src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  88     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  97     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  98     __m256i          vfitab;
  99     __m128i          vfitab_lo,vfitab_hi;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m256           dummy_mask,cutoff_mask;
 104     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 105     __m256           one     = _mm256_set1_ps(1.0);
 106     __m256           two     = _mm256_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm256_set1_ps(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     krf              = _mm256_set1_ps(fr->ic->k_rf);
 121     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 122     crf              = _mm256_set1_ps(fr->ic->c_rf);
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_vdw->data;
 128     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136     j_coord_offsetE = 0;
 137     j_coord_offsetF = 0;
 138     j_coord_offsetG = 0;
 139     j_coord_offsetH = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm256_setzero_ps();
 167         fiy0             = _mm256_setzero_ps();
 168         fiz0             = _mm256_setzero_ps();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 172         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_ps();
 176         vvdwsum          = _mm256_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             jnrE             = jjnr[jidx+4];
 188             jnrF             = jjnr[jidx+5];
 189             jnrG             = jjnr[jidx+6];
 190             jnrH             = jjnr[jidx+7];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195             j_coord_offsetE  = DIM*jnrE;
 196             j_coord_offsetF  = DIM*jnrF;
 197             j_coord_offsetG  = DIM*jnrG;
 198             j_coord_offsetH  = DIM*jnrH;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 204                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_ps(ix0,jx0);
 209             dy00             = _mm256_sub_ps(iy0,jy0);
 210             dz00             = _mm256_sub_ps(iz0,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 214
 215             rinv00           = avx256_invsqrt_f(rsq00);
 216
 217             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                                  charge+jnrC+0,charge+jnrD+0,
 222                                                                  charge+jnrE+0,charge+jnrF+0,
 223                                                                  charge+jnrG+0,charge+jnrH+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228             vdwjidx0E        = 2*vdwtype[jnrE+0];
 229             vdwjidx0F        = 2*vdwtype[jnrF+0];
 230             vdwjidx0G        = 2*vdwtype[jnrG+0];
 231             vdwjidx0H        = 2*vdwtype[jnrH+0];
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _mm256_mul_ps(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm256_mul_ps(iq0,jq0);
 241             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 242                                             vdwioffsetptr0+vdwjidx0B,
 243                                             vdwioffsetptr0+vdwjidx0C,
 244                                             vdwioffsetptr0+vdwjidx0D,
 245                                             vdwioffsetptr0+vdwjidx0E,
 246                                             vdwioffsetptr0+vdwjidx0F,
 247                                             vdwioffsetptr0+vdwjidx0G,
 248                                             vdwioffsetptr0+vdwjidx0H,
 249                                             &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm256_mul_ps(r00,vftabscale);
 253             vfitab           = _mm256_cvttps_epi32(rt);
 254             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 255             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 256             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 257             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 258             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 259             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 260
 261             /* REACTION-FIELD ELECTROSTATICS */
 262             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 263             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 264
 265             /* CUBIC SPLINE TABLE DISPERSION */
 266             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 267                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 268             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 269                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 270             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 271                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 272             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 273                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 274             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 275             Heps             = _mm256_mul_ps(vfeps,H);
 276             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 277             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 278             vvdw6            = _mm256_mul_ps(c6_00,VV);
 279             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 280             fvdw6            = _mm256_mul_ps(c6_00,FF);
 281
 282             /* CUBIC SPLINE TABLE REPULSION */
 283             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 284             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 285             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 286                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 287             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 288                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 289             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 291             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 293             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 294             Heps             = _mm256_mul_ps(vfeps,H);
 295             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 296             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 297             vvdw12           = _mm256_mul_ps(c12_00,VV);
 298             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 299             fvdw12           = _mm256_mul_ps(c12_00,FF);
 300             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 301             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velecsum         = _mm256_add_ps(velecsum,velec);
 305             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 306
 307             fscal            = _mm256_add_ps(felec,fvdw);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm256_mul_ps(fscal,dx00);
 311             ty               = _mm256_mul_ps(fscal,dy00);
 312             tz               = _mm256_mul_ps(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm256_add_ps(fix0,tx);
 316             fiy0             = _mm256_add_ps(fiy0,ty);
 317             fiz0             = _mm256_add_ps(fiz0,tz);
 318
 319             fjptrA             = f+j_coord_offsetA;
 320             fjptrB             = f+j_coord_offsetB;
 321             fjptrC             = f+j_coord_offsetC;
 322             fjptrD             = f+j_coord_offsetD;
 323             fjptrE             = f+j_coord_offsetE;
 324             fjptrF             = f+j_coord_offsetF;
 325             fjptrG             = f+j_coord_offsetG;
 326             fjptrH             = f+j_coord_offsetH;
 327             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 328
 329             /* Inner loop uses 67 flops */
 330         }
 331
 332         if(jidx<j_index_end)
 333         {
 334
 335             /* Get j neighbor index, and coordinate index */
 336             jnrlistA         = jjnr[jidx];
 337             jnrlistB         = jjnr[jidx+1];
 338             jnrlistC         = jjnr[jidx+2];
 339             jnrlistD         = jjnr[jidx+3];
 340             jnrlistE         = jjnr[jidx+4];
 341             jnrlistF         = jjnr[jidx+5];
 342             jnrlistG         = jjnr[jidx+6];
 343             jnrlistH         = jjnr[jidx+7];
 344             /* Sign of each element will be negative for non-real atoms.
