src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     krf              = _mm256_set1_pd(fr->ic->k_rf);
 128     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 129     crf              = _mm256_set1_pd(fr->ic->c_rf);
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_vdw->data;
 135     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 140     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 141     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm256_setzero_pd();
 178         fiy0             = _mm256_setzero_pd();
 179         fiz0             = _mm256_setzero_pd();
 180         fix1             = _mm256_setzero_pd();
 181         fiy1             = _mm256_setzero_pd();
 182         fiz1             = _mm256_setzero_pd();
 183         fix2             = _mm256_setzero_pd();
 184         fiy2             = _mm256_setzero_pd();
 185         fiz2             = _mm256_setzero_pd();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_pd();
 189         vvdwsum          = _mm256_setzero_pd();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204
 205             /* load j atom coordinates */
 206             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 208                                                  &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm256_sub_pd(ix0,jx0);
 212             dy00             = _mm256_sub_pd(iy0,jy0);
 213             dz00             = _mm256_sub_pd(iz0,jz0);
 214             dx10             = _mm256_sub_pd(ix1,jx0);
 215             dy10             = _mm256_sub_pd(iy1,jy0);
 216             dz10             = _mm256_sub_pd(iz1,jz0);
 217             dx20             = _mm256_sub_pd(ix2,jx0);
 218             dy20             = _mm256_sub_pd(iy2,jy0);
 219             dz20             = _mm256_sub_pd(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 223             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 224             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 225
 226             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 227             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 228             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 229
 230             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 231             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 232             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 236                                                                  charge+jnrC+0,charge+jnrD+0);
 237             vdwjidx0A        = 2*vdwtype[jnrA+0];
 238             vdwjidx0B        = 2*vdwtype[jnrB+0];
 239             vdwjidx0C        = 2*vdwtype[jnrC+0];
 240             vdwjidx0D        = 2*vdwtype[jnrD+0];
 241
 242             fjx0             = _mm256_setzero_pd();
 243             fjy0             = _mm256_setzero_pd();
 244             fjz0             = _mm256_setzero_pd();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             r00              = _mm256_mul_pd(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm256_mul_pd(iq0,jq0);
 254             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 255                                             vdwioffsetptr0+vdwjidx0B,
 256                                             vdwioffsetptr0+vdwjidx0C,
 257                                             vdwioffsetptr0+vdwjidx0D,
 258                                             &c6_00,&c12_00);
 259
 260             /* Calculate table index by multiplying r with table scale and truncate to integer */
 261             rt               = _mm256_mul_pd(r00,vftabscale);
 262             vfitab           = _mm256_cvttpd_epi32(rt);
 263             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 264             vfitab           = _mm_slli_epi32(vfitab,3);
 265
 266             /* REACTION-FIELD ELECTROSTATICS */
 267             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 268             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 269
 270             /* CUBIC SPLINE TABLE DISPERSION */
 271             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 272             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 273             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 274             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 275             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 276             Heps             = _mm256_mul_pd(vfeps,H);
 277             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 278             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 279             vvdw6            = _mm256_mul_pd(c6_00,VV);
 280             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 281             fvdw6            = _mm256_mul_pd(c6_00,FF);
 282
 283             /* CUBIC SPLINE TABLE REPULSION */
 284             vfitab           = _mm_add_epi32(vfitab,ifour);
 285             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 286             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 287             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 288             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 289             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 290             Heps             = _mm256_mul_pd(vfeps,H);
 291             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 292             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 293             vvdw12           = _mm256_mul_pd(c12_00,VV);
