src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     real *           vdwioffsetptr3;
  79     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  80     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  81     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  82     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  84     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  85     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  86     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  93     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  94     __m256i          vfitab;
  95     __m128i          vfitab_lo,vfitab_hi;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m256           dummy_mask,cutoff_mask;
 100     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 101     __m256           one     = _mm256_set1_ps(1.0);
 102     __m256           two     = _mm256_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm256_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     krf              = _mm256_set1_ps(fr->ic->k_rf);
 117     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 118     crf              = _mm256_set1_ps(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 129     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 130     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 131     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139     j_coord_offsetE = 0;
 140     j_coord_offsetF = 0;
 141     j_coord_offsetG = 0;
 142     j_coord_offsetH = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 169
 170         fix0             = _mm256_setzero_ps();
 171         fiy0             = _mm256_setzero_ps();
 172         fiz0             = _mm256_setzero_ps();
 173         fix1             = _mm256_setzero_ps();
 174         fiy1             = _mm256_setzero_ps();
 175         fiz1             = _mm256_setzero_ps();
 176         fix2             = _mm256_setzero_ps();
 177         fiy2             = _mm256_setzero_ps();
 178         fiz2             = _mm256_setzero_ps();
 179         fix3             = _mm256_setzero_ps();
 180         fiy3             = _mm256_setzero_ps();
 181         fiz3             = _mm256_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm256_setzero_ps();
 185         vvdwsum          = _mm256_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             jnrE             = jjnr[jidx+4];
 197             jnrF             = jjnr[jidx+5];
 198             jnrG             = jjnr[jidx+6];
 199             jnrH             = jjnr[jidx+7];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204             j_coord_offsetE  = DIM*jnrE;
 205             j_coord_offsetF  = DIM*jnrF;
 206             j_coord_offsetG  = DIM*jnrG;
 207             j_coord_offsetH  = DIM*jnrH;
 208
 209             /* load j atom coordinates */
 210             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 212                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 213                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 214                                                  &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm256_sub_ps(ix0,jx0);
 218             dy00             = _mm256_sub_ps(iy0,jy0);
 219             dz00             = _mm256_sub_ps(iz0,jz0);
 220             dx10             = _mm256_sub_ps(ix1,jx0);
 221             dy10             = _mm256_sub_ps(iy1,jy0);
 222             dz10             = _mm256_sub_ps(iz1,jz0);
 223             dx20             = _mm256_sub_ps(ix2,jx0);
 224             dy20             = _mm256_sub_ps(iy2,jy0);
 225             dz20             = _mm256_sub_ps(iz2,jz0);
 226             dx30             = _mm256_sub_ps(ix3,jx0);
 227             dy30             = _mm256_sub_ps(iy3,jy0);
 228             dz30             = _mm256_sub_ps(iz3,jz0);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 232             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 233             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 234             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 235
 236             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 237             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 238             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 239             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 240
 241             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 242             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 243             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                                  charge+jnrC+0,charge+jnrD+0,
 248                                                                  charge+jnrE+0,charge+jnrF+0,
 249                                                                  charge+jnrG+0,charge+jnrH+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251             vdwjidx0B        = 2*vdwtype[jnrB+0];
 252             vdwjidx0C        = 2*vdwtype[jnrC+0];
 253             vdwjidx0D        = 2*vdwtype[jnrD+0];
 254             vdwjidx0E        = 2*vdwtype[jnrE+0];
 255             vdwjidx0F        = 2*vdwtype[jnrF+0];
 256             vdwjidx0G        = 2*vdwtype[jnrG+0];
 257             vdwjidx0H        = 2*vdwtype[jnrH+0];
 258
 259             fjx0             = _mm256_setzero_ps();
 260             fjy0             = _mm256_setzero_ps();
 261             fjz0             = _mm256_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r00              = _mm256_mul_ps(rsq00,rinv00);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 271                                             vdwioffsetptr0+vdwjidx0B,
 272                                             vdwioffsetptr0+vdwjidx0C,
 273                                             vdwioffsetptr0+vdwjidx0D,
 274                                             vdwioffsetptr0+vdwjidx0E,
 275                                             vdwioffsetptr0+vdwjidx0F,
 276                                             vdwioffsetptr0+vdwjidx0G,
 277                                             vdwioffsetptr0+vdwjidx0H,
 278                                             &c6_00,&c12_00);
 279
 280             /* Calculate table index by multiplying r with table scale and truncate to integer */
 281             rt               = _mm256_mul_ps(r00,vftabscale);
 282             vfitab           = _mm256_cvttps_epi32(rt);
 283             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 284             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 285             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 286             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 287             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 288             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 289
 290             /* CUBIC SPLINE TABLE DISPERSION */