 345              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 346              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 347              */
 348             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 349                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 350
 351             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 352             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 353             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 354             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 355             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 356             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 357             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 358             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 359             j_coord_offsetA  = DIM*jnrA;
 360             j_coord_offsetB  = DIM*jnrB;
 361             j_coord_offsetC  = DIM*jnrC;
 362             j_coord_offsetD  = DIM*jnrD;
 363             j_coord_offsetE  = DIM*jnrE;
 364             j_coord_offsetF  = DIM*jnrF;
 365             j_coord_offsetG  = DIM*jnrG;
 366             j_coord_offsetH  = DIM*jnrH;
 367
 368             /* load j atom coordinates */
 369             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 370                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 371                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 372                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 373                                                  &jx0,&jy0,&jz0);
 374
 375             /* Calculate displacement vector */
 376             dx00             = _mm256_sub_ps(ix0,jx0);
 377             dy00             = _mm256_sub_ps(iy0,jy0);
 378             dz00             = _mm256_sub_ps(iz0,jz0);
 379
 380             /* Calculate squared distance and things based on it */
 381             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 382
 383             rinv00           = avx256_invsqrt_f(rsq00);
 384
 385             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 386
 387             /* Load parameters for j particles */
 388             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 389                                                                  charge+jnrC+0,charge+jnrD+0,
 390                                                                  charge+jnrE+0,charge+jnrF+0,
 391                                                                  charge+jnrG+0,charge+jnrH+0);
 392             vdwjidx0A        = 2*vdwtype[jnrA+0];
 393             vdwjidx0B        = 2*vdwtype[jnrB+0];
 394             vdwjidx0C        = 2*vdwtype[jnrC+0];
 395             vdwjidx0D        = 2*vdwtype[jnrD+0];
 396             vdwjidx0E        = 2*vdwtype[jnrE+0];
 397             vdwjidx0F        = 2*vdwtype[jnrF+0];
 398             vdwjidx0G        = 2*vdwtype[jnrG+0];
 399             vdwjidx0H        = 2*vdwtype[jnrH+0];
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             r00              = _mm256_mul_ps(rsq00,rinv00);
 406             r00              = _mm256_andnot_ps(dummy_mask,r00);
 407
 408             /* Compute parameters for interactions between i and j atoms */
 409             qq00             = _mm256_mul_ps(iq0,jq0);
 410             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 411                                             vdwioffsetptr0+vdwjidx0B,
 412                                             vdwioffsetptr0+vdwjidx0C,
 413                                             vdwioffsetptr0+vdwjidx0D,
 414                                             vdwioffsetptr0+vdwjidx0E,
 415                                             vdwioffsetptr0+vdwjidx0F,
 416                                             vdwioffsetptr0+vdwjidx0G,
 417                                             vdwioffsetptr0+vdwjidx0H,
 418                                             &c6_00,&c12_00);
 419
 420             /* Calculate table index by multiplying r with table scale and truncate to integer */
 421             rt               = _mm256_mul_ps(r00,vftabscale);
 422             vfitab           = _mm256_cvttps_epi32(rt);
 423             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 424             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 425             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 426             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 427             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 428             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 429
 430             /* REACTION-FIELD ELECTROSTATICS */
 431             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 432             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 433
 434             /* CUBIC SPLINE TABLE DISPERSION */
 435             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 436                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 437             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 438                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 439             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 440                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 441             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 442                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 443             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 444             Heps             = _mm256_mul_ps(vfeps,H);
 445             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 446             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 447             vvdw6            = _mm256_mul_ps(c6_00,VV);
 448             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 449             fvdw6            = _mm256_mul_ps(c6_00,FF);
