src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_ElecEw_VdwCSTab_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 104     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 105     __m256i          vfitab;
 106     __m128i          vfitab_lo,vfitab_hi;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m256i          ewitab;
 111     __m128i          ewitab_lo,ewitab_hi;
 112     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 114     real             *ewtab;
 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     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     vftab            = kernel_data->table_vdw->data;
 137     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 138
 139     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 140     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 141     beta2            = _mm256_mul_ps(beta,beta);
 142     beta3            = _mm256_mul_ps(beta,beta2);
 143
 144     ewtab            = fr->ic->tabq_coul_FDV0;
 145     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 146     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 147
 148     /* Setup water-specific parameters */
 149     inr              = nlist->iinr[0];
 150     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 151     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 152     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 153     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159     j_coord_offsetC = 0;
 160     j_coord_offsetD = 0;
 161     j_coord_offsetE = 0;
 162     j_coord_offsetF = 0;
 163     j_coord_offsetG = 0;
 164     j_coord_offsetH = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 191
 192         fix0             = _mm256_setzero_ps();
 193         fiy0             = _mm256_setzero_ps();
 194         fiz0             = _mm256_setzero_ps();
 195         fix1             = _mm256_setzero_ps();
 196         fiy1             = _mm256_setzero_ps();
 197         fiz1             = _mm256_setzero_ps();
 198         fix2             = _mm256_setzero_ps();
 199         fiy2             = _mm256_setzero_ps();
 200         fiz2             = _mm256_setzero_ps();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm256_setzero_ps();
 204         vvdwsum          = _mm256_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             jnrE             = jjnr[jidx+4];
 216             jnrF             = jjnr[jidx+5];
 217             jnrG             = jjnr[jidx+6];
 218             jnrH             = jjnr[jidx+7];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223             j_coord_offsetE  = DIM*jnrE;
 224             j_coord_offsetF  = DIM*jnrF;
 225             j_coord_offsetG  = DIM*jnrG;
 226             j_coord_offsetH  = DIM*jnrH;
 227
 228             /* load j atom coordinates */
 229             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 230                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 231                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 232                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 233                                                  &jx0,&jy0,&jz0);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm256_sub_ps(ix0,jx0);
 237             dy00             = _mm256_sub_ps(iy0,jy0);
 238             dz00             = _mm256_sub_ps(iz0,jz0);
 239             dx10             = _mm256_sub_ps(ix1,jx0);
 240             dy10             = _mm256_sub_ps(iy1,jy0);
 241             dz10             = _mm256_sub_ps(iz1,jz0);
 242             dx20             = _mm256_sub_ps(ix2,jx0);
 243             dy20             = _mm256_sub_ps(iy2,jy0);
 244             dz20             = _mm256_sub_ps(iz2,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
 251             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 252             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 253             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 254
 255             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 256             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 257             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 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             qq00             = _mm256_mul_ps(iq0,jq0);
 285             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 286                                             vdwioffsetptr0+vdwjidx0B,
 287                                             vdwioffsetptr0+vdwjidx0C,
 288                                             vdwioffsetptr0+vdwjidx0D,
 289                                             vdwioffsetptr0+vdwjidx0E,
 290                                             vdwioffsetptr0+vdwjidx0F,
 291                                             vdwioffsetptr0+vdwjidx0G,
 292                                             vdwioffsetptr0+vdwjidx0H,
 293                                             &c6_00,&c12_00);
 294
 295             /* Calculate table index by multiplying r with table scale and truncate to integer */
 296             rt               = _mm256_mul_ps(r00,vftabscale);
 297             vfitab           = _mm256_cvttps_epi32(rt);
 298             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 299             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 300             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 301             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 302             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 303             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 304
 305             /* EWALD ELECTROSTATICS */
 306
 307             /* Analytical PME correction */
 308             zeta2            = _mm256_mul_ps(beta2,rsq00);
 309             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 310             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 311             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 312             felec            = _mm256_mul_ps(qq00,felec);
 313             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 314             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 315             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 316             velec            = _mm256_mul_ps(qq00,velec);
 317
 318             /* CUBIC SPLINE TABLE DISPERSION */
 319             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 320                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 321             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 323             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 325             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 327             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 328             Heps             = _mm256_mul_ps(vfeps,H);
 329             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 330             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 331             vvdw6            = _mm256_mul_ps(c6_00,VV);
