src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_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_ElecCoul_VdwCSTab_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     vftab            = kernel_data->table_vdw->data;
 112     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120     j_coord_offsetE = 0;
 121     j_coord_offsetF = 0;
 122     j_coord_offsetG = 0;
 123     j_coord_offsetH = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     for(iidx=0;iidx<4*DIM;iidx++)
 129     {
 130         scratch[iidx] = 0.0;
 131     }
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 149
 150         fix0             = _mm256_setzero_ps();
 151         fiy0             = _mm256_setzero_ps();
 152         fiz0             = _mm256_setzero_ps();
 153
 154         /* Load parameters for i particles */
 155         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 156         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         velecsum         = _mm256_setzero_ps();
 160         vvdwsum          = _mm256_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             jnrE             = jjnr[jidx+4];
 172             jnrF             = jjnr[jidx+5];
 173             jnrG             = jjnr[jidx+6];
 174             jnrH             = jjnr[jidx+7];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179             j_coord_offsetE  = DIM*jnrE;
 180             j_coord_offsetF  = DIM*jnrF;
 181             j_coord_offsetG  = DIM*jnrG;
 182             j_coord_offsetH  = DIM*jnrH;
 183
 184             /* load j atom coordinates */
 185             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 187                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 188                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 189                                                  &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm256_sub_ps(ix0,jx0);
 193             dy00             = _mm256_sub_ps(iy0,jy0);
 194             dz00             = _mm256_sub_ps(iz0,jz0);
 195
 196             /* Calculate squared distance and things based on it */
 197             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 198
 199             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 200
 201             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 202
 203             /* Load parameters for j particles */
 204             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 205                                                                  charge+jnrC+0,charge+jnrD+0,
 206                                                                  charge+jnrE+0,charge+jnrF+0,
 207                                                                  charge+jnrG+0,charge+jnrH+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210             vdwjidx0C        = 2*vdwtype[jnrC+0];
 211             vdwjidx0D        = 2*vdwtype[jnrD+0];
 212             vdwjidx0E        = 2*vdwtype[jnrE+0];
 213             vdwjidx0F        = 2*vdwtype[jnrF+0];
 214             vdwjidx0G        = 2*vdwtype[jnrG+0];
 215             vdwjidx0H        = 2*vdwtype[jnrH+0];
 216
 217             /**************************
 218              * CALCULATE INTERACTIONS *
 219              **************************/
 220
 221             r00              = _mm256_mul_ps(rsq00,rinv00);
 222
 223             /* Compute parameters for interactions between i and j atoms */
 224             qq00             = _mm256_mul_ps(iq0,jq0);
 225             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 226                                             vdwioffsetptr0+vdwjidx0B,
 227                                             vdwioffsetptr0+vdwjidx0C,
 228                                             vdwioffsetptr0+vdwjidx0D,
 229                                             vdwioffsetptr0+vdwjidx0E,
 230                                             vdwioffsetptr0+vdwjidx0F,
 231                                             vdwioffsetptr0+vdwjidx0G,
 232                                             vdwioffsetptr0+vdwjidx0H,
 233                                             &c6_00,&c12_00);
 234
 235             /* Calculate table index by multiplying r with table scale and truncate to integer */
 236             rt               = _mm256_mul_ps(r00,vftabscale);
 237             vfitab           = _mm256_cvttps_epi32(rt);
 238             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 239             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 240             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 241             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 242             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 243             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 244
 245             /* COULOMB ELECTROSTATICS */
 246             velec            = _mm256_mul_ps(qq00,rinv00);
 247             felec            = _mm256_mul_ps(velec,rinvsq00);
 248
 249             /* CUBIC SPLINE TABLE DISPERSION */
 250             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 251                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 252             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 253                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 254             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 255                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 256             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 257                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 258             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 259             Heps             = _mm256_mul_ps(vfeps,H);
 260             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 261             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 262             vvdw6            = _mm256_mul_ps(c6_00,VV);
 263             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 264             fvdw6            = _mm256_mul_ps(c6_00,FF);
 265
 266             /* CUBIC SPLINE TABLE REPULSION */
 267             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 268             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 269             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 270                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 271             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 272                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 273             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 274                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 275             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 276                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 277             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 278             Heps             = _mm256_mul_ps(vfeps,H);
