src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256i          vfitab;
 111     __m128i          vfitab_lo,vfitab_hi;
 112     __m128i          ifour       = _mm_set1_epi32(4);
 113     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 114     real             *vftab;
 115     __m256           dummy_mask,cutoff_mask;
 116     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 117     __m256           one     = _mm256_set1_ps(1.0);
 118     __m256           two     = _mm256_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm256_set1_ps(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     krf              = _mm256_set1_ps(fr->ic->k_rf);
 133     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 134     crf              = _mm256_set1_ps(fr->ic->c_rf);
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138
 139     vftab            = kernel_data->table_vdw->data;
 140     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 145     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 146     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 147     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153     j_coord_offsetC = 0;
 154     j_coord_offsetD = 0;
 155     j_coord_offsetE = 0;
 156     j_coord_offsetF = 0;
 157     j_coord_offsetG = 0;
 158     j_coord_offsetH = 0;
 159
 160     outeriter        = 0;
 161     inneriter        = 0;
 162
 163     for(iidx=0;iidx<4*DIM;iidx++)
 164     {
 165         scratch[iidx] = 0.0;
 166     }
 167
 168     /* Start outer loop over neighborlists */
 169     for(iidx=0; iidx<nri; iidx++)
 170     {
 171         /* Load shift vector for this list */
 172         i_shift_offset   = DIM*shiftidx[iidx];
 173
 174         /* Load limits for loop over neighbors */
 175         j_index_start    = jindex[iidx];
 176         j_index_end      = jindex[iidx+1];
 177
 178         /* Get outer coordinate index */
 179         inr              = iinr[iidx];
 180         i_coord_offset   = DIM*inr;
 181
 182         /* Load i particle coords and add shift vector */
 183         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 184                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 185
 186         fix0             = _mm256_setzero_ps();
 187         fiy0             = _mm256_setzero_ps();
 188         fiz0             = _mm256_setzero_ps();
 189         fix1             = _mm256_setzero_ps();
 190         fiy1             = _mm256_setzero_ps();
 191         fiz1             = _mm256_setzero_ps();
 192         fix2             = _mm256_setzero_ps();
 193         fiy2             = _mm256_setzero_ps();
 194         fiz2             = _mm256_setzero_ps();
 195         fix3             = _mm256_setzero_ps();
 196         fiy3             = _mm256_setzero_ps();
 197         fiz3             = _mm256_setzero_ps();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm256_setzero_ps();
 201         vvdwsum          = _mm256_setzero_ps();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             jnrC             = jjnr[jidx+2];
 211             jnrD             = jjnr[jidx+3];
 212             jnrE             = jjnr[jidx+4];
 213             jnrF             = jjnr[jidx+5];
 214             jnrG             = jjnr[jidx+6];
 215             jnrH             = jjnr[jidx+7];
 216             j_coord_offsetA  = DIM*jnrA;
 217             j_coord_offsetB  = DIM*jnrB;
 218             j_coord_offsetC  = DIM*jnrC;
 219             j_coord_offsetD  = DIM*jnrD;
 220             j_coord_offsetE  = DIM*jnrE;
 221             j_coord_offsetF  = DIM*jnrF;
 222             j_coord_offsetG  = DIM*jnrG;
 223             j_coord_offsetH  = DIM*jnrH;
 224
 225             /* load j atom coordinates */
 226             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 227                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 228                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 229                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 230                                                  &jx0,&jy0,&jz0);
 231
 232             /* Calculate displacement vector */
 233             dx00             = _mm256_sub_ps(ix0,jx0);
 234             dy00             = _mm256_sub_ps(iy0,jy0);
 235             dz00             = _mm256_sub_ps(iz0,jz0);
 236             dx10             = _mm256_sub_ps(ix1,jx0);
 237             dy10             = _mm256_sub_ps(iy1,jy0);
 238             dz10             = _mm256_sub_ps(iz1,jz0);
 239             dx20             = _mm256_sub_ps(ix2,jx0);
 240             dy20             = _mm256_sub_ps(iy2,jy0);
 241             dz20             = _mm256_sub_ps(iz2,jz0);
 242             dx30             = _mm256_sub_ps(ix3,jx0);
 243             dy30             = _mm256_sub_ps(iy3,jy0);
 244             dz30             = _mm256_sub_ps(iz3,jz0);
 245
 246             /* Calculate squared distance and things based on it */
 247             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 248             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 249             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 250             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 251
 252             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 253             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 254             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 255             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 256
 257             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 258             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 259             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 260
 261             /* Load parameters for j particles */
 262             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 263                                                                  charge+jnrC+0,charge+jnrD+0,
 264                                                                  charge+jnrE+0,charge+jnrF+0,
 265                                                                  charge+jnrG+0,charge+jnrH+0);
 266             vdwjidx0A        = 2*vdwtype[jnrA+0];
 267             vdwjidx0B        = 2*vdwtype[jnrB+0];
 268             vdwjidx0C        = 2*vdwtype[jnrC+0];
 269             vdwjidx0D        = 2*vdwtype[jnrD+0];
 270             vdwjidx0E        = 2*vdwtype[jnrE+0];
 271             vdwjidx0F        = 2*vdwtype[jnrF+0];
 272             vdwjidx0G        = 2*vdwtype[jnrG+0];
 273             vdwjidx0H        = 2*vdwtype[jnrH+0];
 274
 275             fjx0             = _mm256_setzero_ps();
 276             fjy0             = _mm256_setzero_ps();
 277             fjz0             = _mm256_setzero_ps();
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r00              = _mm256_mul_ps(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 287                                             vdwioffsetptr0+vdwjidx0B,
 288                                             vdwioffsetptr0+vdwjidx0C,
 289                                             vdwioffsetptr0+vdwjidx0D,
 290                                             vdwioffsetptr0+vdwjidx0E,
 291                                             vdwioffsetptr0+vdwjidx0F,
 292                                             vdwioffsetptr0+vdwjidx0G,
 293                                             vdwioffsetptr0+vdwjidx0H,
 294                                             &c6_00,&c12_00);
 295
 296             /* Calculate table index by multiplying r with table scale and truncate to integer */
