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