
 997
 998             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 999             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1000             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1001             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1002             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1003             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1004             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1005             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1006             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1007
1008             /* Inner loop uses 83 flops */
1009         }
1010
1011         /* End of innermost loop */
1012
1013         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1014                                                  f+i_coord_offset,fshift+i_shift_offset);
1015
1016         /* Increment number of inner iterations */
1017         inneriter                  += j_index_end - j_index_start;
1018
1019         /* Outer loop uses 7 flops */
1020     }
1021
1022     /* Increment number of outer iterations */
1023     outeriter        += nri;
1024
1025     /* Update outer/inner flops */
1026
1027     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
1028 }