src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_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_elec_vdw->data;
 112     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_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             /* Load parameters for j particles */
 202             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 203                                                                  charge+jnrC+0,charge+jnrD+0,
 204                                                                  charge+jnrE+0,charge+jnrF+0,
 205                                                                  charge+jnrG+0,charge+jnrH+0);
 206             vdwjidx0A        = 2*vdwtype[jnrA+0];
 207             vdwjidx0B        = 2*vdwtype[jnrB+0];
 208             vdwjidx0C        = 2*vdwtype[jnrC+0];
 209             vdwjidx0D        = 2*vdwtype[jnrD+0];
 210             vdwjidx0E        = 2*vdwtype[jnrE+0];
 211             vdwjidx0F        = 2*vdwtype[jnrF+0];
 212             vdwjidx0G        = 2*vdwtype[jnrG+0];
 213             vdwjidx0H        = 2*vdwtype[jnrH+0];
 214
 215             /**************************
 216              * CALCULATE INTERACTIONS *
 217              **************************/
 218
 219             r00              = _mm256_mul_ps(rsq00,rinv00);
 220
 221             /* Compute parameters for interactions between i and j atoms */
 222             qq00             = _mm256_mul_ps(iq0,jq0);
 223             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 224                                             vdwioffsetptr0+vdwjidx0B,
 225                                             vdwioffsetptr0+vdwjidx0C,
 226                                             vdwioffsetptr0+vdwjidx0D,
 227                                             vdwioffsetptr0+vdwjidx0E,
 228                                             vdwioffsetptr0+vdwjidx0F,
 229                                             vdwioffsetptr0+vdwjidx0G,
 230                                             vdwioffsetptr0+vdwjidx0H,
 231                                             &c6_00,&c12_00);
 232
 233             /* Calculate table index by multiplying r with table scale and truncate to integer */
 234             rt               = _mm256_mul_ps(r00,vftabscale);
 235             vfitab           = _mm256_cvttps_epi32(rt);
 236             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 237             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 238             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 239             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 240             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 241             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 242
 243             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 244             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 245                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 246             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 247                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 248             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 249                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 250             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 251                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 252             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 253             Heps             = _mm256_mul_ps(vfeps,H);
 254             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 255             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 256             velec            = _mm256_mul_ps(qq00,VV);
 257             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 258             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 259
 260             /* CUBIC SPLINE TABLE DISPERSION */
 261             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 262             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 263             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 264                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 265             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 266                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 267             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 268                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 269             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 270                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 271             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 272             Heps             = _mm256_mul_ps(vfeps,H);
 273             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 274             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 275             vvdw6            = _mm256_mul_ps(c6_00,VV);
 276             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 277             fvdw6            = _mm256_mul_ps(c6_00,FF);
 278
 279             /* CUBIC SPLINE TABLE REPULSION */
 280             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 281             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 282             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 283                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 284             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 285                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 286             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 287                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 288             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 289                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 290             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 291             Heps             = _mm256_mul_ps(vfeps,H);
 292             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 293             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 294             vvdw12           = _mm256_mul_ps(c12_00,VV);
 295             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 296             fvdw12           = _mm256_mul_ps(c12_00,FF);
 297             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 298             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm256_add_ps(velecsum,velec);
 302             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm256_add_ps(felec,fvdw);
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm256_mul_ps(fscal,dx00);
 308             ty               = _mm256_mul_ps(fscal,dy00);
 309             tz               = _mm256_mul_ps(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm256_add_ps(fix0,tx);
 313             fiy0             = _mm256_add_ps(fiy0,ty);
 314             fiz0             = _mm256_add_ps(fiz0,tz);
 315
 316             fjptrA             = f+j_coord_offsetA;
 317             fjptrB             = f+j_coord_offsetB;
 318             fjptrC             = f+j_coord_offsetC;
 319             fjptrD             = f+j_coord_offsetD;
 320             fjptrE             = f+j_coord_offsetE;
 321             fjptrF             = f+j_coord_offsetF;
 322             fjptrG             = f+j_coord_offsetG;
 323             fjptrH             = f+j_coord_offsetH;
 324             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 325
 326             /* Inner loop uses 73 flops */
 327         }
 328
 329         if(jidx<j_index_end)
 330         {
 331
 332             /* Get j neighbor index, and coordinate index */
 333             jnrlistA         = jjnr[jidx];
 334             jnrlistB         = jjnr[jidx+1];
 335             jnrlistC         = jjnr[jidx+2];
 336             jnrlistD         = jjnr[jidx+3];
 337             jnrlistE         = jjnr[jidx+4];
 338             jnrlistF         = jjnr[jidx+5];
 339             jnrlistG         = jjnr[jidx+6];
 340             jnrlistH         = jjnr[jidx+7];
 341             /* Sign of each element will be negative for non-real atoms.