 291             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 293             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 295             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 297             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 298                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 299             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 300             Heps             = _mm256_mul_ps(vfeps,H);
 301             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 302             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 303             vvdw6            = _mm256_mul_ps(c6_00,VV);
 304             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 305             fvdw6            = _mm256_mul_ps(c6_00,FF);
 306
 307             /* CUBIC SPLINE TABLE REPULSION */
 308             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 309             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 310             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 312             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 314             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 316             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 317                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 318             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 319             Heps             = _mm256_mul_ps(vfeps,H);
 320             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 321             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 322             vvdw12           = _mm256_mul_ps(c12_00,VV);
 323             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 324             fvdw12           = _mm256_mul_ps(c12_00,FF);
 325             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 326             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 330
 331             fscal            = fvdw;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm256_mul_ps(fscal,dx00);
 335             ty               = _mm256_mul_ps(fscal,dy00);
 336             tz               = _mm256_mul_ps(fscal,dz00);
 337
 338             /* Update vectorial force */
 339             fix0             = _mm256_add_ps(fix0,tx);
 340             fiy0             = _mm256_add_ps(fiy0,ty);
 341             fiz0             = _mm256_add_ps(fiz0,tz);
 342
 343             fjx0             = _mm256_add_ps(fjx0,tx);
 344             fjy0             = _mm256_add_ps(fjy0,ty);
 345             fjz0             = _mm256_add_ps(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq10             = _mm256_mul_ps(iq1,jq0);
 353
 354             /* REACTION-FIELD ELECTROSTATICS */
 355             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 356             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm256_add_ps(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm256_mul_ps(fscal,dx10);
 365             ty               = _mm256_mul_ps(fscal,dy10);
 366             tz               = _mm256_mul_ps(fscal,dz10);
 367
 368             /* Update vectorial force */
 369             fix1             = _mm256_add_ps(fix1,tx);
 370             fiy1             = _mm256_add_ps(fiy1,ty);
 371             fiz1             = _mm256_add_ps(fiz1,tz);
 372
 373             fjx0             = _mm256_add_ps(fjx0,tx);
 374             fjy0             = _mm256_add_ps(fjy0,ty);
 375             fjz0             = _mm256_add_ps(fjz0,tz);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq20             = _mm256_mul_ps(iq2,jq0);
 383
 384             /* REACTION-FIELD ELECTROSTATICS */
 385             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 386             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm256_add_ps(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm256_mul_ps(fscal,dx20);
 395             ty               = _mm256_mul_ps(fscal,dy20);
 396             tz               = _mm256_mul_ps(fscal,dz20);
 397
 398             /* Update vectorial force */
 399             fix2             = _mm256_add_ps(fix2,tx);
 400             fiy2             = _mm256_add_ps(fiy2,ty);
 401             fiz2             = _mm256_add_ps(fiz2,tz);
 402
 403             fjx0             = _mm256_add_ps(fjx0,tx);
 404             fjy0             = _mm256_add_ps(fjy0,ty);
 405             fjz0             = _mm256_add_ps(fjz0,tz);
 406
 407             /**************************
 408              * CALCULATE INTERACTIONS *
 409              **************************/
 410
 411             /* Compute parameters for interactions between i and j atoms */
 412             qq30             = _mm256_mul_ps(iq3,jq0);
 413
 414             /* REACTION-FIELD ELECTROSTATICS */
 415             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 416             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velecsum         = _mm256_add_ps(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             /* Calculate temporary vectorial force */
 424             tx               = _mm256_mul_ps(fscal,dx30);
 425             ty               = _mm256_mul_ps(fscal,dy30);
 426             tz               = _mm256_mul_ps(fscal,dz30);
 427
 428             /* Update vectorial force */
 429             fix3             = _mm256_add_ps(fix3,tx);
 430             fiy3             = _mm256_add_ps(fiy3,ty);
 431             fiz3             = _mm256_add_ps(fiz3,tz);
 432
 433             fjx0             = _mm256_add_ps(fjx0,tx);
 434             fjy0             = _mm256_add_ps(fjy0,ty);
 435             fjz0             = _mm256_add_ps(fjz0,tz);
 436
 437             fjptrA             = f+j_coord_offsetA;
 438             fjptrB             = f+j_coord_offsetB;
 439             fjptrC             = f+j_coord_offsetC;
 440             fjptrD             = f+j_coord_offsetD;
 441             fjptrE             = f+j_coord_offsetE;
 442             fjptrF             = f+j_coord_offsetF;
 443             fjptrG             = f+j_coord_offsetG;
 444             fjptrH             = f+j_coord_offsetH;
 445
 446             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 447
 448             /* Inner loop uses 155 flops */
 449         }
 450
 451         if(jidx<j_index_end)
 452         {
 453
 454             /* Get j neighbor index, and coordinate index */
 455             jnrlistA         = jjnr[jidx];
 456             jnrlistB         = jjnr[jidx+1];
 457             jnrlistC         = jjnr[jidx+2];
 458             jnrlistD         = jjnr[jidx+3];
 459             jnrlistE         = jjnr[jidx+4];
 460             jnrlistF         = jjnr[jidx+5];
 461             jnrlistG         = jjnr[jidx+6];
 462             jnrlistH         = jjnr[jidx+7];
 463             /* Sign of each element will be negative for non-real atoms.
