src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  75     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  84     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  85     __m256i          vfitab;
  86     __m128i          vfitab_lo,vfitab_hi;
  87     __m128i          ifour       = _mm_set1_epi32(4);
  88     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  89     real             *vftab;
  90     __m256           dummy_mask,cutoff_mask;
  91     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  92     __m256           one     = _mm256_set1_ps(1.0);
  93     __m256           two     = _mm256_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm256_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     krf              = _mm256_set1_ps(fr->ic->k_rf);
 108     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 109     crf              = _mm256_set1_ps(fr->ic->c_rf);
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     vftab            = kernel_data->table_vdw->data;
 115     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123     j_coord_offsetE = 0;
 124     j_coord_offsetF = 0;
 125     j_coord_offsetG = 0;
 126     j_coord_offsetH = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm256_setzero_ps();
 154         fiy0             = _mm256_setzero_ps();
 155         fiz0             = _mm256_setzero_ps();
 156
 157         /* Load parameters for i particles */
 158         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 159         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm256_setzero_ps();
 163         vvdwsum          = _mm256_setzero_ps();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174             jnrE             = jjnr[jidx+4];
 175             jnrF             = jjnr[jidx+5];
 176             jnrG             = jjnr[jidx+6];
 177             jnrH             = jjnr[jidx+7];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182             j_coord_offsetE  = DIM*jnrE;
 183             j_coord_offsetF  = DIM*jnrF;
 184             j_coord_offsetG  = DIM*jnrG;
 185             j_coord_offsetH  = DIM*jnrH;
 186
 187             /* load j atom coordinates */
 188             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 190                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 191                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 192                                                  &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm256_sub_ps(ix0,jx0);
 196             dy00             = _mm256_sub_ps(iy0,jy0);
 197             dz00             = _mm256_sub_ps(iz0,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 201
 202             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 203
 204             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 208                                                                  charge+jnrC+0,charge+jnrD+0,
 209                                                                  charge+jnrE+0,charge+jnrF+0,
 210                                                                  charge+jnrG+0,charge+jnrH+0);
 211             vdwjidx0A        = 2*vdwtype[jnrA+0];
 212             vdwjidx0B        = 2*vdwtype[jnrB+0];
 213             vdwjidx0C        = 2*vdwtype[jnrC+0];
 214             vdwjidx0D        = 2*vdwtype[jnrD+0];
 215             vdwjidx0E        = 2*vdwtype[jnrE+0];
 216             vdwjidx0F        = 2*vdwtype[jnrF+0];
 217             vdwjidx0G        = 2*vdwtype[jnrG+0];
 218             vdwjidx0H        = 2*vdwtype[jnrH+0];
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             r00              = _mm256_mul_ps(rsq00,rinv00);
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm256_mul_ps(iq0,jq0);
 228             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 229                                             vdwioffsetptr0+vdwjidx0B,
 230                                             vdwioffsetptr0+vdwjidx0C,
 231                                             vdwioffsetptr0+vdwjidx0D,
 232                                             vdwioffsetptr0+vdwjidx0E,
 233                                             vdwioffsetptr0+vdwjidx0F,
 234                                             vdwioffsetptr0+vdwjidx0G,
 235                                             vdwioffsetptr0+vdwjidx0H,
 236                                             &c6_00,&c12_00);
 237
 238             /* Calculate table index by multiplying r with table scale and truncate to integer */
 239             rt               = _mm256_mul_ps(r00,vftabscale);
 240             vfitab           = _mm256_cvttps_epi32(rt);
 241             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 242             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 243             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 244             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 245             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 246             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 247
 248             /* REACTION-FIELD ELECTROSTATICS */
 249             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 250             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 251
 252             /* CUBIC SPLINE TABLE DISPERSION */
 253             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 254                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 255             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 256                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 257             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 258                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 259             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 260                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 261             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 262             Heps             = _mm256_mul_ps(vfeps,H);
 263             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 264             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 265             vvdw6            = _mm256_mul_ps(c6_00,VV);
 266             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 267             fvdw6            = _mm256_mul_ps(c6_00,FF);
 268
 269             /* CUBIC SPLINE TABLE REPULSION */