 294             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 295             fvdw12           = _mm256_mul_pd(c12_00,FF);
 296             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 297             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm256_add_pd(velecsum,velec);
 301             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 302
 303             fscal            = _mm256_add_pd(felec,fvdw);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm256_mul_pd(fscal,dx00);
 307             ty               = _mm256_mul_pd(fscal,dy00);
 308             tz               = _mm256_mul_pd(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm256_add_pd(fix0,tx);
 312             fiy0             = _mm256_add_pd(fiy0,ty);
 313             fiz0             = _mm256_add_pd(fiz0,tz);
 314
 315             fjx0             = _mm256_add_pd(fjx0,tx);
 316             fjy0             = _mm256_add_pd(fjy0,ty);
 317             fjz0             = _mm256_add_pd(fjz0,tz);
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm256_mul_pd(iq1,jq0);
 325
 326             /* REACTION-FIELD ELECTROSTATICS */
 327             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 328             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm256_add_pd(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_pd(fscal,dx10);
 337             ty               = _mm256_mul_pd(fscal,dy10);
 338             tz               = _mm256_mul_pd(fscal,dz10);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm256_add_pd(fix1,tx);
 342             fiy1             = _mm256_add_pd(fiy1,ty);
 343             fiz1             = _mm256_add_pd(fiz1,tz);
 344
 345             fjx0             = _mm256_add_pd(fjx0,tx);
 346             fjy0             = _mm256_add_pd(fjy0,ty);
 347             fjz0             = _mm256_add_pd(fjz0,tz);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm256_mul_pd(iq2,jq0);
 355
 356             /* REACTION-FIELD ELECTROSTATICS */
 357             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 358             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm256_add_pd(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm256_mul_pd(fscal,dx20);
 367             ty               = _mm256_mul_pd(fscal,dy20);
 368             tz               = _mm256_mul_pd(fscal,dz20);
 369
 370             /* Update vectorial force */
 371             fix2             = _mm256_add_pd(fix2,tx);
 372             fiy2             = _mm256_add_pd(fiy2,ty);
 373             fiz2             = _mm256_add_pd(fiz2,tz);
 374
 375             fjx0             = _mm256_add_pd(fjx0,tx);
 376             fjy0             = _mm256_add_pd(fjy0,ty);
 377             fjz0             = _mm256_add_pd(fjz0,tz);
 378
 379             fjptrA             = f+j_coord_offsetA;
 380             fjptrB             = f+j_coord_offsetB;
 381             fjptrC             = f+j_coord_offsetC;
 382             fjptrD             = f+j_coord_offsetD;
 383
 384             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 385
 386             /* Inner loop uses 134 flops */
 387         }
 388
 389         if(jidx<j_index_end)
 390         {
 391
 392             /* Get j neighbor index, and coordinate index */
 393             jnrlistA         = jjnr[jidx];
 394             jnrlistB         = jjnr[jidx+1];
 395             jnrlistC         = jjnr[jidx+2];
 396             jnrlistD         = jjnr[jidx+3];
 397             /* Sign of each element will be negative for non-real atoms.
 398              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 399              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 400              */
 401             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 402
 403             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 404             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 405             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 406
 407             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 408             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 409             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 410             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 411             j_coord_offsetA  = DIM*jnrA;
 412             j_coord_offsetB  = DIM*jnrB;
 413             j_coord_offsetC  = DIM*jnrC;
 414             j_coord_offsetD  = DIM*jnrD;
 415
 416             /* load j atom coordinates */
 417             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 418                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 419                                                  &jx0,&jy0,&jz0);
 420
 421             /* Calculate displacement vector */
 422             dx00             = _mm256_sub_pd(ix0,jx0);
 423             dy00             = _mm256_sub_pd(iy0,jy0);
 424             dz00             = _mm256_sub_pd(iz0,jz0);
 425             dx10             = _mm256_sub_pd(ix1,jx0);
 426             dy10             = _mm256_sub_pd(iy1,jy0);
 427             dz10             = _mm256_sub_pd(iz1,jz0);
 428             dx20             = _mm256_sub_pd(ix2,jx0);
 429             dy20             = _mm256_sub_pd(iy2,jy0);
 430             dz20             = _mm256_sub_pd(iz2,jz0);
 431
 432             /* Calculate squared distance and things based on it */
 433             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 434             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 435             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 436