 332             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 333             fvdw6            = _mm256_mul_ps(c6_00,FF);
 334
 335             /* CUBIC SPLINE TABLE REPULSION */
 336             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 337             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 338             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 339                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 340             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 341                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 342             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 343                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 344             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 345                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 346             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 347             Heps             = _mm256_mul_ps(vfeps,H);
 348             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 349             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 350             vvdw12           = _mm256_mul_ps(c12_00,VV);
 351             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 352             fvdw12           = _mm256_mul_ps(c12_00,FF);
 353             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 354             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm256_add_ps(velecsum,velec);
 358             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 359
 360             fscal            = _mm256_add_ps(felec,fvdw);
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm256_mul_ps(fscal,dx00);
 364             ty               = _mm256_mul_ps(fscal,dy00);
 365             tz               = _mm256_mul_ps(fscal,dz00);
 366
 367             /* Update vectorial force */
 368             fix0             = _mm256_add_ps(fix0,tx);
 369             fiy0             = _mm256_add_ps(fiy0,ty);
 370             fiz0             = _mm256_add_ps(fiz0,tz);
 371
 372             fjx0             = _mm256_add_ps(fjx0,tx);
 373             fjy0             = _mm256_add_ps(fjy0,ty);
 374             fjz0             = _mm256_add_ps(fjz0,tz);
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             r10              = _mm256_mul_ps(rsq10,rinv10);
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq10             = _mm256_mul_ps(iq1,jq0);
 384
 385             /* EWALD ELECTROSTATICS */
 386
 387             /* Analytical PME correction */
 388             zeta2            = _mm256_mul_ps(beta2,rsq10);
 389             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 390             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 391             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 392             felec            = _mm256_mul_ps(qq10,felec);
 393             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 394             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 395             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 396             velec            = _mm256_mul_ps(qq10,velec);
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velecsum         = _mm256_add_ps(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             /* Calculate temporary vectorial force */
 404             tx               = _mm256_mul_ps(fscal,dx10);
 405             ty               = _mm256_mul_ps(fscal,dy10);
 406             tz               = _mm256_mul_ps(fscal,dz10);
 407
 408             /* Update vectorial force */
 409             fix1             = _mm256_add_ps(fix1,tx);
 410             fiy1             = _mm256_add_ps(fiy1,ty);
 411             fiz1             = _mm256_add_ps(fiz1,tz);
 412
 413             fjx0             = _mm256_add_ps(fjx0,tx);
 414             fjy0             = _mm256_add_ps(fjy0,ty);
 415             fjz0             = _mm256_add_ps(fjz0,tz);
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             r20              = _mm256_mul_ps(rsq20,rinv20);
 422
 423             /* Compute parameters for interactions between i and j atoms */
 424             qq20             = _mm256_mul_ps(iq2,jq0);
 425
 426             /* EWALD ELECTROSTATICS */
 427
 428             /* Analytical PME correction */
 429             zeta2            = _mm256_mul_ps(beta2,rsq20);
 430             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 431             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 432             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 433             felec            = _mm256_mul_ps(qq20,felec);
 434             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 435             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 436             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 437             velec            = _mm256_mul_ps(qq20,velec);
 438
 439             /* Update potential sum for this i atom from the interaction with this j atom. */
 440             velecsum         = _mm256_add_ps(velecsum,velec);
 441
 442             fscal            = felec;
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm256_mul_ps(fscal,dx20);
 446             ty               = _mm256_mul_ps(fscal,dy20);
 447             tz               = _mm256_mul_ps(fscal,dz20);
 448
 449             /* Update vectorial force */
 450             fix2             = _mm256_add_ps(fix2,tx);
 451             fiy2             = _mm256_add_ps(fiy2,ty);
 452             fiz2             = _mm256_add_ps(fiz2,tz);
 453
 454             fjx0             = _mm256_add_ps(fjx0,tx);
 455             fjy0             = _mm256_add_ps(fjy0,ty);
 456             fjz0             = _mm256_add_ps(fjz0,tz);
 457
 458             fjptrA             = f+j_coord_offsetA;
 459             fjptrB             = f+j_coord_offsetB;
 460             fjptrC             = f+j_coord_offsetC;
 461             fjptrD             = f+j_coord_offsetD;
 462             fjptrE             = f+j_coord_offsetE;
 463             fjptrF             = f+j_coord_offsetF;
 464             fjptrG             = f+j_coord_offsetG;
 465             fjptrH             = f+j_coord_offsetH;
 466
 467             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 468
 469             /* Inner loop uses 289 flops */
 470         }
 471
 472         if(jidx<j_index_end)
 473         {
 474
 475             /* Get j neighbor index, and coordinate index */
 476             jnrlistA         = jjnr[jidx];
 477             jnrlistB         = jjnr[jidx+1];
 478             jnrlistC         = jjnr[jidx+2];
 479             jnrlistD         = jjnr[jidx+3];
 480             jnrlistE         = jjnr[jidx+4];
 481             jnrlistF         = jjnr[jidx+5];
 482             jnrlistG         = jjnr[jidx+6];
 483             jnrlistH         = jjnr[jidx+7];
 484             /* Sign of each element will be negative for non-real atoms.
