src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_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_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256i          vfitab;
  92     __m128i          vfitab_lo,vfitab_hi;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m256           dummy_mask,cutoff_mask;
  97     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  98     __m256           one     = _mm256_set1_ps(1.0);
  99     __m256           two     = _mm256_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm256_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     krf              = _mm256_set1_ps(fr->ic->k_rf);
 114     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 115     crf              = _mm256_set1_ps(fr->ic->c_rf);
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_vdw->data;
 121     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 126     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 127     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 128     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 129
 130     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 131     rcutoff_scalar   = fr->rcoulomb;
 132     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 133     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141     j_coord_offsetE = 0;
 142     j_coord_offsetF = 0;
 143     j_coord_offsetG = 0;
 144     j_coord_offsetH = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm256_setzero_ps();
 173         fiy0             = _mm256_setzero_ps();
 174         fiz0             = _mm256_setzero_ps();
 175         fix1             = _mm256_setzero_ps();
 176         fiy1             = _mm256_setzero_ps();
 177         fiz1             = _mm256_setzero_ps();
 178         fix2             = _mm256_setzero_ps();
 179         fiy2             = _mm256_setzero_ps();
 180         fiz2             = _mm256_setzero_ps();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm256_setzero_ps();
 184         vvdwsum          = _mm256_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             jnrE             = jjnr[jidx+4];
 196             jnrF             = jjnr[jidx+5];
 197             jnrG             = jjnr[jidx+6];
 198             jnrH             = jjnr[jidx+7];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203             j_coord_offsetE  = DIM*jnrE;
 204             j_coord_offsetF  = DIM*jnrF;
 205             j_coord_offsetG  = DIM*jnrG;
 206             j_coord_offsetH  = DIM*jnrH;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 212                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 213                                                  &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm256_sub_ps(ix0,jx0);
 217             dy00             = _mm256_sub_ps(iy0,jy0);
 218             dz00             = _mm256_sub_ps(iz0,jz0);
 219             dx10             = _mm256_sub_ps(ix1,jx0);
 220             dy10             = _mm256_sub_ps(iy1,jy0);
 221             dz10             = _mm256_sub_ps(iz1,jz0);
 222             dx20             = _mm256_sub_ps(ix2,jx0);
 223             dy20             = _mm256_sub_ps(iy2,jy0);
 224             dz20             = _mm256_sub_ps(iz2,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 228             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 229             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 230
 231             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 232             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 234
 235             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 236             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 237             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 241                                                                  charge+jnrC+0,charge+jnrD+0,
 242                                                                  charge+jnrE+0,charge+jnrF+0,
 243                                                                  charge+jnrG+0,charge+jnrH+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248             vdwjidx0E        = 2*vdwtype[jnrE+0];
 249             vdwjidx0F        = 2*vdwtype[jnrF+0];
 250             vdwjidx0G        = 2*vdwtype[jnrG+0];
 251             vdwjidx0H        = 2*vdwtype[jnrH+0];
 252
 253             fjx0             = _mm256_setzero_ps();
 254             fjy0             = _mm256_setzero_ps();
 255             fjz0             = _mm256_setzero_ps();
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 262             {
 263
 264             r00              = _mm256_mul_ps(rsq00,rinv00);
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq00             = _mm256_mul_ps(iq0,jq0);
 268             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 269                                             vdwioffsetptr0+vdwjidx0B,
 270                                             vdwioffsetptr0+vdwjidx0C,
 271                                             vdwioffsetptr0+vdwjidx0D,
 272                                             vdwioffsetptr0+vdwjidx0E,
 273                                             vdwioffsetptr0+vdwjidx0F,
 274                                             vdwioffsetptr0+vdwjidx0G,
 275                                             vdwioffsetptr0+vdwjidx0H,
 276                                             &c6_00,&c12_00);
 277
 278             /* Calculate table index by multiplying r with table scale and truncate to integer */
 279             rt               = _mm256_mul_ps(r00,vftabscale);
 280             vfitab           = _mm256_cvttps_epi32(rt);
 281             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 282             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 283             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 284             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 285             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 286             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 287
 288             /* REACTION-FIELD ELECTROSTATICS */
 289             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 290             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 291
 292             /* CUBIC SPLINE TABLE DISPERSION */