 437             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 438             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 439             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 440
 441             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 442             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 443             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 444
 445             /* Load parameters for j particles */
 446             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 447                                                                  charge+jnrC+0,charge+jnrD+0);
 448             vdwjidx0A        = 2*vdwtype[jnrA+0];
 449             vdwjidx0B        = 2*vdwtype[jnrB+0];
 450             vdwjidx0C        = 2*vdwtype[jnrC+0];
 451             vdwjidx0D        = 2*vdwtype[jnrD+0];
 452
 453             fjx0             = _mm256_setzero_pd();
 454             fjy0             = _mm256_setzero_pd();
 455             fjz0             = _mm256_setzero_pd();
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             r00              = _mm256_mul_pd(rsq00,rinv00);
 462             r00              = _mm256_andnot_pd(dummy_mask,r00);
 463
 464             /* Compute parameters for interactions between i and j atoms */
 465             qq00             = _mm256_mul_pd(iq0,jq0);
 466             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 467                                             vdwioffsetptr0+vdwjidx0B,
 468                                             vdwioffsetptr0+vdwjidx0C,
 469                                             vdwioffsetptr0+vdwjidx0D,
 470                                             &c6_00,&c12_00);
 471
 472             /* Calculate table index by multiplying r with table scale and truncate to integer */
 473             rt               = _mm256_mul_pd(r00,vftabscale);
 474             vfitab           = _mm256_cvttpd_epi32(rt);
 475             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 476             vfitab           = _mm_slli_epi32(vfitab,3);
 477
 478             /* REACTION-FIELD ELECTROSTATICS */
 479             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 480             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 481
 482             /* CUBIC SPLINE TABLE DISPERSION */
 483             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 484             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 485             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 486             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 487             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 488             Heps             = _mm256_mul_pd(vfeps,H);
 489             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 490             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 491             vvdw6            = _mm256_mul_pd(c6_00,VV);
 492             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 493             fvdw6            = _mm256_mul_pd(c6_00,FF);
 494
 495             /* CUBIC SPLINE TABLE REPULSION */
 496             vfitab           = _mm_add_epi32(vfitab,ifour);
 497             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 498             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 499             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 500             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 501             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 502             Heps             = _mm256_mul_pd(vfeps,H);
 503             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 504             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 505             vvdw12           = _mm256_mul_pd(c12_00,VV);
 506             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 507             fvdw12           = _mm256_mul_pd(c12_00,FF);
 508             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 509             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm256_andnot_pd(dummy_mask,velec);
 513             velecsum         = _mm256_add_pd(velecsum,velec);
 514             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 515             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 516
 517             fscal            = _mm256_add_pd(felec,fvdw);
 518
 519             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_pd(fscal,dx00);
 523             ty               = _mm256_mul_pd(fscal,dy00);
 524             tz               = _mm256_mul_pd(fscal,dz00);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm256_add_pd(fix0,tx);
 528             fiy0             = _mm256_add_pd(fiy0,ty);
 529             fiz0             = _mm256_add_pd(fiz0,tz);
 530
 531             fjx0             = _mm256_add_pd(fjx0,tx);
 532             fjy0             = _mm256_add_pd(fjy0,ty);
 533             fjz0             = _mm256_add_pd(fjz0,tz);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq10             = _mm256_mul_pd(iq1,jq0);
 541
 542             /* REACTION-FIELD ELECTROSTATICS */
 543             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 544             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             velec            = _mm256_andnot_pd(dummy_mask,velec);
 548             velecsum         = _mm256_add_pd(velecsum,velec);
 549
 550             fscal            = felec;
 551
 552             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm256_mul_pd(fscal,dx10);
