src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_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_ElecRFCut_VdwCSTab_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_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     krf              = _mm256_set1_ps(fr->ic->k_rf);
 122     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 123     crf              = _mm256_set1_ps(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     vftab            = kernel_data->table_vdw->data;
 129     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 130
 131     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 132     rcutoff_scalar   = fr->rcoulomb;
 133     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 134     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142     j_coord_offsetE = 0;
 143     j_coord_offsetF = 0;
 144     j_coord_offsetG = 0;
 145     j_coord_offsetH = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 171
 172         fix0             = _mm256_setzero_ps();
 173         fiy0             = _mm256_setzero_ps();
 174         fiz0             = _mm256_setzero_ps();
 175
 176         /* Load parameters for i particles */
 177         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 178         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_ps();
 182         vvdwsum          = _mm256_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             jnrE             = jjnr[jidx+4];
 194             jnrF             = jjnr[jidx+5];
 195             jnrG             = jjnr[jidx+6];
 196             jnrH             = jjnr[jidx+7];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201             j_coord_offsetE  = DIM*jnrE;
 202             j_coord_offsetF  = DIM*jnrF;
 203             j_coord_offsetG  = DIM*jnrG;
 204             j_coord_offsetH  = DIM*jnrH;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 210                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_ps(ix0,jx0);
 215             dy00             = _mm256_sub_ps(iy0,jy0);
 216             dz00             = _mm256_sub_ps(iz0,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 220
 221             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 222
 223             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                                  charge+jnrC+0,charge+jnrD+0,
 228                                                                  charge+jnrE+0,charge+jnrF+0,
 229                                                                  charge+jnrG+0,charge+jnrH+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234             vdwjidx0E        = 2*vdwtype[jnrE+0];
 235             vdwjidx0F        = 2*vdwtype[jnrF+0];
 236             vdwjidx0G        = 2*vdwtype[jnrG+0];
 237             vdwjidx0H        = 2*vdwtype[jnrH+0];
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 244             {
 245
 246             r00              = _mm256_mul_ps(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             qq00             = _mm256_mul_ps(iq0,jq0);
 250             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 251                                             vdwioffsetptr0+vdwjidx0B,
 252                                             vdwioffsetptr0+vdwjidx0C,
 253                                             vdwioffsetptr0+vdwjidx0D,
 254                                             vdwioffsetptr0+vdwjidx0E,
 255                                             vdwioffsetptr0+vdwjidx0F,
 256                                             vdwioffsetptr0+vdwjidx0G,
 257                                             vdwioffsetptr0+vdwjidx0H,
 258                                             &c6_00,&c12_00);
 259
 260             /* Calculate table index by multiplying r with table scale and truncate to integer */
 261             rt               = _mm256_mul_ps(r00,vftabscale);
 262             vfitab           = _mm256_cvttps_epi32(rt);
 263             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 264             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 265             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 266             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 267             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 268             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 269
 270             /* REACTION-FIELD ELECTROSTATICS */
 271             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 272             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 276                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 277             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 278                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 279             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 280                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 281             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 282                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 283             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 284             Heps             = _mm256_mul_ps(vfeps,H);
 285             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 286             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 287             vvdw6            = _mm256_mul_ps(c6_00,VV);
 288             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 289             fvdw6            = _mm256_mul_ps(c6_00,FF);
 290
 291             /* CUBIC SPLINE TABLE REPULSION */
 292             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 293             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 294             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 296             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 298             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 299                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 300             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 301                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 302             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 303             Heps             = _mm256_mul_ps(vfeps,H);
 304             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 305             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 306             vvdw12           = _mm256_mul_ps(c12_00,VV);
 307             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 308             fvdw12           = _mm256_mul_ps(c12_00,FF);
 309             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 310             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 311
 312             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm256_and_ps(velec,cutoff_mask);
 316             velecsum         = _mm256_add_ps(velecsum,velec);
 317             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 318             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 319
 320             fscal            = _mm256_add_ps(felec,fvdw);
 321
 322             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm256_mul_ps(fscal,dx00);
 326             ty               = _mm256_mul_ps(fscal,dy00);
 327             tz               = _mm256_mul_ps(fscal,dz00);
 328
 329             /* Update vectorial force */
 330             fix0             = _mm256_add_ps(fix0,tx);
 331             fiy0             = _mm256_add_ps(fiy0,ty);
 332             fiz0             = _mm256_add_ps(fiz0,tz);
 333
 334             fjptrA             = f+j_coord_offsetA;
 335             fjptrB             = f+j_coord_offsetB;
 336             fjptrC             = f+j_coord_offsetC;
 337             fjptrD             = f+j_coord_offsetD;
 338             fjptrE             = f+j_coord_offsetE;
 339             fjptrF             = f+j_coord_offsetF;
 340             fjptrG             = f+j_coord_offsetG;
 341             fjptrH             = f+j_coord_offsetH;
 342             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 343
 344             }
 345
 346             /* Inner loop uses 72 flops */
 347         }
 348
 349         if(jidx<j_index_end)
 350         {
 351
 352             /* Get j neighbor index, and coordinate index */
 353             jnrlistA         = jjnr[jidx];
 354             jnrlistB         = jjnr[jidx+1];
 355             jnrlistC         = jjnr[jidx+2];
 356             jnrlistD         = jjnr[jidx+3];
 357             jnrlistE         = jjnr[jidx+4];
 358             jnrlistF         = jjnr[jidx+5];
 359             jnrlistG         = jjnr[jidx+6];
 360             jnrlistH         = jjnr[jidx+7];
 361             /* Sign of each element will be negative for non-real atoms.