 293             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 295             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 297             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 298                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 299             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 300                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 301             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 302             Heps             = _mm256_mul_ps(vfeps,H);
 303             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 304             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 305             vvdw6            = _mm256_mul_ps(c6_00,VV);
 306             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 307             fvdw6            = _mm256_mul_ps(c6_00,FF);
 308
 309             /* CUBIC SPLINE TABLE REPULSION */
 310             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 311             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 312             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 314             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 316             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 317                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 318             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 320             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 321             Heps             = _mm256_mul_ps(vfeps,H);
 322             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 323             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 324             vvdw12           = _mm256_mul_ps(c12_00,VV);
 325             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 326             fvdw12           = _mm256_mul_ps(c12_00,FF);
 327             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 328             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 329
 330             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm256_and_ps(velec,cutoff_mask);
 334             velecsum         = _mm256_add_ps(velecsum,velec);
 335             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 336             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 337
 338             fscal            = _mm256_add_ps(felec,fvdw);
 339
 340             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm256_mul_ps(fscal,dx00);
 344             ty               = _mm256_mul_ps(fscal,dy00);
 345             tz               = _mm256_mul_ps(fscal,dz00);
 346
 347             /* Update vectorial force */
 348             fix0             = _mm256_add_ps(fix0,tx);
 349             fiy0             = _mm256_add_ps(fiy0,ty);
 350             fiz0             = _mm256_add_ps(fiz0,tz);
 351
 352             fjx0             = _mm256_add_ps(fjx0,tx);
 353             fjy0             = _mm256_add_ps(fjy0,ty);
 354             fjz0             = _mm256_add_ps(fjz0,tz);
 355
 356             }
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 363             {
 364
 365             /* Compute parameters for interactions between i and j atoms */
 366             qq10             = _mm256_mul_ps(iq1,jq0);
 367
 368             /* REACTION-FIELD ELECTROSTATICS */
 369             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 370             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 371
 372             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velec            = _mm256_and_ps(velec,cutoff_mask);
 376             velecsum         = _mm256_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm256_mul_ps(fscal,dx10);
 384             ty               = _mm256_mul_ps(fscal,dy10);
 385             tz               = _mm256_mul_ps(fscal,dz10);
 386
 387             /* Update vectorial force */
 388             fix1             = _mm256_add_ps(fix1,tx);
 389             fiy1             = _mm256_add_ps(fiy1,ty);
 390             fiz1             = _mm256_add_ps(fiz1,tz);
 391
 392             fjx0             = _mm256_add_ps(fjx0,tx);
 393             fjy0             = _mm256_add_ps(fjy0,ty);
 394             fjz0             = _mm256_add_ps(fjz0,tz);
 395
 396             }
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 403             {
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq20             = _mm256_mul_ps(iq2,jq0);
 407
 408             /* REACTION-FIELD ELECTROSTATICS */
 409             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 410             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 411
 412             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm256_and_ps(velec,cutoff_mask);
 416             velecsum         = _mm256_add_ps(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 421
 422             /* Calculate temporary vectorial force */
 423             tx               = _mm256_mul_ps(fscal,dx20);
 424             ty               = _mm256_mul_ps(fscal,dy20);
 425             tz               = _mm256_mul_ps(fscal,dz20);
 426
 427             /* Update vectorial force */
 428             fix2             = _mm256_add_ps(fix2,tx);
 429             fiy2             = _mm256_add_ps(fiy2,ty);
 430             fiz2             = _mm256_add_ps(fiz2,tz);
 431
 432             fjx0             = _mm256_add_ps(fjx0,tx);
 433             fjy0             = _mm256_add_ps(fjy0,ty);
 434             fjz0             = _mm256_add_ps(fjz0,tz);
 435
 436             }
 437
 438             fjptrA             = f+j_coord_offsetA;
 439             fjptrB             = f+j_coord_offsetB;
 440             fjptrC             = f+j_coord_offsetC;
 441             fjptrD             = f+j_coord_offsetD;
 442             fjptrE             = f+j_coord_offsetE;
 443             fjptrF             = f+j_coord_offsetF;
 444             fjptrG             = f+j_coord_offsetG;
 445             fjptrH             = f+j_coord_offsetH;
 446
 447             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 448
 449             /* Inner loop uses 147 flops */
 450         }
 451
 452         if(jidx<j_index_end)
 453         {
 454
 455             /* Get j neighbor index, and coordinate index */
 456             jnrlistA         = jjnr[jidx];
 457             jnrlistB         = jjnr[jidx+1];
 458             jnrlistC         = jjnr[jidx+2];
 459             jnrlistD         = jjnr[jidx+3];
 460             jnrlistE         = jjnr[jidx+4];
 461             jnrlistF         = jjnr[jidx+5];
 462             jnrlistG         = jjnr[jidx+6];
 463             jnrlistH         = jjnr[jidx+7];
 464             /* Sign of each element will be negative for non-real atoms.
