src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_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_elec->data;
 126     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->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,2);
 257             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 258
 259             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 260             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 261                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 262             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 263                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 264             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 265                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 266             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 267                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 268             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 269             Heps             = _mm256_mul_ps(vfeps,H);
 270             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 271             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 272             velec            = _mm256_mul_ps(qq00,VV);
 273             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 274             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 275
 276             /* LENNARD-JONES DISPERSION/REPULSION */
 277
 278             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 279             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 280             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 281             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 282             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velecsum         = _mm256_add_ps(velecsum,velec);
 286             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 287
 288             fscal            = _mm256_add_ps(felec,fvdw);
 289
 290             /* Calculate temporary vectorial force */
 291             tx               = _mm256_mul_ps(fscal,dx00);
 292             ty               = _mm256_mul_ps(fscal,dy00);
 293             tz               = _mm256_mul_ps(fscal,dz00);
 294
 295             /* Update vectorial force */
 296             fix0             = _mm256_add_ps(fix0,tx);
 297             fiy0             = _mm256_add_ps(fiy0,ty);
 298             fiz0             = _mm256_add_ps(fiz0,tz);
 299
 300             fjptrA             = f+j_coord_offsetA;
 301             fjptrB             = f+j_coord_offsetB;
 302             fjptrC             = f+j_coord_offsetC;
 303             fjptrD             = f+j_coord_offsetD;
 304             fjptrE             = f+j_coord_offsetE;
 305             fjptrF             = f+j_coord_offsetF;
 306             fjptrG             = f+j_coord_offsetG;
 307             fjptrH             = f+j_coord_offsetH;
 308             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 309
 310             /* Inner loop uses 56 flops */
 311         }
 312
 313         if(jidx<j_index_end)
 314         {
 315
 316             /* Get j neighbor index, and coordinate index */
 317             jnrlistA         = jjnr[jidx];
 318             jnrlistB         = jjnr[jidx+1];
 319             jnrlistC         = jjnr[jidx+2];
 320             jnrlistD         = jjnr[jidx+3];
 321             jnrlistE         = jjnr[jidx+4];
 322             jnrlistF         = jjnr[jidx+5];
 323             jnrlistG         = jjnr[jidx+6];
 324             jnrlistH         = jjnr[jidx+7];
 325             /* Sign of each element will be negative for non-real atoms.
 326              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 327              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 328              */
 329             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 330                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 331
 332             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 333             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 334             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 335             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 336             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 337             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 338             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 339             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 340             j_coord_offsetA  = DIM*jnrA;
 341             j_coord_offsetB  = DIM*jnrB;
 342             j_coord_offsetC  = DIM*jnrC;
 343             j_coord_offsetD  = DIM*jnrD;
 344             j_coord_offsetE  = DIM*jnrE;
 345             j_coord_offsetF  = DIM*jnrF;
 346             j_coord_offsetG  = DIM*jnrG;
 347             j_coord_offsetH  = DIM*jnrH;
 348
 349             /* load j atom coordinates */
 350             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 351                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 352                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 353                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 354                                                  &jx0,&jy0,&jz0);
 355
 356             /* Calculate displacement vector */
 357             dx00             = _mm256_sub_ps(ix0,jx0);
 358             dy00             = _mm256_sub_ps(iy0,jy0);
 359             dz00             = _mm256_sub_ps(iz0,jz0);
 360
 361             /* Calculate squared distance and things based on it */
 362             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 363
 364             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 365
 366             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 367
 368             /* Load parameters for j particles */
 369             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 370                                                                  charge+jnrC+0,charge+jnrD+0,
 371                                                                  charge+jnrE+0,charge+jnrF+0,
 372                                                                  charge+jnrG+0,charge+jnrH+0);
 373             vdwjidx0A        = 2*vdwtype[jnrA+0];
 374             vdwjidx0B        = 2*vdwtype[jnrB+0];
 375             vdwjidx0C        = 2*vdwtype[jnrC+0];
 376             vdwjidx0D        = 2*vdwtype[jnrD+0];
 377             vdwjidx0E        = 2*vdwtype[jnrE+0];
 378             vdwjidx0F        = 2*vdwtype[jnrF+0];
 379             vdwjidx0G        = 2*vdwtype[jnrG+0];
 380             vdwjidx0H        = 2*vdwtype[jnrH+0];