 450
 451             /* CUBIC SPLINE TABLE REPULSION */
 452             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 453             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 454             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 455                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 456             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 457                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 458             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 459                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 460             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 461                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 462             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 463             Heps             = _mm256_mul_ps(vfeps,H);
 464             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 465             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 466             vvdw12           = _mm256_mul_ps(c12_00,VV);
 467             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 468             fvdw12           = _mm256_mul_ps(c12_00,FF);
 469             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 470             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velec            = _mm256_andnot_ps(dummy_mask,velec);
 474             velecsum         = _mm256_add_ps(velecsum,velec);
 475             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 476             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 477
 478             fscal            = _mm256_add_ps(felec,fvdw);
 479
 480             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 481
 482             /* Calculate temporary vectorial force */
 483             tx               = _mm256_mul_ps(fscal,dx00);
 484             ty               = _mm256_mul_ps(fscal,dy00);
 485             tz               = _mm256_mul_ps(fscal,dz00);
 486
 487             /* Update vectorial force */
 488             fix0             = _mm256_add_ps(fix0,tx);
 489             fiy0             = _mm256_add_ps(fiy0,ty);
 490             fiz0             = _mm256_add_ps(fiz0,tz);
 491
 492             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 493             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 494             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 495             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 496             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 497             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 498             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 499             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 500             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 501
 502             /* Inner loop uses 68 flops */
 503         }
 504
 505         /* End of innermost loop */
 506
 507         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 508                                                  f+i_coord_offset,fshift+i_shift_offset);
 509
 510         ggid                        = gid[iidx];
 511         /* Update potential energies */
 512         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 513         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 514
 515         /* Increment number of inner iterations */
 516         inneriter                  += j_index_end - j_index_start;
 517
 518         /* Outer loop uses 9 flops */
 519     }
 520
 521     /* Increment number of outer iterations */
 522     outeriter        += nri;
 523
 524     /* Update outer/inner flops */
 525
 526     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 527 }
 528 /*
 529  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_256_single
 530  * Electrostatics interaction: ReactionField
 531  * VdW interaction:            CubicSplineTable
 532  * Geometry:                   Particle-Particle
 533  * Calculate force/pot:        Force
 534  */
 535 void
 536 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_256_single
 537                     (t_nblist                    * gmx_restrict       nlist,
 538                      rvec                        * gmx_restrict          xx,
 539                      rvec                        * gmx_restrict          ff,
 540                      struct t_forcerec           * gmx_restrict          fr,
 541                      t_mdatoms                   * gmx_restrict     mdatoms,
 542                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 543                      t_nrnb                      * gmx_restrict        nrnb)
 544 {
 545     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 546      * just 0 for non-waters.
 547      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 548      * jnr indices corresponding to data put in the four positions in the SIMD register.
 549      */
 550     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 551     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 552     int              jnrA,jnrB,jnrC,jnrD;
 553     int              jnrE,jnrF,jnrG,jnrH;
 554     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 555     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 556     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 557     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 558     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 559     real             rcutoff_scalar;
 560     real             *shiftvec,*fshift,*x,*f;
 561     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 562     real             scratch[4*DIM];
 563     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 564     real *           vdwioffsetptr0;