 465              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 466              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 467              */
 468             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 469                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 470
 471             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 472             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 473             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 474             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 475             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 476             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 477             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 478             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 479             j_coord_offsetA  = DIM*jnrA;
 480             j_coord_offsetB  = DIM*jnrB;
 481             j_coord_offsetC  = DIM*jnrC;
 482             j_coord_offsetD  = DIM*jnrD;
 483             j_coord_offsetE  = DIM*jnrE;
 484             j_coord_offsetF  = DIM*jnrF;
 485             j_coord_offsetG  = DIM*jnrG;
 486             j_coord_offsetH  = DIM*jnrH;
 487
 488             /* load j atom coordinates */
 489             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 490                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 491                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 492                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 493                                                  &jx0,&jy0,&jz0);
 494
 495             /* Calculate displacement vector */
 496             dx00             = _mm256_sub_ps(ix0,jx0);
 497             dy00             = _mm256_sub_ps(iy0,jy0);
 498             dz00             = _mm256_sub_ps(iz0,jz0);
 499             dx10             = _mm256_sub_ps(ix1,jx0);
 500             dy10             = _mm256_sub_ps(iy1,jy0);
 501             dz10             = _mm256_sub_ps(iz1,jz0);
 502             dx20             = _mm256_sub_ps(ix2,jx0);
 503             dy20             = _mm256_sub_ps(iy2,jy0);
 504             dz20             = _mm256_sub_ps(iz2,jz0);
 505
 506             /* Calculate squared distance and things based on it */
 507             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 508             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 509             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 510
 511             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 512             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 513             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 514
 515             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 516             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 517             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 521                                                                  charge+jnrC+0,charge+jnrD+0,
 522                                                                  charge+jnrE+0,charge+jnrF+0,
 523                                                                  charge+jnrG+0,charge+jnrH+0);
 524             vdwjidx0A        = 2*vdwtype[jnrA+0];
 525             vdwjidx0B        = 2*vdwtype[jnrB+0];
 526             vdwjidx0C        = 2*vdwtype[jnrC+0];
 527             vdwjidx0D        = 2*vdwtype[jnrD+0];
 528             vdwjidx0E        = 2*vdwtype[jnrE+0];
 529             vdwjidx0F        = 2*vdwtype[jnrF+0];
 530             vdwjidx0G        = 2*vdwtype[jnrG+0];
 531             vdwjidx0H        = 2*vdwtype[jnrH+0];
 532
 533             fjx0             = _mm256_setzero_ps();
 534             fjy0             = _mm256_setzero_ps();
 535             fjz0             = _mm256_setzero_ps();
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 542             {
 543
 544             r00              = _mm256_mul_ps(rsq00,rinv00);
 545             r00              = _mm256_andnot_ps(dummy_mask,r00);
 546
 547             /* Compute parameters for interactions between i and j atoms */
 548             qq00             = _mm256_mul_ps(iq0,jq0);
 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             /* REACTION-FIELD ELECTROSTATICS */
 570             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 571             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 572
 573             /* CUBIC SPLINE TABLE DISPERSION */
 574             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 576             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 578             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 579                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 580             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 581                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 582             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 583             Heps             = _mm256_mul_ps(vfeps,H);
 584             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 585             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 586             vvdw6            = _mm256_mul_ps(c6_00,VV);
 587             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 588             fvdw6            = _mm256_mul_ps(c6_00,FF);
 589
 590             /* CUBIC SPLINE TABLE REPULSION */
 591             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 592             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 593             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 594                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 595             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 596                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 597             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 598                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 599             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 600                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 601             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 602             Heps             = _mm256_mul_ps(vfeps,H);
 603             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 604             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 605             vvdw12           = _mm256_mul_ps(c12_00,VV);
 606             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 607             fvdw12           = _mm256_mul_ps(c12_00,FF);
 608             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 609             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 610