 485              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 486              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 487              */
 488             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 489                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 490
 491             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 492             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 493             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 494             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 495             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 496             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 497             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 498             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 499             j_coord_offsetA  = DIM*jnrA;
 500             j_coord_offsetB  = DIM*jnrB;
 501             j_coord_offsetC  = DIM*jnrC;
 502             j_coord_offsetD  = DIM*jnrD;
 503             j_coord_offsetE  = DIM*jnrE;
 504             j_coord_offsetF  = DIM*jnrF;
 505             j_coord_offsetG  = DIM*jnrG;
 506             j_coord_offsetH  = DIM*jnrH;
 507
 508             /* load j atom coordinates */
 509             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 510                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 511                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 512                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 513                                                  &jx0,&jy0,&jz0);
 514
 515             /* Calculate displacement vector */
 516             dx00             = _mm256_sub_ps(ix0,jx0);
 517             dy00             = _mm256_sub_ps(iy0,jy0);
 518             dz00             = _mm256_sub_ps(iz0,jz0);
 519             dx10             = _mm256_sub_ps(ix1,jx0);
 520             dy10             = _mm256_sub_ps(iy1,jy0);
 521             dz10             = _mm256_sub_ps(iz1,jz0);
 522             dx20             = _mm256_sub_ps(ix2,jx0);
 523             dy20             = _mm256_sub_ps(iy2,jy0);
 524             dz20             = _mm256_sub_ps(iz2,jz0);
 525
 526             /* Calculate squared distance and things based on it */
 527             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 528             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 529             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 530
 531             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 532             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 533             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 534
 535             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 536             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 537             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 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             qq00             = _mm256_mul_ps(iq0,jq0);
 566             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 567                                             vdwioffsetptr0+vdwjidx0B,
 568                                             vdwioffsetptr0+vdwjidx0C,
 569                                             vdwioffsetptr0+vdwjidx0D,
 570                                             vdwioffsetptr0+vdwjidx0E,
 571                                             vdwioffsetptr0+vdwjidx0F,
 572                                             vdwioffsetptr0+vdwjidx0G,
 573                                             vdwioffsetptr0+vdwjidx0H,
 574                                             &c6_00,&c12_00);
 575
 576             /* Calculate table index by multiplying r with table scale and truncate to integer */
 577             rt               = _mm256_mul_ps(r00,vftabscale);
 578             vfitab           = _mm256_cvttps_epi32(rt);
 579             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 580             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 581             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 582             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 583             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 584             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 585
 586             /* EWALD ELECTROSTATICS */
 587
 588             /* Analytical PME correction */
 589             zeta2            = _mm256_mul_ps(beta2,rsq00);
 590             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 591             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 592             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 593             felec            = _mm256_mul_ps(qq00,felec);
 594             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 595             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 596             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 597             velec            = _mm256_mul_ps(qq00,velec);
 598
 599             /* CUBIC SPLINE TABLE DISPERSION */
 600             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 601                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 602             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 603                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 604             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 605                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 606             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 607                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 608             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 609             Heps             = _mm256_mul_ps(vfeps,H);
 610             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 611             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 612             vvdw6            = _mm256_mul_ps(c6_00,VV);
 613             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 614             fvdw6            = _mm256_mul_ps(c6_00,FF);
 615
 616             /* CUBIC SPLINE TABLE REPULSION */
 617             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 618             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 619             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 620                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 621             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 622                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 623             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 625             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 627             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 628             Heps             = _mm256_mul_ps(vfeps,H);
 629             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 630             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 631             vvdw12           = _mm256_mul_ps(c12_00,VV);
 632             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 633             fvdw12           = _mm256_mul_ps(c12_00,FF);
 634             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 635             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm256_andnot_ps(dummy_mask,velec);
 639             velecsum         = _mm256_add_ps(velecsum,velec);
 640             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 641             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 642
 643             fscal            = _mm256_add_ps(felec,fvdw);
 644
 645             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm256_mul_ps(fscal,dx00);
 649             ty               = _mm256_mul_ps(fscal,dy00);
 650             tz               = _mm256_mul_ps(fscal,dz00);
 651
 652             /* Update vectorial force */
 653             fix0             = _mm256_add_ps(fix0,tx);
 654             fiy0             = _mm256_add_ps(fiy0,ty);
 655             fiz0             = _mm256_add_ps(fiz0,tz);
 656
 657             fjx0             = _mm256_add_ps(fjx0,tx);