 464              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 465              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 466              */
 467             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 468                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 469
 470             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 471             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 472             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 473             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 474             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 475             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 476             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 477             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 478             j_coord_offsetA  = DIM*jnrA;
 479             j_coord_offsetB  = DIM*jnrB;
 480             j_coord_offsetC  = DIM*jnrC;
 481             j_coord_offsetD  = DIM*jnrD;
 482             j_coord_offsetE  = DIM*jnrE;
 483             j_coord_offsetF  = DIM*jnrF;
 484             j_coord_offsetG  = DIM*jnrG;
 485             j_coord_offsetH  = DIM*jnrH;
 486
 487             /* load j atom coordinates */
 488             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 489                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 490                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 491                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 492                                                  &jx0,&jy0,&jz0);
 493
 494             /* Calculate displacement vector */
 495             dx00             = _mm256_sub_ps(ix0,jx0);
 496             dy00             = _mm256_sub_ps(iy0,jy0);
 497             dz00             = _mm256_sub_ps(iz0,jz0);
 498             dx10             = _mm256_sub_ps(ix1,jx0);
 499             dy10             = _mm256_sub_ps(iy1,jy0);
 500             dz10             = _mm256_sub_ps(iz1,jz0);
 501             dx20             = _mm256_sub_ps(ix2,jx0);
 502             dy20             = _mm256_sub_ps(iy2,jy0);
 503             dz20             = _mm256_sub_ps(iz2,jz0);
 504             dx30             = _mm256_sub_ps(ix3,jx0);
 505             dy30             = _mm256_sub_ps(iy3,jy0);
 506             dz30             = _mm256_sub_ps(iz3,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 510             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 511             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 512             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 513
 514             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 515             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 516             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 517             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 518
 519             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 520             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 521             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 522
 523             /* Load parameters for j particles */
 524             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 525                                                                  charge+jnrC+0,charge+jnrD+0,
 526                                                                  charge+jnrE+0,charge+jnrF+0,
 527                                                                  charge+jnrG+0,charge+jnrH+0);
 528             vdwjidx0A        = 2*vdwtype[jnrA+0];
 529             vdwjidx0B        = 2*vdwtype[jnrB+0];
 530             vdwjidx0C        = 2*vdwtype[jnrC+0];
 531             vdwjidx0D        = 2*vdwtype[jnrD+0];
 532             vdwjidx0E        = 2*vdwtype[jnrE+0];
 533             vdwjidx0F        = 2*vdwtype[jnrF+0];
 534             vdwjidx0G        = 2*vdwtype[jnrG+0];
 535             vdwjidx0H        = 2*vdwtype[jnrH+0];
 536
 537             fjx0             = _mm256_setzero_ps();
 538             fjy0             = _mm256_setzero_ps();
 539             fjz0             = _mm256_setzero_ps();
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             r00              = _mm256_mul_ps(rsq00,rinv00);
 546             r00              = _mm256_andnot_ps(dummy_mask,r00);
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 550                                             vdwioffsetptr0+vdwjidx0B,
 551                                             vdwioffsetptr0+vdwjidx0C,
 552                                             vdwioffsetptr0+vdwjidx0D,
 553                                             vdwioffsetptr0+vdwjidx0E,
 554                                             vdwioffsetptr0+vdwjidx0F,
 555                                             vdwioffsetptr0+vdwjidx0G,
 556                                             vdwioffsetptr0+vdwjidx0H,
 557                                             &c6_00,&c12_00);
 558
 559             /* Calculate table index by multiplying r with table scale and truncate to integer */
 560             rt               = _mm256_mul_ps(r00,vftabscale);
 561             vfitab           = _mm256_cvttps_epi32(rt);
 562             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 563             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 564             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 565             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 566             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 567             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 568
 569             /* CUBIC SPLINE TABLE DISPERSION */
 570             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 571                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 572             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 573                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 574             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 576             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 578             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 579             Heps             = _mm256_mul_ps(vfeps,H);
 580             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 581             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 582             vvdw6            = _mm256_mul_ps(c6_00,VV);
 583             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 584             fvdw6            = _mm256_mul_ps(c6_00,FF);
 585
 586             /* CUBIC SPLINE TABLE REPULSION */
 587             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 588             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 589             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 590                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 591             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 592                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 593             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 594                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 595             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 596                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 597             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 598             Heps             = _mm256_mul_ps(vfeps,H);