 297             rt               = _mm256_mul_ps(r00,vftabscale);
 298             vfitab           = _mm256_cvttps_epi32(rt);
 299             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 300             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 301             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 302             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 303             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 304             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 305
 306             /* CUBIC SPLINE TABLE DISPERSION */
 307             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 309             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 311             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 313             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 314                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 315             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 316             Heps             = _mm256_mul_ps(vfeps,H);
 317             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 318             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 319             vvdw6            = _mm256_mul_ps(c6_00,VV);
 320             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 321             fvdw6            = _mm256_mul_ps(c6_00,FF);
 322
 323             /* CUBIC SPLINE TABLE REPULSION */
 324             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 325             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 326             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 328             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 330             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 332             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 334             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 335             Heps             = _mm256_mul_ps(vfeps,H);
 336             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 337             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 338             vvdw12           = _mm256_mul_ps(c12_00,VV);
 339             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 340             fvdw12           = _mm256_mul_ps(c12_00,FF);
 341             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 342             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 346
 347             fscal            = fvdw;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_ps(fscal,dx00);
 351             ty               = _mm256_mul_ps(fscal,dy00);
 352             tz               = _mm256_mul_ps(fscal,dz00);
 353
 354             /* Update vectorial force */
 355             fix0             = _mm256_add_ps(fix0,tx);
 356             fiy0             = _mm256_add_ps(fiy0,ty);
 357             fiz0             = _mm256_add_ps(fiz0,tz);
 358
 359             fjx0             = _mm256_add_ps(fjx0,tx);
 360             fjy0             = _mm256_add_ps(fjy0,ty);
 361             fjz0             = _mm256_add_ps(fjz0,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq10             = _mm256_mul_ps(iq1,jq0);
 369
 370             /* REACTION-FIELD ELECTROSTATICS */
 371             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 372             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velecsum         = _mm256_add_ps(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm256_mul_ps(fscal,dx10);
 381             ty               = _mm256_mul_ps(fscal,dy10);
 382             tz               = _mm256_mul_ps(fscal,dz10);
 383
 384             /* Update vectorial force */
 385             fix1             = _mm256_add_ps(fix1,tx);
 386             fiy1             = _mm256_add_ps(fiy1,ty);
 387             fiz1             = _mm256_add_ps(fiz1,tz);
 388
 389             fjx0             = _mm256_add_ps(fjx0,tx);
 390             fjy0             = _mm256_add_ps(fjy0,ty);
 391             fjz0             = _mm256_add_ps(fjz0,tz);
 392
 393             /**************************
 394              * CALCULATE INTERACTIONS *
 395              **************************/
 396
 397             /* Compute parameters for interactions between i and j atoms */
 398             qq20             = _mm256_mul_ps(iq2,jq0);
 399
 400             /* REACTION-FIELD ELECTROSTATICS */
 401             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 402             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velecsum         = _mm256_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm256_mul_ps(fscal,dx20);
 411             ty               = _mm256_mul_ps(fscal,dy20);
 412             tz               = _mm256_mul_ps(fscal,dz20);
 413
 414             /* Update vectorial force */
 415             fix2             = _mm256_add_ps(fix2,tx);
 416             fiy2             = _mm256_add_ps(fiy2,ty);
 417             fiz2             = _mm256_add_ps(fiz2,tz);
 418
 419             fjx0             = _mm256_add_ps(fjx0,tx);
 420             fjy0             = _mm256_add_ps(fjy0,ty);
 421             fjz0             = _mm256_add_ps(fjz0,tz);
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             /* Compute parameters for interactions between i and j atoms */
 428             qq30             = _mm256_mul_ps(iq3,jq0);
 429
 430             /* REACTION-FIELD ELECTROSTATICS */
 431             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 432             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velecsum         = _mm256_add_ps(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             /* Calculate temporary vectorial force */
 440             tx               = _mm256_mul_ps(fscal,dx30);
 441             ty               = _mm256_mul_ps(fscal,dy30);
 442             tz               = _mm256_mul_ps(fscal,dz30);
 443
 444             /* Update vectorial force */
 445             fix3             = _mm256_add_ps(fix3,tx);
 446             fiy3             = _mm256_add_ps(fiy3,ty);
 447             fiz3             = _mm256_add_ps(fiz3,tz);
 448
 449             fjx0             = _mm256_add_ps(fjx0,tx);
 450             fjy0             = _mm256_add_ps(fjy0,ty);
 451             fjz0             = _mm256_add_ps(fjz0,tz);
 452
 453             fjptrA             = f+j_coord_offsetA;
 454             fjptrB             = f+j_coord_offsetB;
 455             fjptrC             = f+j_coord_offsetC;
 456             fjptrD             = f+j_coord_offsetD;
 457             fjptrE             = f+j_coord_offsetE;
 458             fjptrF             = f+j_coord_offsetF;
 459             fjptrG             = f+j_coord_offsetG;
 460             fjptrH             = f+j_coord_offsetH;
 461
 462             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 463
 464             /* Inner loop uses 155 flops */
 465         }
 466
 467         if(jidx<j_index_end)
 468         {
 469
 470             /* Get j neighbor index, and coordinate index */
 471             jnrlistA         = jjnr[jidx];
 472             jnrlistB         = jjnr[jidx+1];
 473             jnrlistC         = jjnr[jidx+2];
 474             jnrlistD         = jjnr[jidx+3];
 475             jnrlistE         = jjnr[jidx+4];
 476             jnrlistF         = jjnr[jidx+5];
 477             jnrlistG         = jjnr[jidx+6];
 478             jnrlistH         = jjnr[jidx+7];
 479             /* Sign of each element will be negative for non-real atoms.