 658             fjy0             = _mm256_add_ps(fjy0,ty);
 659             fjz0             = _mm256_add_ps(fjz0,tz);
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             r10              = _mm256_mul_ps(rsq10,rinv10);
 666             r10              = _mm256_andnot_ps(dummy_mask,r10);
 667
 668             /* Compute parameters for interactions between i and j atoms */
 669             qq10             = _mm256_mul_ps(iq1,jq0);
 670
 671             /* EWALD ELECTROSTATICS */
 672
 673             /* Analytical PME correction */
 674             zeta2            = _mm256_mul_ps(beta2,rsq10);
 675             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 676             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 677             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 678             felec            = _mm256_mul_ps(qq10,felec);
 679             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 680             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 681             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 682             velec            = _mm256_mul_ps(qq10,velec);
 683
 684             /* Update potential sum for this i atom from the interaction with this j atom. */
 685             velec            = _mm256_andnot_ps(dummy_mask,velec);
 686             velecsum         = _mm256_add_ps(velecsum,velec);
 687
 688             fscal            = felec;
 689
 690             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 691
 692             /* Calculate temporary vectorial force */
 693             tx               = _mm256_mul_ps(fscal,dx10);
 694             ty               = _mm256_mul_ps(fscal,dy10);
 695             tz               = _mm256_mul_ps(fscal,dz10);
 696
 697             /* Update vectorial force */
 698             fix1             = _mm256_add_ps(fix1,tx);
 699             fiy1             = _mm256_add_ps(fiy1,ty);
 700             fiz1             = _mm256_add_ps(fiz1,tz);
 701
 702             fjx0             = _mm256_add_ps(fjx0,tx);
 703             fjy0             = _mm256_add_ps(fjy0,ty);
 704             fjz0             = _mm256_add_ps(fjz0,tz);
 705
 706             /**************************
 707              * CALCULATE INTERACTIONS *
 708              **************************/
 709
 710             r20              = _mm256_mul_ps(rsq20,rinv20);
 711             r20              = _mm256_andnot_ps(dummy_mask,r20);
 712
 713             /* Compute parameters for interactions between i and j atoms */
 714             qq20             = _mm256_mul_ps(iq2,jq0);
 715
 716             /* EWALD ELECTROSTATICS */
 717
 718             /* Analytical PME correction */
 719             zeta2            = _mm256_mul_ps(beta2,rsq20);
 720             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 721             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 722             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 723             felec            = _mm256_mul_ps(qq20,felec);
 724             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 725             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 726             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 727             velec            = _mm256_mul_ps(qq20,velec);
 728
 729             /* Update potential sum for this i atom from the interaction with this j atom. */
 730             velec            = _mm256_andnot_ps(dummy_mask,velec);
 731             velecsum         = _mm256_add_ps(velecsum,velec);
 732
 733             fscal            = felec;
 734
 735             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 736
 737             /* Calculate temporary vectorial force */
 738             tx               = _mm256_mul_ps(fscal,dx20);
 739             ty               = _mm256_mul_ps(fscal,dy20);
 740             tz               = _mm256_mul_ps(fscal,dz20);
 741
 742             /* Update vectorial force */
 743             fix2             = _mm256_add_ps(fix2,tx);
 744             fiy2             = _mm256_add_ps(fiy2,ty);
 745             fiz2             = _mm256_add_ps(fiz2,tz);
 746
 747             fjx0             = _mm256_add_ps(fjx0,tx);
 748             fjy0             = _mm256_add_ps(fjy0,ty);
 749             fjz0             = _mm256_add_ps(fjz0,tz);
 750
 751             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 752             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 753             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 754             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 755             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 756             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 757             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 758             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 759
 760             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 761
 762             /* Inner loop uses 292 flops */
 763         }
 764
 765         /* End of innermost loop */
 766
 767         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 768                                                  f+i_coord_offset,fshift+i_shift_offset);
 769
 770         ggid                        = gid[iidx];
 771         /* Update potential energies */
 772         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 773         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 774
 775         /* Increment number of inner iterations */
 776         inneriter                  += j_index_end - j_index_start;
 777
 778         /* Outer loop uses 20 flops */
 779     }
 780
 781     /* Increment number of outer iterations */
 782     outeriter        += nri;
 783
 784     /* Update outer/inner flops */
 785
 786     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*292);
 787 }
 788 /*
 789  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_single
 790  * Electrostatics interaction: Ewald
 791  * VdW interaction:            CubicSplineTable
 792  * Geometry:                   Water3-Particle
 793  * Calculate force/pot:        Force
 794  */
 795 void
 796 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_single
 797                     (t_nblist                    * gmx_restrict       nlist,
 798                      rvec                        * gmx_restrict          xx,
 799                      rvec                        * gmx_restrict          ff,
 800                      t_forcerec                  * gmx_restrict          fr,
 801                      t_mdatoms                   * gmx_restrict     mdatoms,
 802                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 803                      t_nrnb                      * gmx_restrict        nrnb)
 804 {
 805     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 806      * just 0 for non-waters.
 807      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 808      * jnr indices corresponding to data put in the four positions in the SIMD register.
 809      */
 810     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 811     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 812     int              jnrA,jnrB,jnrC,jnrD;
 813     int              jnrE,jnrF,jnrG,jnrH;
 814     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 815     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 816     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 817     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 818     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 819     real             rcutoff_scalar;
 820     real             *shiftvec,*fshift,*x,*f;
 821     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 822     real             scratch[4*DIM];
 823     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 824     real *           vdwioffsetptr0;
 825     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 826     real *           vdwioffsetptr1;
 827     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 828     real *           vdwioffsetptr2;