 279             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 280             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 281             vvdw12           = _mm256_mul_ps(c12_00,VV);
 282             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 283             fvdw12           = _mm256_mul_ps(c12_00,FF);
 284             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 285             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             velecsum         = _mm256_add_ps(velecsum,velec);
 289             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 290
 291             fscal            = _mm256_add_ps(felec,fvdw);
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm256_mul_ps(fscal,dx00);
 295             ty               = _mm256_mul_ps(fscal,dy00);
 296             tz               = _mm256_mul_ps(fscal,dz00);
 297
 298             /* Update vectorial force */
 299             fix0             = _mm256_add_ps(fix0,tx);
 300             fiy0             = _mm256_add_ps(fiy0,ty);
 301             fiz0             = _mm256_add_ps(fiz0,tz);
 302
 303             fjptrA             = f+j_coord_offsetA;
 304             fjptrB             = f+j_coord_offsetB;
 305             fjptrC             = f+j_coord_offsetC;
 306             fjptrD             = f+j_coord_offsetD;
 307             fjptrE             = f+j_coord_offsetE;
 308             fjptrF             = f+j_coord_offsetF;
 309             fjptrG             = f+j_coord_offsetG;
 310             fjptrH             = f+j_coord_offsetH;
 311             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 312
 313             /* Inner loop uses 62 flops */
 314         }
 315
 316         if(jidx<j_index_end)
 317         {
 318
 319             /* Get j neighbor index, and coordinate index */
 320             jnrlistA         = jjnr[jidx];
 321             jnrlistB         = jjnr[jidx+1];
 322             jnrlistC         = jjnr[jidx+2];
 323             jnrlistD         = jjnr[jidx+3];
 324             jnrlistE         = jjnr[jidx+4];
 325             jnrlistF         = jjnr[jidx+5];
 326             jnrlistG         = jjnr[jidx+6];
 327             jnrlistH         = jjnr[jidx+7];
 328             /* Sign of each element will be negative for non-real atoms.
 329              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 330              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 331              */
 332             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 333                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 334
 335             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 336             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 337             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 338             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 339             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 340             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 341             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 342             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 343             j_coord_offsetA  = DIM*jnrA;
 344             j_coord_offsetB  = DIM*jnrB;
 345             j_coord_offsetC  = DIM*jnrC;
 346             j_coord_offsetD  = DIM*jnrD;
 347             j_coord_offsetE  = DIM*jnrE;
 348             j_coord_offsetF  = DIM*jnrF;
 349             j_coord_offsetG  = DIM*jnrG;
 350             j_coord_offsetH  = DIM*jnrH;
 351
 352             /* load j atom coordinates */
 353             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 354                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 355                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 356                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 357                                                  &jx0,&jy0,&jz0);
 358
 359             /* Calculate displacement vector */
 360             dx00             = _mm256_sub_ps(ix0,jx0);
 361             dy00             = _mm256_sub_ps(iy0,jy0);
 362             dz00             = _mm256_sub_ps(iz0,jz0);
 363
 364             /* Calculate squared distance and things based on it */
 365             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 366
 367             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 368
 369             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 370
 371             /* Load parameters for j particles */
 372             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 373                                                                  charge+jnrC+0,charge+jnrD+0,
 374                                                                  charge+jnrE+0,charge+jnrF+0,
 375                                                                  charge+jnrG+0,charge+jnrH+0);
 376             vdwjidx0A        = 2*vdwtype[jnrA+0];
 377             vdwjidx0B        = 2*vdwtype[jnrB+0];
 378             vdwjidx0C        = 2*vdwtype[jnrC+0];
 379             vdwjidx0D        = 2*vdwtype[jnrD+0];
 380             vdwjidx0E        = 2*vdwtype[jnrE+0];
 381             vdwjidx0F        = 2*vdwtype[jnrF+0];
 382             vdwjidx0G        = 2*vdwtype[jnrG+0];
 383             vdwjidx0H        = 2*vdwtype[jnrH+0];
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             r00              = _mm256_mul_ps(rsq00,rinv00);
 390             r00              = _mm256_andnot_ps(dummy_mask,r00);
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq00             = _mm256_mul_ps(iq0,jq0);
 394             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 395                                             vdwioffsetptr0+vdwjidx0B,
 396                                             vdwioffsetptr0+vdwjidx0C,
 397                                             vdwioffsetptr0+vdwjidx0D,
 398                                             vdwioffsetptr0+vdwjidx0E,
 399                                             vdwioffsetptr0+vdwjidx0F,
 400                                             vdwioffsetptr0+vdwjidx0G,
 401                                             vdwioffsetptr0+vdwjidx0H,
 402                                             &c6_00,&c12_00);
 403
 404             /* Calculate table index by multiplying r with table scale and truncate to integer */
