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) 2012,2013,2014,2015,2017, 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_ElecEw_VdwCSTab_GeomP1P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  88     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  97     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  98     __m256i          vfitab;
  99     __m128i          vfitab_lo,vfitab_hi;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m256i          ewitab;
 104     __m128i          ewitab_lo,ewitab_hi;
 105     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 107     real             *ewtab;
 108     __m256           dummy_mask,cutoff_mask;
 109     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 110     __m256           one     = _mm256_set1_ps(1.0);
 111     __m256           two     = _mm256_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_ps(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 131
 132     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 134     beta2            = _mm256_mul_ps(beta,beta);
 135     beta3            = _mm256_mul_ps(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 176
 177         fix0             = _mm256_setzero_ps();
 178         fiy0             = _mm256_setzero_ps();
 179         fiz0             = _mm256_setzero_ps();
 180
 181         /* Load parameters for i particles */
 182         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 183         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm256_setzero_ps();
 187         vvdwsum          = _mm256_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             jnrE             = jjnr[jidx+4];
 199             jnrF             = jjnr[jidx+5];
 200             jnrG             = jjnr[jidx+6];
 201             jnrH             = jjnr[jidx+7];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206             j_coord_offsetE  = DIM*jnrE;
 207             j_coord_offsetF  = DIM*jnrF;
 208             j_coord_offsetG  = DIM*jnrG;
 209             j_coord_offsetH  = DIM*jnrH;
 210
 211             /* load j atom coordinates */
 212             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 214                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 215                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 216                                                  &jx0,&jy0,&jz0);
 217
 218             /* Calculate displacement vector */
 219             dx00             = _mm256_sub_ps(ix0,jx0);
 220             dy00             = _mm256_sub_ps(iy0,jy0);
 221             dz00             = _mm256_sub_ps(iz0,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 225
 226             rinv00           = avx256_invsqrt_f(rsq00);
 227
 228             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 232                                                                  charge+jnrC+0,charge+jnrD+0,
 233                                                                  charge+jnrE+0,charge+jnrF+0,
 234                                                                  charge+jnrG+0,charge+jnrH+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239             vdwjidx0E        = 2*vdwtype[jnrE+0];
 240             vdwjidx0F        = 2*vdwtype[jnrF+0];
 241             vdwjidx0G        = 2*vdwtype[jnrG+0];
 242             vdwjidx0H        = 2*vdwtype[jnrH+0];
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             r00              = _mm256_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm256_mul_ps(iq0,jq0);
 252             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 253                                             vdwioffsetptr0+vdwjidx0B,
 254                                             vdwioffsetptr0+vdwjidx0C,
 255                                             vdwioffsetptr0+vdwjidx0D,
 256                                             vdwioffsetptr0+vdwjidx0E,
 257                                             vdwioffsetptr0+vdwjidx0F,
 258                                             vdwioffsetptr0+vdwjidx0G,
 259                                             vdwioffsetptr0+vdwjidx0H,
 260                                             &c6_00,&c12_00);
 261
 262             /* Calculate table index by multiplying r with table scale and truncate to integer */
 263             rt               = _mm256_mul_ps(r00,vftabscale);
 264             vfitab           = _mm256_cvttps_epi32(rt);
 265             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 266             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 267             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 268             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 269             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 270             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 271
 272             /* EWALD ELECTROSTATICS */
 273
 274             /* Analytical PME correction */
 275             zeta2            = _mm256_mul_ps(beta2,rsq00);
 276             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 277             pmecorrF         = avx256_pmecorrF_f(zeta2);
 278             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 279             felec            = _mm256_mul_ps(qq00,felec);
 280             pmecorrV         = avx256_pmecorrV_f(zeta2);
 281             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 282             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 283             velec            = _mm256_mul_ps(qq00,velec);
 284
 285             /* CUBIC SPLINE TABLE DISPERSION */
 286             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 287                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 288             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 289                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 290             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 292             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 294             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 295             Heps             = _mm256_mul_ps(vfeps,H);
 296             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 297             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 298             vvdw6            = _mm256_mul_ps(c6_00,VV);
 299             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 300             fvdw6            = _mm256_mul_ps(c6_00,FF);
 301
 302             /* CUBIC SPLINE TABLE REPULSION */
 303             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 304             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 305             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 307             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 309             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 311             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 313             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 314             Heps             = _mm256_mul_ps(vfeps,H);
