src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_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     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     vftab            = kernel_data->table_vdw->data;
 126     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134     j_coord_offsetE = 0;
 135     j_coord_offsetF = 0;
 136     j_coord_offsetG = 0;
 137     j_coord_offsetH = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 163
 164         fix0             = _mm256_setzero_ps();
 165         fiy0             = _mm256_setzero_ps();
 166         fiz0             = _mm256_setzero_ps();
 167
 168         /* Load parameters for i particles */
 169         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 170         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm256_setzero_ps();
 174         vvdwsum          = _mm256_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             jnrE             = jjnr[jidx+4];
 186             jnrF             = jjnr[jidx+5];
 187             jnrG             = jjnr[jidx+6];
 188             jnrH             = jjnr[jidx+7];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193             j_coord_offsetE  = DIM*jnrE;
 194             j_coord_offsetF  = DIM*jnrF;
 195             j_coord_offsetG  = DIM*jnrG;
 196             j_coord_offsetH  = DIM*jnrH;
 197
 198             /* load j atom coordinates */
 199             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 201                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 202                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 203                                                  &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm256_sub_ps(ix0,jx0);
 207             dy00             = _mm256_sub_ps(iy0,jy0);
 208             dz00             = _mm256_sub_ps(iz0,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 212
 213             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 214
 215             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                                  charge+jnrC+0,charge+jnrD+0,
 220                                                                  charge+jnrE+0,charge+jnrF+0,
 221                                                                  charge+jnrG+0,charge+jnrH+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224             vdwjidx0C        = 2*vdwtype[jnrC+0];
 225             vdwjidx0D        = 2*vdwtype[jnrD+0];
 226             vdwjidx0E        = 2*vdwtype[jnrE+0];
 227             vdwjidx0F        = 2*vdwtype[jnrF+0];
 228             vdwjidx0G        = 2*vdwtype[jnrG+0];
 229             vdwjidx0H        = 2*vdwtype[jnrH+0];
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             r00              = _mm256_mul_ps(rsq00,rinv00);
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq00             = _mm256_mul_ps(iq0,jq0);
 239             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 240                                             vdwioffsetptr0+vdwjidx0B,
 241                                             vdwioffsetptr0+vdwjidx0C,
 242                                             vdwioffsetptr0+vdwjidx0D,
 243                                             vdwioffsetptr0+vdwjidx0E,
 244                                             vdwioffsetptr0+vdwjidx0F,
 245                                             vdwioffsetptr0+vdwjidx0G,
 246                                             vdwioffsetptr0+vdwjidx0H,
 247                                             &c6_00,&c12_00);
 248
 249             /* Calculate table index by multiplying r with table scale and truncate to integer */
 250             rt               = _mm256_mul_ps(r00,vftabscale);
 251             vfitab           = _mm256_cvttps_epi32(rt);
 252             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 253             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 254             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 255             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 256             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 257             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 258
 259             /* COULOMB ELECTROSTATICS */
 260             velec            = _mm256_mul_ps(qq00,rinv00);
 261             felec            = _mm256_mul_ps(velec,rinvsq00);
 262
 263             /* CUBIC SPLINE TABLE DISPERSION */
 264             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 265                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 266             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 267                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 268             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 269                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 270             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 271                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 272             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 273             Heps             = _mm256_mul_ps(vfeps,H);
 274             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 275             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 276             vvdw6            = _mm256_mul_ps(c6_00,VV);
 277             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 278             fvdw6            = _mm256_mul_ps(c6_00,FF);
 279
 280             /* CUBIC SPLINE TABLE REPULSION */
 281             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 282             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 283             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 284                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 285             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 286                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 287             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 288                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 289             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 291             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 292             Heps             = _mm256_mul_ps(vfeps,H);
 293             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 294             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 295             vvdw12           = _mm256_mul_ps(c12_00,VV);
 296             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 297             fvdw12           = _mm256_mul_ps(c12_00,FF);
 298             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 299             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velecsum         = _mm256_add_ps(velecsum,velec);
 303             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 304
 305             fscal            = _mm256_add_ps(felec,fvdw);
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm256_mul_ps(fscal,dx00);
 309             ty               = _mm256_mul_ps(fscal,dy00);
 310             tz               = _mm256_mul_ps(fscal,dz00);
 311
 312             /* Update vectorial force */
 313             fix0             = _mm256_add_ps(fix0,tx);
 314             fiy0             = _mm256_add_ps(fiy0,ty);
 315             fiz0             = _mm256_add_ps(fiz0,tz);
 316
 317             fjptrA             = f+j_coord_offsetA;
 318             fjptrB             = f+j_coord_offsetB;
 319             fjptrC             = f+j_coord_offsetC;
 320             fjptrD             = f+j_coord_offsetD;
 321             fjptrE             = f+j_coord_offsetE;
 322             fjptrF             = f+j_coord_offsetF;
 323             fjptrG             = f+j_coord_offsetG;
 324             fjptrH             = f+j_coord_offsetH;
 325             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 326
 327             /* Inner loop uses 62 flops */
 328         }
 329
 330         if(jidx<j_index_end)
 331         {
 332
 333             /* Get j neighbor index, and coordinate index */
 334             jnrlistA         = jjnr[jidx];
 335             jnrlistB         = jjnr[jidx+1];
 336             jnrlistC         = jjnr[jidx+2];
 337             jnrlistD         = jjnr[jidx+3];
 338             jnrlistE         = jjnr[jidx+4];
 339             jnrlistF         = jjnr[jidx+5];
 340             jnrlistG         = jjnr[jidx+6];
 341             jnrlistH         = jjnr[jidx+7];
 342             /* Sign of each element will be negative for non-real atoms.