 362              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 363              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 364              */
 365             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 366                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 367
 368             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 369             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 370             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 371             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 372             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 373             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 374             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 375             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 376             j_coord_offsetA  = DIM*jnrA;
 377             j_coord_offsetB  = DIM*jnrB;
 378             j_coord_offsetC  = DIM*jnrC;
 379             j_coord_offsetD  = DIM*jnrD;
 380             j_coord_offsetE  = DIM*jnrE;
 381             j_coord_offsetF  = DIM*jnrF;
 382             j_coord_offsetG  = DIM*jnrG;
 383             j_coord_offsetH  = DIM*jnrH;
 384
 385             /* load j atom coordinates */
 386             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 387                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 388                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 389                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 390                                                  &jx0,&jy0,&jz0);
 391
 392             /* Calculate displacement vector */
 393             dx00             = _mm256_sub_ps(ix0,jx0);
 394             dy00             = _mm256_sub_ps(iy0,jy0);
 395             dz00             = _mm256_sub_ps(iz0,jz0);
 396
 397             /* Calculate squared distance and things based on it */
 398             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 399
 400             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 401
 402             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 403
 404             /* Load parameters for j particles */
 405             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 406                                                                  charge+jnrC+0,charge+jnrD+0,
 407                                                                  charge+jnrE+0,charge+jnrF+0,
 408                                                                  charge+jnrG+0,charge+jnrH+0);
 409             vdwjidx0A        = 2*vdwtype[jnrA+0];
 410             vdwjidx0B        = 2*vdwtype[jnrB+0];
 411             vdwjidx0C        = 2*vdwtype[jnrC+0];
 412             vdwjidx0D        = 2*vdwtype[jnrD+0];
 413             vdwjidx0E        = 2*vdwtype[jnrE+0];
 414             vdwjidx0F        = 2*vdwtype[jnrF+0];
 415             vdwjidx0G        = 2*vdwtype[jnrG+0];
 416             vdwjidx0H        = 2*vdwtype[jnrH+0];
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 423             {
 424
 425             r00              = _mm256_mul_ps(rsq00,rinv00);
 426             r00              = _mm256_andnot_ps(dummy_mask,r00);
 427
 428             /* Compute parameters for interactions between i and j atoms */
 429             qq00             = _mm256_mul_ps(iq0,jq0);
 430             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 431                                             vdwioffsetptr0+vdwjidx0B,
 432                                             vdwioffsetptr0+vdwjidx0C,
 433                                             vdwioffsetptr0+vdwjidx0D,
 434                                             vdwioffsetptr0+vdwjidx0E,
 435                                             vdwioffsetptr0+vdwjidx0F,
 436                                             vdwioffsetptr0+vdwjidx0G,
 437                                             vdwioffsetptr0+vdwjidx0H,
 438                                             &c6_00,&c12_00);
 439
 440             /* Calculate table index by multiplying r with table scale and truncate to integer */
 441             rt               = _mm256_mul_ps(r00,vftabscale);
 442             vfitab           = _mm256_cvttps_epi32(rt);
 443             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 444             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 445             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 446             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 447             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 448             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 449
 450             /* REACTION-FIELD ELECTROSTATICS */
 451             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 452             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 453
 454             /* CUBIC SPLINE TABLE DISPERSION */
 455             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 456                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 457             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 458                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 459             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 460                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 461             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 462                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 463             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 464             Heps             = _mm256_mul_ps(vfeps,H);