 565     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 566     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 567     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 568     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 569     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 570     real             *charge;
 571     int              nvdwtype;
 572     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 573     int              *vdwtype;
 574     real             *vdwparam;
 575     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 576     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 577     __m256i          vfitab;
 578     __m128i          vfitab_lo,vfitab_hi;
 579     __m128i          ifour       = _mm_set1_epi32(4);
 580     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 581     real             *vftab;
 582     __m256           dummy_mask,cutoff_mask;
 583     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 584     __m256           one     = _mm256_set1_ps(1.0);
 585     __m256           two     = _mm256_set1_ps(2.0);
 586     x                = xx[0];
 587     f                = ff[0];
 588
 589     nri              = nlist->nri;
 590     iinr             = nlist->iinr;
 591     jindex           = nlist->jindex;
 592     jjnr             = nlist->jjnr;
 593     shiftidx         = nlist->shift;
 594     gid              = nlist->gid;
 595     shiftvec         = fr->shift_vec[0];
 596     fshift           = fr->fshift[0];
 597     facel            = _mm256_set1_ps(fr->ic->epsfac);
 598     charge           = mdatoms->chargeA;
 599     krf              = _mm256_set1_ps(fr->ic->k_rf);
 600     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 601     crf              = _mm256_set1_ps(fr->ic->c_rf);
 602     nvdwtype         = fr->ntype;
 603     vdwparam         = fr->nbfp;
 604     vdwtype          = mdatoms->typeA;
 605
 606     vftab            = kernel_data->table_vdw->data;
 607     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 608
 609     /* Avoid stupid compiler warnings */
 610     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 611     j_coord_offsetA = 0;
 612     j_coord_offsetB = 0;
 613     j_coord_offsetC = 0;
 614     j_coord_offsetD = 0;
 615     j_coord_offsetE = 0;
 616     j_coord_offsetF = 0;
 617     j_coord_offsetG = 0;
 618     j_coord_offsetH = 0;
 619
 620     outeriter        = 0;
 621     inneriter        = 0;
 622
 623     for(iidx=0;iidx<4*DIM;iidx++)
 624     {
 625         scratch[iidx] = 0.0;
 626     }
 627
 628     /* Start outer loop over neighborlists */
 629     for(iidx=0; iidx<nri; iidx++)
 630     {
 631         /* Load shift vector for this list */
 632         i_shift_offset   = DIM*shiftidx[iidx];
 633
 634         /* Load limits for loop over neighbors */
 635         j_index_start    = jindex[iidx];
 636         j_index_end      = jindex[iidx+1];
 637
 638         /* Get outer coordinate index */
 639         inr              = iinr[iidx];
 640         i_coord_offset   = DIM*inr;
 641
 642         /* Load i particle coords and add shift vector */
 643         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 644
 645         fix0             = _mm256_setzero_ps();
 646         fiy0             = _mm256_setzero_ps();
 647         fiz0             = _mm256_setzero_ps();
 648
 649         /* Load parameters for i particles */
 650         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 651         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 652
 653         /* Start inner kernel loop */
 654         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 655         {
 656
 657             /* Get j neighbor index, and coordinate index */
 658             jnrA             = jjnr[jidx];
 659             jnrB             = jjnr[jidx+1];
 660             jnrC             = jjnr[jidx+2];
 661             jnrD             = jjnr[jidx+3];
 662             jnrE             = jjnr[jidx+4];
 663             jnrF             = jjnr[jidx+5];
 664             jnrG             = jjnr[jidx+6];
 665             jnrH             = jjnr[jidx+7];
 666             j_coord_offsetA  = DIM*jnrA;
 667             j_coord_offsetB  = DIM*jnrB;
 668             j_coord_offsetC  = DIM*jnrC;
 669             j_coord_offsetD  = DIM*jnrD;
 670             j_coord_offsetE  = DIM*jnrE;
 671             j_coord_offsetF  = DIM*jnrF;
 672             j_coord_offsetG  = DIM*jnrG;
 673             j_coord_offsetH  = DIM*jnrH;
 674
 675             /* load j atom coordinates */
 676             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 677                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 678                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 679                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 680                                                  &jx0,&jy0,&jz0);
 681
 682             /* Calculate displacement vector */
 683             dx00             = _mm256_sub_ps(ix0,jx0);
 684             dy00             = _mm256_sub_ps(iy0,jy0);
 685             dz00             = _mm256_sub_ps(iz0,jz0);
 686
 687             /* Calculate squared distance and things based on it */
 688             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 689
 690             rinv00           = avx256_invsqrt_f(rsq00);
 691
 692             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 693
 694             /* Load parameters for j particles */
 695             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 696                                                                  charge+jnrC+0,charge+jnrD+0,
 697                                                                  charge+jnrE+0,charge+jnrF+0,
 698                                                                  charge+jnrG+0,charge+jnrH+0);
 699             vdwjidx0A        = 2*vdwtype[jnrA+0];
 700             vdwjidx0B        = 2*vdwtype[jnrB+0];
 701             vdwjidx0C        = 2*vdwtype[jnrC+0];