 270             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 271             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 272             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 273                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 274             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 275                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 276             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 277                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 278             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 279                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 280             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 281             Heps             = _mm256_mul_ps(vfeps,H);
 282             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 283             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 284             vvdw12           = _mm256_mul_ps(c12_00,VV);
 285             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 286             fvdw12           = _mm256_mul_ps(c12_00,FF);
 287             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 288             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velecsum         = _mm256_add_ps(velecsum,velec);
 292             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 293
 294             fscal            = _mm256_add_ps(felec,fvdw);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm256_mul_ps(fscal,dx00);
 298             ty               = _mm256_mul_ps(fscal,dy00);
 299             tz               = _mm256_mul_ps(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm256_add_ps(fix0,tx);
 303             fiy0             = _mm256_add_ps(fiy0,ty);
 304             fiz0             = _mm256_add_ps(fiz0,tz);
 305
 306             fjptrA             = f+j_coord_offsetA;
 307             fjptrB             = f+j_coord_offsetB;
 308             fjptrC             = f+j_coord_offsetC;
 309             fjptrD             = f+j_coord_offsetD;
 310             fjptrE             = f+j_coord_offsetE;
 311             fjptrF             = f+j_coord_offsetF;
 312             fjptrG             = f+j_coord_offsetG;
 313             fjptrH             = f+j_coord_offsetH;
 314             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 315
 316             /* Inner loop uses 67 flops */
 317         }
 318
 319         if(jidx<j_index_end)
 320         {
 321
 322             /* Get j neighbor index, and coordinate index */
 323             jnrlistA         = jjnr[jidx];
 324             jnrlistB         = jjnr[jidx+1];
 325             jnrlistC         = jjnr[jidx+2];
 326             jnrlistD         = jjnr[jidx+3];
 327             jnrlistE         = jjnr[jidx+4];
 328             jnrlistF         = jjnr[jidx+5];
 329             jnrlistG         = jjnr[jidx+6];
 330             jnrlistH         = jjnr[jidx+7];
 331             /* Sign of each element will be negative for non-real atoms.
 332              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 333              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 334              */
 335             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 336                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 337
 338             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 339             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 340             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 341             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 342             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 343             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 344             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 345             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 346             j_coord_offsetA  = DIM*jnrA;
 347             j_coord_offsetB  = DIM*jnrB;
 348             j_coord_offsetC  = DIM*jnrC;
 349             j_coord_offsetD  = DIM*jnrD;
 350             j_coord_offsetE  = DIM*jnrE;
 351             j_coord_offsetF  = DIM*jnrF;
 352             j_coord_offsetG  = DIM*jnrG;
 353             j_coord_offsetH  = DIM*jnrH;
 354
 355             /* load j atom coordinates */
 356             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 357                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 358                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 359                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 360                                                  &jx0,&jy0,&jz0);
 361
 362             /* Calculate displacement vector */
 363             dx00             = _mm256_sub_ps(ix0,jx0);
 364             dy00             = _mm256_sub_ps(iy0,jy0);
 365             dz00             = _mm256_sub_ps(iz0,jz0);
 366
 367             /* Calculate squared distance and things based on it */
 368             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 369
 370             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 371
 372             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 373
 374             /* Load parameters for j particles */
 375             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 376                                                                  charge+jnrC+0,charge+jnrD+0,
 377                                                                  charge+jnrE+0,charge+jnrF+0,
 378                                                                  charge+jnrG+0,charge+jnrH+0);
 379             vdwjidx0A        = 2*vdwtype[jnrA+0];
 380             vdwjidx0B        = 2*vdwtype[jnrB+0];
 381             vdwjidx0C        = 2*vdwtype[jnrC+0];
 382             vdwjidx0D        = 2*vdwtype[jnrD+0];
 383             vdwjidx0E        = 2*vdwtype[jnrE+0];
 384             vdwjidx0F        = 2*vdwtype[jnrF+0];
 385             vdwjidx0G        = 2*vdwtype[jnrG+0];
 386             vdwjidx0H        = 2*vdwtype[jnrH+0];
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             r00              = _mm256_mul_ps(rsq00,rinv00);
 393             r00              = _mm256_andnot_ps(dummy_mask,r00);
 394
 395             /* Compute parameters for interactions between i and j atoms */
 396             qq00             = _mm256_mul_ps(iq0,jq0);
 397             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 398                                             vdwioffsetptr0+vdwjidx0B,
 399                                             vdwioffsetptr0+vdwjidx0C,
 400                                             vdwioffsetptr0+vdwjidx0D,
 401                                             vdwioffsetptr0+vdwjidx0E,
 402                                             vdwioffsetptr0+vdwjidx0F,
 403                                             vdwioffsetptr0+vdwjidx0G,
 404                                             vdwioffsetptr0+vdwjidx0H,
 405                                             &c6_00,&c12_00);