 599             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 600             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 601             vvdw12           = _mm256_mul_ps(c12_00,VV);
 602             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 603             fvdw12           = _mm256_mul_ps(c12_00,FF);
 604             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 605             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 609             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 610
 611             fscal            = fvdw;
 612
 613             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm256_mul_ps(fscal,dx00);
 617             ty               = _mm256_mul_ps(fscal,dy00);
 618             tz               = _mm256_mul_ps(fscal,dz00);
 619
 620             /* Update vectorial force */
 621             fix0             = _mm256_add_ps(fix0,tx);
 622             fiy0             = _mm256_add_ps(fiy0,ty);
 623             fiz0             = _mm256_add_ps(fiz0,tz);
 624
 625             fjx0             = _mm256_add_ps(fjx0,tx);
 626             fjy0             = _mm256_add_ps(fjy0,ty);
 627             fjz0             = _mm256_add_ps(fjz0,tz);
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq10             = _mm256_mul_ps(iq1,jq0);
 635
 636             /* REACTION-FIELD ELECTROSTATICS */
 637             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 638             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velec            = _mm256_andnot_ps(dummy_mask,velec);
 642             velecsum         = _mm256_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm256_mul_ps(fscal,dx10);
 650             ty               = _mm256_mul_ps(fscal,dy10);
 651             tz               = _mm256_mul_ps(fscal,dz10);
 652
 653             /* Update vectorial force */
 654             fix1             = _mm256_add_ps(fix1,tx);
 655             fiy1             = _mm256_add_ps(fiy1,ty);
 656             fiz1             = _mm256_add_ps(fiz1,tz);
 657
 658             fjx0             = _mm256_add_ps(fjx0,tx);
 659             fjy0             = _mm256_add_ps(fjy0,ty);
 660             fjz0             = _mm256_add_ps(fjz0,tz);
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq20             = _mm256_mul_ps(iq2,jq0);
 668
 669             /* REACTION-FIELD ELECTROSTATICS */
 670             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 671             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm256_andnot_ps(dummy_mask,velec);
 675             velecsum         = _mm256_add_ps(velecsum,velec);
 676
 677             fscal            = felec;
 678
 679             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm256_mul_ps(fscal,dx20);
 683             ty               = _mm256_mul_ps(fscal,dy20);
 684             tz               = _mm256_mul_ps(fscal,dz20);
 685
 686             /* Update vectorial force */
 687             fix2             = _mm256_add_ps(fix2,tx);
 688             fiy2             = _mm256_add_ps(fiy2,ty);
 689             fiz2             = _mm256_add_ps(fiz2,tz);
 690
 691             fjx0             = _mm256_add_ps(fjx0,tx);
 692             fjy0             = _mm256_add_ps(fjy0,ty);
 693             fjz0             = _mm256_add_ps(fjz0,tz);
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             /* Compute parameters for interactions between i and j atoms */
 700             qq30             = _mm256_mul_ps(iq3,jq0);
 701
 702             /* REACTION-FIELD ELECTROSTATICS */
 703             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 704             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 705
 706             /* Update potential sum for this i atom from the interaction with this j atom. */
 707             velec            = _mm256_andnot_ps(dummy_mask,velec);
 708             velecsum         = _mm256_add_ps(velecsum,velec);
 709
 710             fscal            = felec;
 711
 712             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 713
 714             /* Calculate temporary vectorial force */
 715             tx               = _mm256_mul_ps(fscal,dx30);
 716             ty               = _mm256_mul_ps(fscal,dy30);
 717             tz               = _mm256_mul_ps(fscal,dz30);
 718
 719             /* Update vectorial force */
 720             fix3             = _mm256_add_ps(fix3,tx);
 721             fiy3             = _mm256_add_ps(fiy3,ty);
 722             fiz3             = _mm256_add_ps(fiz3,tz);
 723
 724             fjx0             = _mm256_add_ps(fjx0,tx);
 725             fjy0             = _mm256_add_ps(fjy0,ty);
 726             fjz0             = _mm256_add_ps(fjz0,tz);
 727
 728             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 729             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 730             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 731             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 732             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 733             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 734             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 735             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 736
 737             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 738
 739             /* Inner loop uses 156 flops */
 740         }
 741
 742         /* End of innermost loop */
 743
 744         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 745                                                  f+i_coord_offset,fshift+i_shift_offset);
 746
 747         ggid                        = gid[iidx];
 748         /* Update potential energies */
 749         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 750         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 751
 752         /* Increment number of inner iterations */
 753         inneriter                  += j_index_end - j_index_start;
 754
 755         /* Outer loop uses 26 flops */
 756     }
 757
 758     /* Increment number of outer iterations */
 759     outeriter        += nri;
 760
 761     /* Update outer/inner flops */
 762
 763     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*156);
 764 }
 765 /*
 766  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_single
 767  * Electrostatics interaction: ReactionField
 768  * VdW interaction:            CubicSplineTable
 769  * Geometry:                   Water4-Particle
 770  * Calculate force/pot:        Force
 771  */
 772 void
 773 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_single
 774                     (t_nblist * gmx_restrict                nlist,
 775                      rvec * gmx_restrict                    xx,
 776                      rvec * gmx_restrict                    ff,
 777                      t_forcerec * gmx_restrict              fr,
 778                      t_mdatoms * gmx_restrict               mdatoms,
 779                      nb_kernel_data_t * gmx_restrict        kernel_data,
 780                      t_nrnb * gmx_restrict                  nrnb)
 781 {
 782     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 783      * just 0 for non-waters.
 784      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 785      * jnr indices corresponding to data put in the four positions in the SIMD register.