 343              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 344              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 345              */
 346             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 347                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 348
 349             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 350             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 351             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 352             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 353             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 354             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 355             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 356             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 357             j_coord_offsetA  = DIM*jnrA;
 358             j_coord_offsetB  = DIM*jnrB;
 359             j_coord_offsetC  = DIM*jnrC;
 360             j_coord_offsetD  = DIM*jnrD;
 361             j_coord_offsetE  = DIM*jnrE;
 362             j_coord_offsetF  = DIM*jnrF;
 363             j_coord_offsetG  = DIM*jnrG;
 364             j_coord_offsetH  = DIM*jnrH;
 365
 366             /* load j atom coordinates */
 367             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 368                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 369                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 370                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 371                                                  &jx0,&jy0,&jz0);
 372
 373             /* Calculate displacement vector */
 374             dx00             = _mm256_sub_ps(ix0,jx0);
 375             dy00             = _mm256_sub_ps(iy0,jy0);
 376             dz00             = _mm256_sub_ps(iz0,jz0);
 377
 378             /* Calculate squared distance and things based on it */
 379             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 380
 381             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 382
 383             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 384
 385             /* Load parameters for j particles */
 386             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 387                                                                  charge+jnrC+0,charge+jnrD+0,
 388                                                                  charge+jnrE+0,charge+jnrF+0,
 389                                                                  charge+jnrG+0,charge+jnrH+0);
 390             vdwjidx0A        = 2*vdwtype[jnrA+0];
 391             vdwjidx0B        = 2*vdwtype[jnrB+0];
 392             vdwjidx0C        = 2*vdwtype[jnrC+0];
 393             vdwjidx0D        = 2*vdwtype[jnrD+0];
 394             vdwjidx0E        = 2*vdwtype[jnrE+0];
 395             vdwjidx0F        = 2*vdwtype[jnrF+0];
 396             vdwjidx0G        = 2*vdwtype[jnrG+0];
 397             vdwjidx0H        = 2*vdwtype[jnrH+0];
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             r00              = _mm256_mul_ps(rsq00,rinv00);
 404             r00              = _mm256_andnot_ps(dummy_mask,r00);
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq00             = _mm256_mul_ps(iq0,jq0);
 408             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 409                                             vdwioffsetptr0+vdwjidx0B,
 410                                             vdwioffsetptr0+vdwjidx0C,
 411                                             vdwioffsetptr0+vdwjidx0D,
 412                                             vdwioffsetptr0+vdwjidx0E,
 413                                             vdwioffsetptr0+vdwjidx0F,
 414                                             vdwioffsetptr0+vdwjidx0G,
 415                                             vdwioffsetptr0+vdwjidx0H,
 416                                             &c6_00,&c12_00);
 417
 418             /* Calculate table index by multiplying r with table scale and truncate to integer */
 419             rt               = _mm256_mul_ps(r00,vftabscale);
 420             vfitab           = _mm256_cvttps_epi32(rt);
 421             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 422             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 423             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 424             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 425             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 426             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 427
 428             /* COULOMB ELECTROSTATICS */
 429             velec            = _mm256_mul_ps(qq00,rinv00);
 430             felec            = _mm256_mul_ps(velec,rinvsq00);
 431
 432             /* CUBIC SPLINE TABLE DISPERSION */
 433             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 434                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 435             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 436                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 437             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 438                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 439             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 440                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 441             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 442             Heps             = _mm256_mul_ps(vfeps,H);