 406
 407             /* Calculate table index by multiplying r with table scale and truncate to integer */
 408             rt               = _mm256_mul_ps(r00,vftabscale);
 409             vfitab           = _mm256_cvttps_epi32(rt);
 410             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 411             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 412             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 413             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 414             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 415             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 416
 417             /* REACTION-FIELD ELECTROSTATICS */
 418             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 419             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 420
 421             /* CUBIC SPLINE TABLE DISPERSION */
 422             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 423                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 424             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 425                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 426             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 427                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 428             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 429                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 430             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 431             Heps             = _mm256_mul_ps(vfeps,H);
 432             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 433             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 434             vvdw6            = _mm256_mul_ps(c6_00,VV);
 435             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 436             fvdw6            = _mm256_mul_ps(c6_00,FF);
 437
 438             /* CUBIC SPLINE TABLE REPULSION */
 439             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 440             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 441             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 442                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 443             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 444                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 445             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 446                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 447             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 448                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 449             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 450             Heps             = _mm256_mul_ps(vfeps,H);
 451             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 452             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 453             vvdw12           = _mm256_mul_ps(c12_00,VV);
 454             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 455             fvdw12           = _mm256_mul_ps(c12_00,FF);
 456             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 457             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velec            = _mm256_andnot_ps(dummy_mask,velec);
 461             velecsum         = _mm256_add_ps(velecsum,velec);
 462             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 463             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 464
 465             fscal            = _mm256_add_ps(felec,fvdw);
 466
 467             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 468
 469             /* Calculate temporary vectorial force */
 470             tx               = _mm256_mul_ps(fscal,dx00);
 471             ty               = _mm256_mul_ps(fscal,dy00);
 472             tz               = _mm256_mul_ps(fscal,dz00);
 473
 474             /* Update vectorial force */
 475             fix0             = _mm256_add_ps(fix0,tx);
 476             fiy0             = _mm256_add_ps(fiy0,ty);
 477             fiz0             = _mm256_add_ps(fiz0,tz);
 478
 479             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 480             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 481             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 482             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 483             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 484             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 485             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 486             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 487             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 488
 489             /* Inner loop uses 68 flops */
 490         }
 491
 492         /* End of innermost loop */
 493
 494         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 495                                                  f+i_coord_offset,fshift+i_shift_offset);
 496
 497         ggid                        = gid[iidx];
 498         /* Update potential energies */
 499         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 500         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 501
 502         /* Increment number of inner iterations */
 503         inneriter                  += j_index_end - j_index_start;
 504
 505         /* Outer loop uses 9 flops */
 506     }
 507
 508     /* Increment number of outer iterations */
 509     outeriter        += nri;
 510
 511     /* Update outer/inner flops */
 512
 513     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 514 }
 515 /*
 516  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_256_single
 517  * Electrostatics interaction: ReactionField
 518  * VdW interaction:            CubicSplineTable
 519  * Geometry:                   Particle-Particle
 520  * Calculate force/pot:        Force
 521  */
 522 void
 523 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_256_single
 524                     (t_nblist * gmx_restrict                nlist,
 525                      rvec * gmx_restrict                    xx,
 526                      rvec * gmx_restrict                    ff,
 527                      t_forcerec * gmx_restrict              fr,
 528                      t_mdatoms * gmx_restrict               mdatoms,
 529                      nb_kernel_data_t * gmx_restrict        kernel_data,
 530                      t_nrnb * gmx_restrict                  nrnb)
 531 {
 532     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 533      * just 0 for non-waters.