 342              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 343              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 344              */
 345             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 346                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 347
 348             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 349             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 350             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 351             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 352             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 353             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 354             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 355             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 356             j_coord_offsetA  = DIM*jnrA;
 357             j_coord_offsetB  = DIM*jnrB;
 358             j_coord_offsetC  = DIM*jnrC;
 359             j_coord_offsetD  = DIM*jnrD;
 360             j_coord_offsetE  = DIM*jnrE;
 361             j_coord_offsetF  = DIM*jnrF;
 362             j_coord_offsetG  = DIM*jnrG;
 363             j_coord_offsetH  = DIM*jnrH;
 364
 365             /* load j atom coordinates */
 366             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 367                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 368                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 369                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 370                                                  &jx0,&jy0,&jz0);
 371
 372             /* Calculate displacement vector */
 373             dx00             = _mm256_sub_ps(ix0,jx0);
 374             dy00             = _mm256_sub_ps(iy0,jy0);
 375             dz00             = _mm256_sub_ps(iz0,jz0);
 376
 377             /* Calculate squared distance and things based on it */
 378             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 379
 380             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 381
 382             /* Load parameters for j particles */
 383             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 384                                                                  charge+jnrC+0,charge+jnrD+0,
 385                                                                  charge+jnrE+0,charge+jnrF+0,
 386                                                                  charge+jnrG+0,charge+jnrH+0);
 387             vdwjidx0A        = 2*vdwtype[jnrA+0];
 388             vdwjidx0B        = 2*vdwtype[jnrB+0];
 389             vdwjidx0C        = 2*vdwtype[jnrC+0];
 390             vdwjidx0D        = 2*vdwtype[jnrD+0];
 391             vdwjidx0E        = 2*vdwtype[jnrE+0];
 392             vdwjidx0F        = 2*vdwtype[jnrF+0];
 393             vdwjidx0G        = 2*vdwtype[jnrG+0];
 394             vdwjidx0H        = 2*vdwtype[jnrH+0];
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             r00              = _mm256_mul_ps(rsq00,rinv00);
 401             r00              = _mm256_andnot_ps(dummy_mask,r00);
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq00             = _mm256_mul_ps(iq0,jq0);
 405             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 406                                             vdwioffsetptr0+vdwjidx0B,
 407                                             vdwioffsetptr0+vdwjidx0C,
 408                                             vdwioffsetptr0+vdwjidx0D,
 409                                             vdwioffsetptr0+vdwjidx0E,
 410                                             vdwioffsetptr0+vdwjidx0F,
 411                                             vdwioffsetptr0+vdwjidx0G,
 412                                             vdwioffsetptr0+vdwjidx0H,
 413                                             &c6_00,&c12_00);
 414
 415             /* Calculate table index by multiplying r with table scale and truncate to integer */
 416             rt               = _mm256_mul_ps(r00,vftabscale);
 417             vfitab           = _mm256_cvttps_epi32(rt);
 418             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 419             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 420             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 421             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 422             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 423             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 424
 425             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 426             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 427                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 428             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 429                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 430             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 431                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 432             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 433                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 434             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 435             Heps             = _mm256_mul_ps(vfeps,H);
 436             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 437             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 438             velec            = _mm256_mul_ps(qq00,VV);
 439             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 440             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 441
 442             /* CUBIC SPLINE TABLE DISPERSION */
 443             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 444             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 445             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 446                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 447             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 448                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 449             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 450                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 451             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 452                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 453             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 454             Heps             = _mm256_mul_ps(vfeps,H);