 611             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velec            = _mm256_and_ps(velec,cutoff_mask);
 615             velec            = _mm256_andnot_ps(dummy_mask,velec);
 616             velecsum         = _mm256_add_ps(velecsum,velec);
 617             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 618             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 619             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 620
 621             fscal            = _mm256_add_ps(felec,fvdw);
 622
 623             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 624
 625             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm256_mul_ps(fscal,dx00);
 629             ty               = _mm256_mul_ps(fscal,dy00);
 630             tz               = _mm256_mul_ps(fscal,dz00);
 631
 632             /* Update vectorial force */
 633             fix0             = _mm256_add_ps(fix0,tx);
 634             fiy0             = _mm256_add_ps(fiy0,ty);
 635             fiz0             = _mm256_add_ps(fiz0,tz);
 636
 637             fjx0             = _mm256_add_ps(fjx0,tx);
 638             fjy0             = _mm256_add_ps(fjy0,ty);
 639             fjz0             = _mm256_add_ps(fjz0,tz);
 640
 641             }
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 648             {
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq10             = _mm256_mul_ps(iq1,jq0);
 652
 653             /* REACTION-FIELD ELECTROSTATICS */
 654             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 655             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 656
 657             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 658
 659             /* Update potential sum for this i atom from the interaction with this j atom. */
 660             velec            = _mm256_and_ps(velec,cutoff_mask);
 661             velec            = _mm256_andnot_ps(dummy_mask,velec);
 662             velecsum         = _mm256_add_ps(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 667
 668             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm256_mul_ps(fscal,dx10);
 672             ty               = _mm256_mul_ps(fscal,dy10);
 673             tz               = _mm256_mul_ps(fscal,dz10);
 674
 675             /* Update vectorial force */
 676             fix1             = _mm256_add_ps(fix1,tx);
 677             fiy1             = _mm256_add_ps(fiy1,ty);
 678             fiz1             = _mm256_add_ps(fiz1,tz);
 679
 680             fjx0             = _mm256_add_ps(fjx0,tx);
 681             fjy0             = _mm256_add_ps(fjy0,ty);
 682             fjz0             = _mm256_add_ps(fjz0,tz);
 683
 684             }
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 691             {
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq20             = _mm256_mul_ps(iq2,jq0);
 695
 696             /* REACTION-FIELD ELECTROSTATICS */
 697             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 698             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 699
 700             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velec            = _mm256_and_ps(velec,cutoff_mask);
 704             velec            = _mm256_andnot_ps(dummy_mask,velec);
 705             velecsum         = _mm256_add_ps(velecsum,velec);
 706
 707             fscal            = felec;
 708
 709             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 710
 711             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 712
 713             /* Calculate temporary vectorial force */
 714             tx               = _mm256_mul_ps(fscal,dx20);
 715             ty               = _mm256_mul_ps(fscal,dy20);
 716             tz               = _mm256_mul_ps(fscal,dz20);
 717
 718             /* Update vectorial force */
 719             fix2             = _mm256_add_ps(fix2,tx);
 720             fiy2             = _mm256_add_ps(fiy2,ty);
 721             fiz2             = _mm256_add_ps(fiz2,tz);
 722
 723             fjx0             = _mm256_add_ps(fjx0,tx);
 724             fjy0             = _mm256_add_ps(fjy0,ty);
 725             fjz0             = _mm256_add_ps(fjz0,tz);
 726
 727             }
 728
 729             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 730             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 731             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 732             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 733             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 734             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 735             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 736             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 737
 738             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 739
 740             /* Inner loop uses 148 flops */
 741         }
 742
 743         /* End of innermost loop */
 744
 745         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 746                                                  f+i_coord_offset,fshift+i_shift_offset);
 747
 748         ggid                        = gid[iidx];
 749         /* Update potential energies */
 750         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 751         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 752
 753         /* Increment number of inner iterations */
 754         inneriter                  += j_index_end - j_index_start;
 755
 756         /* Outer loop uses 20 flops */
 757     }
 758
 759     /* Increment number of outer iterations */
 760     outeriter        += nri;
 761
 762     /* Update outer/inner flops */
 763
 764     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*148);
 765 }
 766 /*
 767  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_single
 768  * Electrostatics interaction: ReactionField
 769  * VdW interaction:            CubicSplineTable
 770  * Geometry:                   Water3-Particle
 771  * Calculate force/pot:        Force
 772  */
 773 void
 774 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_256_single
 775                     (t_nblist * gmx_restrict                nlist,
 776                      rvec * gmx_restrict                    xx,
 777                      rvec * gmx_restrict                    ff,
 778                      t_forcerec * gmx_restrict              fr,
 779                      t_mdatoms * gmx_restrict               mdatoms,
 780                      nb_kernel_data_t * gmx_restrict        kernel_data,
 781                      t_nrnb * gmx_restrict                  nrnb)
 782 {
 783     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 784      * just 0 for non-waters.