 786      */
 787     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 788     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 789     int              jnrA,jnrB,jnrC,jnrD;
 790     int              jnrE,jnrF,jnrG,jnrH;
 791     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 792     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 793     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 794     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 795     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 796     real             rcutoff_scalar;
 797     real             *shiftvec,*fshift,*x,*f;
 798     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 799     real             scratch[4*DIM];
 800     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 801     real *           vdwioffsetptr0;
 802     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 803     real *           vdwioffsetptr1;
 804     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 805     real *           vdwioffsetptr2;
 806     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 807     real *           vdwioffsetptr3;
 808     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 809     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 810     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 811     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 812     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 813     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 814     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 815     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 816     real             *charge;
 817     int              nvdwtype;
 818     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 819     int              *vdwtype;
 820     real             *vdwparam;
 821     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 822     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 823     __m256i          vfitab;
 824     __m128i          vfitab_lo,vfitab_hi;
 825     __m128i          ifour       = _mm_set1_epi32(4);
 826     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 827     real             *vftab;
 828     __m256           dummy_mask,cutoff_mask;
 829     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 830     __m256           one     = _mm256_set1_ps(1.0);
 831     __m256           two     = _mm256_set1_ps(2.0);
 832     x                = xx[0];
 833     f                = ff[0];
 834
 835     nri              = nlist->nri;
 836     iinr             = nlist->iinr;
 837     jindex           = nlist->jindex;
 838     jjnr             = nlist->jjnr;
 839     shiftidx         = nlist->shift;
 840     gid              = nlist->gid;
 841     shiftvec         = fr->shift_vec[0];
 842     fshift           = fr->fshift[0];
 843     facel            = _mm256_set1_ps(fr->epsfac);
 844     charge           = mdatoms->chargeA;
 845     krf              = _mm256_set1_ps(fr->ic->k_rf);
 846     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 847     crf              = _mm256_set1_ps(fr->ic->c_rf);
 848     nvdwtype         = fr->ntype;
 849     vdwparam         = fr->nbfp;
 850     vdwtype          = mdatoms->typeA;
 851
 852     vftab            = kernel_data->table_vdw->data;
 853     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 854
 855     /* Setup water-specific parameters */
 856     inr              = nlist->iinr[0];
 857     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 858     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 859     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 860     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 861
 862     /* Avoid stupid compiler warnings */
 863     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 864     j_coord_offsetA = 0;
 865     j_coord_offsetB = 0;
 866     j_coord_offsetC = 0;
 867     j_coord_offsetD = 0;
 868     j_coord_offsetE = 0;
 869     j_coord_offsetF = 0;
 870     j_coord_offsetG = 0;
 871     j_coord_offsetH = 0;
 872
 873     outeriter        = 0;
 874     inneriter        = 0;
 875
 876     for(iidx=0;iidx<4*DIM;iidx++)
 877     {
 878         scratch[iidx] = 0.0;
 879     }
 880
 881     /* Start outer loop over neighborlists */
 882     for(iidx=0; iidx<nri; iidx++)
 883     {
 884         /* Load shift vector for this list */
 885         i_shift_offset   = DIM*shiftidx[iidx];
 886
 887         /* Load limits for loop over neighbors */
 888         j_index_start    = jindex[iidx];
 889         j_index_end      = jindex[iidx+1];
 890
 891         /* Get outer coordinate index */
 892         inr              = iinr[iidx];
 893         i_coord_offset   = DIM*inr;
 894
 895         /* Load i particle coords and add shift vector */
 896         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 897                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 898
 899         fix0             = _mm256_setzero_ps();
 900         fiy0             = _mm256_setzero_ps();
 901         fiz0             = _mm256_setzero_ps();
 902         fix1             = _mm256_setzero_ps();
 903         fiy1             = _mm256_setzero_ps();
 904         fiz1             = _mm256_setzero_ps();
 905         fix2             = _mm256_setzero_ps();
 906         fiy2             = _mm256_setzero_ps();
 907         fiz2             = _mm256_setzero_ps();
 908         fix3             = _mm256_setzero_ps();
 909         fiy3             = _mm256_setzero_ps();
 910         fiz3             = _mm256_setzero_ps();
 911
 912         /* Start inner kernel loop */
 913         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrA             = jjnr[jidx];
 918             jnrB             = jjnr[jidx+1];
 919             jnrC             = jjnr[jidx+2];
 920             jnrD             = jjnr[jidx+3];
 921             jnrE             = jjnr[jidx+4];
 922             jnrF             = jjnr[jidx+5];
 923             jnrG             = jjnr[jidx+6];
 924             jnrH             = jjnr[jidx+7];
 925             j_coord_offsetA  = DIM*jnrA;
 926             j_coord_offsetB  = DIM*jnrB;
 927             j_coord_offsetC  = DIM*jnrC;
 928             j_coord_offsetD  = DIM*jnrD;
 929             j_coord_offsetE  = DIM*jnrE;
 930             j_coord_offsetF  = DIM*jnrF;
 931             j_coord_offsetG  = DIM*jnrG;
 932             j_coord_offsetH  = DIM*jnrH;
 933
 934             /* load j atom coordinates */
 935             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 936                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 937                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 938                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 939                                                  &jx0,&jy0,&jz0);
 940
 941             /* Calculate displacement vector */
 942             dx00             = _mm256_sub_ps(ix0,jx0);
 943             dy00             = _mm256_sub_ps(iy0,jy0);
 944             dz00             = _mm256_sub_ps(iz0,jz0);
 945             dx10             = _mm256_sub_ps(ix1,jx0);
 946             dy10             = _mm256_sub_ps(iy1,jy0);
 947             dz10             = _mm256_sub_ps(iz1,jz0);
 948             dx20             = _mm256_sub_ps(ix2,jx0);
 949             dy20             = _mm256_sub_ps(iy2,jy0);
 950             dz20             = _mm256_sub_ps(iz2,jz0);
 951             dx30             = _mm256_sub_ps(ix3,jx0);
 952             dy30             = _mm256_sub_ps(iy3,jy0);
 953             dz30             = _mm256_sub_ps(iz3,jz0);
 954
 955             /* Calculate squared distance and things based on it */
 956             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 957             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 958             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 959             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 960
 961             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 962             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 963             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 964             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 965
 966             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 967             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 968             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 969
 970             /* Load parameters for j particles */
 971             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 972                                                                  charge+jnrC+0,charge+jnrD+0,
 973                                                                  charge+jnrE+0,charge+jnrF+0,
 974                                                                  charge+jnrG+0,charge+jnrH+0);