 455             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 456             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 457             vvdw6            = _mm256_mul_ps(c6_00,VV);
 458             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 459             fvdw6            = _mm256_mul_ps(c6_00,FF);
 460
 461             /* CUBIC SPLINE TABLE REPULSION */
 462             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 463             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 464             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 465                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 466             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 467                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 468             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 469                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 470             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 471                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 472             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 473             Heps             = _mm256_mul_ps(vfeps,H);
 474             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 475             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 476             vvdw12           = _mm256_mul_ps(c12_00,VV);
 477             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 478             fvdw12           = _mm256_mul_ps(c12_00,FF);
 479             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 480             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm256_andnot_ps(dummy_mask,velec);
 484             velecsum         = _mm256_add_ps(velecsum,velec);
 485             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 486             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 487
 488             fscal            = _mm256_add_ps(felec,fvdw);
 489
 490             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 491
 492             /* Calculate temporary vectorial force */
 493             tx               = _mm256_mul_ps(fscal,dx00);
 494             ty               = _mm256_mul_ps(fscal,dy00);
 495             tz               = _mm256_mul_ps(fscal,dz00);
 496
 497             /* Update vectorial force */
 498             fix0             = _mm256_add_ps(fix0,tx);
 499             fiy0             = _mm256_add_ps(fiy0,ty);
 500             fiz0             = _mm256_add_ps(fiz0,tz);
 501
 502             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 503             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 504             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 505             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 506             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 507             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 508             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 509             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 510             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 511
 512             /* Inner loop uses 74 flops */
 513         }
 514
 515         /* End of innermost loop */
 516
 517         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 518                                                  f+i_coord_offset,fshift+i_shift_offset);
 519
 520         ggid                        = gid[iidx];
 521         /* Update potential energies */
 522         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 523         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 524
 525         /* Increment number of inner iterations */
 526         inneriter                  += j_index_end - j_index_start;
 527
 528         /* Outer loop uses 9 flops */
 529     }
 530
 531     /* Increment number of outer iterations */
 532     outeriter        += nri;
 533
 534     /* Update outer/inner flops */
 535
 536     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*74);
 537 }
 538 /*
 539  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_single
 540  * Electrostatics interaction: CubicSplineTable
 541  * VdW interaction:            CubicSplineTable
 542  * Geometry:                   Particle-Particle
 543  * Calculate force/pot:        Force
 544  */
 545 void
 546 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_256_single
 547                     (t_nblist * gmx_restrict                nlist,
 548                      rvec * gmx_restrict                    xx,
 549                      rvec * gmx_restrict                    ff,
 550                      t_forcerec * gmx_restrict              fr,
 551                      t_mdatoms * gmx_restrict               mdatoms,
 552                      nb_kernel_data_t * gmx_restrict        kernel_data,
 553                      t_nrnb * gmx_restrict                  nrnb)
 554 {
 555     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 556      * just 0 for non-waters.
 557      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 558      * jnr indices corresponding to data put in the four positions in the SIMD register.
 559      */
 560     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 561     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 562     int              jnrA,jnrB,jnrC,jnrD;
 563     int              jnrE,jnrF,jnrG,jnrH;
 564     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 565     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 566     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 567     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 568     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 569     real             rcutoff_scalar;
 570     real             *shiftvec,*fshift,*x,*f;
 571     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 572     real             scratch[4*DIM];
 573     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 574     real *           vdwioffsetptr0;