 702             vdwjidx0D        = 2*vdwtype[jnrD+0];
 703             vdwjidx0E        = 2*vdwtype[jnrE+0];
 704             vdwjidx0F        = 2*vdwtype[jnrF+0];
 705             vdwjidx0G        = 2*vdwtype[jnrG+0];
 706             vdwjidx0H        = 2*vdwtype[jnrH+0];
 707
 708             /**************************
 709              * CALCULATE INTERACTIONS *
 710              **************************/
 711
 712             r00              = _mm256_mul_ps(rsq00,rinv00);
 713
 714             /* Compute parameters for interactions between i and j atoms */
 715             qq00             = _mm256_mul_ps(iq0,jq0);
 716             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 717                                             vdwioffsetptr0+vdwjidx0B,
 718                                             vdwioffsetptr0+vdwjidx0C,
 719                                             vdwioffsetptr0+vdwjidx0D,
 720                                             vdwioffsetptr0+vdwjidx0E,
 721                                             vdwioffsetptr0+vdwjidx0F,
 722                                             vdwioffsetptr0+vdwjidx0G,
 723                                             vdwioffsetptr0+vdwjidx0H,
 724                                             &c6_00,&c12_00);
 725
 726             /* Calculate table index by multiplying r with table scale and truncate to integer */
 727             rt               = _mm256_mul_ps(r00,vftabscale);
 728             vfitab           = _mm256_cvttps_epi32(rt);
 729             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 730             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 731             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 732             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 733             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 734             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 735
 736             /* REACTION-FIELD ELECTROSTATICS */
 737             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 738
 739             /* CUBIC SPLINE TABLE DISPERSION */
 740             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 741                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 742             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 743                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 744             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 745                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 746             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 747                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 748             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 749             Heps             = _mm256_mul_ps(vfeps,H);
 750             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 751             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 752             fvdw6            = _mm256_mul_ps(c6_00,FF);
 753
 754             /* CUBIC SPLINE TABLE REPULSION */
 755             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 756             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 757             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 758                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 759             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 760                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 761             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 762                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 763             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 764                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 765             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 766             Heps             = _mm256_mul_ps(vfeps,H);
 767             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 768             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 769             fvdw12           = _mm256_mul_ps(c12_00,FF);
 770             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 771
 772             fscal            = _mm256_add_ps(felec,fvdw);
 773
 774             /* Calculate temporary vectorial force */
 775             tx               = _mm256_mul_ps(fscal,dx00);
 776             ty               = _mm256_mul_ps(fscal,dy00);
 777             tz               = _mm256_mul_ps(fscal,dz00);
 778
 779             /* Update vectorial force */
 780             fix0             = _mm256_add_ps(fix0,tx);
 781             fiy0             = _mm256_add_ps(fiy0,ty);
 782             fiz0             = _mm256_add_ps(fiz0,tz);
 783
 784             fjptrA             = f+j_coord_offsetA;
 785             fjptrB             = f+j_coord_offsetB;
 786             fjptrC             = f+j_coord_offsetC;
 787             fjptrD             = f+j_coord_offsetD;
 788             fjptrE             = f+j_coord_offsetE;
 789             fjptrF             = f+j_coord_offsetF;
 790             fjptrG             = f+j_coord_offsetG;
 791             fjptrH             = f+j_coord_offsetH;
 792             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 793
 794             /* Inner loop uses 54 flops */
 795         }
 796
 797         if(jidx<j_index_end)
 798         {
 799
 800             /* Get j neighbor index, and coordinate index */
 801             jnrlistA         = jjnr[jidx];
 802             jnrlistB         = jjnr[jidx+1];
 803             jnrlistC         = jjnr[jidx+2];
 804             jnrlistD         = jjnr[jidx+3];
 805             jnrlistE         = jjnr[jidx+4];
 806             jnrlistF         = jjnr[jidx+5];
 807             jnrlistG         = jjnr[jidx+6];
 808             jnrlistH         = jjnr[jidx+7];
 809             /* Sign of each element will be negative for non-real atoms.