 443             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 444             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 445             vvdw6            = _mm256_mul_ps(c6_00,VV);
 446             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 447             fvdw6            = _mm256_mul_ps(c6_00,FF);
 448
 449             /* CUBIC SPLINE TABLE REPULSION */
 450             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 451             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 452             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 453                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 454             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 455                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 456             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 457                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 458             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 459                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 460             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 461             Heps             = _mm256_mul_ps(vfeps,H);
 462             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 463             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 464             vvdw12           = _mm256_mul_ps(c12_00,VV);
 465             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 466             fvdw12           = _mm256_mul_ps(c12_00,FF);
 467             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 468             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             velec            = _mm256_andnot_ps(dummy_mask,velec);
 472             velecsum         = _mm256_add_ps(velecsum,velec);
 473             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 474             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 475
 476             fscal            = _mm256_add_ps(felec,fvdw);
 477
 478             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm256_mul_ps(fscal,dx00);
 482             ty               = _mm256_mul_ps(fscal,dy00);
 483             tz               = _mm256_mul_ps(fscal,dz00);
 484
 485             /* Update vectorial force */
 486             fix0             = _mm256_add_ps(fix0,tx);
 487             fiy0             = _mm256_add_ps(fiy0,ty);
 488             fiz0             = _mm256_add_ps(fiz0,tz);
 489
 490             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 491             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 492             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 493             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 494             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 495             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 496             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 497             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 498             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 499
 500             /* Inner loop uses 63 flops */
 501         }
 502
 503         /* End of innermost loop */
 504
 505         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 506                                                  f+i_coord_offset,fshift+i_shift_offset);
 507
 508         ggid                        = gid[iidx];
 509         /* Update potential energies */
 510         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 511         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 512
 513         /* Increment number of inner iterations */
 514         inneriter                  += j_index_end - j_index_start;
 515
 516         /* Outer loop uses 9 flops */
 517     }
 518
 519     /* Increment number of outer iterations */
 520     outeriter        += nri;
 521
 522     /* Update outer/inner flops */
 523
 524     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*63);
 525 }
 526 /*
 527  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_256_single
 528  * Electrostatics interaction: Coulomb
 529  * VdW interaction:            CubicSplineTable
 530  * Geometry:                   Particle-Particle
 531  * Calculate force/pot:        Force
 532  */
 533 void
 534 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_256_single
 535                     (t_nblist                    * gmx_restrict       nlist,
 536                      rvec                        * gmx_restrict          xx,
 537                      rvec                        * gmx_restrict          ff,
 538                      t_forcerec                  * gmx_restrict          fr,
 539                      t_mdatoms                   * gmx_restrict     mdatoms,
 540                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 541                      t_nrnb                      * gmx_restrict        nrnb)
 542 {
 543     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 544      * just 0 for non-waters.
 545      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 546      * jnr indices corresponding to data put in the four positions in the SIMD register.