 381
 382             /**************************
 383              * CALCULATE INTERACTIONS *
 384              **************************/
 385
 386             r00              = _mm256_mul_ps(rsq00,rinv00);
 387             r00              = _mm256_andnot_ps(dummy_mask,r00);
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq00             = _mm256_mul_ps(iq0,jq0);
 391             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 392                                             vdwioffsetptr0+vdwjidx0B,
 393                                             vdwioffsetptr0+vdwjidx0C,
 394                                             vdwioffsetptr0+vdwjidx0D,
 395                                             vdwioffsetptr0+vdwjidx0E,
 396                                             vdwioffsetptr0+vdwjidx0F,
 397                                             vdwioffsetptr0+vdwjidx0G,
 398                                             vdwioffsetptr0+vdwjidx0H,
 399                                             &c6_00,&c12_00);
 400
 401             /* Calculate table index by multiplying r with table scale and truncate to integer */
 402             rt               = _mm256_mul_ps(r00,vftabscale);
 403             vfitab           = _mm256_cvttps_epi32(rt);
 404             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 405             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 406             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 407             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 408             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 409             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 410
 411             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 412             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 413                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 414             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 415                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 416             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 417                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 418             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 419                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 420             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 421             Heps             = _mm256_mul_ps(vfeps,H);
 422             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 423             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 424             velec            = _mm256_mul_ps(qq00,VV);
 425             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 426             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 427
 428             /* LENNARD-JONES DISPERSION/REPULSION */
 429
 430             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 431             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 432             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 433             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 434             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velec            = _mm256_andnot_ps(dummy_mask,velec);
 438             velecsum         = _mm256_add_ps(velecsum,velec);
 439             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 440             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 441
 442             fscal            = _mm256_add_ps(felec,fvdw);
 443
 444             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm256_mul_ps(fscal,dx00);
 448             ty               = _mm256_mul_ps(fscal,dy00);
 449             tz               = _mm256_mul_ps(fscal,dz00);
 450
 451             /* Update vectorial force */
 452             fix0             = _mm256_add_ps(fix0,tx);
 453             fiy0             = _mm256_add_ps(fiy0,ty);
 454             fiz0             = _mm256_add_ps(fiz0,tz);
 455
 456             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 457             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 458             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 459             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 460             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 461             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 462             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 463             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 464             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 465
 466             /* Inner loop uses 57 flops */
 467         }
 468
 469         /* End of innermost loop */
 470
 471         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 472                                                  f+i_coord_offset,fshift+i_shift_offset);
 473
 474         ggid                        = gid[iidx];
 475         /* Update potential energies */
 476         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 477         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 478
 479         /* Increment number of inner iterations */
 480         inneriter                  += j_index_end - j_index_start;
 481
 482         /* Outer loop uses 9 flops */
 483     }
 484
 485     /* Increment number of outer iterations */
 486     outeriter        += nri;
 487
 488     /* Update outer/inner flops */
 489
 490     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 491 }
 492 /*
 493  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_256_single
 494  * Electrostatics interaction: CubicSplineTable
 495  * VdW interaction:            LennardJones
 496  * Geometry:                   Particle-Particle
 497  * Calculate force/pot:        Force
 498  */
 499 void
 500 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_256_single
 501                     (t_nblist                    * gmx_restrict       nlist,
 502                      rvec                        * gmx_restrict          xx,
 503                      rvec                        * gmx_restrict          ff,
 504                      t_forcerec                  * gmx_restrict          fr,
 505                      t_mdatoms                   * gmx_restrict     mdatoms,
 506                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 507                      t_nrnb                      * gmx_restrict        nrnb)
 508 {
 509     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 510      * just 0 for non-waters.
 511      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 512      * jnr indices corresponding to data put in the four positions in the SIMD register.