 405             rt               = _mm256_mul_ps(r00,vftabscale);
 406             vfitab           = _mm256_cvttps_epi32(rt);
 407             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 408             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 409             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 410             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 411             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 412             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 413
 414             /* COULOMB ELECTROSTATICS */
 415             velec            = _mm256_mul_ps(qq00,rinv00);
 416             felec            = _mm256_mul_ps(velec,rinvsq00);
 417
 418             /* CUBIC SPLINE TABLE DISPERSION */
 419             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 420                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 421             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 422                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 423             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 424                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 425             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 426                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 427             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 428             Heps             = _mm256_mul_ps(vfeps,H);
 429             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 430             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 431             vvdw6            = _mm256_mul_ps(c6_00,VV);
 432             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 433             fvdw6            = _mm256_mul_ps(c6_00,FF);
 434
 435             /* CUBIC SPLINE TABLE REPULSION */
 436             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 437             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 438             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 439                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 440             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 441                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 442             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 443                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 444             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 445                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 446             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 447             Heps             = _mm256_mul_ps(vfeps,H);
 448             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 449             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 450             vvdw12           = _mm256_mul_ps(c12_00,VV);
 451             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 452             fvdw12           = _mm256_mul_ps(c12_00,FF);
 453             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 454             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm256_andnot_ps(dummy_mask,velec);
 458             velecsum         = _mm256_add_ps(velecsum,velec);
 459             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 460             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 461
 462             fscal            = _mm256_add_ps(felec,fvdw);
 463
 464             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm256_mul_ps(fscal,dx00);
 468             ty               = _mm256_mul_ps(fscal,dy00);
 469             tz               = _mm256_mul_ps(fscal,dz00);
 470
 471             /* Update vectorial force */
 472             fix0             = _mm256_add_ps(fix0,tx);
 473             fiy0             = _mm256_add_ps(fiy0,ty);
 474             fiz0             = _mm256_add_ps(fiz0,tz);
 475
 476             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 477             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 478             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 479             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 480             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 481             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 482             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 483             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 484             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 485
 486             /* Inner loop uses 63 flops */
 487         }
 488
 489         /* End of innermost loop */
 490
 491         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 492                                                  f+i_coord_offset,fshift+i_shift_offset);
 493
 494         ggid                        = gid[iidx];
 495         /* Update potential energies */
 496         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 497         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 498
 499         /* Increment number of inner iterations */
 500         inneriter                  += j_index_end - j_index_start;
 501
 502         /* Outer loop uses 9 flops */
 503     }
 504
 505     /* Increment number of outer iterations */
 506     outeriter        += nri;
 507
 508     /* Update outer/inner flops */
 509
 510     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*63);
 511 }
 512 /*
 513  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_256_single
 514  * Electrostatics interaction: Coulomb
 515  * VdW interaction:            CubicSplineTable
 516  * Geometry:                   Particle-Particle
 517  * Calculate force/pot:        Force
 518  */
 519 void
 520 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_256_single
 521                     (t_nblist * gmx_restrict                nlist,
 522                      rvec * gmx_restrict                    xx,
 523                      rvec * gmx_restrict                    ff,
 524                      t_forcerec * gmx_restrict              fr,
 525                      t_mdatoms * gmx_restrict               mdatoms,
 526                      nb_kernel_data_t * gmx_restrict        kernel_data,
 527                      t_nrnb * gmx_restrict                  nrnb)
 528 {
 529     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 530      * just 0 for non-waters.
 531      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 532      * jnr indices corresponding to data put in the four positions in the SIMD register.