 547      */
 548     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 549     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 550     int              jnrA,jnrB,jnrC,jnrD;
 551     int              jnrE,jnrF,jnrG,jnrH;
 552     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 553     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 554     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 555     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 556     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 557     real             rcutoff_scalar;
 558     real             *shiftvec,*fshift,*x,*f;
 559     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 560     real             scratch[4*DIM];
 561     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 562     real *           vdwioffsetptr0;
 563     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 564     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 565     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 566     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 567     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 568     real             *charge;
 569     int              nvdwtype;
 570     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 571     int              *vdwtype;
 572     real             *vdwparam;
 573     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 574     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 575     __m256i          vfitab;
 576     __m128i          vfitab_lo,vfitab_hi;
 577     __m128i          ifour       = _mm_set1_epi32(4);
 578     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 579     real             *vftab;
 580     __m256           dummy_mask,cutoff_mask;
 581     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 582     __m256           one     = _mm256_set1_ps(1.0);
 583     __m256           two     = _mm256_set1_ps(2.0);
 584     x                = xx[0];
 585     f                = ff[0];
 586
 587     nri              = nlist->nri;
 588     iinr             = nlist->iinr;
 589     jindex           = nlist->jindex;
 590     jjnr             = nlist->jjnr;
 591     shiftidx         = nlist->shift;
 592     gid              = nlist->gid;
 593     shiftvec         = fr->shift_vec[0];
 594     fshift           = fr->fshift[0];
 595     facel            = _mm256_set1_ps(fr->epsfac);
 596     charge           = mdatoms->chargeA;
 597     nvdwtype         = fr->ntype;
 598     vdwparam         = fr->nbfp;
 599     vdwtype          = mdatoms->typeA;
 600
 601     vftab            = kernel_data->table_vdw->data;
 602     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 603
 604     /* Avoid stupid compiler warnings */
 605     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 606     j_coord_offsetA = 0;
 607     j_coord_offsetB = 0;
 608     j_coord_offsetC = 0;
 609     j_coord_offsetD = 0;
 610     j_coord_offsetE = 0;
 611     j_coord_offsetF = 0;
 612     j_coord_offsetG = 0;
 613     j_coord_offsetH = 0;
 614
 615     outeriter        = 0;
 616     inneriter        = 0;
 617
 618     for(iidx=0;iidx<4*DIM;iidx++)
 619     {
 620         scratch[iidx] = 0.0;
 621     }
 622
 623     /* Start outer loop over neighborlists */
 624     for(iidx=0; iidx<nri; iidx++)
 625     {
 626         /* Load shift vector for this list */
 627         i_shift_offset   = DIM*shiftidx[iidx];
 628
 629         /* Load limits for loop over neighbors */
 630         j_index_start    = jindex[iidx];
 631         j_index_end      = jindex[iidx+1];
 632
 633         /* Get outer coordinate index */
 634         inr              = iinr[iidx];
 635         i_coord_offset   = DIM*inr;
 636
 637         /* Load i particle coords and add shift vector */
 638         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 639
 640         fix0             = _mm256_setzero_ps();
 641         fiy0             = _mm256_setzero_ps();
 642         fiz0             = _mm256_setzero_ps();
 643
 644         /* Load parameters for i particles */
 645         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 646         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 647
 648         /* Start inner kernel loop */
 649         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 650         {
 651
 652             /* Get j neighbor index, and coordinate index */
 653             jnrA             = jjnr[jidx];
 654             jnrB             = jjnr[jidx+1];
 655             jnrC             = jjnr[jidx+2];
 656             jnrD             = jjnr[jidx+3];
 657             jnrE             = jjnr[jidx+4];
 658             jnrF             = jjnr[jidx+5];
 659             jnrG             = jjnr[jidx+6];
 660             jnrH             = jjnr[jidx+7];
 661             j_coord_offsetA  = DIM*jnrA;
 662             j_coord_offsetB  = DIM*jnrB;
 663             j_coord_offsetC  = DIM*jnrC;
 664             j_coord_offsetD  = DIM*jnrD;
 665             j_coord_offsetE  = DIM*jnrE;
 666             j_coord_offsetF  = DIM*jnrF;
 667             j_coord_offsetG  = DIM*jnrG;
 668             j_coord_offsetH  = DIM*jnrH;
 669
 670             /* load j atom coordinates */
 671             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 672                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 673                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 674                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 675                                                  &jx0,&jy0,&jz0);
 676
 677             /* Calculate displacement vector */
 678             dx00             = _mm256_sub_ps(ix0,jx0);
 679             dy00             = _mm256_sub_ps(iy0,jy0);
 680             dz00             = _mm256_sub_ps(iz0,jz0);
 681
 682             /* Calculate squared distance and things based on it */
 683             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 684
 685             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 686
 687             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 688
 689             /* Load parameters for j particles */
 690             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 691                                                                  charge+jnrC+0,charge+jnrD+0,