 534      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 535      * jnr indices corresponding to data put in the four positions in the SIMD register.
 536      */
 537     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 538     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 539     int              jnrA,jnrB,jnrC,jnrD;
 540     int              jnrE,jnrF,jnrG,jnrH;
 541     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 542     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 543     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 544     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 545     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 546     real             rcutoff_scalar;
 547     real             *shiftvec,*fshift,*x,*f;
 548     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 549     real             scratch[4*DIM];
 550     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 551     real *           vdwioffsetptr0;
 552     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 553     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 554     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 555     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 556     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 557     real             *charge;
 558     int              nvdwtype;
 559     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 560     int              *vdwtype;
 561     real             *vdwparam;
 562     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 563     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 564     __m256i          vfitab;
 565     __m128i          vfitab_lo,vfitab_hi;
 566     __m128i          ifour       = _mm_set1_epi32(4);
 567     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 568     real             *vftab;
 569     __m256           dummy_mask,cutoff_mask;
 570     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 571     __m256           one     = _mm256_set1_ps(1.0);
 572     __m256           two     = _mm256_set1_ps(2.0);
 573     x                = xx[0];
 574     f                = ff[0];
 575
 576     nri              = nlist->nri;
 577     iinr             = nlist->iinr;
 578     jindex           = nlist->jindex;
 579     jjnr             = nlist->jjnr;
 580     shiftidx         = nlist->shift;
 581     gid              = nlist->gid;
 582     shiftvec         = fr->shift_vec[0];
 583     fshift           = fr->fshift[0];
 584     facel            = _mm256_set1_ps(fr->epsfac);
 585     charge           = mdatoms->chargeA;
 586     krf              = _mm256_set1_ps(fr->ic->k_rf);
 587     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 588     crf              = _mm256_set1_ps(fr->ic->c_rf);
 589     nvdwtype         = fr->ntype;
 590     vdwparam         = fr->nbfp;
 591     vdwtype          = mdatoms->typeA;
 592
 593     vftab            = kernel_data->table_vdw->data;
 594     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 595
 596     /* Avoid stupid compiler warnings */
 597     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 598     j_coord_offsetA = 0;
 599     j_coord_offsetB = 0;
 600     j_coord_offsetC = 0;
 601     j_coord_offsetD = 0;
 602     j_coord_offsetE = 0;
 603     j_coord_offsetF = 0;
 604     j_coord_offsetG = 0;
 605     j_coord_offsetH = 0;
 606
 607     outeriter        = 0;
 608     inneriter        = 0;
 609
 610     for(iidx=0;iidx<4*DIM;iidx++)
 611     {
 612         scratch[iidx] = 0.0;
 613     }
 614
 615     /* Start outer loop over neighborlists */
 616     for(iidx=0; iidx<nri; iidx++)
 617     {
 618         /* Load shift vector for this list */
 619         i_shift_offset   = DIM*shiftidx[iidx];
 620
 621         /* Load limits for loop over neighbors */
 622         j_index_start    = jindex[iidx];
 623         j_index_end      = jindex[iidx+1];
 624
 625         /* Get outer coordinate index */
 626         inr              = iinr[iidx];
 627         i_coord_offset   = DIM*inr;
 628
 629         /* Load i particle coords and add shift vector */
 630         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 631
 632         fix0             = _mm256_setzero_ps();
 633         fiy0             = _mm256_setzero_ps();
 634         fiz0             = _mm256_setzero_ps();
 635
 636         /* Load parameters for i particles */
 637         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 638         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 639
 640         /* Start inner kernel loop */
 641         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 642         {
 643
 644             /* Get j neighbor index, and coordinate index */
 645             jnrA             = jjnr[jidx];
 646             jnrB             = jjnr[jidx+1];
 647             jnrC             = jjnr[jidx+2];
 648             jnrD             = jjnr[jidx+3];
 649             jnrE             = jjnr[jidx+4];
 650             jnrF             = jjnr[jidx+5];
 651             jnrG             = jjnr[jidx+6];
 652             jnrH             = jjnr[jidx+7];
 653             j_coord_offsetA  = DIM*jnrA;
 654             j_coord_offsetB  = DIM*jnrB;
 655             j_coord_offsetC  = DIM*jnrC;
 656             j_coord_offsetD  = DIM*jnrD;
 657             j_coord_offsetE  = DIM*jnrE;
 658             j_coord_offsetF  = DIM*jnrF;
 659             j_coord_offsetG  = DIM*jnrG;