 465             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 466             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 467             vvdw6            = _mm256_mul_ps(c6_00,VV);
 468             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 469             fvdw6            = _mm256_mul_ps(c6_00,FF);
 470
 471             /* CUBIC SPLINE TABLE REPULSION */
 472             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 473             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 474             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 475                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 476             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 477                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 478             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 479                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 480             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 481                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 482             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 483             Heps             = _mm256_mul_ps(vfeps,H);
 484             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 485             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 486             vvdw12           = _mm256_mul_ps(c12_00,VV);
 487             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 488             fvdw12           = _mm256_mul_ps(c12_00,FF);
 489             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 490             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 491
 492             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm256_and_ps(velec,cutoff_mask);
 496             velec            = _mm256_andnot_ps(dummy_mask,velec);
 497             velecsum         = _mm256_add_ps(velecsum,velec);
 498             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 499             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 500             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 501
 502             fscal            = _mm256_add_ps(felec,fvdw);
 503
 504             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 505
 506             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm256_mul_ps(fscal,dx00);
 510             ty               = _mm256_mul_ps(fscal,dy00);
 511             tz               = _mm256_mul_ps(fscal,dz00);
 512
 513             /* Update vectorial force */
 514             fix0             = _mm256_add_ps(fix0,tx);
 515             fiy0             = _mm256_add_ps(fiy0,ty);
 516             fiz0             = _mm256_add_ps(fiz0,tz);
 517
 518             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 519             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 520             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 521             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 522             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 523             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 524             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 525             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 526             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 527
 528             }
 529
 530             /* Inner loop uses 73 flops */
 531         }
 532
 533         /* End of innermost loop */
 534
 535         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 536                                                  f+i_coord_offset,fshift+i_shift_offset);
 537
 538         ggid                        = gid[iidx];
 539         /* Update potential energies */
 540         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 541         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 542
 543         /* Increment number of inner iterations */
 544         inneriter                  += j_index_end - j_index_start;
 545
 546         /* Outer loop uses 9 flops */
 547     }
 548
 549     /* Increment number of outer iterations */
 550     outeriter        += nri;
 551
 552     /* Update outer/inner flops */
 553
 554     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 555 }
 556 /*
 557  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_256_single
 558  * Electrostatics interaction: ReactionField
 559  * VdW interaction:            CubicSplineTable
 560  * Geometry:                   Particle-Particle
 561  * Calculate force/pot:        Force
 562  */
 563 void
 564 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_256_single
 565                     (t_nblist                    * gmx_restrict       nlist,
 566                      rvec                        * gmx_restrict          xx,
 567                      rvec                        * gmx_restrict          ff,
 568                      t_forcerec                  * gmx_restrict          fr,
 569                      t_mdatoms                   * gmx_restrict     mdatoms,
 570                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 571                      t_nrnb                      * gmx_restrict        nrnb)
 572 {
 573     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 574      * just 0 for non-waters.
 575      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 576      * jnr indices corresponding to data put in the four positions in the SIMD register.