 315             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 316             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 317             vvdw12           = _mm256_mul_ps(c12_00,VV);
 318             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 319             fvdw12           = _mm256_mul_ps(c12_00,FF);
 320             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 321             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velecsum         = _mm256_add_ps(velecsum,velec);
 325             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 326
 327             fscal            = _mm256_add_ps(felec,fvdw);
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm256_mul_ps(fscal,dx00);
 331             ty               = _mm256_mul_ps(fscal,dy00);
 332             tz               = _mm256_mul_ps(fscal,dz00);
 333
 334             /* Update vectorial force */
 335             fix0             = _mm256_add_ps(fix0,tx);
 336             fiy0             = _mm256_add_ps(fiy0,ty);
 337             fiz0             = _mm256_add_ps(fiz0,tz);
 338
 339             fjptrA             = f+j_coord_offsetA;
 340             fjptrB             = f+j_coord_offsetB;
 341             fjptrC             = f+j_coord_offsetC;
 342             fjptrD             = f+j_coord_offsetD;
 343             fjptrE             = f+j_coord_offsetE;
 344             fjptrF             = f+j_coord_offsetF;
 345             fjptrG             = f+j_coord_offsetG;
 346             fjptrH             = f+j_coord_offsetH;
 347             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 348
 349             /* Inner loop uses 118 flops */
 350         }
 351
 352         if(jidx<j_index_end)
 353         {
 354
 355             /* Get j neighbor index, and coordinate index */
 356             jnrlistA         = jjnr[jidx];
 357             jnrlistB         = jjnr[jidx+1];
 358             jnrlistC         = jjnr[jidx+2];
 359             jnrlistD         = jjnr[jidx+3];
 360             jnrlistE         = jjnr[jidx+4];
 361             jnrlistF         = jjnr[jidx+5];
 362             jnrlistG         = jjnr[jidx+6];
 363             jnrlistH         = jjnr[jidx+7];
 364             /* Sign of each element will be negative for non-real atoms.
 365              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 366              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 367              */
 368             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 369                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 370
 371             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 372             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 373             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 374             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 375             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 376             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 377             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 378             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 379             j_coord_offsetA  = DIM*jnrA;
 380             j_coord_offsetB  = DIM*jnrB;
 381             j_coord_offsetC  = DIM*jnrC;
 382             j_coord_offsetD  = DIM*jnrD;
 383             j_coord_offsetE  = DIM*jnrE;
 384             j_coord_offsetF  = DIM*jnrF;
 385             j_coord_offsetG  = DIM*jnrG;
 386             j_coord_offsetH  = DIM*jnrH;
 387
 388             /* load j atom coordinates */
 389             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 390                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 391                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 392                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 393                                                  &jx0,&jy0,&jz0);
 394
 395             /* Calculate displacement vector */
 396             dx00             = _mm256_sub_ps(ix0,jx0);
 397             dy00             = _mm256_sub_ps(iy0,jy0);
 398             dz00             = _mm256_sub_ps(iz0,jz0);
 399
 400             /* Calculate squared distance and things based on it */
 401             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 402
 403             rinv00           = avx256_invsqrt_f(rsq00);
 404
 405             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 406
 407             /* Load parameters for j particles */
 408             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 409                                                                  charge+jnrC+0,charge+jnrD+0,
 410                                                                  charge+jnrE+0,charge+jnrF+0,
 411                                                                  charge+jnrG+0,charge+jnrH+0);
 412             vdwjidx0A        = 2*vdwtype[jnrA+0];
 413             vdwjidx0B        = 2*vdwtype[jnrB+0];
 414             vdwjidx0C        = 2*vdwtype[jnrC+0];
 415             vdwjidx0D        = 2*vdwtype[jnrD+0];
 416             vdwjidx0E        = 2*vdwtype[jnrE+0];
 417             vdwjidx0F        = 2*vdwtype[jnrF+0];
 418             vdwjidx0G        = 2*vdwtype[jnrG+0];
 419             vdwjidx0H        = 2*vdwtype[jnrH+0];
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             r00              = _mm256_mul_ps(rsq00,rinv00);
 426             r00              = _mm256_andnot_ps(dummy_mask,r00);
 427
 428             /* Compute parameters for interactions between i and j atoms */
 429             qq00             = _mm256_mul_ps(iq0,jq0);
 430             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 431                                             vdwioffsetptr0+vdwjidx0B,
 432                                             vdwioffsetptr0+vdwjidx0C,
 433                                             vdwioffsetptr0+vdwjidx0D,
 434                                             vdwioffsetptr0+vdwjidx0E,
 435                                             vdwioffsetptr0+vdwjidx0F,
 436                                             vdwioffsetptr0+vdwjidx0G,
 437                                             vdwioffsetptr0+vdwjidx0H,
 438                                             &c6_00,&c12_00);
 439
 440             /* Calculate table index by multiplying r with table scale and truncate to integer */
 441             rt               = _mm256_mul_ps(r00,vftabscale);
 442             vfitab           = _mm256_cvttps_epi32(rt);