 513      */
 514     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 515     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 516     int              jnrA,jnrB,jnrC,jnrD;
 517     int              jnrE,jnrF,jnrG,jnrH;
 518     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 519     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 520     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 521     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 522     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 523     real             rcutoff_scalar;
 524     real             *shiftvec,*fshift,*x,*f;
 525     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 526     real             scratch[4*DIM];
 527     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 528     real *           vdwioffsetptr0;
 529     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 530     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 531     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 532     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 533     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 534     real             *charge;
 535     int              nvdwtype;
 536     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 537     int              *vdwtype;
 538     real             *vdwparam;
 539     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 540     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 541     __m256i          vfitab;
 542     __m128i          vfitab_lo,vfitab_hi;
 543     __m128i          ifour       = _mm_set1_epi32(4);
 544     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 545     real             *vftab;
 546     __m256           dummy_mask,cutoff_mask;
 547     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 548     __m256           one     = _mm256_set1_ps(1.0);
 549     __m256           two     = _mm256_set1_ps(2.0);
 550     x                = xx[0];
 551     f                = ff[0];
 552
 553     nri              = nlist->nri;
 554     iinr             = nlist->iinr;
 555     jindex           = nlist->jindex;
 556     jjnr             = nlist->jjnr;
 557     shiftidx         = nlist->shift;
 558     gid              = nlist->gid;
 559     shiftvec         = fr->shift_vec[0];
 560     fshift           = fr->fshift[0];
 561     facel            = _mm256_set1_ps(fr->epsfac);
 562     charge           = mdatoms->chargeA;
 563     nvdwtype         = fr->ntype;
 564     vdwparam         = fr->nbfp;
 565     vdwtype          = mdatoms->typeA;
 566
 567     vftab            = kernel_data->table_elec->data;
 568     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 569
 570     /* Avoid stupid compiler warnings */
 571     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 572     j_coord_offsetA = 0;
 573     j_coord_offsetB = 0;
 574     j_coord_offsetC = 0;
 575     j_coord_offsetD = 0;
 576     j_coord_offsetE = 0;
 577     j_coord_offsetF = 0;
 578     j_coord_offsetG = 0;
 579     j_coord_offsetH = 0;
 580
 581     outeriter        = 0;
 582     inneriter        = 0;
 583
 584     for(iidx=0;iidx<4*DIM;iidx++)
 585     {
 586         scratch[iidx] = 0.0;
 587     }
 588
 589     /* Start outer loop over neighborlists */
 590     for(iidx=0; iidx<nri; iidx++)
 591     {
 592         /* Load shift vector for this list */
 593         i_shift_offset   = DIM*shiftidx[iidx];
 594
 595         /* Load limits for loop over neighbors */
 596         j_index_start    = jindex[iidx];
 597         j_index_end      = jindex[iidx+1];
 598
 599         /* Get outer coordinate index */
 600         inr              = iinr[iidx];
 601         i_coord_offset   = DIM*inr;
 602
 603         /* Load i particle coords and add shift vector */
 604         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 605
 606         fix0             = _mm256_setzero_ps();
 607         fiy0             = _mm256_setzero_ps();
 608         fiz0             = _mm256_setzero_ps();
 609
 610         /* Load parameters for i particles */
 611         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 612         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 613
 614         /* Start inner kernel loop */
 615         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 616         {
 617
 618             /* Get j neighbor index, and coordinate index */
 619             jnrA             = jjnr[jidx];
 620             jnrB             = jjnr[jidx+1];
 621             jnrC             = jjnr[jidx+2];
 622             jnrD             = jjnr[jidx+3];
 623             jnrE             = jjnr[jidx+4];
 624             jnrF             = jjnr[jidx+5];
 625             jnrG             = jjnr[jidx+6];
 626             jnrH             = jjnr[jidx+7];
 627             j_coord_offsetA  = DIM*jnrA;
 628             j_coord_offsetB  = DIM*jnrB;
 629             j_coord_offsetC  = DIM*jnrC;
 630             j_coord_offsetD  = DIM*jnrD;
 631             j_coord_offsetE  = DIM*jnrE;
 632             j_coord_offsetF  = DIM*jnrF;
 633             j_coord_offsetG  = DIM*jnrG;
 634             j_coord_offsetH  = DIM*jnrH;
 635
 636             /* load j atom coordinates */
 637             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 638                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 639                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 640                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 641                                                  &jx0,&jy0,&jz0);
 642
 643             /* Calculate displacement vector */
 644             dx00             = _mm256_sub_ps(ix0,jx0);
 645             dy00             = _mm256_sub_ps(iy0,jy0);