 660             j_coord_offsetH  = DIM*jnrH;
 661
 662             /* load j atom coordinates */
 663             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 664                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 665                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 666                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 667                                                  &jx0,&jy0,&jz0);
 668
 669             /* Calculate displacement vector */
 670             dx00             = _mm256_sub_ps(ix0,jx0);
 671             dy00             = _mm256_sub_ps(iy0,jy0);
 672             dz00             = _mm256_sub_ps(iz0,jz0);
 673
 674             /* Calculate squared distance and things based on it */
 675             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 676
 677             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 678
 679             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 680
 681             /* Load parameters for j particles */
 682             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 683                                                                  charge+jnrC+0,charge+jnrD+0,
 684                                                                  charge+jnrE+0,charge+jnrF+0,
 685                                                                  charge+jnrG+0,charge+jnrH+0);
 686             vdwjidx0A        = 2*vdwtype[jnrA+0];
 687             vdwjidx0B        = 2*vdwtype[jnrB+0];
 688             vdwjidx0C        = 2*vdwtype[jnrC+0];
 689             vdwjidx0D        = 2*vdwtype[jnrD+0];
 690             vdwjidx0E        = 2*vdwtype[jnrE+0];
 691             vdwjidx0F        = 2*vdwtype[jnrF+0];
 692             vdwjidx0G        = 2*vdwtype[jnrG+0];
 693             vdwjidx0H        = 2*vdwtype[jnrH+0];
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             r00              = _mm256_mul_ps(rsq00,rinv00);
 700
 701             /* Compute parameters for interactions between i and j atoms */
 702             qq00             = _mm256_mul_ps(iq0,jq0);
 703             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 704                                             vdwioffsetptr0+vdwjidx0B,
 705                                             vdwioffsetptr0+vdwjidx0C,
 706                                             vdwioffsetptr0+vdwjidx0D,
 707                                             vdwioffsetptr0+vdwjidx0E,
 708                                             vdwioffsetptr0+vdwjidx0F,
 709                                             vdwioffsetptr0+vdwjidx0G,
 710                                             vdwioffsetptr0+vdwjidx0H,
 711                                             &c6_00,&c12_00);
 712
 713             /* Calculate table index by multiplying r with table scale and truncate to integer */
 714             rt               = _mm256_mul_ps(r00,vftabscale);
 715             vfitab           = _mm256_cvttps_epi32(rt);
 716             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 717             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 718             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 719             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 720             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 721             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 722
 723             /* REACTION-FIELD ELECTROSTATICS */
 724             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 725
 726             /* CUBIC SPLINE TABLE DISPERSION */
 727             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 728                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 729             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 730                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 731             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 732                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 733             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 734                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 735             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 736             Heps             = _mm256_mul_ps(vfeps,H);
 737             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 738             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 739             fvdw6            = _mm256_mul_ps(c6_00,FF);
 740
 741             /* CUBIC SPLINE TABLE REPULSION */
 742             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 743             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 744             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 745                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 746             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 747                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 748             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 749                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 750             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 751                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 752             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 753             Heps             = _mm256_mul_ps(vfeps,H);