 575     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 576     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 577     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 578     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 579     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 580     real             *charge;
 581     int              nvdwtype;
 582     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 583     int              *vdwtype;
 584     real             *vdwparam;
 585     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 586     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 587     __m256i          vfitab;
 588     __m128i          vfitab_lo,vfitab_hi;
 589     __m128i          ifour       = _mm_set1_epi32(4);
 590     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 591     real             *vftab;
 592     __m256           dummy_mask,cutoff_mask;
 593     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 594     __m256           one     = _mm256_set1_ps(1.0);
 595     __m256           two     = _mm256_set1_ps(2.0);
 596     x                = xx[0];
 597     f                = ff[0];
 598
 599     nri              = nlist->nri;
 600     iinr             = nlist->iinr;
 601     jindex           = nlist->jindex;
 602     jjnr             = nlist->jjnr;
 603     shiftidx         = nlist->shift;
 604     gid              = nlist->gid;
 605     shiftvec         = fr->shift_vec[0];
 606     fshift           = fr->fshift[0];
 607     facel            = _mm256_set1_ps(fr->epsfac);
 608     charge           = mdatoms->chargeA;
 609     nvdwtype         = fr->ntype;
 610     vdwparam         = fr->nbfp;
 611     vdwtype          = mdatoms->typeA;
 612
 613     vftab            = kernel_data->table_elec_vdw->data;
 614     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 615
 616     /* Avoid stupid compiler warnings */
 617     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 618     j_coord_offsetA = 0;
 619     j_coord_offsetB = 0;
 620     j_coord_offsetC = 0;
 621     j_coord_offsetD = 0;
 622     j_coord_offsetE = 0;
 623     j_coord_offsetF = 0;
 624     j_coord_offsetG = 0;
 625     j_coord_offsetH = 0;
 626
 627     outeriter        = 0;
 628     inneriter        = 0;
 629
 630     for(iidx=0;iidx<4*DIM;iidx++)
 631     {
 632         scratch[iidx] = 0.0;
 633     }
 634
 635     /* Start outer loop over neighborlists */
 636     for(iidx=0; iidx<nri; iidx++)
 637     {
 638         /* Load shift vector for this list */
 639         i_shift_offset   = DIM*shiftidx[iidx];
 640
 641         /* Load limits for loop over neighbors */
 642         j_index_start    = jindex[iidx];
 643         j_index_end      = jindex[iidx+1];
 644
 645         /* Get outer coordinate index */
 646         inr              = iinr[iidx];
 647         i_coord_offset   = DIM*inr;
 648
 649         /* Load i particle coords and add shift vector */
 650         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 651
 652         fix0             = _mm256_setzero_ps();
 653         fiy0             = _mm256_setzero_ps();
 654         fiz0             = _mm256_setzero_ps();
 655
 656         /* Load parameters for i particles */
 657         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 658         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 659
 660         /* Start inner kernel loop */
 661         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 662         {
 663
 664             /* Get j neighbor index, and coordinate index */
 665             jnrA             = jjnr[jidx];
 666             jnrB             = jjnr[jidx+1];
 667             jnrC             = jjnr[jidx+2];
 668             jnrD             = jjnr[jidx+3];
 669             jnrE             = jjnr[jidx+4];
 670             jnrF             = jjnr[jidx+5];
 671             jnrG             = jjnr[jidx+6];
 672             jnrH             = jjnr[jidx+7];
 673             j_coord_offsetA  = DIM*jnrA;
 674             j_coord_offsetB  = DIM*jnrB;
 675             j_coord_offsetC  = DIM*jnrC;
 676             j_coord_offsetD  = DIM*jnrD;
 677             j_coord_offsetE  = DIM*jnrE;
 678             j_coord_offsetF  = DIM*jnrF;
 679             j_coord_offsetG  = DIM*jnrG;
 680             j_coord_offsetH  = DIM*jnrH;
 681
 682             /* load j atom coordinates */
 683             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 684                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 685                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 686                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 687                                                  &jx0,&jy0,&jz0);
 688
 689             /* Calculate displacement vector */
 690             dx00             = _mm256_sub_ps(ix0,jx0);
 691             dy00             = _mm256_sub_ps(iy0,jy0);
 692             dz00             = _mm256_sub_ps(iz0,jz0);
 693
 694             /* Calculate squared distance and things based on it */
 695             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 696
 697             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 698
 699             /* Load parameters for j particles */
 700             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 701                                                                  charge+jnrC+0,charge+jnrD+0,
 702                                                                  charge+jnrE+0,charge+jnrF+0,
 703                                                                  charge+jnrG+0,charge+jnrH+0);
 704             vdwjidx0A        = 2*vdwtype[jnrA+0];
 705             vdwjidx0B        = 2*vdwtype[jnrB+0];
 706             vdwjidx0C        = 2*vdwtype[jnrC+0];
 707             vdwjidx0D        = 2*vdwtype[jnrD+0];
 708             vdwjidx0E        = 2*vdwtype[jnrE+0];
 709             vdwjidx0F        = 2*vdwtype[jnrF+0];
 710             vdwjidx0G        = 2*vdwtype[jnrG+0];
 711             vdwjidx0H        = 2*vdwtype[jnrH+0];
 712
 713             /**************************
 714              * CALCULATE INTERACTIONS *
 715              **************************/
 716
 717             r00              = _mm256_mul_ps(rsq00,rinv00);
 718
 719             /* Compute parameters for interactions between i and j atoms */