 443             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 444             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 445             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 446             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 447             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 448             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Analytical PME correction */
 453             zeta2            = _mm256_mul_ps(beta2,rsq00);
 454             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 455             pmecorrF         = avx256_pmecorrF_f(zeta2);
 456             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 457             felec            = _mm256_mul_ps(qq00,felec);
 458             pmecorrV         = avx256_pmecorrV_f(zeta2);
 459             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 460             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 461             velec            = _mm256_mul_ps(qq00,velec);
 462
 463             /* CUBIC SPLINE TABLE DISPERSION */
 464             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 465                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 466             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 467                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 468             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 469                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 470             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 471                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 472             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 473             Heps             = _mm256_mul_ps(vfeps,H);
 474             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 475             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 476             vvdw6            = _mm256_mul_ps(c6_00,VV);
 477             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 478             fvdw6            = _mm256_mul_ps(c6_00,FF);
 479
 480             /* CUBIC SPLINE TABLE REPULSION */
 481             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 482             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 483             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 484                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 485             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 486                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 487             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 488                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 489             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 490                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 491             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 492             Heps             = _mm256_mul_ps(vfeps,H);
 493             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 494             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 495             vvdw12           = _mm256_mul_ps(c12_00,VV);
 496             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 497             fvdw12           = _mm256_mul_ps(c12_00,FF);
 498             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 499             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velec            = _mm256_andnot_ps(dummy_mask,velec);
 503             velecsum         = _mm256_add_ps(velecsum,velec);
 504             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 505             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 506
 507             fscal            = _mm256_add_ps(felec,fvdw);
 508
 509             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm256_mul_ps(fscal,dx00);
 513             ty               = _mm256_mul_ps(fscal,dy00);
 514             tz               = _mm256_mul_ps(fscal,dz00);
 515
 516             /* Update vectorial force */
 517             fix0             = _mm256_add_ps(fix0,tx);
 518             fiy0             = _mm256_add_ps(fiy0,ty);
 519             fiz0             = _mm256_add_ps(fiz0,tz);
 520
 521             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 522             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 523             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 524             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 525             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 526             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 527             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 528             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 529             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 530
 531             /* Inner loop uses 119 flops */
 532         }
 533
 534         /* End of innermost loop */
 535
 536         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 537                                                  f+i_coord_offset,fshift+i_shift_offset);
 538
 539         ggid                        = gid[iidx];
 540         /* Update potential energies */
 541         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 542         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 543
 544         /* Increment number of inner iterations */
 545         inneriter                  += j_index_end - j_index_start;
 546
 547         /* Outer loop uses 9 flops */
 548     }
 549
 550     /* Increment number of outer iterations */
 551     outeriter        += nri;
 552
 553     /* Update outer/inner flops */
 554
 555     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*119);
 556 }
 557 /*
 558  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_256_single
 559  * Electrostatics interaction: Ewald
 560  * VdW interaction:            CubicSplineTable
 561  * Geometry:                   Particle-Particle
 562  * Calculate force/pot:        Force
 563  */
 564 void
 565 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_256_single
 566                     (t_nblist                    * gmx_restrict       nlist,
 567                      rvec                        * gmx_restrict          xx,
 568                      rvec                        * gmx_restrict          ff,
 569                      struct t_forcerec           * gmx_restrict          fr,
 570                      t_mdatoms                   * gmx_restrict     mdatoms,
 571                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 572                      t_nrnb                      * gmx_restrict        nrnb)
 573 {
 574     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 575      * just 0 for non-waters.