 754             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 755             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 756             fvdw12           = _mm256_mul_ps(c12_00,FF);
 757             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 758
 759             fscal            = _mm256_add_ps(felec,fvdw);
 760
 761             /* Calculate temporary vectorial force */
 762             tx               = _mm256_mul_ps(fscal,dx00);
 763             ty               = _mm256_mul_ps(fscal,dy00);
 764             tz               = _mm256_mul_ps(fscal,dz00);
 765
 766             /* Update vectorial force */
 767             fix0             = _mm256_add_ps(fix0,tx);
 768             fiy0             = _mm256_add_ps(fiy0,ty);
 769             fiz0             = _mm256_add_ps(fiz0,tz);
 770
 771             fjptrA             = f+j_coord_offsetA;
 772             fjptrB             = f+j_coord_offsetB;
 773             fjptrC             = f+j_coord_offsetC;
 774             fjptrD             = f+j_coord_offsetD;
 775             fjptrE             = f+j_coord_offsetE;
 776             fjptrF             = f+j_coord_offsetF;
 777             fjptrG             = f+j_coord_offsetG;
 778             fjptrH             = f+j_coord_offsetH;
 779             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 780
 781             /* Inner loop uses 54 flops */
 782         }
 783
 784         if(jidx<j_index_end)
 785         {
 786
 787             /* Get j neighbor index, and coordinate index */
 788             jnrlistA         = jjnr[jidx];
 789             jnrlistB         = jjnr[jidx+1];
 790             jnrlistC         = jjnr[jidx+2];
 791             jnrlistD         = jjnr[jidx+3];
 792             jnrlistE         = jjnr[jidx+4];
 793             jnrlistF         = jjnr[jidx+5];
 794             jnrlistG         = jjnr[jidx+6];
 795             jnrlistH         = jjnr[jidx+7];
 796             /* Sign of each element will be negative for non-real atoms.
 797              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 798              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 799              */
 800             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 801                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 802
 803             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 804             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 805             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 806             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 807             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 808             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 809             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 810             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 811             j_coord_offsetA  = DIM*jnrA;
 812             j_coord_offsetB  = DIM*jnrB;
 813             j_coord_offsetC  = DIM*jnrC;
 814             j_coord_offsetD  = DIM*jnrD;
 815             j_coord_offsetE  = DIM*jnrE;
 816             j_coord_offsetF  = DIM*jnrF;
 817             j_coord_offsetG  = DIM*jnrG;
 818             j_coord_offsetH  = DIM*jnrH;
 819
 820             /* load j atom coordinates */
 821             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 822                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 823                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 824                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 825                                                  &jx0,&jy0,&jz0);
 826
 827             /* Calculate displacement vector */
 828             dx00             = _mm256_sub_ps(ix0,jx0);
 829             dy00             = _mm256_sub_ps(iy0,jy0);
 830             dz00             = _mm256_sub_ps(iz0,jz0);
 831
 832             /* Calculate squared distance and things based on it */
 833             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 834
 835             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 836
 837             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 838
 839             /* Load parameters for j particles */
 840             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 841                                                                  charge+jnrC+0,charge+jnrD+0,
 842                                                                  charge+jnrE+0,charge+jnrF+0,
 843                                                                  charge+jnrG+0,charge+jnrH+0);
 844             vdwjidx0A        = 2*vdwtype[jnrA+0];
 845             vdwjidx0B        = 2*vdwtype[jnrB+0];
 846             vdwjidx0C        = 2*vdwtype[jnrC+0];
 847             vdwjidx0D        = 2*vdwtype[jnrD+0];
 848             vdwjidx0E        = 2*vdwtype[jnrE+0];
 849             vdwjidx0F        = 2*vdwtype[jnrF+0];
 850             vdwjidx0G        = 2*vdwtype[jnrG+0];
 851             vdwjidx0H        = 2*vdwtype[jnrH+0];
 852
 853             /**************************
 854              * CALCULATE INTERACTIONS *
 855              **************************/
 856
 857             r00              = _mm256_mul_ps(rsq00,rinv00);
 858             r00              = _mm256_andnot_ps(dummy_mask,r00);
 859
 860             /* Compute parameters for interactions between i and j atoms */