 785      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 786      * jnr indices corresponding to data put in the four positions in the SIMD register.
 787      */
 788     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 789     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 790     int              jnrA,jnrB,jnrC,jnrD;
 791     int              jnrE,jnrF,jnrG,jnrH;
 792     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 793     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 794     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 795     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 796     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 797     real             rcutoff_scalar;
 798     real             *shiftvec,*fshift,*x,*f;
 799     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 800     real             scratch[4*DIM];
 801     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 802     real *           vdwioffsetptr0;
 803     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 804     real *           vdwioffsetptr1;
 805     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 806     real *           vdwioffsetptr2;
 807     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 808     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 809     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 810     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 811     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 812     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 813     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 814     real             *charge;
 815     int              nvdwtype;
 816     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 817     int              *vdwtype;
 818     real             *vdwparam;
 819     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 820     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 821     __m256i          vfitab;
 822     __m128i          vfitab_lo,vfitab_hi;
 823     __m128i          ifour       = _mm_set1_epi32(4);
 824     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 825     real             *vftab;
 826     __m256           dummy_mask,cutoff_mask;
 827     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 828     __m256           one     = _mm256_set1_ps(1.0);
 829     __m256           two     = _mm256_set1_ps(2.0);
 830     x                = xx[0];
 831     f                = ff[0];
 832
 833     nri              = nlist->nri;
 834     iinr             = nlist->iinr;
 835     jindex           = nlist->jindex;
 836     jjnr             = nlist->jjnr;
 837     shiftidx         = nlist->shift;
 838     gid              = nlist->gid;
 839     shiftvec         = fr->shift_vec[0];
 840     fshift           = fr->fshift[0];
 841     facel            = _mm256_set1_ps(fr->epsfac);
 842     charge           = mdatoms->chargeA;
 843     krf              = _mm256_set1_ps(fr->ic->k_rf);
 844     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 845     crf              = _mm256_set1_ps(fr->ic->c_rf);
 846     nvdwtype         = fr->ntype;
 847     vdwparam         = fr->nbfp;
 848     vdwtype          = mdatoms->typeA;
 849
 850     vftab            = kernel_data->table_vdw->data;
 851     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 852
 853     /* Setup water-specific parameters */
 854     inr              = nlist->iinr[0];
 855     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 856     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 857     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 858     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 859
 860     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 861     rcutoff_scalar   = fr->rcoulomb;
 862     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 863     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 864
 865     /* Avoid stupid compiler warnings */
 866     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 867     j_coord_offsetA = 0;
 868     j_coord_offsetB = 0;
 869     j_coord_offsetC = 0;
 870     j_coord_offsetD = 0;
 871     j_coord_offsetE = 0;
 872     j_coord_offsetF = 0;
 873     j_coord_offsetG = 0;
 874     j_coord_offsetH = 0;
 875
 876     outeriter        = 0;
 877     inneriter        = 0;
 878
 879     for(iidx=0;iidx<4*DIM;iidx++)
 880     {
 881         scratch[iidx] = 0.0;
 882     }
 883
 884     /* Start outer loop over neighborlists */
 885     for(iidx=0; iidx<nri; iidx++)
 886     {
 887         /* Load shift vector for this list */
 888         i_shift_offset   = DIM*shiftidx[iidx];
 889
 890         /* Load limits for loop over neighbors */
 891         j_index_start    = jindex[iidx];
 892         j_index_end      = jindex[iidx+1];
 893
 894         /* Get outer coordinate index */
 895         inr              = iinr[iidx];
 896         i_coord_offset   = DIM*inr;
 897
 898         /* Load i particle coords and add shift vector */
 899         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 900                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 901
 902         fix0             = _mm256_setzero_ps();
 903         fiy0             = _mm256_setzero_ps();
 904         fiz0             = _mm256_setzero_ps();
 905         fix1             = _mm256_setzero_ps();
 906         fiy1             = _mm256_setzero_ps();
 907         fiz1             = _mm256_setzero_ps();
 908         fix2             = _mm256_setzero_ps();
 909         fiy2             = _mm256_setzero_ps();
 910         fiz2             = _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
 952             /* Calculate squared distance and things based on it */
 953             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 954             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 955             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 956
 957             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 958             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 959             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 960
 961             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 962             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 963             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 964