 480              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 481              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 482              */
 483             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 484                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 485
 486             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 487             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 488             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 489             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 490             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 491             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 492             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 493             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 494             j_coord_offsetA  = DIM*jnrA;
 495             j_coord_offsetB  = DIM*jnrB;
 496             j_coord_offsetC  = DIM*jnrC;
 497             j_coord_offsetD  = DIM*jnrD;
 498             j_coord_offsetE  = DIM*jnrE;
 499             j_coord_offsetF  = DIM*jnrF;
 500             j_coord_offsetG  = DIM*jnrG;
 501             j_coord_offsetH  = DIM*jnrH;
 502
 503             /* load j atom coordinates */
 504             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 505                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 506                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 507                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 508                                                  &jx0,&jy0,&jz0);
 509
 510             /* Calculate displacement vector */
 511             dx00             = _mm256_sub_ps(ix0,jx0);
 512             dy00             = _mm256_sub_ps(iy0,jy0);
 513             dz00             = _mm256_sub_ps(iz0,jz0);
 514             dx10             = _mm256_sub_ps(ix1,jx0);
 515             dy10             = _mm256_sub_ps(iy1,jy0);
 516             dz10             = _mm256_sub_ps(iz1,jz0);
 517             dx20             = _mm256_sub_ps(ix2,jx0);
 518             dy20             = _mm256_sub_ps(iy2,jy0);
 519             dz20             = _mm256_sub_ps(iz2,jz0);
 520             dx30             = _mm256_sub_ps(ix3,jx0);
 521             dy30             = _mm256_sub_ps(iy3,jy0);
 522             dz30             = _mm256_sub_ps(iz3,jz0);
 523
 524             /* Calculate squared distance and things based on it */
 525             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 526             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 527             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 528             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 529
 530             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 531             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 532             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 533             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 534
 535             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 536             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 537             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 538
 539             /* Load parameters for j particles */
 540             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 541                                                                  charge+jnrC+0,charge+jnrD+0,
 542                                                                  charge+jnrE+0,charge+jnrF+0,
 543                                                                  charge+jnrG+0,charge+jnrH+0);
 544             vdwjidx0A        = 2*vdwtype[jnrA+0];
 545             vdwjidx0B        = 2*vdwtype[jnrB+0];
 546             vdwjidx0C        = 2*vdwtype[jnrC+0];
 547             vdwjidx0D        = 2*vdwtype[jnrD+0];
 548             vdwjidx0E        = 2*vdwtype[jnrE+0];
 549             vdwjidx0F        = 2*vdwtype[jnrF+0];
 550             vdwjidx0G        = 2*vdwtype[jnrG+0];
 551             vdwjidx0H        = 2*vdwtype[jnrH+0];
 552
 553             fjx0             = _mm256_setzero_ps();
 554             fjy0             = _mm256_setzero_ps();
 555             fjz0             = _mm256_setzero_ps();
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             r00              = _mm256_mul_ps(rsq00,rinv00);
 562             r00              = _mm256_andnot_ps(dummy_mask,r00);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 566                                             vdwioffsetptr0+vdwjidx0B,
 567                                             vdwioffsetptr0+vdwjidx0C,
 568                                             vdwioffsetptr0+vdwjidx0D,
 569                                             vdwioffsetptr0+vdwjidx0E,
 570                                             vdwioffsetptr0+vdwjidx0F,
 571                                             vdwioffsetptr0+vdwjidx0G,
 572                                             vdwioffsetptr0+vdwjidx0H,
 573                                             &c6_00,&c12_00);
 574
 575             /* Calculate table index by multiplying r with table scale and truncate to integer */
 576             rt               = _mm256_mul_ps(r00,vftabscale);
 577             vfitab           = _mm256_cvttps_epi32(rt);
 578             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 579             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 580             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 581             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 582             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 583             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 584
 585             /* CUBIC SPLINE TABLE DISPERSION */
 586             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 587                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 588             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 589                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 590             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 591                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 592             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 593                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 594             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 595             Heps             = _mm256_mul_ps(vfeps,H);
 596             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 597             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 598             vvdw6            = _mm256_mul_ps(c6_00,VV);
 599             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 600             fvdw6            = _mm256_mul_ps(c6_00,FF);
 601
 602             /* CUBIC SPLINE TABLE REPULSION */
 603             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 604             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 605             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 606                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 607             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 608                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 609             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 610                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 611             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 612                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 613             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 614             Heps             = _mm256_mul_ps(vfeps,H);