 829     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 830     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 831     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 832     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 833     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 834     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 835     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 836     real             *charge;
 837     int              nvdwtype;
 838     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 839     int              *vdwtype;
 840     real             *vdwparam;
 841     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 842     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 843     __m256i          vfitab;
 844     __m128i          vfitab_lo,vfitab_hi;
 845     __m128i          ifour       = _mm_set1_epi32(4);
 846     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 847     real             *vftab;
 848     __m256i          ewitab;
 849     __m128i          ewitab_lo,ewitab_hi;
 850     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 851     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 852     real             *ewtab;
 853     __m256           dummy_mask,cutoff_mask;
 854     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 855     __m256           one     = _mm256_set1_ps(1.0);
 856     __m256           two     = _mm256_set1_ps(2.0);
 857     x                = xx[0];
 858     f                = ff[0];
 859
 860     nri              = nlist->nri;
 861     iinr             = nlist->iinr;
 862     jindex           = nlist->jindex;
 863     jjnr             = nlist->jjnr;
 864     shiftidx         = nlist->shift;
 865     gid              = nlist->gid;
 866     shiftvec         = fr->shift_vec[0];
 867     fshift           = fr->fshift[0];
 868     facel            = _mm256_set1_ps(fr->epsfac);
 869     charge           = mdatoms->chargeA;
 870     nvdwtype         = fr->ntype;
 871     vdwparam         = fr->nbfp;
 872     vdwtype          = mdatoms->typeA;
 873
 874     vftab            = kernel_data->table_vdw->data;
 875     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 876
 877     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 878     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 879     beta2            = _mm256_mul_ps(beta,beta);
 880     beta3            = _mm256_mul_ps(beta,beta2);
 881
 882     ewtab            = fr->ic->tabq_coul_F;
 883     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 884     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 885
 886     /* Setup water-specific parameters */
 887     inr              = nlist->iinr[0];
 888     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 889     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 890     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 891     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 892
 893     /* Avoid stupid compiler warnings */
 894     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 895     j_coord_offsetA = 0;
 896     j_coord_offsetB = 0;
 897     j_coord_offsetC = 0;
 898     j_coord_offsetD = 0;
 899     j_coord_offsetE = 0;
 900     j_coord_offsetF = 0;
 901     j_coord_offsetG = 0;
 902     j_coord_offsetH = 0;
 903
 904     outeriter        = 0;
 905     inneriter        = 0;
 906
 907     for(iidx=0;iidx<4*DIM;iidx++)
 908     {
 909         scratch[iidx] = 0.0;
 910     }
 911
 912     /* Start outer loop over neighborlists */
 913     for(iidx=0; iidx<nri; iidx++)
 914     {
 915         /* Load shift vector for this list */
 916         i_shift_offset   = DIM*shiftidx[iidx];
 917
 918         /* Load limits for loop over neighbors */
 919         j_index_start    = jindex[iidx];
 920         j_index_end      = jindex[iidx+1];
 921
 922         /* Get outer coordinate index */
 923         inr              = iinr[iidx];
 924         i_coord_offset   = DIM*inr;
 925
 926         /* Load i particle coords and add shift vector */
 927         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 928                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 929
 930         fix0             = _mm256_setzero_ps();
 931         fiy0             = _mm256_setzero_ps();
 932         fiz0             = _mm256_setzero_ps();
 933         fix1             = _mm256_setzero_ps();
 934         fiy1             = _mm256_setzero_ps();
 935         fiz1             = _mm256_setzero_ps();
 936         fix2             = _mm256_setzero_ps();
 937         fiy2             = _mm256_setzero_ps();
 938         fiz2             = _mm256_setzero_ps();
 939
 940         /* Start inner kernel loop */
 941         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 942         {
 943
 944             /* Get j neighbor index, and coordinate index */
 945             jnrA             = jjnr[jidx];
 946             jnrB             = jjnr[jidx+1];
 947             jnrC             = jjnr[jidx+2];
 948             jnrD             = jjnr[jidx+3];
 949             jnrE             = jjnr[jidx+4];
 950             jnrF             = jjnr[jidx+5];
 951             jnrG             = jjnr[jidx+6];
 952             jnrH             = jjnr[jidx+7];
 953             j_coord_offsetA  = DIM*jnrA;
 954             j_coord_offsetB  = DIM*jnrB;
 955             j_coord_offsetC  = DIM*jnrC;
 956             j_coord_offsetD  = DIM*jnrD;
 957             j_coord_offsetE  = DIM*jnrE;
 958             j_coord_offsetF  = DIM*jnrF;
 959             j_coord_offsetG  = DIM*jnrG;
 960             j_coord_offsetH  = DIM*jnrH;
 961
 962             /* load j atom coordinates */
 963             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 964                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 965                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 966                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 967                                                  &jx0,&jy0,&jz0);
 968
 969             /* Calculate displacement vector */
 970             dx00             = _mm256_sub_ps(ix0,jx0);
 971             dy00             = _mm256_sub_ps(iy0,jy0);
 972             dz00             = _mm256_sub_ps(iz0,jz0);
 973             dx10             = _mm256_sub_ps(ix1,jx0);
 974             dy10             = _mm256_sub_ps(iy1,jy0);
 975             dz10             = _mm256_sub_ps(iz1,jz0);
 976             dx20             = _mm256_sub_ps(ix2,jx0);
 977             dy20             = _mm256_sub_ps(iy2,jy0);
 978             dz20             = _mm256_sub_ps(iz2,jz0);
 979
 980             /* Calculate squared distance and things based on it */
 981             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 982             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 983             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 984
 985             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 986             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 987             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 988
 989             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 990             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 991             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 992
 993             /* Load parameters for j particles */
 994             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 995                                                                  charge+jnrC+0,charge+jnrD+0,
 996                                                                  charge+jnrE+0,charge+jnrF+0,
 997                                                                  charge+jnrG+0,charge+jnrH+0);
 998             vdwjidx0A        = 2*vdwtype[jnrA+0];
 999             vdwjidx0B        = 2*vdwtype[jnrB+0];
1000             vdwjidx0C        = 2*vdwtype[jnrC+0];
1001             vdwjidx0D        = 2*vdwtype[jnrD+0];
1002             vdwjidx0E        = 2*vdwtype[jnrE+0];
1003             vdwjidx0F        = 2*vdwtype[jnrF+0];
1004             vdwjidx0G        = 2*vdwtype[jnrG+0];
1005             vdwjidx0H        = 2*vdwtype[jnrH+0];
1006
1007             fjx0             = _mm256_setzero_ps();
1008             fjy0             = _mm256_setzero_ps();