 692                                                                  charge+jnrE+0,charge+jnrF+0,
 693                                                                  charge+jnrG+0,charge+jnrH+0);
 694             vdwjidx0A        = 2*vdwtype[jnrA+0];
 695             vdwjidx0B        = 2*vdwtype[jnrB+0];
 696             vdwjidx0C        = 2*vdwtype[jnrC+0];
 697             vdwjidx0D        = 2*vdwtype[jnrD+0];
 698             vdwjidx0E        = 2*vdwtype[jnrE+0];
 699             vdwjidx0F        = 2*vdwtype[jnrF+0];
 700             vdwjidx0G        = 2*vdwtype[jnrG+0];
 701             vdwjidx0H        = 2*vdwtype[jnrH+0];
 702
 703             /**************************
 704              * CALCULATE INTERACTIONS *
 705              **************************/
 706
 707             r00              = _mm256_mul_ps(rsq00,rinv00);
 708
 709             /* Compute parameters for interactions between i and j atoms */
 710             qq00             = _mm256_mul_ps(iq0,jq0);
 711             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 712                                             vdwioffsetptr0+vdwjidx0B,
 713                                             vdwioffsetptr0+vdwjidx0C,
 714                                             vdwioffsetptr0+vdwjidx0D,
 715                                             vdwioffsetptr0+vdwjidx0E,
 716                                             vdwioffsetptr0+vdwjidx0F,
 717                                             vdwioffsetptr0+vdwjidx0G,
 718                                             vdwioffsetptr0+vdwjidx0H,
 719                                             &c6_00,&c12_00);
 720
 721             /* Calculate table index by multiplying r with table scale and truncate to integer */
 722             rt               = _mm256_mul_ps(r00,vftabscale);
 723             vfitab           = _mm256_cvttps_epi32(rt);
 724             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 725             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 726             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 727             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 728             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 729             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 730
 731             /* COULOMB ELECTROSTATICS */
 732             velec            = _mm256_mul_ps(qq00,rinv00);
 733             felec            = _mm256_mul_ps(velec,rinvsq00);
 734
 735             /* CUBIC SPLINE TABLE DISPERSION */
 736             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 737                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 738             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 739                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 740             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 741                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 742             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 743                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 744             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 745             Heps             = _mm256_mul_ps(vfeps,H);
 746             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 747             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 748             fvdw6            = _mm256_mul_ps(c6_00,FF);
 749
 750             /* CUBIC SPLINE TABLE REPULSION */
 751             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 752             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 753             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 754                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 755             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 756                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 757             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 758                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 759             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 760                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 761             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 762             Heps             = _mm256_mul_ps(vfeps,H);
 763             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 764             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 765             fvdw12           = _mm256_mul_ps(c12_00,FF);
 766             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 767
 768             fscal            = _mm256_add_ps(felec,fvdw);
 769
 770             /* Calculate temporary vectorial force */
 771             tx               = _mm256_mul_ps(fscal,dx00);
 772             ty               = _mm256_mul_ps(fscal,dy00);
 773             tz               = _mm256_mul_ps(fscal,dz00);
 774
 775             /* Update vectorial force */
 776             fix0             = _mm256_add_ps(fix0,tx);
 777             fiy0             = _mm256_add_ps(fiy0,ty);
 778             fiz0             = _mm256_add_ps(fiz0,tz);
 779
 780             fjptrA             = f+j_coord_offsetA;
 781             fjptrB             = f+j_coord_offsetB;
 782             fjptrC             = f+j_coord_offsetC;
 783             fjptrD             = f+j_coord_offsetD;
 784             fjptrE             = f+j_coord_offsetE;
 785             fjptrF             = f+j_coord_offsetF;
 786             fjptrG             = f+j_coord_offsetG;
 787             fjptrH             = f+j_coord_offsetH;
 788             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 789
 790             /* Inner loop uses 53 flops */
 791         }
 792
 793         if(jidx<j_index_end)
 794         {
 795
 796             /* Get j neighbor index, and coordinate index */
 797             jnrlistA         = jjnr[jidx];
 798             jnrlistB         = jjnr[jidx+1];
 799             jnrlistC         = jjnr[jidx+2];
 800             jnrlistD         = jjnr[jidx+3];
 801             jnrlistE         = jjnr[jidx+4];
 802             jnrlistF         = jjnr[jidx+5];
 803             jnrlistG         = jjnr[jidx+6];
 804             jnrlistH         = jjnr[jidx+7];
 805             /* Sign of each element will be negative for non-real atoms.