 720             qq00             = _mm256_mul_ps(iq0,jq0);
 721             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 722                                             vdwioffsetptr0+vdwjidx0B,
 723                                             vdwioffsetptr0+vdwjidx0C,
 724                                             vdwioffsetptr0+vdwjidx0D,
 725                                             vdwioffsetptr0+vdwjidx0E,
 726                                             vdwioffsetptr0+vdwjidx0F,
 727                                             vdwioffsetptr0+vdwjidx0G,
 728                                             vdwioffsetptr0+vdwjidx0H,
 729                                             &c6_00,&c12_00);
 730
 731             /* Calculate table index by multiplying r with table scale and truncate to integer */
 732             rt               = _mm256_mul_ps(r00,vftabscale);
 733             vfitab           = _mm256_cvttps_epi32(rt);
 734             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 735             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 736             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 737             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 738             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 739             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 740
 741             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 742             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 743                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 744             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 745                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 746             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 747                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 748             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 749                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 750             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 751             Heps             = _mm256_mul_ps(vfeps,H);
 752             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 753             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 754             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 755
 756             /* CUBIC SPLINE TABLE DISPERSION */
 757             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 758             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 759             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 760                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 761             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 762                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 763             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 764                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 765             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 766                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 767             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 768             Heps             = _mm256_mul_ps(vfeps,H);
 769             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 770             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 771             fvdw6            = _mm256_mul_ps(c6_00,FF);
 772
 773             /* CUBIC SPLINE TABLE REPULSION */
 774             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 775             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 776             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 777                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 778             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 779                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 780             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 781                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 782             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 783                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 784             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 785             Heps             = _mm256_mul_ps(vfeps,H);
 786             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 787             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 788             fvdw12           = _mm256_mul_ps(c12_00,FF);
 789             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 790
 791             fscal            = _mm256_add_ps(felec,fvdw);
 792
 793             /* Calculate temporary vectorial force */
 794             tx               = _mm256_mul_ps(fscal,dx00);
 795             ty               = _mm256_mul_ps(fscal,dy00);
 796             tz               = _mm256_mul_ps(fscal,dz00);
 797
 798             /* Update vectorial force */
 799             fix0             = _mm256_add_ps(fix0,tx);
 800             fiy0             = _mm256_add_ps(fiy0,ty);
 801             fiz0             = _mm256_add_ps(fiz0,tz);
 802
 803             fjptrA             = f+j_coord_offsetA;
 804             fjptrB             = f+j_coord_offsetB;
 805             fjptrC             = f+j_coord_offsetC;
 806             fjptrD             = f+j_coord_offsetD;
 807             fjptrE             = f+j_coord_offsetE;
 808             fjptrF             = f+j_coord_offsetF;
 809             fjptrG             = f+j_coord_offsetG;
 810             fjptrH             = f+j_coord_offsetH;
 811             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 812
 813             /* Inner loop uses 61 flops */
 814         }
 815
 816         if(jidx<j_index_end)
 817         {
 818
 819             /* Get j neighbor index, and coordinate index */
 820             jnrlistA         = jjnr[jidx];
 821             jnrlistB         = jjnr[jidx+1];
 822             jnrlistC         = jjnr[jidx+2];
 823             jnrlistD         = jjnr[jidx+3];
 824             jnrlistE         = jjnr[jidx+4];
 825             jnrlistF         = jjnr[jidx+5];
 826             jnrlistG         = jjnr[jidx+6];
 827             jnrlistH         = jjnr[jidx+7];
 828             /* Sign of each element will be negative for non-real atoms.