 810              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 811              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 812              */
 813             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 814                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 815
 816             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 817             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 818             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 819             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 820             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 821             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 822             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 823             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 824             j_coord_offsetA  = DIM*jnrA;
 825             j_coord_offsetB  = DIM*jnrB;
 826             j_coord_offsetC  = DIM*jnrC;
 827             j_coord_offsetD  = DIM*jnrD;
 828             j_coord_offsetE  = DIM*jnrE;
 829             j_coord_offsetF  = DIM*jnrF;
 830             j_coord_offsetG  = DIM*jnrG;
 831             j_coord_offsetH  = DIM*jnrH;
 832
 833             /* load j atom coordinates */
 834             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 835                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 836                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 837                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 838                                                  &jx0,&jy0,&jz0);
 839
 840             /* Calculate displacement vector */
 841             dx00             = _mm256_sub_ps(ix0,jx0);
 842             dy00             = _mm256_sub_ps(iy0,jy0);
 843             dz00             = _mm256_sub_ps(iz0,jz0);
 844
 845             /* Calculate squared distance and things based on it */
 846             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 847
 848             rinv00           = avx256_invsqrt_f(rsq00);
 849
 850             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 851
 852             /* Load parameters for j particles */
 853             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 854                                                                  charge+jnrC+0,charge+jnrD+0,
 855                                                                  charge+jnrE+0,charge+jnrF+0,
 856                                                                  charge+jnrG+0,charge+jnrH+0);
 857             vdwjidx0A        = 2*vdwtype[jnrA+0];
 858             vdwjidx0B        = 2*vdwtype[jnrB+0];
 859             vdwjidx0C        = 2*vdwtype[jnrC+0];
 860             vdwjidx0D        = 2*vdwtype[jnrD+0];
 861             vdwjidx0E        = 2*vdwtype[jnrE+0];
 862             vdwjidx0F        = 2*vdwtype[jnrF+0];
 863             vdwjidx0G        = 2*vdwtype[jnrG+0];
 864             vdwjidx0H        = 2*vdwtype[jnrH+0];
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             r00              = _mm256_mul_ps(rsq00,rinv00);
 871             r00              = _mm256_andnot_ps(dummy_mask,r00);
 872
 873             /* Compute parameters for interactions between i and j atoms */
 874             qq00             = _mm256_mul_ps(iq0,jq0);
 875             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 876                                             vdwioffsetptr0+vdwjidx0B,
 877                                             vdwioffsetptr0+vdwjidx0C,
 878                                             vdwioffsetptr0+vdwjidx0D,
 879                                             vdwioffsetptr0+vdwjidx0E,
 880                                             vdwioffsetptr0+vdwjidx0F,
 881                                             vdwioffsetptr0+vdwjidx0G,
 882                                             vdwioffsetptr0+vdwjidx0H,
 883                                             &c6_00,&c12_00);
 884
 885             /* Calculate table index by multiplying r with table scale and truncate to integer */
 886             rt               = _mm256_mul_ps(r00,vftabscale);
 887             vfitab           = _mm256_cvttps_epi32(rt);
 888             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 889             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 890             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 891             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 892             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 893             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 894
 895             /* REACTION-FIELD ELECTROSTATICS */
 896             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 897
 898             /* CUBIC SPLINE TABLE DISPERSION */
 899             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 900                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 901             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 902                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 903             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 904                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 905             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 906                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 907             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 908             Heps             = _mm256_mul_ps(vfeps,H);
 909             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 910             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 911             fvdw6            = _mm256_mul_ps(c6_00,FF);
 912
 913             /* CUBIC SPLINE TABLE REPULSION */
 914             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 915             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 916             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 917                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 918             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 919                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 920             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 921                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 922             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 923                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 924             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 925             Heps             = _mm256_mul_ps(vfeps,H);
 926             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 927             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 928             fvdw12           = _mm256_mul_ps(c12_00,FF);
 929             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 930
 931             fscal            = _mm256_add_ps(felec,fvdw);
 932
 933             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm256_mul_ps(fscal,dx00);
 937             ty               = _mm256_mul_ps(fscal,dy00);
 938             tz               = _mm256_mul_ps(fscal,dz00);
 939
 940             /* Update vectorial force */
 941             fix0             = _mm256_add_ps(fix0,tx);
 942             fiy0             = _mm256_add_ps(fiy0,ty);
 943             fiz0             = _mm256_add_ps(fiz0,tz);
 944
 945             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 946             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 947             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 948             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 949             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 950             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 951             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 952             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 953             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 954
 955             /* Inner loop uses 55 flops */
 956         }
 957
 958         /* End of innermost loop */
 959
 960         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 961                                                  f+i_coord_offset,fshift+i_shift_offset);
 962
 963         /* Increment number of inner iterations */
 964         inneriter                  += j_index_end - j_index_start;
 965
 966         /* Outer loop uses 7 flops */
 967     }
 968
 969     /* Increment number of outer iterations */
 970     outeriter        += nri;
 971
 972     /* Update outer/inner flops */
 973
 974     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
 975 }