 646             dz00             = _mm256_sub_ps(iz0,jz0);
 647
 648             /* Calculate squared distance and things based on it */
 649             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 650
 651             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 652
 653             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 654
 655             /* Load parameters for j particles */
 656             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 657                                                                  charge+jnrC+0,charge+jnrD+0,
 658                                                                  charge+jnrE+0,charge+jnrF+0,
 659                                                                  charge+jnrG+0,charge+jnrH+0);
 660             vdwjidx0A        = 2*vdwtype[jnrA+0];
 661             vdwjidx0B        = 2*vdwtype[jnrB+0];
 662             vdwjidx0C        = 2*vdwtype[jnrC+0];
 663             vdwjidx0D        = 2*vdwtype[jnrD+0];
 664             vdwjidx0E        = 2*vdwtype[jnrE+0];
 665             vdwjidx0F        = 2*vdwtype[jnrF+0];
 666             vdwjidx0G        = 2*vdwtype[jnrG+0];
 667             vdwjidx0H        = 2*vdwtype[jnrH+0];
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             r00              = _mm256_mul_ps(rsq00,rinv00);
 674
 675             /* Compute parameters for interactions between i and j atoms */
 676             qq00             = _mm256_mul_ps(iq0,jq0);
 677             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 678                                             vdwioffsetptr0+vdwjidx0B,
 679                                             vdwioffsetptr0+vdwjidx0C,
 680                                             vdwioffsetptr0+vdwjidx0D,
 681                                             vdwioffsetptr0+vdwjidx0E,
 682                                             vdwioffsetptr0+vdwjidx0F,
 683                                             vdwioffsetptr0+vdwjidx0G,
 684                                             vdwioffsetptr0+vdwjidx0H,
 685                                             &c6_00,&c12_00);
 686
 687             /* Calculate table index by multiplying r with table scale and truncate to integer */
 688             rt               = _mm256_mul_ps(r00,vftabscale);
 689             vfitab           = _mm256_cvttps_epi32(rt);
 690             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 691             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 692             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 693             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 694             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 695             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 696
 697             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 698             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 699                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 700             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 701                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 702             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 703                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 704             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 705                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 706             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 707             Heps             = _mm256_mul_ps(vfeps,H);
 708             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 709             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 710             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 711
 712             /* LENNARD-JONES DISPERSION/REPULSION */
 713
 714             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 715             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 716
 717             fscal            = _mm256_add_ps(felec,fvdw);
 718
 719             /* Calculate temporary vectorial force */
 720             tx               = _mm256_mul_ps(fscal,dx00);
 721             ty               = _mm256_mul_ps(fscal,dy00);
 722             tz               = _mm256_mul_ps(fscal,dz00);
 723
 724             /* Update vectorial force */
 725             fix0             = _mm256_add_ps(fix0,tx);
 726             fiy0             = _mm256_add_ps(fiy0,ty);
 727             fiz0             = _mm256_add_ps(fiz0,tz);
 728
 729             fjptrA             = f+j_coord_offsetA;
 730             fjptrB             = f+j_coord_offsetB;
 731             fjptrC             = f+j_coord_offsetC;
 732             fjptrD             = f+j_coord_offsetD;
 733             fjptrE             = f+j_coord_offsetE;
 734             fjptrF             = f+j_coord_offsetF;
 735             fjptrG             = f+j_coord_offsetG;
 736             fjptrH             = f+j_coord_offsetH;
 737             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 738
 739             /* Inner loop uses 47 flops */
 740         }
 741
 742         if(jidx<j_index_end)
 743         {
 744
 745             /* Get j neighbor index, and coordinate index */
 746             jnrlistA         = jjnr[jidx];
 747             jnrlistB         = jjnr[jidx+1];
 748             jnrlistC         = jjnr[jidx+2];
 749             jnrlistD         = jjnr[jidx+3];
 750             jnrlistE         = jjnr[jidx+4];
 751             jnrlistF         = jjnr[jidx+5];
 752             jnrlistG         = jjnr[jidx+6];
 753             jnrlistH         = jjnr[jidx+7];
 754             /* Sign of each element will be negative for non-real atoms.