 556             ty               = _mm256_mul_pd(fscal,dy10);
 557             tz               = _mm256_mul_pd(fscal,dz10);
 558
 559             /* Update vectorial force */
 560             fix1             = _mm256_add_pd(fix1,tx);
 561             fiy1             = _mm256_add_pd(fiy1,ty);
 562             fiz1             = _mm256_add_pd(fiz1,tz);
 563
 564             fjx0             = _mm256_add_pd(fjx0,tx);
 565             fjy0             = _mm256_add_pd(fjy0,ty);
 566             fjz0             = _mm256_add_pd(fjz0,tz);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm256_mul_pd(iq2,jq0);
 574
 575             /* REACTION-FIELD ELECTROSTATICS */
 576             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 577             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velec            = _mm256_andnot_pd(dummy_mask,velec);
 581             velecsum         = _mm256_add_pd(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm256_mul_pd(fscal,dx20);
 589             ty               = _mm256_mul_pd(fscal,dy20);
 590             tz               = _mm256_mul_pd(fscal,dz20);
 591
 592             /* Update vectorial force */
 593             fix2             = _mm256_add_pd(fix2,tx);
 594             fiy2             = _mm256_add_pd(fiy2,ty);
 595             fiz2             = _mm256_add_pd(fiz2,tz);
 596
 597             fjx0             = _mm256_add_pd(fjx0,tx);
 598             fjy0             = _mm256_add_pd(fjy0,ty);
 599             fjz0             = _mm256_add_pd(fjz0,tz);
 600
 601             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 602             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 603             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 604             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 605
 606             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 607
 608             /* Inner loop uses 135 flops */
 609         }
 610
 611         /* End of innermost loop */
 612
 613         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 614                                                  f+i_coord_offset,fshift+i_shift_offset);
 615
 616         ggid                        = gid[iidx];
 617         /* Update potential energies */
 618         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 619         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 20 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*135);
 633 }
 634 /*
 635  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_double
 636  * Electrostatics interaction: ReactionField
 637  * VdW interaction:            CubicSplineTable
 638  * Geometry:                   Water3-Particle
 639  * Calculate force/pot:        Force
 640  */
 641 void
 642 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_double
 643                     (t_nblist                    * gmx_restrict       nlist,
 644                      rvec                        * gmx_restrict          xx,
 645                      rvec                        * gmx_restrict          ff,
 646                      t_forcerec                  * gmx_restrict          fr,
 647                      t_mdatoms                   * gmx_restrict     mdatoms,
 648                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 649                      t_nrnb                      * gmx_restrict        nrnb)
 650 {
 651     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 652      * just 0 for non-waters.
 653      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 654      * jnr indices corresponding to data put in the four positions in the SIMD register.
 655      */
 656     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 657     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 658     int              jnrA,jnrB,jnrC,jnrD;
 659     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 660     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 661     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 662     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 663     real             rcutoff_scalar;
 664     real             *shiftvec,*fshift,*x,*f;
 665     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 666     real             scratch[4*DIM];
 667     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 668     real *           vdwioffsetptr0;
 669     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 670     real *           vdwioffsetptr1;
 671     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 672     real *           vdwioffsetptr2;
 673     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 674     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 675     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 676     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 677     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 678     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 679     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 680     real             *charge;
 681     int              nvdwtype;
 682     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 683     int              *vdwtype;
 684     real             *vdwparam;
 685     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 686     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 687     __m128i          vfitab;
 688     __m128i          ifour       = _mm_set1_epi32(4);
 689     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 690     real             *vftab;
 691     __m256d          dummy_mask,cutoff_mask;