 861             qq00             = _mm256_mul_ps(iq0,jq0);
 862             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 863                                             vdwioffsetptr0+vdwjidx0B,
 864                                             vdwioffsetptr0+vdwjidx0C,
 865                                             vdwioffsetptr0+vdwjidx0D,
 866                                             vdwioffsetptr0+vdwjidx0E,
 867                                             vdwioffsetptr0+vdwjidx0F,
 868                                             vdwioffsetptr0+vdwjidx0G,
 869                                             vdwioffsetptr0+vdwjidx0H,
 870                                             &c6_00,&c12_00);
 871
 872             /* Calculate table index by multiplying r with table scale and truncate to integer */
 873             rt               = _mm256_mul_ps(r00,vftabscale);
 874             vfitab           = _mm256_cvttps_epi32(rt);
 875             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 876             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 877             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 878             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 879             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 880             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 881
 882             /* REACTION-FIELD ELECTROSTATICS */
 883             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 884
 885             /* CUBIC SPLINE TABLE DISPERSION */
 886             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 887                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 888             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 889                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 890             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 891                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 892             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 893                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 894             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 895             Heps             = _mm256_mul_ps(vfeps,H);
 896             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 897             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 898             fvdw6            = _mm256_mul_ps(c6_00,FF);
 899
 900             /* CUBIC SPLINE TABLE REPULSION */
 901             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 902             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 903             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 904                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 905             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 906                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 907             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 908                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 909             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 910                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 911             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 912             Heps             = _mm256_mul_ps(vfeps,H);
 913             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 914             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 915             fvdw12           = _mm256_mul_ps(c12_00,FF);
 916             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 917
 918             fscal            = _mm256_add_ps(felec,fvdw);
 919
 920             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm256_mul_ps(fscal,dx00);
 924             ty               = _mm256_mul_ps(fscal,dy00);
 925             tz               = _mm256_mul_ps(fscal,dz00);
 926
 927             /* Update vectorial force */
 928             fix0             = _mm256_add_ps(fix0,tx);
 929             fiy0             = _mm256_add_ps(fiy0,ty);
 930             fiz0             = _mm256_add_ps(fiz0,tz);
 931
 932             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 933             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 934             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 935             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 936             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 937             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 938             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 939             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 940             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 941
 942             /* Inner loop uses 55 flops */
 943         }
 944
 945         /* End of innermost loop */
 946
 947         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 948                                                  f+i_coord_offset,fshift+i_shift_offset);
 949
 950         /* Increment number of inner iterations */
 951         inneriter                  += j_index_end - j_index_start;
 952
 953         /* Outer loop uses 7 flops */
 954     }
 955
 956     /* Increment number of outer iterations */
 957     outeriter        += nri;
 958
 959     /* Update outer/inner flops */
 960
 961     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
 962 }