 533      */
 534     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 535     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 536     int              jnrA,jnrB,jnrC,jnrD;
 537     int              jnrE,jnrF,jnrG,jnrH;
 538     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 539     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 540     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 541     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 542     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 543     real             rcutoff_scalar;
 544     real             *shiftvec,*fshift,*x,*f;
 545     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 546     real             scratch[4*DIM];
 547     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 548     real *           vdwioffsetptr0;
 549     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 550     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 551     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 552     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 553     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 554     real             *charge;
 555     int              nvdwtype;
 556     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 557     int              *vdwtype;
 558     real             *vdwparam;
 559     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 560     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 561     __m256i          vfitab;
 562     __m128i          vfitab_lo,vfitab_hi;
 563     __m128i          ifour       = _mm_set1_epi32(4);
 564     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 565     real             *vftab;
 566     __m256           dummy_mask,cutoff_mask;
 567     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 568     __m256           one     = _mm256_set1_ps(1.0);
 569     __m256           two     = _mm256_set1_ps(2.0);
 570     x                = xx[0];
 571     f                = ff[0];
 572
 573     nri              = nlist->nri;
 574     iinr             = nlist->iinr;
 575     jindex           = nlist->jindex;
 576     jjnr             = nlist->jjnr;
 577     shiftidx         = nlist->shift;
 578     gid              = nlist->gid;
 579     shiftvec         = fr->shift_vec[0];
 580     fshift           = fr->fshift[0];
 581     facel            = _mm256_set1_ps(fr->epsfac);
 582     charge           = mdatoms->chargeA;
 583     nvdwtype         = fr->ntype;
 584     vdwparam         = fr->nbfp;
 585     vdwtype          = mdatoms->typeA;
 586
 587     vftab            = kernel_data->table_vdw->data;
 588     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 589
 590     /* Avoid stupid compiler warnings */
 591     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 592     j_coord_offsetA = 0;
 593     j_coord_offsetB = 0;
 594     j_coord_offsetC = 0;
 595     j_coord_offsetD = 0;
 596     j_coord_offsetE = 0;
 597     j_coord_offsetF = 0;
 598     j_coord_offsetG = 0;
 599     j_coord_offsetH = 0;
 600
 601     outeriter        = 0;
 602     inneriter        = 0;
 603
 604     for(iidx=0;iidx<4*DIM;iidx++)
 605     {
 606         scratch[iidx] = 0.0;
 607     }
 608
 609     /* Start outer loop over neighborlists */
 610     for(iidx=0; iidx<nri; iidx++)
 611     {
 612         /* Load shift vector for this list */
 613         i_shift_offset   = DIM*shiftidx[iidx];
 614
 615         /* Load limits for loop over neighbors */
 616         j_index_start    = jindex[iidx];
 617         j_index_end      = jindex[iidx+1];
 618
 619         /* Get outer coordinate index */
 620         inr              = iinr[iidx];
 621         i_coord_offset   = DIM*inr;
 622
 623         /* Load i particle coords and add shift vector */
 624         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 625
 626         fix0             = _mm256_setzero_ps();
 627         fiy0             = _mm256_setzero_ps();
 628         fiz0             = _mm256_setzero_ps();
 629
 630         /* Load parameters for i particles */
 631         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 632         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 633
 634         /* Start inner kernel loop */
 635         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 636         {
 637
 638             /* Get j neighbor index, and coordinate index */
 639             jnrA             = jjnr[jidx];
 640             jnrB             = jjnr[jidx+1];
 641             jnrC             = jjnr[jidx+2];
 642             jnrD             = jjnr[jidx+3];
 643             jnrE             = jjnr[jidx+4];
 644             jnrF             = jjnr[jidx+5];
 645             jnrG             = jjnr[jidx+6];
 646             jnrH             = jjnr[jidx+7];
 647             j_coord_offsetA  = DIM*jnrA;
 648             j_coord_offsetB  = DIM*jnrB;
 649             j_coord_offsetC  = DIM*jnrC;
 650             j_coord_offsetD  = DIM*jnrD;
 651             j_coord_offsetE  = DIM*jnrE;
 652             j_coord_offsetF  = DIM*jnrF;
 653             j_coord_offsetG  = DIM*jnrG;
 654             j_coord_offsetH  = DIM*jnrH;
 655
 656             /* load j atom coordinates */
 657             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 658                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 659                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 660                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 661                                                  &jx0,&jy0,&jz0);
 662
 663             /* Calculate displacement vector */
 664             dx00             = _mm256_sub_ps(ix0,jx0);
 665             dy00             = _mm256_sub_ps(iy0,jy0);
 666             dz00             = _mm256_sub_ps(iz0,jz0);
 667
 668             /* Calculate squared distance and things based on it */
 669             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 670
 671             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 672
 673             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 674
 675             /* Load parameters for j particles */
 676             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 677                                                                  charge+jnrC+0,charge+jnrD+0,