 975             vdwjidx0A        = 2*vdwtype[jnrA+0];
 976             vdwjidx0B        = 2*vdwtype[jnrB+0];
 977             vdwjidx0C        = 2*vdwtype[jnrC+0];
 978             vdwjidx0D        = 2*vdwtype[jnrD+0];
 979             vdwjidx0E        = 2*vdwtype[jnrE+0];
 980             vdwjidx0F        = 2*vdwtype[jnrF+0];
 981             vdwjidx0G        = 2*vdwtype[jnrG+0];
 982             vdwjidx0H        = 2*vdwtype[jnrH+0];
 983
 984             fjx0             = _mm256_setzero_ps();
 985             fjy0             = _mm256_setzero_ps();
 986             fjz0             = _mm256_setzero_ps();
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             r00              = _mm256_mul_ps(rsq00,rinv00);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 996                                             vdwioffsetptr0+vdwjidx0B,
 997                                             vdwioffsetptr0+vdwjidx0C,
 998                                             vdwioffsetptr0+vdwjidx0D,
 999                                             vdwioffsetptr0+vdwjidx0E,
1000                                             vdwioffsetptr0+vdwjidx0F,
1001                                             vdwioffsetptr0+vdwjidx0G,
1002                                             vdwioffsetptr0+vdwjidx0H,
1003                                             &c6_00,&c12_00);
1004
1005             /* Calculate table index by multiplying r with table scale and truncate to integer */
1006             rt               = _mm256_mul_ps(r00,vftabscale);
1007             vfitab           = _mm256_cvttps_epi32(rt);
1008             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1009             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1010             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1011             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1012             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1013             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1014
1015             /* CUBIC SPLINE TABLE DISPERSION */
1016             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1017                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1018             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1019                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1020             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1021                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1022             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1023                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1024             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1025             Heps             = _mm256_mul_ps(vfeps,H);
1026             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1027             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1028             fvdw6            = _mm256_mul_ps(c6_00,FF);
1029
1030             /* CUBIC SPLINE TABLE REPULSION */
1031             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1032             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1033             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1034                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1035             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1036                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1037             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1038                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1039             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1040                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1041             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1042             Heps             = _mm256_mul_ps(vfeps,H);
1043             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1044             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1045             fvdw12           = _mm256_mul_ps(c12_00,FF);
1046             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1047
1048             fscal            = fvdw;
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm256_mul_ps(fscal,dx00);
1052             ty               = _mm256_mul_ps(fscal,dy00);
1053             tz               = _mm256_mul_ps(fscal,dz00);
1054
1055             /* Update vectorial force */
1056             fix0             = _mm256_add_ps(fix0,tx);
1057             fiy0             = _mm256_add_ps(fiy0,ty);
1058             fiz0             = _mm256_add_ps(fiz0,tz);
1059
1060             fjx0             = _mm256_add_ps(fjx0,tx);
1061             fjy0             = _mm256_add_ps(fjy0,ty);
1062             fjz0             = _mm256_add_ps(fjz0,tz);
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq10             = _mm256_mul_ps(iq1,jq0);
1070
1071             /* REACTION-FIELD ELECTROSTATICS */
1072             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1073
1074             fscal            = felec;
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm256_mul_ps(fscal,dx10);
1078             ty               = _mm256_mul_ps(fscal,dy10);
1079             tz               = _mm256_mul_ps(fscal,dz10);
1080
1081             /* Update vectorial force */
1082             fix1             = _mm256_add_ps(fix1,tx);
1083             fiy1             = _mm256_add_ps(fiy1,ty);
1084             fiz1             = _mm256_add_ps(fiz1,tz);
1085
1086             fjx0             = _mm256_add_ps(fjx0,tx);
1087             fjy0             = _mm256_add_ps(fjy0,ty);
1088             fjz0             = _mm256_add_ps(fjz0,tz);
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
1092              **************************/
1093
1094             /* Compute parameters for interactions between i and j atoms */
1095             qq20             = _mm256_mul_ps(iq2,jq0);
1096
1097             /* REACTION-FIELD ELECTROSTATICS */
1098             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1099
1100             fscal            = felec;
1101
1102             /* Calculate temporary vectorial force */
1103             tx               = _mm256_mul_ps(fscal,dx20);
1104             ty               = _mm256_mul_ps(fscal,dy20);
1105             tz               = _mm256_mul_ps(fscal,dz20);
1106
1107             /* Update vectorial force */
1108             fix2             = _mm256_add_ps(fix2,tx);
1109             fiy2             = _mm256_add_ps(fiy2,ty);
1110             fiz2             = _mm256_add_ps(fiz2,tz);
1111
1112             fjx0             = _mm256_add_ps(fjx0,tx);
1113             fjy0             = _mm256_add_ps(fjy0,ty);
1114             fjz0             = _mm256_add_ps(fjz0,tz);
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             qq30             = _mm256_mul_ps(iq3,jq0);
1122
1123             /* REACTION-FIELD ELECTROSTATICS */
1124             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1125
1126             fscal            = felec;
1127
1128             /* Calculate temporary vectorial force */
1129             tx               = _mm256_mul_ps(fscal,dx30);
1130             ty               = _mm256_mul_ps(fscal,dy30);
1131             tz               = _mm256_mul_ps(fscal,dz30);
1132
1133             /* Update vectorial force */
1134             fix3             = _mm256_add_ps(fix3,tx);
1135             fiy3             = _mm256_add_ps(fiy3,ty);
1136             fiz3             = _mm256_add_ps(fiz3,tz);
1137
1138             fjx0             = _mm256_add_ps(fjx0,tx);
1139             fjy0             = _mm256_add_ps(fjy0,ty);
1140             fjz0             = _mm256_add_ps(fjz0,tz);
1141
1142             fjptrA             = f+j_coord_offsetA;
1143             fjptrB             = f+j_coord_offsetB;
1144             fjptrC             = f+j_coord_offsetC;
1145             fjptrD             = f+j_coord_offsetD;
1146             fjptrE             = f+j_coord_offsetE;
1147             fjptrF             = f+j_coord_offsetF;
1148             fjptrG             = f+j_coord_offsetG;
1149             fjptrH             = f+j_coord_offsetH;
1150
1151             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1152
1153             /* Inner loop uses 132 flops */
1154         }
1155
1156         if(jidx<j_index_end)
1157         {
1158
1159             /* Get j neighbor index, and coordinate index */
1160             jnrlistA         = jjnr[jidx];
1161             jnrlistB         = jjnr[jidx+1];
1162             jnrlistC         = jjnr[jidx+2];
1163             jnrlistD         = jjnr[jidx+3];
1164             jnrlistE         = jjnr[jidx+4];
1165             jnrlistF         = jjnr[jidx+5];
1166             jnrlistG         = jjnr[jidx+6];
1167             jnrlistH         = jjnr[jidx+7];
1168             /* Sign of each element will be negative for non-real atoms.