 965             /* Load parameters for j particles */
 966             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 967                                                                  charge+jnrC+0,charge+jnrD+0,
 968                                                                  charge+jnrE+0,charge+jnrF+0,
 969                                                                  charge+jnrG+0,charge+jnrH+0);
 970             vdwjidx0A        = 2*vdwtype[jnrA+0];
 971             vdwjidx0B        = 2*vdwtype[jnrB+0];
 972             vdwjidx0C        = 2*vdwtype[jnrC+0];
 973             vdwjidx0D        = 2*vdwtype[jnrD+0];
 974             vdwjidx0E        = 2*vdwtype[jnrE+0];
 975             vdwjidx0F        = 2*vdwtype[jnrF+0];
 976             vdwjidx0G        = 2*vdwtype[jnrG+0];
 977             vdwjidx0H        = 2*vdwtype[jnrH+0];
 978
 979             fjx0             = _mm256_setzero_ps();
 980             fjy0             = _mm256_setzero_ps();
 981             fjz0             = _mm256_setzero_ps();
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 988             {
 989
 990             r00              = _mm256_mul_ps(rsq00,rinv00);
 991
 992             /* Compute parameters for interactions between i and j atoms */
 993             qq00             = _mm256_mul_ps(iq0,jq0);
 994             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 995                                             vdwioffsetptr0+vdwjidx0B,
 996                                             vdwioffsetptr0+vdwjidx0C,
 997                                             vdwioffsetptr0+vdwjidx0D,
 998                                             vdwioffsetptr0+vdwjidx0E,
 999                                             vdwioffsetptr0+vdwjidx0F,
1000                                             vdwioffsetptr0+vdwjidx0G,
1001                                             vdwioffsetptr0+vdwjidx0H,
1002                                             &c6_00,&c12_00);
1003
1004             /* Calculate table index by multiplying r with table scale and truncate to integer */
1005             rt               = _mm256_mul_ps(r00,vftabscale);
1006             vfitab           = _mm256_cvttps_epi32(rt);
1007             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1008             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1009             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1010             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1011             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1012             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1013
1014             /* REACTION-FIELD ELECTROSTATICS */
1015             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1016
1017             /* CUBIC SPLINE TABLE DISPERSION */
1018             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1019                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1020             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1021                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1022             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1023                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1024             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1025                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1026             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1027             Heps             = _mm256_mul_ps(vfeps,H);
1028             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1029             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1030             fvdw6            = _mm256_mul_ps(c6_00,FF);
1031
1032             /* CUBIC SPLINE TABLE REPULSION */
1033             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1034             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1035             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1036                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1037             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1038                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1039             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1040                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1041             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1042                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1043             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1044             Heps             = _mm256_mul_ps(vfeps,H);
1045             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1046             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1047             fvdw12           = _mm256_mul_ps(c12_00,FF);
1048             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1049
1050             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1051
1052             fscal            = _mm256_add_ps(felec,fvdw);
1053
1054             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1055
1056             /* Calculate temporary vectorial force */
1057             tx               = _mm256_mul_ps(fscal,dx00);
1058             ty               = _mm256_mul_ps(fscal,dy00);
1059             tz               = _mm256_mul_ps(fscal,dz00);
1060
1061             /* Update vectorial force */
1062             fix0             = _mm256_add_ps(fix0,tx);
1063             fiy0             = _mm256_add_ps(fiy0,ty);
1064             fiz0             = _mm256_add_ps(fiz0,tz);
1065
1066             fjx0             = _mm256_add_ps(fjx0,tx);
1067             fjy0             = _mm256_add_ps(fjy0,ty);
1068             fjz0             = _mm256_add_ps(fjz0,tz);
1069
1070             }
1071
1072             /**************************
1073              * CALCULATE INTERACTIONS *
1074              **************************/
1075
1076             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1077             {
1078
1079             /* Compute parameters for interactions between i and j atoms */
1080             qq10             = _mm256_mul_ps(iq1,jq0);
1081
1082             /* REACTION-FIELD ELECTROSTATICS */
1083             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1084
1085             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1090