1009             fjz0             = _mm256_setzero_ps();
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             r00              = _mm256_mul_ps(rsq00,rinv00);
1016
1017             /* Compute parameters for interactions between i and j atoms */
1018             qq00             = _mm256_mul_ps(iq0,jq0);
1019             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1020                                             vdwioffsetptr0+vdwjidx0B,
1021                                             vdwioffsetptr0+vdwjidx0C,
1022                                             vdwioffsetptr0+vdwjidx0D,
1023                                             vdwioffsetptr0+vdwjidx0E,
1024                                             vdwioffsetptr0+vdwjidx0F,
1025                                             vdwioffsetptr0+vdwjidx0G,
1026                                             vdwioffsetptr0+vdwjidx0H,
1027                                             &c6_00,&c12_00);
1028
1029             /* Calculate table index by multiplying r with table scale and truncate to integer */
1030             rt               = _mm256_mul_ps(r00,vftabscale);
1031             vfitab           = _mm256_cvttps_epi32(rt);
1032             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1033             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1034             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1035             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1036             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1037             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1038
1039             /* EWALD ELECTROSTATICS */
1040
1041             /* Analytical PME correction */
1042             zeta2            = _mm256_mul_ps(beta2,rsq00);
1043             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1044             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1045             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1046             felec            = _mm256_mul_ps(qq00,felec);
1047
1048             /* CUBIC SPLINE TABLE DISPERSION */
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             fvdw6            = _mm256_mul_ps(c6_00,FF);
1062
1063             /* CUBIC SPLINE TABLE REPULSION */
1064             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1065             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1066             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1067                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1068             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1069                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1070             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1071                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1072             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1073                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1074             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1075             Heps             = _mm256_mul_ps(vfeps,H);
1076             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1077             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1078             fvdw12           = _mm256_mul_ps(c12_00,FF);
1079             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1080
1081             fscal            = _mm256_add_ps(felec,fvdw);
1082
1083             /* Calculate temporary vectorial force */
1084             tx               = _mm256_mul_ps(fscal,dx00);
1085             ty               = _mm256_mul_ps(fscal,dy00);
1086             tz               = _mm256_mul_ps(fscal,dz00);
1087
1088             /* Update vectorial force */
1089             fix0             = _mm256_add_ps(fix0,tx);
1090             fiy0             = _mm256_add_ps(fiy0,ty);
1091             fiz0             = _mm256_add_ps(fiz0,tz);
1092
1093             fjx0             = _mm256_add_ps(fjx0,tx);
1094             fjy0             = _mm256_add_ps(fjy0,ty);
1095             fjz0             = _mm256_add_ps(fjz0,tz);
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             r10              = _mm256_mul_ps(rsq10,rinv10);
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq10             = _mm256_mul_ps(iq1,jq0);
1105
1106             /* EWALD ELECTROSTATICS */
1107
1108             /* Analytical PME correction */
1109             zeta2            = _mm256_mul_ps(beta2,rsq10);
1110             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1111             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1112             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1113             felec            = _mm256_mul_ps(qq10,felec);
1114
1115             fscal            = felec;
1116
1117             /* Calculate temporary vectorial force */
1118             tx               = _mm256_mul_ps(fscal,dx10);
1119             ty               = _mm256_mul_ps(fscal,dy10);
1120             tz               = _mm256_mul_ps(fscal,dz10);
1121
1122             /* Update vectorial force */
1123             fix1             = _mm256_add_ps(fix1,tx);
1124             fiy1             = _mm256_add_ps(fiy1,ty);
1125             fiz1             = _mm256_add_ps(fiz1,tz);
1126
1127             fjx0             = _mm256_add_ps(fjx0,tx);
1128             fjy0             = _mm256_add_ps(fjy0,ty);
1129             fjz0             = _mm256_add_ps(fjz0,tz);
1130
1131             /**************************
1132              * CALCULATE INTERACTIONS *
1133              **************************/
1134
1135             r20              = _mm256_mul_ps(rsq20,rinv20);
1136
1137             /* Compute parameters for interactions between i and j atoms */
1138             qq20             = _mm256_mul_ps(iq2,jq0);
1139
1140             /* EWALD ELECTROSTATICS */
1141
1142             /* Analytical PME correction */
1143             zeta2            = _mm256_mul_ps(beta2,rsq20);
1144             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1145             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1146             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1147             felec            = _mm256_mul_ps(qq20,felec);
1148
1149             fscal            = felec;
1150
1151             /* Calculate temporary vectorial force */
1152             tx               = _mm256_mul_ps(fscal,dx20);
1153             ty               = _mm256_mul_ps(fscal,dy20);
1154             tz               = _mm256_mul_ps(fscal,dz20);
1155
1156             /* Update vectorial force */
1157             fix2             = _mm256_add_ps(fix2,tx);
1158             fiy2             = _mm256_add_ps(fiy2,ty);
1159             fiz2             = _mm256_add_ps(fiz2,tz);
1160
1161             fjx0             = _mm256_add_ps(fjx0,tx);
1162             fjy0             = _mm256_add_ps(fjy0,ty);
1163             fjz0             = _mm256_add_ps(fjz0,tz);
1164
1165             fjptrA             = f+j_coord_offsetA;
1166             fjptrB             = f+j_coord_offsetB;
1167             fjptrC             = f+j_coord_offsetC;
1168             fjptrD             = f+j_coord_offsetD;
1169             fjptrE             = f+j_coord_offsetE;
1170             fjptrF             = f+j_coord_offsetF;
1171             fjptrG             = f+j_coord_offsetG;
1172             fjptrH             = f+j_coord_offsetH;
1173
1174             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1175
1176             /* Inner loop uses 197 flops */
1177         }
1178
1179         if(jidx<j_index_end)
1180         {
1181
1182             /* Get j neighbor index, and coordinate index */
1183             jnrlistA         = jjnr[jidx];
1184             jnrlistB         = jjnr[jidx+1];
1185             jnrlistC         = jjnr[jidx+2];
1186             jnrlistD         = jjnr[jidx+3];
1187             jnrlistE         = jjnr[jidx+4];
1188             jnrlistF         = jjnr[jidx+5];
1189             jnrlistG         = jjnr[jidx+6];
1190             jnrlistH         = jjnr[jidx+7];
1191             /* Sign of each element will be negative for non-real atoms.