 576      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 577      * jnr indices corresponding to data put in the four positions in the SIMD register.
 578      */
 579     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 580     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 581     int              jnrA,jnrB,jnrC,jnrD;
 582     int              jnrE,jnrF,jnrG,jnrH;
 583     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 584     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 585     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 586     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 587     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 588     real             rcutoff_scalar;
 589     real             *shiftvec,*fshift,*x,*f;
 590     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 591     real             scratch[4*DIM];
 592     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 593     real *           vdwioffsetptr0;
 594     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 595     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 596     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 597     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 598     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 599     real             *charge;
 600     int              nvdwtype;
 601     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 602     int              *vdwtype;
 603     real             *vdwparam;
 604     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 605     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 606     __m256i          vfitab;
 607     __m128i          vfitab_lo,vfitab_hi;
 608     __m128i          ifour       = _mm_set1_epi32(4);
 609     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 610     real             *vftab;
 611     __m256i          ewitab;
 612     __m128i          ewitab_lo,ewitab_hi;
 613     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 614     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 615     real             *ewtab;
 616     __m256           dummy_mask,cutoff_mask;
 617     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 618     __m256           one     = _mm256_set1_ps(1.0);
 619     __m256           two     = _mm256_set1_ps(2.0);
 620     x                = xx[0];
 621     f                = ff[0];
 622
 623     nri              = nlist->nri;
 624     iinr             = nlist->iinr;
 625     jindex           = nlist->jindex;
 626     jjnr             = nlist->jjnr;
 627     shiftidx         = nlist->shift;
 628     gid              = nlist->gid;
 629     shiftvec         = fr->shift_vec[0];
 630     fshift           = fr->fshift[0];
 631     facel            = _mm256_set1_ps(fr->ic->epsfac);
 632     charge           = mdatoms->chargeA;
 633     nvdwtype         = fr->ntype;
 634     vdwparam         = fr->nbfp;
 635     vdwtype          = mdatoms->typeA;
 636
 637     vftab            = kernel_data->table_vdw->data;
 638     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 639
 640     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 641     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 642     beta2            = _mm256_mul_ps(beta,beta);
 643     beta3            = _mm256_mul_ps(beta,beta2);
 644
 645     ewtab            = fr->ic->tabq_coul_F;
 646     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 647     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 648
 649     /* Avoid stupid compiler warnings */
 650     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 651     j_coord_offsetA = 0;
 652     j_coord_offsetB = 0;
 653     j_coord_offsetC = 0;
 654     j_coord_offsetD = 0;
 655     j_coord_offsetE = 0;
 656     j_coord_offsetF = 0;
 657     j_coord_offsetG = 0;
 658     j_coord_offsetH = 0;
 659
 660     outeriter        = 0;
 661     inneriter        = 0;
 662
 663     for(iidx=0;iidx<4*DIM;iidx++)
 664     {
 665         scratch[iidx] = 0.0;
 666     }
 667
 668     /* Start outer loop over neighborlists */
 669     for(iidx=0; iidx<nri; iidx++)
 670     {
 671         /* Load shift vector for this list */
 672         i_shift_offset   = DIM*shiftidx[iidx];
 673
 674         /* Load limits for loop over neighbors */
 675         j_index_start    = jindex[iidx];
 676         j_index_end      = jindex[iidx+1];
 677
 678         /* Get outer coordinate index */
 679         inr              = iinr[iidx];
 680         i_coord_offset   = DIM*inr;
 681
 682         /* Load i particle coords and add shift vector */
 683         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 684
 685         fix0             = _mm256_setzero_ps();
 686         fiy0             = _mm256_setzero_ps();
 687         fiz0             = _mm256_setzero_ps();
 688
 689         /* Load parameters for i particles */
 690         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 691         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 692
 693         /* Start inner kernel loop */
 694         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 695         {
 696
 697             /* Get j neighbor index, and coordinate index */
 698             jnrA             = jjnr[jidx];
 699             jnrB             = jjnr[jidx+1];
 700             jnrC             = jjnr[jidx+2];
 701             jnrD             = jjnr[jidx+3];
 702             jnrE             = jjnr[jidx+4];
 703             jnrF             = jjnr[jidx+5];
 704             jnrG             = jjnr[jidx+6];
 705             jnrH             = jjnr[jidx+7];
 706             j_coord_offsetA  = DIM*jnrA;
 707             j_coord_offsetB  = DIM*jnrB;
 708             j_coord_offsetC  = DIM*jnrC;