 577      */
 578     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 579     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 580     int              jnrA,jnrB,jnrC,jnrD;
 581     int              jnrE,jnrF,jnrG,jnrH;
 582     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 583     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 584     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 585     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 586     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 587     real             rcutoff_scalar;
 588     real             *shiftvec,*fshift,*x,*f;
 589     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 590     real             scratch[4*DIM];
 591     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 592     real *           vdwioffsetptr0;
 593     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 594     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 595     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 596     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 597     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 598     real             *charge;
 599     int              nvdwtype;
 600     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 601     int              *vdwtype;
 602     real             *vdwparam;
 603     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 604     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 605     __m256i          vfitab;
 606     __m128i          vfitab_lo,vfitab_hi;
 607     __m128i          ifour       = _mm_set1_epi32(4);
 608     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 609     real             *vftab;
 610     __m256           dummy_mask,cutoff_mask;
 611     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 612     __m256           one     = _mm256_set1_ps(1.0);
 613     __m256           two     = _mm256_set1_ps(2.0);
 614     x                = xx[0];
 615     f                = ff[0];
 616
 617     nri              = nlist->nri;
 618     iinr             = nlist->iinr;
 619     jindex           = nlist->jindex;
 620     jjnr             = nlist->jjnr;
 621     shiftidx         = nlist->shift;
 622     gid              = nlist->gid;
 623     shiftvec         = fr->shift_vec[0];
 624     fshift           = fr->fshift[0];
 625     facel            = _mm256_set1_ps(fr->epsfac);
 626     charge           = mdatoms->chargeA;
 627     krf              = _mm256_set1_ps(fr->ic->k_rf);
 628     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 629     crf              = _mm256_set1_ps(fr->ic->c_rf);
 630     nvdwtype         = fr->ntype;
 631     vdwparam         = fr->nbfp;
 632     vdwtype          = mdatoms->typeA;
 633
 634     vftab            = kernel_data->table_vdw->data;
 635     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 636
 637     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 638     rcutoff_scalar   = fr->rcoulomb;
 639     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 640     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 641
 642     /* Avoid stupid compiler warnings */
 643     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 644     j_coord_offsetA = 0;
 645     j_coord_offsetB = 0;
 646     j_coord_offsetC = 0;
 647     j_coord_offsetD = 0;
 648     j_coord_offsetE = 0;
 649     j_coord_offsetF = 0;
 650     j_coord_offsetG = 0;
 651     j_coord_offsetH = 0;
 652
 653     outeriter        = 0;
 654     inneriter        = 0;
 655
 656     for(iidx=0;iidx<4*DIM;iidx++)
 657     {
 658         scratch[iidx] = 0.0;
 659     }
 660
 661     /* Start outer loop over neighborlists */
 662     for(iidx=0; iidx<nri; iidx++)
 663     {
 664         /* Load shift vector for this list */
 665         i_shift_offset   = DIM*shiftidx[iidx];
 666
 667         /* Load limits for loop over neighbors */
 668         j_index_start    = jindex[iidx];
 669         j_index_end      = jindex[iidx+1];
 670
 671         /* Get outer coordinate index */
 672         inr              = iinr[iidx];
 673         i_coord_offset   = DIM*inr;
 674
 675         /* Load i particle coords and add shift vector */
 676         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 677
 678         fix0             = _mm256_setzero_ps();
 679         fiy0             = _mm256_setzero_ps();
 680         fiz0             = _mm256_setzero_ps();
 681
 682         /* Load parameters for i particles */
 683         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 684         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 685
 686         /* Start inner kernel loop */
 687         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 688         {
 689
 690             /* Get j neighbor index, and coordinate index */
 691             jnrA             = jjnr[jidx];
 692             jnrB             = jjnr[jidx+1];
 693             jnrC             = jjnr[jidx+2];
 694             jnrD             = jjnr[jidx+3];
 695             jnrE             = jjnr[jidx+4];
 696             jnrF             = jjnr[jidx+5];
 697             jnrG             = jjnr[jidx+6];
 698             jnrH             = jjnr[jidx+7];
 699             j_coord_offsetA  = DIM*jnrA;
 700             j_coord_offsetB  = DIM*jnrB;
 701             j_coord_offsetC  = DIM*jnrC;
 702             j_coord_offsetD  = DIM*jnrD;
 703             j_coord_offsetE  = DIM*jnrE;
 704             j_coord_offsetF  = DIM*jnrF;
 705             j_coord_offsetG  = DIM*jnrG;
 706             j_coord_offsetH  = DIM*jnrH;
 707
 708             /* load j atom coordinates */
 709             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 710                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 711                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 712                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 713                                                  &jx0,&jy0,&jz0);
 714
 715             /* Calculate displacement vector */
 716             dx00             = _mm256_sub_ps(ix0,jx0);
 717             dy00             = _mm256_sub_ps(iy0,jy0);
 718             dz00             = _mm256_sub_ps(iz0,jz0);
 719
 720             /* Calculate squared distance and things based on it */
 721             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 722
 723             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 724
 725             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 726
 727             /* Load parameters for j particles */