1192              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1193              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1194              */
1195             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1196                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1197
1198             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1199             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1200             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1201             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1202             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1203             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1204             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1205             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1206             j_coord_offsetA  = DIM*jnrA;
1207             j_coord_offsetB  = DIM*jnrB;
1208             j_coord_offsetC  = DIM*jnrC;
1209             j_coord_offsetD  = DIM*jnrD;
1210             j_coord_offsetE  = DIM*jnrE;
1211             j_coord_offsetF  = DIM*jnrF;
1212             j_coord_offsetG  = DIM*jnrG;
1213             j_coord_offsetH  = DIM*jnrH;
1214
1215             /* load j atom coordinates */
1216             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1217                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1218                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1219                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1220                                                  &jx0,&jy0,&jz0);
1221
1222             /* Calculate displacement vector */
1223             dx00             = _mm256_sub_ps(ix0,jx0);
1224             dy00             = _mm256_sub_ps(iy0,jy0);
1225             dz00             = _mm256_sub_ps(iz0,jz0);
1226             dx10             = _mm256_sub_ps(ix1,jx0);
1227             dy10             = _mm256_sub_ps(iy1,jy0);
1228             dz10             = _mm256_sub_ps(iz1,jz0);
1229             dx20             = _mm256_sub_ps(ix2,jx0);
1230             dy20             = _mm256_sub_ps(iy2,jy0);
1231             dz20             = _mm256_sub_ps(iz2,jz0);
1232
1233             /* Calculate squared distance and things based on it */
1234             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1235             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1236             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1237
1238             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1239             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1240             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1241
1242             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1243             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1244             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1245
1246             /* Load parameters for j particles */
1247             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1248                                                                  charge+jnrC+0,charge+jnrD+0,
1249                                                                  charge+jnrE+0,charge+jnrF+0,
1250                                                                  charge+jnrG+0,charge+jnrH+0);
1251             vdwjidx0A        = 2*vdwtype[jnrA+0];
1252             vdwjidx0B        = 2*vdwtype[jnrB+0];
1253             vdwjidx0C        = 2*vdwtype[jnrC+0];
1254             vdwjidx0D        = 2*vdwtype[jnrD+0];
1255             vdwjidx0E        = 2*vdwtype[jnrE+0];
1256             vdwjidx0F        = 2*vdwtype[jnrF+0];
1257             vdwjidx0G        = 2*vdwtype[jnrG+0];
1258             vdwjidx0H        = 2*vdwtype[jnrH+0];
1259
1260             fjx0             = _mm256_setzero_ps();
1261             fjy0             = _mm256_setzero_ps();
1262             fjz0             = _mm256_setzero_ps();
1263
1264             /**************************
1265              * CALCULATE INTERACTIONS *
1266              **************************/
1267
1268             r00              = _mm256_mul_ps(rsq00,rinv00);
1269             r00              = _mm256_andnot_ps(dummy_mask,r00);
1270
1271             /* Compute parameters for interactions between i and j atoms */
1272             qq00             = _mm256_mul_ps(iq0,jq0);
1273             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1274                                             vdwioffsetptr0+vdwjidx0B,
1275                                             vdwioffsetptr0+vdwjidx0C,
1276                                             vdwioffsetptr0+vdwjidx0D,
1277                                             vdwioffsetptr0+vdwjidx0E,
1278                                             vdwioffsetptr0+vdwjidx0F,
1279                                             vdwioffsetptr0+vdwjidx0G,
1280                                             vdwioffsetptr0+vdwjidx0H,
1281                                             &c6_00,&c12_00);
1282
1283             /* Calculate table index by multiplying r with table scale and truncate to integer */
1284             rt               = _mm256_mul_ps(r00,vftabscale);
1285             vfitab           = _mm256_cvttps_epi32(rt);
1286             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1287             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1288             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1289             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1290             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1291             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1292
1293             /* EWALD ELECTROSTATICS */
1294
1295             /* Analytical PME correction */
1296             zeta2            = _mm256_mul_ps(beta2,rsq00);
1297             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1298             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1299             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1300             felec            = _mm256_mul_ps(qq00,felec);
1301
1302             /* CUBIC SPLINE TABLE DISPERSION */
1303             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1304                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1305             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1306                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1307             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1309             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1311             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1312             Heps             = _mm256_mul_ps(vfeps,H);