 709             j_coord_offsetD  = DIM*jnrD;
 710             j_coord_offsetE  = DIM*jnrE;
 711             j_coord_offsetF  = DIM*jnrF;
 712             j_coord_offsetG  = DIM*jnrG;
 713             j_coord_offsetH  = DIM*jnrH;
 714
 715             /* load j atom coordinates */
 716             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 717                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 718                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 719                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 720                                                  &jx0,&jy0,&jz0);
 721
 722             /* Calculate displacement vector */
 723             dx00             = _mm256_sub_ps(ix0,jx0);
 724             dy00             = _mm256_sub_ps(iy0,jy0);
 725             dz00             = _mm256_sub_ps(iz0,jz0);
 726
 727             /* Calculate squared distance and things based on it */
 728             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 729
 730             rinv00           = avx256_invsqrt_f(rsq00);
 731
 732             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 733
 734             /* Load parameters for j particles */
 735             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 736                                                                  charge+jnrC+0,charge+jnrD+0,
 737                                                                  charge+jnrE+0,charge+jnrF+0,
 738                                                                  charge+jnrG+0,charge+jnrH+0);
 739             vdwjidx0A        = 2*vdwtype[jnrA+0];
 740             vdwjidx0B        = 2*vdwtype[jnrB+0];
 741             vdwjidx0C        = 2*vdwtype[jnrC+0];
 742             vdwjidx0D        = 2*vdwtype[jnrD+0];
 743             vdwjidx0E        = 2*vdwtype[jnrE+0];
 744             vdwjidx0F        = 2*vdwtype[jnrF+0];
 745             vdwjidx0G        = 2*vdwtype[jnrG+0];
 746             vdwjidx0H        = 2*vdwtype[jnrH+0];
 747
 748             /**************************
 749              * CALCULATE INTERACTIONS *
 750              **************************/
 751
 752             r00              = _mm256_mul_ps(rsq00,rinv00);
 753
 754             /* Compute parameters for interactions between i and j atoms */
 755             qq00             = _mm256_mul_ps(iq0,jq0);
 756             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 757                                             vdwioffsetptr0+vdwjidx0B,
 758                                             vdwioffsetptr0+vdwjidx0C,
 759                                             vdwioffsetptr0+vdwjidx0D,
 760                                             vdwioffsetptr0+vdwjidx0E,
 761                                             vdwioffsetptr0+vdwjidx0F,
 762                                             vdwioffsetptr0+vdwjidx0G,
 763                                             vdwioffsetptr0+vdwjidx0H,
 764                                             &c6_00,&c12_00);
 765
 766             /* Calculate table index by multiplying r with table scale and truncate to integer */
 767             rt               = _mm256_mul_ps(r00,vftabscale);
 768             vfitab           = _mm256_cvttps_epi32(rt);
 769             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 770             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 771             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 772             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 773             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 774             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 775
 776             /* EWALD ELECTROSTATICS */
 777
 778             /* Analytical PME correction */
 779             zeta2            = _mm256_mul_ps(beta2,rsq00);
 780             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 781             pmecorrF         = avx256_pmecorrF_f(zeta2);
 782             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 783             felec            = _mm256_mul_ps(qq00,felec);
 784
 785             /* CUBIC SPLINE TABLE DISPERSION */
 786             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 787                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 788             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 789                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 790             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 791                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 792             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 793                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 794             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 795             Heps             = _mm256_mul_ps(vfeps,H);
 796             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 797             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 798             fvdw6            = _mm256_mul_ps(c6_00,FF);
 799
 800             /* CUBIC SPLINE TABLE REPULSION */
 801             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 802             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 803             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 804                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 805             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 806                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 807             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 808                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 809             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 810                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 811             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 812             Heps             = _mm256_mul_ps(vfeps,H);