 728             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 729                                                                  charge+jnrC+0,charge+jnrD+0,
 730                                                                  charge+jnrE+0,charge+jnrF+0,
 731                                                                  charge+jnrG+0,charge+jnrH+0);
 732             vdwjidx0A        = 2*vdwtype[jnrA+0];
 733             vdwjidx0B        = 2*vdwtype[jnrB+0];
 734             vdwjidx0C        = 2*vdwtype[jnrC+0];
 735             vdwjidx0D        = 2*vdwtype[jnrD+0];
 736             vdwjidx0E        = 2*vdwtype[jnrE+0];
 737             vdwjidx0F        = 2*vdwtype[jnrF+0];
 738             vdwjidx0G        = 2*vdwtype[jnrG+0];
 739             vdwjidx0H        = 2*vdwtype[jnrH+0];
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 746             {
 747
 748             r00              = _mm256_mul_ps(rsq00,rinv00);
 749
 750             /* Compute parameters for interactions between i and j atoms */
 751             qq00             = _mm256_mul_ps(iq0,jq0);
 752             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 753                                             vdwioffsetptr0+vdwjidx0B,
 754                                             vdwioffsetptr0+vdwjidx0C,
 755                                             vdwioffsetptr0+vdwjidx0D,
 756                                             vdwioffsetptr0+vdwjidx0E,
 757                                             vdwioffsetptr0+vdwjidx0F,
 758                                             vdwioffsetptr0+vdwjidx0G,
 759                                             vdwioffsetptr0+vdwjidx0H,
 760                                             &c6_00,&c12_00);
 761
 762             /* Calculate table index by multiplying r with table scale and truncate to integer */
 763             rt               = _mm256_mul_ps(r00,vftabscale);
 764             vfitab           = _mm256_cvttps_epi32(rt);
 765             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 766             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 767             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 768             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 769             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 770             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 771
 772             /* REACTION-FIELD ELECTROSTATICS */
 773             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 774
 775             /* CUBIC SPLINE TABLE DISPERSION */
 776             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 777                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 778             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 779                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 780             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 781                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 782             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 783                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 784             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 785             Heps             = _mm256_mul_ps(vfeps,H);
 786             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 787             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 788             fvdw6            = _mm256_mul_ps(c6_00,FF);
 789
 790             /* CUBIC SPLINE TABLE REPULSION */
 791             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 792             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 793             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 794                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 795             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 796                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 797             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 798                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 799             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 800                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 801             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 802             Heps             = _mm256_mul_ps(vfeps,H);
 803             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 804             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 805             fvdw12           = _mm256_mul_ps(c12_00,FF);
 806             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 807
 808             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 809
 810             fscal            = _mm256_add_ps(felec,fvdw);
 811
 812             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 813
 814             /* Calculate temporary vectorial force */
 815             tx               = _mm256_mul_ps(fscal,dx00);
 816             ty               = _mm256_mul_ps(fscal,dy00);
 817             tz               = _mm256_mul_ps(fscal,dz00);
 818
 819             /* Update vectorial force */
 820             fix0             = _mm256_add_ps(fix0,tx);
 821             fiy0             = _mm256_add_ps(fiy0,ty);
 822             fiz0             = _mm256_add_ps(fiz0,tz);
 823
 824             fjptrA             = f+j_coord_offsetA;
 825             fjptrB             = f+j_coord_offsetB;
 826             fjptrC             = f+j_coord_offsetC;
 827             fjptrD             = f+j_coord_offsetD;
 828             fjptrE             = f+j_coord_offsetE;
 829             fjptrF             = f+j_coord_offsetF;
 830             fjptrG             = f+j_coord_offsetG;
 831             fjptrH             = f+j_coord_offsetH;
 832             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 833
 834             }
 835
 836             /* Inner loop uses 57 flops */
 837         }
 838
 839         if(jidx<j_index_end)
 840         {
 841
 842             /* Get j neighbor index, and coordinate index */
 843             jnrlistA         = jjnr[jidx];
 844             jnrlistB         = jjnr[jidx+1];
 845             jnrlistC         = jjnr[jidx+2];
 846             jnrlistD         = jjnr[jidx+3];
 847             jnrlistE         = jjnr[jidx+4];
 848             jnrlistF         = jjnr[jidx+5];
 849             jnrlistG         = jjnr[jidx+6];
 850             jnrlistH         = jjnr[jidx+7];
 851             /* Sign of each element will be negative for non-real atoms.