1169              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1170              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1171              */
1172             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1173                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1174
1175             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1176             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1177             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1178             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1179             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1180             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1181             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1182             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1183             j_coord_offsetA  = DIM*jnrA;
1184             j_coord_offsetB  = DIM*jnrB;
1185             j_coord_offsetC  = DIM*jnrC;
1186             j_coord_offsetD  = DIM*jnrD;
1187             j_coord_offsetE  = DIM*jnrE;
1188             j_coord_offsetF  = DIM*jnrF;
1189             j_coord_offsetG  = DIM*jnrG;
1190             j_coord_offsetH  = DIM*jnrH;
1191
1192             /* load j atom coordinates */
1193             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1195                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1196                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1197                                                  &jx0,&jy0,&jz0);
1198
1199             /* Calculate displacement vector */
1200             dx00             = _mm256_sub_ps(ix0,jx0);
1201             dy00             = _mm256_sub_ps(iy0,jy0);
1202             dz00             = _mm256_sub_ps(iz0,jz0);
1203             dx10             = _mm256_sub_ps(ix1,jx0);
1204             dy10             = _mm256_sub_ps(iy1,jy0);
1205             dz10             = _mm256_sub_ps(iz1,jz0);
1206             dx20             = _mm256_sub_ps(ix2,jx0);
1207             dy20             = _mm256_sub_ps(iy2,jy0);
1208             dz20             = _mm256_sub_ps(iz2,jz0);
1209             dx30             = _mm256_sub_ps(ix3,jx0);
1210             dy30             = _mm256_sub_ps(iy3,jy0);
1211             dz30             = _mm256_sub_ps(iz3,jz0);
1212
1213             /* Calculate squared distance and things based on it */
1214             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1215             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1216             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1217             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1218
1219             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1220             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1221             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1222             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1223
1224             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1225             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1226             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1227
1228             /* Load parameters for j particles */
1229             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1230                                                                  charge+jnrC+0,charge+jnrD+0,
1231                                                                  charge+jnrE+0,charge+jnrF+0,
1232                                                                  charge+jnrG+0,charge+jnrH+0);
1233             vdwjidx0A        = 2*vdwtype[jnrA+0];
1234             vdwjidx0B        = 2*vdwtype[jnrB+0];
1235             vdwjidx0C        = 2*vdwtype[jnrC+0];
1236             vdwjidx0D        = 2*vdwtype[jnrD+0];
1237             vdwjidx0E        = 2*vdwtype[jnrE+0];
1238             vdwjidx0F        = 2*vdwtype[jnrF+0];
1239             vdwjidx0G        = 2*vdwtype[jnrG+0];
1240             vdwjidx0H        = 2*vdwtype[jnrH+0];
1241
1242             fjx0             = _mm256_setzero_ps();
1243             fjy0             = _mm256_setzero_ps();
1244             fjz0             = _mm256_setzero_ps();
1245
1246             /**************************
1247              * CALCULATE INTERACTIONS *
1248              **************************/
1249
1250             r00              = _mm256_mul_ps(rsq00,rinv00);
1251             r00              = _mm256_andnot_ps(dummy_mask,r00);
1252
1253             /* Compute parameters for interactions between i and j atoms */
1254             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1255                                             vdwioffsetptr0+vdwjidx0B,
1256                                             vdwioffsetptr0+vdwjidx0C,
1257                                             vdwioffsetptr0+vdwjidx0D,
1258                                             vdwioffsetptr0+vdwjidx0E,
1259                                             vdwioffsetptr0+vdwjidx0F,
1260                                             vdwioffsetptr0+vdwjidx0G,
1261                                             vdwioffsetptr0+vdwjidx0H,
1262                                             &c6_00,&c12_00);
1263
1264             /* Calculate table index by multiplying r with table scale and truncate to integer */
1265             rt               = _mm256_mul_ps(r00,vftabscale);
1266             vfitab           = _mm256_cvttps_epi32(rt);
1267             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1268             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1269             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1270             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1271             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1272             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1273
1274             /* CUBIC SPLINE TABLE DISPERSION */
1275             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1276                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1277             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1278                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1279             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1280                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1281             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1282                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1283             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1284             Heps             = _mm256_mul_ps(vfeps,H);