 813             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 814             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 815             fvdw12           = _mm256_mul_ps(c12_00,FF);
 816             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 817
 818             fscal            = _mm256_add_ps(felec,fvdw);
 819
 820             /* Calculate temporary vectorial force */
 821             tx               = _mm256_mul_ps(fscal,dx00);
 822             ty               = _mm256_mul_ps(fscal,dy00);
 823             tz               = _mm256_mul_ps(fscal,dz00);
 824
 825             /* Update vectorial force */
 826             fix0             = _mm256_add_ps(fix0,tx);
 827             fiy0             = _mm256_add_ps(fiy0,ty);
 828             fiz0             = _mm256_add_ps(fiz0,tz);
 829
 830             fjptrA             = f+j_coord_offsetA;
 831             fjptrB             = f+j_coord_offsetB;
 832             fjptrC             = f+j_coord_offsetC;
 833             fjptrD             = f+j_coord_offsetD;
 834             fjptrE             = f+j_coord_offsetE;
 835             fjptrF             = f+j_coord_offsetF;
 836             fjptrG             = f+j_coord_offsetG;
 837             fjptrH             = f+j_coord_offsetH;
 838             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 839
 840             /* Inner loop uses 82 flops */
 841         }
 842
 843         if(jidx<j_index_end)
 844         {
 845
 846             /* Get j neighbor index, and coordinate index */
 847             jnrlistA         = jjnr[jidx];
 848             jnrlistB         = jjnr[jidx+1];
 849             jnrlistC         = jjnr[jidx+2];
 850             jnrlistD         = jjnr[jidx+3];
 851             jnrlistE         = jjnr[jidx+4];
 852             jnrlistF         = jjnr[jidx+5];
 853             jnrlistG         = jjnr[jidx+6];
 854             jnrlistH         = jjnr[jidx+7];
 855             /* Sign of each element will be negative for non-real atoms.
 856              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 857              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 858              */
 859             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 860                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 861
 862             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 863             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 864             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 865             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 866             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 867             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 868             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 869             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 870             j_coord_offsetA  = DIM*jnrA;
 871             j_coord_offsetB  = DIM*jnrB;
 872             j_coord_offsetC  = DIM*jnrC;
 873             j_coord_offsetD  = DIM*jnrD;
 874             j_coord_offsetE  = DIM*jnrE;
 875             j_coord_offsetF  = DIM*jnrF;
 876             j_coord_offsetG  = DIM*jnrG;
 877             j_coord_offsetH  = DIM*jnrH;
 878
 879             /* load j atom coordinates */
 880             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 881                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 882                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 883                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 884                                                  &jx0,&jy0,&jz0);
 885
 886             /* Calculate displacement vector */
 887             dx00             = _mm256_sub_ps(ix0,jx0);
 888             dy00             = _mm256_sub_ps(iy0,jy0);
 889             dz00             = _mm256_sub_ps(iz0,jz0);
 890
 891             /* Calculate squared distance and things based on it */
 892             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 893
 894             rinv00           = avx256_invsqrt_f(rsq00);
 895
 896             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 897
 898             /* Load parameters for j particles */
 899             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 900                                                                  charge+jnrC+0,charge+jnrD+0,
 901                                                                  charge+jnrE+0,charge+jnrF+0,
 902                                                                  charge+jnrG+0,charge+jnrH+0);
 903             vdwjidx0A        = 2*vdwtype[jnrA+0];
 904             vdwjidx0B        = 2*vdwtype[jnrB+0];
 905             vdwjidx0C        = 2*vdwtype[jnrC+0];
 906             vdwjidx0D        = 2*vdwtype[jnrD+0];
 907             vdwjidx0E        = 2*vdwtype[jnrE+0];
 908             vdwjidx0F        = 2*vdwtype[jnrF+0];
 909             vdwjidx0G        = 2*vdwtype[jnrG+0];
 910             vdwjidx0H        = 2*vdwtype[jnrH+0];
 911
 912             /**************************
 913              * CALCULATE INTERACTIONS *
 914              **************************/
 915
 916             r00              = _mm256_mul_ps(rsq00,rinv00);
 917             r00              = _mm256_andnot_ps(dummy_mask,r00);
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq00             = _mm256_mul_ps(iq0,jq0);