 806              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 807              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 808              */
 809             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 810                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 811
 812             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 813             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 814             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 815             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 816             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 817             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 818             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 819             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 820             j_coord_offsetA  = DIM*jnrA;
 821             j_coord_offsetB  = DIM*jnrB;
 822             j_coord_offsetC  = DIM*jnrC;
 823             j_coord_offsetD  = DIM*jnrD;
 824             j_coord_offsetE  = DIM*jnrE;
 825             j_coord_offsetF  = DIM*jnrF;
 826             j_coord_offsetG  = DIM*jnrG;
 827             j_coord_offsetH  = DIM*jnrH;
 828
 829             /* load j atom coordinates */
 830             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 831                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 832                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 833                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 834                                                  &jx0,&jy0,&jz0);
 835
 836             /* Calculate displacement vector */
 837             dx00             = _mm256_sub_ps(ix0,jx0);
 838             dy00             = _mm256_sub_ps(iy0,jy0);
 839             dz00             = _mm256_sub_ps(iz0,jz0);
 840
 841             /* Calculate squared distance and things based on it */
 842             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 843
 844             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 845
 846             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 847
 848             /* Load parameters for j particles */
 849             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 850                                                                  charge+jnrC+0,charge+jnrD+0,
 851                                                                  charge+jnrE+0,charge+jnrF+0,
 852                                                                  charge+jnrG+0,charge+jnrH+0);
 853             vdwjidx0A        = 2*vdwtype[jnrA+0];
 854             vdwjidx0B        = 2*vdwtype[jnrB+0];
 855             vdwjidx0C        = 2*vdwtype[jnrC+0];
 856             vdwjidx0D        = 2*vdwtype[jnrD+0];
 857             vdwjidx0E        = 2*vdwtype[jnrE+0];
 858             vdwjidx0F        = 2*vdwtype[jnrF+0];
 859             vdwjidx0G        = 2*vdwtype[jnrG+0];
 860             vdwjidx0H        = 2*vdwtype[jnrH+0];
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             r00              = _mm256_mul_ps(rsq00,rinv00);
 867             r00              = _mm256_andnot_ps(dummy_mask,r00);
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq00             = _mm256_mul_ps(iq0,jq0);
 871             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 872                                             vdwioffsetptr0+vdwjidx0B,
 873                                             vdwioffsetptr0+vdwjidx0C,
 874                                             vdwioffsetptr0+vdwjidx0D,
 875                                             vdwioffsetptr0+vdwjidx0E,
 876                                             vdwioffsetptr0+vdwjidx0F,
 877                                             vdwioffsetptr0+vdwjidx0G,
 878                                             vdwioffsetptr0+vdwjidx0H,
 879                                             &c6_00,&c12_00);
 880
 881             /* Calculate table index by multiplying r with table scale and truncate to integer */
 882             rt               = _mm256_mul_ps(r00,vftabscale);
 883             vfitab           = _mm256_cvttps_epi32(rt);
 884             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 885             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 886             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 887             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 888             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 889             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 890
 891             /* COULOMB ELECTROSTATICS */
 892             velec            = _mm256_mul_ps(qq00,rinv00);
 893             felec            = _mm256_mul_ps(velec,rinvsq00);
 894
 895             /* CUBIC SPLINE TABLE DISPERSION */
 896             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 897                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 898             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 899                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 900             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 901                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 902             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 903                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 904             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 905             Heps             = _mm256_mul_ps(vfeps,H);
 906             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 907             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 908             fvdw6            = _mm256_mul_ps(c6_00,FF);
 909
 910             /* CUBIC SPLINE TABLE REPULSION */
 911             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 912             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 913             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 914                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 915             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 916                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 917             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 918                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 919             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 920                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 921             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 922             Heps             = _mm256_mul_ps(vfeps,H);
 923             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 924             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 925             fvdw12           = _mm256_mul_ps(c12_00,FF);
 926             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 927
 928             fscal            = _mm256_add_ps(felec,fvdw);
 929
 930             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 931
 932             /* Calculate temporary vectorial force */
 933             tx               = _mm256_mul_ps(fscal,dx00);
 934             ty               = _mm256_mul_ps(fscal,dy00);
 935             tz               = _mm256_mul_ps(fscal,dz00);
 936
 937             /* Update vectorial force */
 938             fix0             = _mm256_add_ps(fix0,tx);
 939             fiy0             = _mm256_add_ps(fiy0,ty);
 940             fiz0             = _mm256_add_ps(fiz0,tz);
 941
 942             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 943             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 944             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 945             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 946             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 947             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 948             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 949             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 950             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 951
 952             /* Inner loop uses 54 flops */
 953         }
 954
 955         /* End of innermost loop */
 956
 957         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 958                                                  f+i_coord_offset,fshift+i_shift_offset);
 959
 960         /* Increment number of inner iterations */
 961         inneriter                  += j_index_end - j_index_start;
 962
 963         /* Outer loop uses 7 flops */
 964     }
 965
 966     /* Increment number of outer iterations */
 967     outeriter        += nri;
 968
 969     /* Update outer/inner flops */
 970
 971     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);
 972 }