 615             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 616             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 617             vvdw12           = _mm256_mul_ps(c12_00,VV);
 618             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 619             fvdw12           = _mm256_mul_ps(c12_00,FF);
 620             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 621             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 625             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 626
 627             fscal            = fvdw;
 628
 629             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm256_mul_ps(fscal,dx00);
 633             ty               = _mm256_mul_ps(fscal,dy00);
 634             tz               = _mm256_mul_ps(fscal,dz00);
 635
 636             /* Update vectorial force */
 637             fix0             = _mm256_add_ps(fix0,tx);
 638             fiy0             = _mm256_add_ps(fiy0,ty);
 639             fiz0             = _mm256_add_ps(fiz0,tz);
 640
 641             fjx0             = _mm256_add_ps(fjx0,tx);
 642             fjy0             = _mm256_add_ps(fjy0,ty);
 643             fjz0             = _mm256_add_ps(fjz0,tz);
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq10             = _mm256_mul_ps(iq1,jq0);
 651
 652             /* REACTION-FIELD ELECTROSTATICS */
 653             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 654             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             velec            = _mm256_andnot_ps(dummy_mask,velec);
 658             velecsum         = _mm256_add_ps(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm256_mul_ps(fscal,dx10);
 666             ty               = _mm256_mul_ps(fscal,dy10);
 667             tz               = _mm256_mul_ps(fscal,dz10);
 668
 669             /* Update vectorial force */
 670             fix1             = _mm256_add_ps(fix1,tx);
 671             fiy1             = _mm256_add_ps(fiy1,ty);
 672             fiz1             = _mm256_add_ps(fiz1,tz);
 673
 674             fjx0             = _mm256_add_ps(fjx0,tx);
 675             fjy0             = _mm256_add_ps(fjy0,ty);
 676             fjz0             = _mm256_add_ps(fjz0,tz);
 677
 678             /**************************
 679              * CALCULATE INTERACTIONS *
 680              **************************/
 681
 682             /* Compute parameters for interactions between i and j atoms */
 683             qq20             = _mm256_mul_ps(iq2,jq0);
 684
 685             /* REACTION-FIELD ELECTROSTATICS */
 686             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 687             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 688
 689             /* Update potential sum for this i atom from the interaction with this j atom. */
 690             velec            = _mm256_andnot_ps(dummy_mask,velec);
 691             velecsum         = _mm256_add_ps(velecsum,velec);
 692
 693             fscal            = felec;
 694
 695             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 696
 697             /* Calculate temporary vectorial force */
 698             tx               = _mm256_mul_ps(fscal,dx20);
 699             ty               = _mm256_mul_ps(fscal,dy20);
 700             tz               = _mm256_mul_ps(fscal,dz20);
 701
 702             /* Update vectorial force */
 703             fix2             = _mm256_add_ps(fix2,tx);
 704             fiy2             = _mm256_add_ps(fiy2,ty);
 705             fiz2             = _mm256_add_ps(fiz2,tz);
 706
 707             fjx0             = _mm256_add_ps(fjx0,tx);
 708             fjy0             = _mm256_add_ps(fjy0,ty);
 709             fjz0             = _mm256_add_ps(fjz0,tz);
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             /* Compute parameters for interactions between i and j atoms */
 716             qq30             = _mm256_mul_ps(iq3,jq0);
 717
 718             /* REACTION-FIELD ELECTROSTATICS */
 719             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 720             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 721
 722             /* Update potential sum for this i atom from the interaction with this j atom. */
 723             velec            = _mm256_andnot_ps(dummy_mask,velec);
 724             velecsum         = _mm256_add_ps(velecsum,velec);
 725
 726             fscal            = felec;
 727
 728             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 729
 730             /* Calculate temporary vectorial force */
 731             tx               = _mm256_mul_ps(fscal,dx30);
 732             ty               = _mm256_mul_ps(fscal,dy30);
 733             tz               = _mm256_mul_ps(fscal,dz30);
 734
 735             /* Update vectorial force */
 736             fix3             = _mm256_add_ps(fix3,tx);
 737             fiy3             = _mm256_add_ps(fiy3,ty);
 738             fiz3             = _mm256_add_ps(fiz3,tz);
 739
 740             fjx0             = _mm256_add_ps(fjx0,tx);
 741             fjy0             = _mm256_add_ps(fjy0,ty);
 742             fjz0             = _mm256_add_ps(fjz0,tz);
 743
 744             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 745             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 746             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 747             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 748             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 749             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 750             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 751             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 752
 753             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 754
 755             /* Inner loop uses 156 flops */
 756         }
 757
 758         /* End of innermost loop */
 759
 760         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 761                                                  f+i_coord_offset,fshift+i_shift_offset);
 762
 763         ggid                        = gid[iidx];
 764         /* Update potential energies */
 765         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 766         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 767
 768         /* Increment number of inner iterations */
 769         inneriter                  += j_index_end - j_index_start;
 770
 771         /* Outer loop uses 26 flops */
 772     }
 773
 774     /* Increment number of outer iterations */
 775     outeriter        += nri;
 776
 777     /* Update outer/inner flops */
 778
 779     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*156);
 780 }
 781 /*
 782  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_single
 783  * Electrostatics interaction: ReactionField
 784  * VdW interaction:            CubicSplineTable
 785  * Geometry:                   Water4-Particle
 786  * Calculate force/pot:        Force
 787  */
 788 void
 789 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_256_single
 790                     (t_nblist                    * gmx_restrict       nlist,
 791                      rvec                        * gmx_restrict          xx,
 792                      rvec                        * gmx_restrict          ff,
 793                      t_forcerec                  * gmx_restrict          fr,
 794                      t_mdatoms                   * gmx_restrict     mdatoms,
 795                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 796                      t_nrnb                      * gmx_restrict        nrnb)
 797 {
 798     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 799      * just 0 for non-waters.
 800      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 801      * jnr indices corresponding to data put in the four positions in the SIMD register.