1313             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1314             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1315             fvdw6            = _mm256_mul_ps(c6_00,FF);
1316
1317             /* CUBIC SPLINE TABLE REPULSION */
1318             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1319             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1320             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1322             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1324             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1326             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1328             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1329             Heps             = _mm256_mul_ps(vfeps,H);
1330             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1331             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1332             fvdw12           = _mm256_mul_ps(c12_00,FF);
1333             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1334
1335             fscal            = _mm256_add_ps(felec,fvdw);
1336
1337             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1338
1339             /* Calculate temporary vectorial force */
1340             tx               = _mm256_mul_ps(fscal,dx00);
1341             ty               = _mm256_mul_ps(fscal,dy00);
1342             tz               = _mm256_mul_ps(fscal,dz00);
1343
1344             /* Update vectorial force */
1345             fix0             = _mm256_add_ps(fix0,tx);
1346             fiy0             = _mm256_add_ps(fiy0,ty);
1347             fiz0             = _mm256_add_ps(fiz0,tz);
1348
1349             fjx0             = _mm256_add_ps(fjx0,tx);
1350             fjy0             = _mm256_add_ps(fjy0,ty);
1351             fjz0             = _mm256_add_ps(fjz0,tz);
1352
1353             /**************************
1354              * CALCULATE INTERACTIONS *
1355              **************************/
1356
1357             r10              = _mm256_mul_ps(rsq10,rinv10);
1358             r10              = _mm256_andnot_ps(dummy_mask,r10);
1359
1360             /* Compute parameters for interactions between i and j atoms */
1361             qq10             = _mm256_mul_ps(iq1,jq0);
1362
1363             /* EWALD ELECTROSTATICS */
1364
1365             /* Analytical PME correction */
1366             zeta2            = _mm256_mul_ps(beta2,rsq10);
1367             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1368             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1369             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1370             felec            = _mm256_mul_ps(qq10,felec);
1371
1372             fscal            = felec;
1373
1374             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1375
1376             /* Calculate temporary vectorial force */
1377             tx               = _mm256_mul_ps(fscal,dx10);
1378             ty               = _mm256_mul_ps(fscal,dy10);
1379             tz               = _mm256_mul_ps(fscal,dz10);
1380
1381             /* Update vectorial force */
1382             fix1             = _mm256_add_ps(fix1,tx);
1383             fiy1             = _mm256_add_ps(fiy1,ty);
1384             fiz1             = _mm256_add_ps(fiz1,tz);
1385
1386             fjx0             = _mm256_add_ps(fjx0,tx);
1387             fjy0             = _mm256_add_ps(fjy0,ty);
1388             fjz0             = _mm256_add_ps(fjz0,tz);
1389
1390             /**************************
1391              * CALCULATE INTERACTIONS *
1392              **************************/
1393
1394             r20              = _mm256_mul_ps(rsq20,rinv20);
1395             r20              = _mm256_andnot_ps(dummy_mask,r20);
1396
1397             /* Compute parameters for interactions between i and j atoms */
1398             qq20             = _mm256_mul_ps(iq2,jq0);
1399
1400             /* EWALD ELECTROSTATICS */
1401
1402             /* Analytical PME correction */
1403             zeta2            = _mm256_mul_ps(beta2,rsq20);
1404             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1405             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1406             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1407             felec            = _mm256_mul_ps(qq20,felec);
1408
1409             fscal            = felec;
1410
1411             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1412
1413             /* Calculate temporary vectorial force */
1414             tx               = _mm256_mul_ps(fscal,dx20);
1415             ty               = _mm256_mul_ps(fscal,dy20);
1416             tz               = _mm256_mul_ps(fscal,dz20);
1417
1418             /* Update vectorial force */
1419             fix2             = _mm256_add_ps(fix2,tx);
1420             fiy2             = _mm256_add_ps(fiy2,ty);
1421             fiz2             = _mm256_add_ps(fiz2,tz);
1422
1423             fjx0             = _mm256_add_ps(fjx0,tx);
1424             fjy0             = _mm256_add_ps(fjy0,ty);
1425             fjz0             = _mm256_add_ps(fjz0,tz);
1426
1427             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1428             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1429             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1430             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1431             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1432             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1433             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1434             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1435
1436             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1437
1438             /* Inner loop uses 200 flops */
1439         }
1440
1441         /* End of innermost loop */
1442
1443         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1444                                                  f+i_coord_offset,fshift+i_shift_offset);
1445
1446         /* Increment number of inner iterations */
1447         inneriter                  += j_index_end - j_index_start;
1448
1449         /* Outer loop uses 18 flops */
1450     }
1451
1452     /* Increment number of outer iterations */
1453     outeriter        += nri;
1454
1455     /* Update outer/inner flops */
1456
1457     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*200);
1458 }