1091             /* Calculate temporary vectorial force */
1092             tx               = _mm256_mul_ps(fscal,dx10);
1093             ty               = _mm256_mul_ps(fscal,dy10);
1094             tz               = _mm256_mul_ps(fscal,dz10);
1095
1096             /* Update vectorial force */
1097             fix1             = _mm256_add_ps(fix1,tx);
1098             fiy1             = _mm256_add_ps(fiy1,ty);
1099             fiz1             = _mm256_add_ps(fiz1,tz);
1100
1101             fjx0             = _mm256_add_ps(fjx0,tx);
1102             fjy0             = _mm256_add_ps(fjy0,ty);
1103             fjz0             = _mm256_add_ps(fjz0,tz);
1104
1105             }
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1112             {
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq20             = _mm256_mul_ps(iq2,jq0);
1116
1117             /* REACTION-FIELD ELECTROSTATICS */
1118             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1119
1120             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1121
1122             fscal            = felec;
1123
1124             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm256_mul_ps(fscal,dx20);
1128             ty               = _mm256_mul_ps(fscal,dy20);
1129             tz               = _mm256_mul_ps(fscal,dz20);
1130
1131             /* Update vectorial force */
1132             fix2             = _mm256_add_ps(fix2,tx);
1133             fiy2             = _mm256_add_ps(fiy2,ty);
1134             fiz2             = _mm256_add_ps(fiz2,tz);
1135
1136             fjx0             = _mm256_add_ps(fjx0,tx);
1137             fjy0             = _mm256_add_ps(fjy0,ty);
1138             fjz0             = _mm256_add_ps(fjz0,tz);
1139
1140             }
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 120 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
1210             /* Calculate squared distance and things based on it */
1211             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1212             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1213             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1214
1215             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1216             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1217             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1218
1219             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1220             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1221             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1222
1223             /* Load parameters for j particles */
1224             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1225                                                                  charge+jnrC+0,charge+jnrD+0,
1226                                                                  charge+jnrE+0,charge+jnrF+0,
1227                                                                  charge+jnrG+0,charge+jnrH+0);
1228             vdwjidx0A        = 2*vdwtype[jnrA+0];
1229             vdwjidx0B        = 2*vdwtype[jnrB+0];
1230             vdwjidx0C        = 2*vdwtype[jnrC+0];
1231             vdwjidx0D        = 2*vdwtype[jnrD+0];
1232             vdwjidx0E        = 2*vdwtype[jnrE+0];
1233             vdwjidx0F        = 2*vdwtype[jnrF+0];
1234             vdwjidx0G        = 2*vdwtype[jnrG+0];
1235             vdwjidx0H        = 2*vdwtype[jnrH+0];
1236
1237             fjx0             = _mm256_setzero_ps();
1238             fjy0             = _mm256_setzero_ps();
1239             fjz0             = _mm256_setzero_ps();
1240
1241             /**************************
1242              * CALCULATE INTERACTIONS *
1243              **************************/
1244
1245             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1246             {
1247
1248             r00              = _mm256_mul_ps(rsq00,rinv00);
1249             r00              = _mm256_andnot_ps(dummy_mask,r00);
1250
1251             /* Compute parameters for interactions between i and j atoms */
1252             qq00             = _mm256_mul_ps(iq0,jq0);
1253             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1254                                             vdwioffsetptr0+vdwjidx0B,
1255                                             vdwioffsetptr0+vdwjidx0C,
1256                                             vdwioffsetptr0+vdwjidx0D,
1257                                             vdwioffsetptr0+vdwjidx0E,
1258                                             vdwioffsetptr0+vdwjidx0F,
1259                                             vdwioffsetptr0+vdwjidx0G,
1260                                             vdwioffsetptr0+vdwjidx0H,
1261                                             &c6_00,&c12_00);
1262
1263             /* Calculate table index by multiplying r with table scale and truncate to integer */
1264             rt               = _mm256_mul_ps(r00,vftabscale);
1265             vfitab           = _mm256_cvttps_epi32(rt);
1266             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1267             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1268             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1269             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1270             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1271             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1272
1273             /* REACTION-FIELD ELECTROSTATICS */
1274             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1275
1276             /* CUBIC SPLINE TABLE DISPERSION */
1277             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1278                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1279             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1280                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1281             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1282                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1283             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1284                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1285             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1286             Heps             = _mm256_mul_ps(vfeps,H);