 802      */
 803     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 804     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 805     int              jnrA,jnrB,jnrC,jnrD;
 806     int              jnrE,jnrF,jnrG,jnrH;
 807     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 808     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 809     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 810     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 811     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 812     real             rcutoff_scalar;
 813     real             *shiftvec,*fshift,*x,*f;
 814     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 815     real             scratch[4*DIM];
 816     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 817     real *           vdwioffsetptr0;
 818     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 819     real *           vdwioffsetptr1;
 820     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 821     real *           vdwioffsetptr2;
 822     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 823     real *           vdwioffsetptr3;
 824     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 825     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 826     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 827     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 828     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 829     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 830     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 831     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 832     real             *charge;
 833     int              nvdwtype;
 834     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 835     int              *vdwtype;
 836     real             *vdwparam;
 837     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 838     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 839     __m256i          vfitab;
 840     __m128i          vfitab_lo,vfitab_hi;
 841     __m128i          ifour       = _mm_set1_epi32(4);
 842     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 843     real             *vftab;
 844     __m256           dummy_mask,cutoff_mask;
 845     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 846     __m256           one     = _mm256_set1_ps(1.0);
 847     __m256           two     = _mm256_set1_ps(2.0);
 848     x                = xx[0];
 849     f                = ff[0];
 850
 851     nri              = nlist->nri;
 852     iinr             = nlist->iinr;
 853     jindex           = nlist->jindex;
 854     jjnr             = nlist->jjnr;
 855     shiftidx         = nlist->shift;
 856     gid              = nlist->gid;
 857     shiftvec         = fr->shift_vec[0];
 858     fshift           = fr->fshift[0];
 859     facel            = _mm256_set1_ps(fr->epsfac);
 860     charge           = mdatoms->chargeA;
 861     krf              = _mm256_set1_ps(fr->ic->k_rf);
 862     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 863     crf              = _mm256_set1_ps(fr->ic->c_rf);
 864     nvdwtype         = fr->ntype;
 865     vdwparam         = fr->nbfp;
 866     vdwtype          = mdatoms->typeA;
 867
 868     vftab            = kernel_data->table_vdw->data;
 869     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 870
 871     /* Setup water-specific parameters */
 872     inr              = nlist->iinr[0];
 873     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 874     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 875     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 876     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 877
 878     /* Avoid stupid compiler warnings */
 879     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 880     j_coord_offsetA = 0;
 881     j_coord_offsetB = 0;
 882     j_coord_offsetC = 0;
 883     j_coord_offsetD = 0;
 884     j_coord_offsetE = 0;
 885     j_coord_offsetF = 0;
 886     j_coord_offsetG = 0;
 887     j_coord_offsetH = 0;
 888
 889     outeriter        = 0;
 890     inneriter        = 0;
 891
 892     for(iidx=0;iidx<4*DIM;iidx++)
 893     {
 894         scratch[iidx] = 0.0;
 895     }
 896
 897     /* Start outer loop over neighborlists */
 898     for(iidx=0; iidx<nri; iidx++)
 899     {
 900         /* Load shift vector for this list */
 901         i_shift_offset   = DIM*shiftidx[iidx];
 902
 903         /* Load limits for loop over neighbors */
 904         j_index_start    = jindex[iidx];
 905         j_index_end      = jindex[iidx+1];
 906
 907         /* Get outer coordinate index */
 908         inr              = iinr[iidx];
 909         i_coord_offset   = DIM*inr;
 910
 911         /* Load i particle coords and add shift vector */
 912         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 913                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 914
 915         fix0             = _mm256_setzero_ps();
 916         fiy0             = _mm256_setzero_ps();
 917         fiz0             = _mm256_setzero_ps();
 918         fix1             = _mm256_setzero_ps();
 919         fiy1             = _mm256_setzero_ps();
 920         fiz1             = _mm256_setzero_ps();
 921         fix2             = _mm256_setzero_ps();
 922         fiy2             = _mm256_setzero_ps();
 923         fiz2             = _mm256_setzero_ps();
 924         fix3             = _mm256_setzero_ps();
 925         fiy3             = _mm256_setzero_ps();
 926         fiz3             = _mm256_setzero_ps();
 927
 928         /* Start inner kernel loop */
 929         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 930         {
 931
 932             /* Get j neighbor index, and coordinate index */
 933             jnrA             = jjnr[jidx];
 934             jnrB             = jjnr[jidx+1];
 935             jnrC             = jjnr[jidx+2];
 936             jnrD             = jjnr[jidx+3];
 937             jnrE             = jjnr[jidx+4];
 938             jnrF             = jjnr[jidx+5];
 939             jnrG             = jjnr[jidx+6];
 940             jnrH             = jjnr[jidx+7];
 941             j_coord_offsetA  = DIM*jnrA;
 942             j_coord_offsetB  = DIM*jnrB;
 943             j_coord_offsetC  = DIM*jnrC;
 944             j_coord_offsetD  = DIM*jnrD;
 945             j_coord_offsetE  = DIM*jnrE;
 946             j_coord_offsetF  = DIM*jnrF;
 947             j_coord_offsetG  = DIM*jnrG;
 948             j_coord_offsetH  = DIM*jnrH;
 949
 950             /* load j atom coordinates */
 951             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 952                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 953                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 954                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 955                                                  &jx0,&jy0,&jz0);
 956
 957             /* Calculate displacement vector */
 958             dx00             = _mm256_sub_ps(ix0,jx0);
 959             dy00             = _mm256_sub_ps(iy0,jy0);
 960             dz00             = _mm256_sub_ps(iz0,jz0);
 961             dx10             = _mm256_sub_ps(ix1,jx0);
 962             dy10             = _mm256_sub_ps(iy1,jy0);
 963             dz10             = _mm256_sub_ps(iz1,jz0);
 964             dx20             = _mm256_sub_ps(ix2,jx0);
 965             dy20             = _mm256_sub_ps(iy2,jy0);
 966             dz20             = _mm256_sub_ps(iz2,jz0);
 967             dx30             = _mm256_sub_ps(ix3,jx0);
 968             dy30             = _mm256_sub_ps(iy3,jy0);
 969             dz30             = _mm256_sub_ps(iz3,jz0);
 970
 971             /* Calculate squared distance and things based on it */