1285             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1286             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1287             fvdw6            = _mm256_mul_ps(c6_00,FF);
1288
1289             /* CUBIC SPLINE TABLE REPULSION */
1290             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1291             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1292             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1294             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1296             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1298             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1299                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1300             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1301             Heps             = _mm256_mul_ps(vfeps,H);
1302             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1303             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1304             fvdw12           = _mm256_mul_ps(c12_00,FF);
1305             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1306
1307             fscal            = fvdw;
1308
1309             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1310
1311             /* Calculate temporary vectorial force */
1312             tx               = _mm256_mul_ps(fscal,dx00);
1313             ty               = _mm256_mul_ps(fscal,dy00);
1314             tz               = _mm256_mul_ps(fscal,dz00);
1315
1316             /* Update vectorial force */
1317             fix0             = _mm256_add_ps(fix0,tx);
1318             fiy0             = _mm256_add_ps(fiy0,ty);
1319             fiz0             = _mm256_add_ps(fiz0,tz);
1320
1321             fjx0             = _mm256_add_ps(fjx0,tx);
1322             fjy0             = _mm256_add_ps(fjy0,ty);
1323             fjz0             = _mm256_add_ps(fjz0,tz);
1324
1325             /**************************
1326              * CALCULATE INTERACTIONS *
1327              **************************/
1328
1329             /* Compute parameters for interactions between i and j atoms */
1330             qq10             = _mm256_mul_ps(iq1,jq0);
1331
1332             /* REACTION-FIELD ELECTROSTATICS */
1333             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1334
1335             fscal            = felec;
1336
1337             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1338
1339             /* Calculate temporary vectorial force */
1340             tx               = _mm256_mul_ps(fscal,dx10);
1341             ty               = _mm256_mul_ps(fscal,dy10);
1342             tz               = _mm256_mul_ps(fscal,dz10);
1343
1344             /* Update vectorial force */
1345             fix1             = _mm256_add_ps(fix1,tx);
1346             fiy1             = _mm256_add_ps(fiy1,ty);
1347             fiz1             = _mm256_add_ps(fiz1,tz);
1348
1349             fjx0             = _mm256_add_ps(fjx0,tx);
1350             fjy0             = _mm256_add_ps(fjy0,ty);
1351             fjz0             = _mm256_add_ps(fjz0,tz);
1352
1353             /**************************
1354              * CALCULATE INTERACTIONS *
1355              **************************/
1356
1357             /* Compute parameters for interactions between i and j atoms */
1358             qq20             = _mm256_mul_ps(iq2,jq0);
1359
1360             /* REACTION-FIELD ELECTROSTATICS */
1361             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1362
1363             fscal            = felec;
1364
1365             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1366
1367             /* Calculate temporary vectorial force */
1368             tx               = _mm256_mul_ps(fscal,dx20);
1369             ty               = _mm256_mul_ps(fscal,dy20);
1370             tz               = _mm256_mul_ps(fscal,dz20);
1371
1372             /* Update vectorial force */
1373             fix2             = _mm256_add_ps(fix2,tx);
1374             fiy2             = _mm256_add_ps(fiy2,ty);
1375             fiz2             = _mm256_add_ps(fiz2,tz);
1376
1377             fjx0             = _mm256_add_ps(fjx0,tx);
1378             fjy0             = _mm256_add_ps(fjy0,ty);
1379             fjz0             = _mm256_add_ps(fjz0,tz);
1380
1381             /**************************
1382              * CALCULATE INTERACTIONS *
1383              **************************/
1384
1385             /* Compute parameters for interactions between i and j atoms */
1386             qq30             = _mm256_mul_ps(iq3,jq0);
1387
1388             /* REACTION-FIELD ELECTROSTATICS */
1389             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1390
1391             fscal            = felec;
1392
1393             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1394
1395             /* Calculate temporary vectorial force */
1396             tx               = _mm256_mul_ps(fscal,dx30);
1397             ty               = _mm256_mul_ps(fscal,dy30);
1398             tz               = _mm256_mul_ps(fscal,dz30);
1399
1400             /* Update vectorial force */
1401             fix3             = _mm256_add_ps(fix3,tx);
1402             fiy3             = _mm256_add_ps(fiy3,ty);
1403             fiz3             = _mm256_add_ps(fiz3,tz);
1404
1405             fjx0             = _mm256_add_ps(fjx0,tx);
1406             fjy0             = _mm256_add_ps(fjy0,ty);
1407             fjz0             = _mm256_add_ps(fjz0,tz);
1408
1409             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1410             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1411             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1412             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1413             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1414             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1415             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1416             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1417
1418             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1419
1420             /* Inner loop uses 133 flops */
1421         }
1422
1423         /* End of innermost loop */
1424
1425         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1426                                                  f+i_coord_offset,fshift+i_shift_offset);
1427
1428         /* Increment number of inner iterations */
1429         inneriter                  += j_index_end - j_index_start;
1430
1431         /* Outer loop uses 24 flops */
1432     }
1433
1434     /* Increment number of outer iterations */
1435     outeriter        += nri;
1436
1437     /* Update outer/inner flops */
1438
1439     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*133);
1440 }