1287             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1288             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1289             fvdw6            = _mm256_mul_ps(c6_00,FF);
1290
1291             /* CUBIC SPLINE TABLE REPULSION */
1292             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1293             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1294             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1296             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1298             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1299                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1300             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1301                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1302             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1303             Heps             = _mm256_mul_ps(vfeps,H);
1304             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1305             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1306             fvdw12           = _mm256_mul_ps(c12_00,FF);
1307             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1308
1309             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1310
1311             fscal            = _mm256_add_ps(felec,fvdw);
1312
1313             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1314
1315             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1316
1317             /* Calculate temporary vectorial force */
1318             tx               = _mm256_mul_ps(fscal,dx00);
1319             ty               = _mm256_mul_ps(fscal,dy00);
1320             tz               = _mm256_mul_ps(fscal,dz00);
1321
1322             /* Update vectorial force */
1323             fix0             = _mm256_add_ps(fix0,tx);
1324             fiy0             = _mm256_add_ps(fiy0,ty);
1325             fiz0             = _mm256_add_ps(fiz0,tz);
1326
1327             fjx0             = _mm256_add_ps(fjx0,tx);
1328             fjy0             = _mm256_add_ps(fjy0,ty);
1329             fjz0             = _mm256_add_ps(fjz0,tz);
1330
1331             }
1332
1333             /**************************
1334              * CALCULATE INTERACTIONS *
1335              **************************/
1336
1337             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1338             {
1339
1340             /* Compute parameters for interactions between i and j atoms */
1341             qq10             = _mm256_mul_ps(iq1,jq0);
1342
1343             /* REACTION-FIELD ELECTROSTATICS */
1344             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1345
1346             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1347
1348             fscal            = felec;
1349
1350             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1351
1352             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1353
1354             /* Calculate temporary vectorial force */
1355             tx               = _mm256_mul_ps(fscal,dx10);
1356             ty               = _mm256_mul_ps(fscal,dy10);
1357             tz               = _mm256_mul_ps(fscal,dz10);
1358
1359             /* Update vectorial force */
1360             fix1             = _mm256_add_ps(fix1,tx);
1361             fiy1             = _mm256_add_ps(fiy1,ty);
1362             fiz1             = _mm256_add_ps(fiz1,tz);
1363
1364             fjx0             = _mm256_add_ps(fjx0,tx);
1365             fjy0             = _mm256_add_ps(fjy0,ty);
1366             fjz0             = _mm256_add_ps(fjz0,tz);
1367
1368             }
1369
1370             /**************************
1371              * CALCULATE INTERACTIONS *
1372              **************************/
1373
1374             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1375             {
1376
1377             /* Compute parameters for interactions between i and j atoms */
1378             qq20             = _mm256_mul_ps(iq2,jq0);
1379
1380             /* REACTION-FIELD ELECTROSTATICS */
1381             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1382
1383             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1384
1385             fscal            = felec;
1386
1387             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1388
1389             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1390
1391             /* Calculate temporary vectorial force */
1392             tx               = _mm256_mul_ps(fscal,dx20);
1393             ty               = _mm256_mul_ps(fscal,dy20);
1394             tz               = _mm256_mul_ps(fscal,dz20);
1395
1396             /* Update vectorial force */
1397             fix2             = _mm256_add_ps(fix2,tx);
1398             fiy2             = _mm256_add_ps(fiy2,ty);
1399             fiz2             = _mm256_add_ps(fiz2,tz);
1400
1401             fjx0             = _mm256_add_ps(fjx0,tx);
1402             fjy0             = _mm256_add_ps(fjy0,ty);
1403             fjz0             = _mm256_add_ps(fjz0,tz);
1404
1405             }
1406
1407             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1408             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1409             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1410             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1411             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1412             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1413             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1414             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1415
1416             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1417
1418             /* Inner loop uses 121 flops */
1419         }
1420
1421         /* End of innermost loop */
1422
1423         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1424                                                  f+i_coord_offset,fshift+i_shift_offset);
1425
1426         /* Increment number of inner iterations */
1427         inneriter                  += j_index_end - j_index_start;
1428
1429         /* Outer loop uses 18 flops */
1430     }
1431
1432     /* Increment number of outer iterations */
1433     outeriter        += nri;
1434
1435     /* Update outer/inner flops */
1436
1437     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1438 }