 972             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 973             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 974             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 975             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 976
 977             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 978             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 979             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 980             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 981
 982             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 983             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 984             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 985
 986             /* Load parameters for j particles */
 987             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 988                                                                  charge+jnrC+0,charge+jnrD+0,
 989                                                                  charge+jnrE+0,charge+jnrF+0,
 990                                                                  charge+jnrG+0,charge+jnrH+0);
 991             vdwjidx0A        = 2*vdwtype[jnrA+0];
 992             vdwjidx0B        = 2*vdwtype[jnrB+0];
 993             vdwjidx0C        = 2*vdwtype[jnrC+0];
 994             vdwjidx0D        = 2*vdwtype[jnrD+0];
 995             vdwjidx0E        = 2*vdwtype[jnrE+0];
 996             vdwjidx0F        = 2*vdwtype[jnrF+0];
 997             vdwjidx0G        = 2*vdwtype[jnrG+0];
 998             vdwjidx0H        = 2*vdwtype[jnrH+0];
 999
1000             fjx0             = _mm256_setzero_ps();
1001             fjy0             = _mm256_setzero_ps();
1002             fjz0             = _mm256_setzero_ps();
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r00              = _mm256_mul_ps(rsq00,rinv00);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1012                                             vdwioffsetptr0+vdwjidx0B,
1013                                             vdwioffsetptr0+vdwjidx0C,
1014                                             vdwioffsetptr0+vdwjidx0D,
1015                                             vdwioffsetptr0+vdwjidx0E,
1016                                             vdwioffsetptr0+vdwjidx0F,
1017                                             vdwioffsetptr0+vdwjidx0G,
1018                                             vdwioffsetptr0+vdwjidx0H,
1019                                             &c6_00,&c12_00);
1020
1021             /* Calculate table index by multiplying r with table scale and truncate to integer */
1022             rt               = _mm256_mul_ps(r00,vftabscale);
1023             vfitab           = _mm256_cvttps_epi32(rt);
1024             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1025             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1026             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1027             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1028             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1029             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1030
1031             /* CUBIC SPLINE TABLE DISPERSION */
1032             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1033                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1034             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1035                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1036             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1037                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1038             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1039                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1040             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1041             Heps             = _mm256_mul_ps(vfeps,H);
1042             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1043             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1044             fvdw6            = _mm256_mul_ps(c6_00,FF);
1045
1046             /* CUBIC SPLINE TABLE REPULSION */
1047             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1048             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1049             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1050                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1051             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1052                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1053             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1054                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1055             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1056                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1057             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1058             Heps             = _mm256_mul_ps(vfeps,H);
1059             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1060             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1061             fvdw12           = _mm256_mul_ps(c12_00,FF);
1062             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1063
1064             fscal            = fvdw;
1065
1066             /* Calculate temporary vectorial force */
1067             tx               = _mm256_mul_ps(fscal,dx00);
1068             ty               = _mm256_mul_ps(fscal,dy00);
1069             tz               = _mm256_mul_ps(fscal,dz00);
1070
1071             /* Update vectorial force */
1072             fix0             = _mm256_add_ps(fix0,tx);
1073             fiy0             = _mm256_add_ps(fiy0,ty);
1074             fiz0             = _mm256_add_ps(fiz0,tz);
1075
1076             fjx0             = _mm256_add_ps(fjx0,tx);
1077             fjy0             = _mm256_add_ps(fjy0,ty);
1078             fjz0             = _mm256_add_ps(fjz0,tz);
1079
1080             /**************************
1081              * CALCULATE INTERACTIONS *
1082              **************************/
1083
1084             /* Compute parameters for interactions between i and j atoms */
1085             qq10             = _mm256_mul_ps(iq1,jq0);
1086
1087             /* REACTION-FIELD ELECTROSTATICS */
1088             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1089
1090             fscal            = felec;
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_ps(fscal,dx10);
1094             ty               = _mm256_mul_ps(fscal,dy10);
1095             tz               = _mm256_mul_ps(fscal,dz10);
1096
1097             /* Update vectorial force */
1098             fix1             = _mm256_add_ps(fix1,tx);
1099             fiy1             = _mm256_add_ps(fiy1,ty);
1100             fiz1             = _mm256_add_ps(fiz1,tz);
1101
1102             fjx0             = _mm256_add_ps(fjx0,tx);
1103             fjy0             = _mm256_add_ps(fjy0,ty);
1104             fjz0             = _mm256_add_ps(fjz0,tz);
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             qq20             = _mm256_mul_ps(iq2,jq0);
1112
1113             /* REACTION-FIELD ELECTROSTATICS */
1114             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1115
1116             fscal            = felec;
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm256_mul_ps(fscal,dx20);
1120             ty               = _mm256_mul_ps(fscal,dy20);
1121             tz               = _mm256_mul_ps(fscal,dz20);
1122
1123             /* Update vectorial force */
1124             fix2             = _mm256_add_ps(fix2,tx);
1125             fiy2             = _mm256_add_ps(fiy2,ty);