 921             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 922                                             vdwioffsetptr0+vdwjidx0B,
 923                                             vdwioffsetptr0+vdwjidx0C,
 924                                             vdwioffsetptr0+vdwjidx0D,
 925                                             vdwioffsetptr0+vdwjidx0E,
 926                                             vdwioffsetptr0+vdwjidx0F,
 927                                             vdwioffsetptr0+vdwjidx0G,
 928                                             vdwioffsetptr0+vdwjidx0H,
 929                                             &c6_00,&c12_00);
 930
 931             /* Calculate table index by multiplying r with table scale and truncate to integer */
 932             rt               = _mm256_mul_ps(r00,vftabscale);
 933             vfitab           = _mm256_cvttps_epi32(rt);
 934             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 935             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 936             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 937             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 938             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 939             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 940
 941             /* EWALD ELECTROSTATICS */
 942
 943             /* Analytical PME correction */
 944             zeta2            = _mm256_mul_ps(beta2,rsq00);
 945             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 946             pmecorrF         = avx256_pmecorrF_f(zeta2);
 947             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 948             felec            = _mm256_mul_ps(qq00,felec);
 949
 950             /* CUBIC SPLINE TABLE DISPERSION */
 951             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 952                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 953             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 954                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 955             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 956                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 957             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 958                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 959             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 960             Heps             = _mm256_mul_ps(vfeps,H);
 961             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 962             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 963             fvdw6            = _mm256_mul_ps(c6_00,FF);
 964
 965             /* CUBIC SPLINE TABLE REPULSION */
 966             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 967             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 968             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 969                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 970             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 971                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 972             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 973                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 974             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 975                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 976             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 977             Heps             = _mm256_mul_ps(vfeps,H);
 978             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 979             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 980             fvdw12           = _mm256_mul_ps(c12_00,FF);
 981             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 982
 983             fscal            = _mm256_add_ps(felec,fvdw);
 984
 985             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 986
 987             /* Calculate temporary vectorial force */
 988             tx               = _mm256_mul_ps(fscal,dx00);
 989             ty               = _mm256_mul_ps(fscal,dy00);
 990             tz               = _mm256_mul_ps(fscal,dz00);
 991
 992             /* Update vectorial force */
 993             fix0             = _mm256_add_ps(fix0,tx);
 994             fiy0             = _mm256_add_ps(fiy0,ty);
 995             fiz0             = _mm256_add_ps(fiz0,tz);
 996
 997             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 998             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 999             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1000             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1001             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1002             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1003             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1004             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1005             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1006
1007             /* Inner loop uses 83 flops */
1008         }
1009
1010         /* End of innermost loop */
1011
1012         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1013                                                  f+i_coord_offset,fshift+i_shift_offset);
1014
1015         /* Increment number of inner iterations */
1016         inneriter                  += j_index_end - j_index_start;
1017
1018         /* Outer loop uses 7 flops */
1019     }
1020
1021     /* Increment number of outer iterations */
1022     outeriter        += nri;
1023
1024     /* Update outer/inner flops */
1025
1026     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
1027 }