 678                                                                  charge+jnrE+0,charge+jnrF+0,
 679                                                                  charge+jnrG+0,charge+jnrH+0);
 680             vdwjidx0A        = 2*vdwtype[jnrA+0];
 681             vdwjidx0B        = 2*vdwtype[jnrB+0];
 682             vdwjidx0C        = 2*vdwtype[jnrC+0];
 683             vdwjidx0D        = 2*vdwtype[jnrD+0];
 684             vdwjidx0E        = 2*vdwtype[jnrE+0];
 685             vdwjidx0F        = 2*vdwtype[jnrF+0];
 686             vdwjidx0G        = 2*vdwtype[jnrG+0];
 687             vdwjidx0H        = 2*vdwtype[jnrH+0];
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             r00              = _mm256_mul_ps(rsq00,rinv00);
 694
 695             /* Compute parameters for interactions between i and j atoms */
 696             qq00             = _mm256_mul_ps(iq0,jq0);
 697             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 698                                             vdwioffsetptr0+vdwjidx0B,
 699                                             vdwioffsetptr0+vdwjidx0C,
 700                                             vdwioffsetptr0+vdwjidx0D,
 701                                             vdwioffsetptr0+vdwjidx0E,
 702                                             vdwioffsetptr0+vdwjidx0F,
 703                                             vdwioffsetptr0+vdwjidx0G,
 704                                             vdwioffsetptr0+vdwjidx0H,
 705                                             &c6_00,&c12_00);
 706
 707             /* Calculate table index by multiplying r with table scale and truncate to integer */
 708             rt               = _mm256_mul_ps(r00,vftabscale);
 709             vfitab           = _mm256_cvttps_epi32(rt);
 710             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 711             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 712             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 713             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 714             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 715             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 716
 717             /* COULOMB ELECTROSTATICS */
 718             velec            = _mm256_mul_ps(qq00,rinv00);
 719             felec            = _mm256_mul_ps(velec,rinvsq00);
 720
 721             /* CUBIC SPLINE TABLE DISPERSION */
 722             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 723                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 724             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 725                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 726             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 727                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 728             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 729                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 730             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 731             Heps             = _mm256_mul_ps(vfeps,H);
 732             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 733             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 734             fvdw6            = _mm256_mul_ps(c6_00,FF);
 735
 736             /* CUBIC SPLINE TABLE REPULSION */
 737             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 738             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 739             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 740                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 741             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 742                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 743             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 744                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 745             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 746                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 747             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 748             Heps             = _mm256_mul_ps(vfeps,H);
 749             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 750             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 751             fvdw12           = _mm256_mul_ps(c12_00,FF);
 752             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 753
 754             fscal            = _mm256_add_ps(felec,fvdw);
 755
 756             /* Calculate temporary vectorial force */
 757             tx               = _mm256_mul_ps(fscal,dx00);
 758             ty               = _mm256_mul_ps(fscal,dy00);
 759             tz               = _mm256_mul_ps(fscal,dz00);
 760
 761             /* Update vectorial force */
 762             fix0             = _mm256_add_ps(fix0,tx);
 763             fiy0             = _mm256_add_ps(fiy0,ty);
 764             fiz0             = _mm256_add_ps(fiz0,tz);
 765
 766             fjptrA             = f+j_coord_offsetA;
 767             fjptrB             = f+j_coord_offsetB;
 768             fjptrC             = f+j_coord_offsetC;
 769             fjptrD             = f+j_coord_offsetD;
 770             fjptrE             = f+j_coord_offsetE;
 771             fjptrF             = f+j_coord_offsetF;
 772             fjptrG             = f+j_coord_offsetG;
 773             fjptrH             = f+j_coord_offsetH;
 774             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 775
 776             /* Inner loop uses 53 flops */
 777         }
 778
 779         if(jidx<j_index_end)
 780         {
 781
 782             /* Get j neighbor index, and coordinate index */
 783             jnrlistA         = jjnr[jidx];
 784             jnrlistB         = jjnr[jidx+1];
 785             jnrlistC         = jjnr[jidx+2];
 786             jnrlistD         = jjnr[jidx+3];
 787             jnrlistE         = jjnr[jidx+4];
 788             jnrlistF         = jjnr[jidx+5];
 789             jnrlistG         = jjnr[jidx+6];
 790             jnrlistH         = jjnr[jidx+7];
 791             /* Sign of each element will be negative for non-real atoms.