1126             fiz2             = _mm256_add_ps(fiz2,tz);
1127
1128             fjx0             = _mm256_add_ps(fjx0,tx);
1129             fjy0             = _mm256_add_ps(fjy0,ty);
1130             fjz0             = _mm256_add_ps(fjz0,tz);
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq30             = _mm256_mul_ps(iq3,jq0);
1138
1139             /* REACTION-FIELD ELECTROSTATICS */
1140             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1141
1142             fscal            = felec;
1143
1144             /* Calculate temporary vectorial force */
1145             tx               = _mm256_mul_ps(fscal,dx30);
1146             ty               = _mm256_mul_ps(fscal,dy30);
1147             tz               = _mm256_mul_ps(fscal,dz30);
1148
1149             /* Update vectorial force */
1150             fix3             = _mm256_add_ps(fix3,tx);
1151             fiy3             = _mm256_add_ps(fiy3,ty);
1152             fiz3             = _mm256_add_ps(fiz3,tz);
1153
1154             fjx0             = _mm256_add_ps(fjx0,tx);
1155             fjy0             = _mm256_add_ps(fjy0,ty);
1156             fjz0             = _mm256_add_ps(fjz0,tz);
1157
1158             fjptrA             = f+j_coord_offsetA;
1159             fjptrB             = f+j_coord_offsetB;
1160             fjptrC             = f+j_coord_offsetC;
1161             fjptrD             = f+j_coord_offsetD;
1162             fjptrE             = f+j_coord_offsetE;
1163             fjptrF             = f+j_coord_offsetF;
1164             fjptrG             = f+j_coord_offsetG;
1165             fjptrH             = f+j_coord_offsetH;
1166
1167             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1168
1169             /* Inner loop uses 132 flops */
1170         }
1171
1172         if(jidx<j_index_end)
1173         {
1174
1175             /* Get j neighbor index, and coordinate index */
1176             jnrlistA         = jjnr[jidx];
1177             jnrlistB         = jjnr[jidx+1];
1178             jnrlistC         = jjnr[jidx+2];
1179             jnrlistD         = jjnr[jidx+3];
1180             jnrlistE         = jjnr[jidx+4];
1181             jnrlistF         = jjnr[jidx+5];
1182             jnrlistG         = jjnr[jidx+6];
1183             jnrlistH         = jjnr[jidx+7];
1184             /* Sign of each element will be negative for non-real atoms.
1185              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1186              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1187              */
1188             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1189                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1190
1191             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1192             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1193             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1194             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1195             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1196             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1197             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1198             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1199             j_coord_offsetA  = DIM*jnrA;
1200             j_coord_offsetB  = DIM*jnrB;
1201             j_coord_offsetC  = DIM*jnrC;
1202             j_coord_offsetD  = DIM*jnrD;
1203             j_coord_offsetE  = DIM*jnrE;
1204             j_coord_offsetF  = DIM*jnrF;
1205             j_coord_offsetG  = DIM*jnrG;
1206             j_coord_offsetH  = DIM*jnrH;
1207
1208             /* load j atom coordinates */
1209             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1211                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1212                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1213                                                  &jx0,&jy0,&jz0);
1214
1215             /* Calculate displacement vector */
1216             dx00             = _mm256_sub_ps(ix0,jx0);
1217             dy00             = _mm256_sub_ps(iy0,jy0);
1218             dz00             = _mm256_sub_ps(iz0,jz0);
1219             dx10             = _mm256_sub_ps(ix1,jx0);
1220             dy10             = _mm256_sub_ps(iy1,jy0);
1221             dz10             = _mm256_sub_ps(iz1,jz0);
1222             dx20             = _mm256_sub_ps(ix2,jx0);
1223             dy20             = _mm256_sub_ps(iy2,jy0);
1224             dz20             = _mm256_sub_ps(iz2,jz0);
1225             dx30             = _mm256_sub_ps(ix3,jx0);
1226             dy30             = _mm256_sub_ps(iy3,jy0);
1227             dz30             = _mm256_sub_ps(iz3,jz0);
1228
1229             /* Calculate squared distance and things based on it */
1230             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1231             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1232             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1233             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1234
1235             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1236             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1237             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1238             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1239
1240             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1241             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1242             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1243
1244             /* Load parameters for j particles */
1245             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1246                                                                  charge+jnrC+0,charge+jnrD+0,
1247                                                                  charge+jnrE+0,charge+jnrF+0,
1248                                                                  charge+jnrG+0,charge+jnrH+0);
1249             vdwjidx0A        = 2*vdwtype[jnrA+0];
1250             vdwjidx0B        = 2*vdwtype[jnrB+0];
1251             vdwjidx0C        = 2*vdwtype[jnrC+0];
1252             vdwjidx0D        = 2*vdwtype[jnrD+0];
1253             vdwjidx0E        = 2*vdwtype[jnrE+0];
1254             vdwjidx0F        = 2*vdwtype[jnrF+0];
1255             vdwjidx0G        = 2*vdwtype[jnrG+0];
1256             vdwjidx0H        = 2*vdwtype[jnrH+0];
1257
1258             fjx0             = _mm256_setzero_ps();
1259             fjy0             = _mm256_setzero_ps();
1260             fjz0             = _mm256_setzero_ps();
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             r00              = _mm256_mul_ps(rsq00,rinv00);
1267             r00              = _mm256_andnot_ps(dummy_mask,r00);
1268
1269             /* Compute parameters for interactions between i and j atoms */
1270             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1271                                             vdwioffsetptr0+vdwjidx0B,
1272                                             vdwioffsetptr0+vdwjidx0C,
1273                                             vdwioffsetptr0+vdwjidx0D,
1274                                             vdwioffsetptr0+vdwjidx0E,
1275                                             vdwioffsetptr0+vdwjidx0F,
1276                                             vdwioffsetptr0+vdwjidx0G,
1277                                             vdwioffsetptr0+vdwjidx0H,
1278                                             &c6_00,&c12_00);
1279
1280             /* Calculate table index by multiplying r with table scale and truncate to integer */
1281             rt               = _mm256_mul_ps(r00,vftabscale);
1282             vfitab           = _mm256_cvttps_epi32(rt);