 755              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 756              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 757              */
 758             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 759                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 760
 761             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 762             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 763             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 764             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 765             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 766             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 767             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 768             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 769             j_coord_offsetA  = DIM*jnrA;
 770             j_coord_offsetB  = DIM*jnrB;
 771             j_coord_offsetC  = DIM*jnrC;
 772             j_coord_offsetD  = DIM*jnrD;
 773             j_coord_offsetE  = DIM*jnrE;
 774             j_coord_offsetF  = DIM*jnrF;
 775             j_coord_offsetG  = DIM*jnrG;
 776             j_coord_offsetH  = DIM*jnrH;
 777
 778             /* load j atom coordinates */
 779             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 780                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 781                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 782                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 783                                                  &jx0,&jy0,&jz0);
 784
 785             /* Calculate displacement vector */
 786             dx00             = _mm256_sub_ps(ix0,jx0);
 787             dy00             = _mm256_sub_ps(iy0,jy0);
 788             dz00             = _mm256_sub_ps(iz0,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 792
 793             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 794
 795             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 796
 797             /* Load parameters for j particles */
 798             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 799                                                                  charge+jnrC+0,charge+jnrD+0,
 800                                                                  charge+jnrE+0,charge+jnrF+0,
 801                                                                  charge+jnrG+0,charge+jnrH+0);
 802             vdwjidx0A        = 2*vdwtype[jnrA+0];
 803             vdwjidx0B        = 2*vdwtype[jnrB+0];
 804             vdwjidx0C        = 2*vdwtype[jnrC+0];
 805             vdwjidx0D        = 2*vdwtype[jnrD+0];
 806             vdwjidx0E        = 2*vdwtype[jnrE+0];
 807             vdwjidx0F        = 2*vdwtype[jnrF+0];
 808             vdwjidx0G        = 2*vdwtype[jnrG+0];
 809             vdwjidx0H        = 2*vdwtype[jnrH+0];
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             r00              = _mm256_mul_ps(rsq00,rinv00);
 816             r00              = _mm256_andnot_ps(dummy_mask,r00);
 817
 818             /* Compute parameters for interactions between i and j atoms */
 819             qq00             = _mm256_mul_ps(iq0,jq0);
 820             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 821                                             vdwioffsetptr0+vdwjidx0B,
 822                                             vdwioffsetptr0+vdwjidx0C,
 823                                             vdwioffsetptr0+vdwjidx0D,
 824                                             vdwioffsetptr0+vdwjidx0E,
 825                                             vdwioffsetptr0+vdwjidx0F,
 826                                             vdwioffsetptr0+vdwjidx0G,
 827                                             vdwioffsetptr0+vdwjidx0H,
 828                                             &c6_00,&c12_00);
 829
 830             /* Calculate table index by multiplying r with table scale and truncate to integer */
 831             rt               = _mm256_mul_ps(r00,vftabscale);
 832             vfitab           = _mm256_cvttps_epi32(rt);
 833             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 834             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 835             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 836             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 837             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 838             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 839
 840             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 841             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 842                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 843             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 844                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 845             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 846                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 847             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 848                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 849             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 850             Heps             = _mm256_mul_ps(vfeps,H);
 851             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 852             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 853             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 854
 855             /* LENNARD-JONES DISPERSION/REPULSION */
 856
 857             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 858             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 859
 860             fscal            = _mm256_add_ps(felec,fvdw);
 861
 862             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 863
 864             /* Calculate temporary vectorial force */
 865             tx               = _mm256_mul_ps(fscal,dx00);
 866             ty               = _mm256_mul_ps(fscal,dy00);
 867             tz               = _mm256_mul_ps(fscal,dz00);
 868
 869             /* Update vectorial force */
 870             fix0             = _mm256_add_ps(fix0,tx);
 871             fiy0             = _mm256_add_ps(fiy0,ty);
 872             fiz0             = _mm256_add_ps(fiz0,tz);
 873
 874             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 875             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 876             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 877             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 878             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 879             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 880             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 881             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 882             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 883
 884             /* Inner loop uses 48 flops */
 885         }
 886
 887         /* End of innermost loop */
 888
 889         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 890                                                  f+i_coord_offset,fshift+i_shift_offset);
 891
 892         /* Increment number of inner iterations */
 893         inneriter                  += j_index_end - j_index_start;
 894
 895         /* Outer loop uses 7 flops */
 896     }
 897
 898     /* Increment number of outer iterations */
 899     outeriter        += nri;
 900
 901     /* Update outer/inner flops */
 902
 903     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*48);
 904 }