 829              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 830              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 831              */
 832             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 833                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 834
 835             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 836             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 837             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 838             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 839             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 840             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 841             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 842             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 843             j_coord_offsetA  = DIM*jnrA;
 844             j_coord_offsetB  = DIM*jnrB;
 845             j_coord_offsetC  = DIM*jnrC;
 846             j_coord_offsetD  = DIM*jnrD;
 847             j_coord_offsetE  = DIM*jnrE;
 848             j_coord_offsetF  = DIM*jnrF;
 849             j_coord_offsetG  = DIM*jnrG;
 850             j_coord_offsetH  = DIM*jnrH;
 851
 852             /* load j atom coordinates */
 853             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 854                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 855                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 856                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 857                                                  &jx0,&jy0,&jz0);
 858
 859             /* Calculate displacement vector */
 860             dx00             = _mm256_sub_ps(ix0,jx0);
 861             dy00             = _mm256_sub_ps(iy0,jy0);
 862             dz00             = _mm256_sub_ps(iz0,jz0);
 863
 864             /* Calculate squared distance and things based on it */
 865             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 866
 867             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 868
 869             /* Load parameters for j particles */
 870             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 871                                                                  charge+jnrC+0,charge+jnrD+0,
 872                                                                  charge+jnrE+0,charge+jnrF+0,
 873                                                                  charge+jnrG+0,charge+jnrH+0);
 874             vdwjidx0A        = 2*vdwtype[jnrA+0];
 875             vdwjidx0B        = 2*vdwtype[jnrB+0];
 876             vdwjidx0C        = 2*vdwtype[jnrC+0];
 877             vdwjidx0D        = 2*vdwtype[jnrD+0];
 878             vdwjidx0E        = 2*vdwtype[jnrE+0];
 879             vdwjidx0F        = 2*vdwtype[jnrF+0];
 880             vdwjidx0G        = 2*vdwtype[jnrG+0];
 881             vdwjidx0H        = 2*vdwtype[jnrH+0];
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             r00              = _mm256_mul_ps(rsq00,rinv00);
 888             r00              = _mm256_andnot_ps(dummy_mask,r00);
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             qq00             = _mm256_mul_ps(iq0,jq0);
 892             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 893                                             vdwioffsetptr0+vdwjidx0B,
 894                                             vdwioffsetptr0+vdwjidx0C,
 895                                             vdwioffsetptr0+vdwjidx0D,
 896                                             vdwioffsetptr0+vdwjidx0E,
 897                                             vdwioffsetptr0+vdwjidx0F,
 898                                             vdwioffsetptr0+vdwjidx0G,
 899                                             vdwioffsetptr0+vdwjidx0H,
 900                                             &c6_00,&c12_00);
 901
 902             /* Calculate table index by multiplying r with table scale and truncate to integer */
 903             rt               = _mm256_mul_ps(r00,vftabscale);
 904             vfitab           = _mm256_cvttps_epi32(rt);
 905             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 906             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 907             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 908             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 909             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 910             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 911
 912             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 913             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 914                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 915             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 916                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 917             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 918                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 919             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 920                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 921             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 922             Heps             = _mm256_mul_ps(vfeps,H);
 923             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 924             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 925             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 926
 927             /* CUBIC SPLINE TABLE DISPERSION */
 928             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 929             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 930             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 931                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 932             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 933                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 934             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 935                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 936             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 937                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 938             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 939             Heps             = _mm256_mul_ps(vfeps,H);
 940             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 941             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 942             fvdw6            = _mm256_mul_ps(c6_00,FF);
 943
 944             /* CUBIC SPLINE TABLE REPULSION */
 945             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 946             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 947             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 948                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 949             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 950                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 951             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 952                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 953             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 954                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 955             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 956             Heps             = _mm256_mul_ps(vfeps,H);
 957             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 958             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 959             fvdw12           = _mm256_mul_ps(c12_00,FF);
 960             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 961
 962             fscal            = _mm256_add_ps(felec,fvdw);
 963
 964             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 965
 966             /* Calculate temporary vectorial force */
 967             tx               = _mm256_mul_ps(fscal,dx00);
 968             ty               = _mm256_mul_ps(fscal,dy00);
 969             tz               = _mm256_mul_ps(fscal,dz00);
 970
 971             /* Update vectorial force */
 972             fix0             = _mm256_add_ps(fix0,tx);
 973             fiy0             = _mm256_add_ps(fiy0,ty);
 974             fiz0             = _mm256_add_ps(fiz0,tz);
 975
 976             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 977             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 978             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 979             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 980             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 981             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 982             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 983             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 984             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 985
 986             /* Inner loop uses 62 flops */
 987         }
 988
 989         /* End of innermost loop */
 990
 991         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 992                                                  f+i_coord_offset,fshift+i_shift_offset);
 993
 994         /* Increment number of inner iterations */
 995         inneriter                  += j_index_end - j_index_start;
 996
 997         /* Outer loop uses 7 flops */
 998     }
 999
1000     /* Increment number of outer iterations */
1001     outeriter        += nri;
1002
1003     /* Update outer/inner flops */
1004
1005     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
1006 }