 852              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 853              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 854              */
 855             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 856                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 857
 858             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 859             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 860             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 861             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 862             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 863             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 864             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 865             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 866             j_coord_offsetA  = DIM*jnrA;
 867             j_coord_offsetB  = DIM*jnrB;
 868             j_coord_offsetC  = DIM*jnrC;
 869             j_coord_offsetD  = DIM*jnrD;
 870             j_coord_offsetE  = DIM*jnrE;
 871             j_coord_offsetF  = DIM*jnrF;
 872             j_coord_offsetG  = DIM*jnrG;
 873             j_coord_offsetH  = DIM*jnrH;
 874
 875             /* load j atom coordinates */
 876             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 877                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 878                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 879                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 880                                                  &jx0,&jy0,&jz0);
 881
 882             /* Calculate displacement vector */
 883             dx00             = _mm256_sub_ps(ix0,jx0);
 884             dy00             = _mm256_sub_ps(iy0,jy0);
 885             dz00             = _mm256_sub_ps(iz0,jz0);
 886
 887             /* Calculate squared distance and things based on it */
 888             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 889
 890             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 891
 892             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 893
 894             /* Load parameters for j particles */
 895             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 896                                                                  charge+jnrC+0,charge+jnrD+0,
 897                                                                  charge+jnrE+0,charge+jnrF+0,
 898                                                                  charge+jnrG+0,charge+jnrH+0);
 899             vdwjidx0A        = 2*vdwtype[jnrA+0];
 900             vdwjidx0B        = 2*vdwtype[jnrB+0];
 901             vdwjidx0C        = 2*vdwtype[jnrC+0];
 902             vdwjidx0D        = 2*vdwtype[jnrD+0];
 903             vdwjidx0E        = 2*vdwtype[jnrE+0];
 904             vdwjidx0F        = 2*vdwtype[jnrF+0];
 905             vdwjidx0G        = 2*vdwtype[jnrG+0];
 906             vdwjidx0H        = 2*vdwtype[jnrH+0];
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 913             {
 914
 915             r00              = _mm256_mul_ps(rsq00,rinv00);
 916             r00              = _mm256_andnot_ps(dummy_mask,r00);
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             qq00             = _mm256_mul_ps(iq0,jq0);
 920             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 921                                             vdwioffsetptr0+vdwjidx0B,
 922                                             vdwioffsetptr0+vdwjidx0C,
 923                                             vdwioffsetptr0+vdwjidx0D,
 924                                             vdwioffsetptr0+vdwjidx0E,
 925                                             vdwioffsetptr0+vdwjidx0F,
 926                                             vdwioffsetptr0+vdwjidx0G,
 927                                             vdwioffsetptr0+vdwjidx0H,
 928                                             &c6_00,&c12_00);
 929
 930             /* Calculate table index by multiplying r with table scale and truncate to integer */
 931             rt               = _mm256_mul_ps(r00,vftabscale);
 932             vfitab           = _mm256_cvttps_epi32(rt);
 933             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 934             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 935             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 936             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 937             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 938             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 939
 940             /* REACTION-FIELD ELECTROSTATICS */
 941             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 942
 943             /* CUBIC SPLINE TABLE DISPERSION */
 944             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 945                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 946             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 947                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 948             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 949                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 950             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 951                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 952             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 953             Heps             = _mm256_mul_ps(vfeps,H);
 954             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 955             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 956             fvdw6            = _mm256_mul_ps(c6_00,FF);
 957
 958             /* CUBIC SPLINE TABLE REPULSION */
 959             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 960             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 961             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 962                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 963             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 964                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 965             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 966                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 967             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 968                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 969             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 970             Heps             = _mm256_mul_ps(vfeps,H);
 971             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 972             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 973             fvdw12           = _mm256_mul_ps(c12_00,FF);
 974             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 975
 976             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 977
 978             fscal            = _mm256_add_ps(felec,fvdw);
 979
 980             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 981
 982             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 983
 984             /* Calculate temporary vectorial force */
 985             tx               = _mm256_mul_ps(fscal,dx00);
 986             ty               = _mm256_mul_ps(fscal,dy00);
 987             tz               = _mm256_mul_ps(fscal,dz00);
 988
 989             /* Update vectorial force */
 990             fix0             = _mm256_add_ps(fix0,tx);
 991             fiy0             = _mm256_add_ps(fiy0,ty);
 992             fiz0             = _mm256_add_ps(fiz0,tz);
 993
 994             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 995             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 996             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 997             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 998             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 999             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1000             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1001             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1002             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1003
1004             }
1005
1006             /* Inner loop uses 58 flops */
1007         }
1008
1009         /* End of innermost loop */
1010
1011         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1012                                                  f+i_coord_offset,fshift+i_shift_offset);
1013
1014         /* Increment number of inner iterations */
1015         inneriter                  += j_index_end - j_index_start;
1016
1017         /* Outer loop uses 7 flops */
1018     }
1019
1020     /* Increment number of outer iterations */
1021     outeriter        += nri;
1022
1023     /* Update outer/inner flops */
1024
1025     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*58);
1026 }