 792              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 793              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 794              */
 795             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 796                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 797
 798             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 799             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 800             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 801             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 802             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 803             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 804             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 805             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 806             j_coord_offsetA  = DIM*jnrA;
 807             j_coord_offsetB  = DIM*jnrB;
 808             j_coord_offsetC  = DIM*jnrC;
 809             j_coord_offsetD  = DIM*jnrD;
 810             j_coord_offsetE  = DIM*jnrE;
 811             j_coord_offsetF  = DIM*jnrF;
 812             j_coord_offsetG  = DIM*jnrG;
 813             j_coord_offsetH  = DIM*jnrH;
 814
 815             /* load j atom coordinates */
 816             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 817                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 818                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 819                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 820                                                  &jx0,&jy0,&jz0);
 821
 822             /* Calculate displacement vector */
 823             dx00             = _mm256_sub_ps(ix0,jx0);
 824             dy00             = _mm256_sub_ps(iy0,jy0);
 825             dz00             = _mm256_sub_ps(iz0,jz0);
 826
 827             /* Calculate squared distance and things based on it */
 828             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 829
 830             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 831
 832             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 833
 834             /* Load parameters for j particles */
 835             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 836                                                                  charge+jnrC+0,charge+jnrD+0,
 837                                                                  charge+jnrE+0,charge+jnrF+0,
 838                                                                  charge+jnrG+0,charge+jnrH+0);
 839             vdwjidx0A        = 2*vdwtype[jnrA+0];
 840             vdwjidx0B        = 2*vdwtype[jnrB+0];
 841             vdwjidx0C        = 2*vdwtype[jnrC+0];
 842             vdwjidx0D        = 2*vdwtype[jnrD+0];
 843             vdwjidx0E        = 2*vdwtype[jnrE+0];
 844             vdwjidx0F        = 2*vdwtype[jnrF+0];
 845             vdwjidx0G        = 2*vdwtype[jnrG+0];
 846             vdwjidx0H        = 2*vdwtype[jnrH+0];
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             r00              = _mm256_mul_ps(rsq00,rinv00);
 853             r00              = _mm256_andnot_ps(dummy_mask,r00);
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             qq00             = _mm256_mul_ps(iq0,jq0);
 857             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 858                                             vdwioffsetptr0+vdwjidx0B,
 859                                             vdwioffsetptr0+vdwjidx0C,
 860                                             vdwioffsetptr0+vdwjidx0D,
 861                                             vdwioffsetptr0+vdwjidx0E,
 862                                             vdwioffsetptr0+vdwjidx0F,
 863                                             vdwioffsetptr0+vdwjidx0G,
 864                                             vdwioffsetptr0+vdwjidx0H,
 865                                             &c6_00,&c12_00);
 866
 867             /* Calculate table index by multiplying r with table scale and truncate to integer */
 868             rt               = _mm256_mul_ps(r00,vftabscale);
 869             vfitab           = _mm256_cvttps_epi32(rt);
 870             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 871             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 872             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 873             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 874             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 875             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 876
 877             /* COULOMB ELECTROSTATICS */
 878             velec            = _mm256_mul_ps(qq00,rinv00);
 879             felec            = _mm256_mul_ps(velec,rinvsq00);
 880
 881             /* CUBIC SPLINE TABLE DISPERSION */
 882             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 883                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 884             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 885                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 886             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 887                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 888             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 889                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 890             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 891             Heps             = _mm256_mul_ps(vfeps,H);
 892             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 893             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 894             fvdw6            = _mm256_mul_ps(c6_00,FF);
 895
 896             /* CUBIC SPLINE TABLE REPULSION */
 897             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 898             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 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             fvdw12           = _mm256_mul_ps(c12_00,FF);
 912             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 913
 914             fscal            = _mm256_add_ps(felec,fvdw);
 915
 916             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 917
 918             /* Calculate temporary vectorial force */
 919             tx               = _mm256_mul_ps(fscal,dx00);
 920             ty               = _mm256_mul_ps(fscal,dy00);
 921             tz               = _mm256_mul_ps(fscal,dz00);
 922
 923             /* Update vectorial force */
 924             fix0             = _mm256_add_ps(fix0,tx);
 925             fiy0             = _mm256_add_ps(fiy0,ty);
 926             fiz0             = _mm256_add_ps(fiz0,tz);
 927
 928             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 929             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 930             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 931             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 932             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 933             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 934             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 935             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 936             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 937
 938             /* Inner loop uses 54 flops */
 939         }
 940
 941         /* End of innermost loop */
 942
 943         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 944                                                  f+i_coord_offset,fshift+i_shift_offset);
 945
 946         /* Increment number of inner iterations */
 947         inneriter                  += j_index_end - j_index_start;
 948
 949         /* Outer loop uses 7 flops */
 950     }
 951
 952     /* Increment number of outer iterations */
 953     outeriter        += nri;
 954
 955     /* Update outer/inner flops */
 956
 957     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);
 958 }