1283             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1284             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1285             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1286             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1287             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1288             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1289
1290             /* CUBIC SPLINE TABLE DISPERSION */
1291             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1293             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1295             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1297             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1298                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1299             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1300             Heps             = _mm256_mul_ps(vfeps,H);
1301             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1302             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1303             fvdw6            = _mm256_mul_ps(c6_00,FF);
1304
1305             /* CUBIC SPLINE TABLE REPULSION */
1306             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1307             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1308             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1310             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1312             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1314             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1316             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1317             Heps             = _mm256_mul_ps(vfeps,H);
1318             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1319             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1320             fvdw12           = _mm256_mul_ps(c12_00,FF);
1321             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1322
1323             fscal            = fvdw;
1324
1325             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1326
1327             /* Calculate temporary vectorial force */
1328             tx               = _mm256_mul_ps(fscal,dx00);
1329             ty               = _mm256_mul_ps(fscal,dy00);
1330             tz               = _mm256_mul_ps(fscal,dz00);
1331
1332             /* Update vectorial force */
1333             fix0             = _mm256_add_ps(fix0,tx);
1334             fiy0             = _mm256_add_ps(fiy0,ty);
1335             fiz0             = _mm256_add_ps(fiz0,tz);
1336
1337             fjx0             = _mm256_add_ps(fjx0,tx);
1338             fjy0             = _mm256_add_ps(fjy0,ty);
1339             fjz0             = _mm256_add_ps(fjz0,tz);
1340
1341             /**************************
1342              * CALCULATE INTERACTIONS *
1343              **************************/
1344
1345             /* Compute parameters for interactions between i and j atoms */
1346             qq10             = _mm256_mul_ps(iq1,jq0);
1347
1348             /* REACTION-FIELD ELECTROSTATICS */
1349             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1350
1351             fscal            = felec;
1352
1353             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1354
1355             /* Calculate temporary vectorial force */
1356             tx               = _mm256_mul_ps(fscal,dx10);
1357             ty               = _mm256_mul_ps(fscal,dy10);
1358             tz               = _mm256_mul_ps(fscal,dz10);
1359
1360             /* Update vectorial force */
1361             fix1             = _mm256_add_ps(fix1,tx);
1362             fiy1             = _mm256_add_ps(fiy1,ty);
1363             fiz1             = _mm256_add_ps(fiz1,tz);
1364
1365             fjx0             = _mm256_add_ps(fjx0,tx);
1366             fjy0             = _mm256_add_ps(fjy0,ty);
1367             fjz0             = _mm256_add_ps(fjz0,tz);
1368
1369             /**************************
1370              * CALCULATE INTERACTIONS *
1371              **************************/
1372
1373             /* Compute parameters for interactions between i and j atoms */
1374             qq20             = _mm256_mul_ps(iq2,jq0);
1375
1376             /* REACTION-FIELD ELECTROSTATICS */
1377             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1378
1379             fscal            = felec;
1380
1381             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1382
1383             /* Calculate temporary vectorial force */
1384             tx               = _mm256_mul_ps(fscal,dx20);
1385             ty               = _mm256_mul_ps(fscal,dy20);
1386             tz               = _mm256_mul_ps(fscal,dz20);
1387
1388             /* Update vectorial force */
1389             fix2             = _mm256_add_ps(fix2,tx);
1390             fiy2             = _mm256_add_ps(fiy2,ty);
1391             fiz2             = _mm256_add_ps(fiz2,tz);
1392
1393             fjx0             = _mm256_add_ps(fjx0,tx);
1394             fjy0             = _mm256_add_ps(fjy0,ty);
1395             fjz0             = _mm256_add_ps(fjz0,tz);
1396
1397             /**************************
1398              * CALCULATE INTERACTIONS *
1399              **************************/
1400
1401             /* Compute parameters for interactions between i and j atoms */
1402             qq30             = _mm256_mul_ps(iq3,jq0);
1403
1404             /* REACTION-FIELD ELECTROSTATICS */
1405             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1406
1407             fscal            = felec;
1408
1409             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1410
1411             /* Calculate temporary vectorial force */
1412             tx               = _mm256_mul_ps(fscal,dx30);
1413             ty               = _mm256_mul_ps(fscal,dy30);
1414             tz               = _mm256_mul_ps(fscal,dz30);
1415
1416             /* Update vectorial force */
1417             fix3             = _mm256_add_ps(fix3,tx);
1418             fiy3             = _mm256_add_ps(fiy3,ty);
1419             fiz3             = _mm256_add_ps(fiz3,tz);
1420
1421             fjx0             = _mm256_add_ps(fjx0,tx);
1422             fjy0             = _mm256_add_ps(fjy0,ty);
1423             fjz0             = _mm256_add_ps(fjz0,tz);
1424
1425             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1426             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1427             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1428             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1429             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1430             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1431             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1432             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1433
1434             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1435
1436             /* Inner loop uses 133 flops */
1437         }
1438
1439         /* End of innermost loop */
1440
1441         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1442                                                  f+i_coord_offset,fshift+i_shift_offset);
1443
1444         /* Increment number of inner iterations */
1445         inneriter                  += j_index_end - j_index_start;
1446
1447         /* Outer loop uses 24 flops */
1448     }
1449
1450     /* Increment number of outer iterations */
1451     outeriter        += nri;
1452
1453     /* Update outer/inner flops */
1454
1455     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*133);
1456 }