 692     __m128           tmpmask0,tmpmask1;
 693     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 694     __m256d          one     = _mm256_set1_pd(1.0);
 695     __m256d          two     = _mm256_set1_pd(2.0);
 696     x                = xx[0];
 697     f                = ff[0];
 698
 699     nri              = nlist->nri;
 700     iinr             = nlist->iinr;
 701     jindex           = nlist->jindex;
 702     jjnr             = nlist->jjnr;
 703     shiftidx         = nlist->shift;
 704     gid              = nlist->gid;
 705     shiftvec         = fr->shift_vec[0];
 706     fshift           = fr->fshift[0];
 707     facel            = _mm256_set1_pd(fr->epsfac);
 708     charge           = mdatoms->chargeA;
 709     krf              = _mm256_set1_pd(fr->ic->k_rf);
 710     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 711     crf              = _mm256_set1_pd(fr->ic->c_rf);
 712     nvdwtype         = fr->ntype;
 713     vdwparam         = fr->nbfp;
 714     vdwtype          = mdatoms->typeA;
 715
 716     vftab            = kernel_data->table_vdw->data;
 717     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 718
 719     /* Setup water-specific parameters */
 720     inr              = nlist->iinr[0];
 721     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 722     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 723     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 724     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 725
 726     /* Avoid stupid compiler warnings */
 727     jnrA = jnrB = jnrC = jnrD = 0;
 728     j_coord_offsetA = 0;
 729     j_coord_offsetB = 0;
 730     j_coord_offsetC = 0;
 731     j_coord_offsetD = 0;
 732
 733     outeriter        = 0;
 734     inneriter        = 0;
 735
 736     for(iidx=0;iidx<4*DIM;iidx++)
 737     {
 738         scratch[iidx] = 0.0;
 739     }
 740
 741     /* Start outer loop over neighborlists */
 742     for(iidx=0; iidx<nri; iidx++)
 743     {
 744         /* Load shift vector for this list */
 745         i_shift_offset   = DIM*shiftidx[iidx];
 746
 747         /* Load limits for loop over neighbors */
 748         j_index_start    = jindex[iidx];
 749         j_index_end      = jindex[iidx+1];
 750
 751         /* Get outer coordinate index */
 752         inr              = iinr[iidx];
 753         i_coord_offset   = DIM*inr;
 754
 755         /* Load i particle coords and add shift vector */
 756         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 757                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 758
 759         fix0             = _mm256_setzero_pd();
 760         fiy0             = _mm256_setzero_pd();
 761         fiz0             = _mm256_setzero_pd();
 762         fix1             = _mm256_setzero_pd();
 763         fiy1             = _mm256_setzero_pd();
 764         fiz1             = _mm256_setzero_pd();
 765         fix2             = _mm256_setzero_pd();
 766         fiy2             = _mm256_setzero_pd();
 767         fiz2             = _mm256_setzero_pd();
 768
 769         /* Start inner kernel loop */
 770         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 771         {
 772
 773             /* Get j neighbor index, and coordinate index */
 774             jnrA             = jjnr[jidx];
 775             jnrB             = jjnr[jidx+1];
 776             jnrC             = jjnr[jidx+2];
 777             jnrD             = jjnr[jidx+3];
 778             j_coord_offsetA  = DIM*jnrA;
 779             j_coord_offsetB  = DIM*jnrB;
 780             j_coord_offsetC  = DIM*jnrC;
 781             j_coord_offsetD  = DIM*jnrD;
 782
 783             /* load j atom coordinates */
 784             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 785                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 786                                                  &jx0,&jy0,&jz0);
 787
 788             /* Calculate displacement vector */
 789             dx00             = _mm256_sub_pd(ix0,jx0);
 790             dy00             = _mm256_sub_pd(iy0,jy0);
 791             dz00             = _mm256_sub_pd(iz0,jz0);
 792             dx10             = _mm256_sub_pd(ix1,jx0);
 793             dy10             = _mm256_sub_pd(iy1,jy0);
 794             dz10             = _mm256_sub_pd(iz1,jz0);
 795             dx20             = _mm256_sub_pd(ix2,jx0);
 796             dy20             = _mm256_sub_pd(iy2,jy0);
 797             dz20             = _mm256_sub_pd(iz2,jz0);
 798
 799             /* Calculate squared distance and things based on it */
 800             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 801             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 802             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 803
 804             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 805             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 806             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 807
 808             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 809             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 810             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 811
 812             /* Load parameters for j particles */
 813             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 814                                                                  charge+jnrC+0,charge+jnrD+0);
 815             vdwjidx0A        = 2*vdwtype[jnrA+0];
 816             vdwjidx0B        = 2*vdwtype[jnrB+0];
 817             vdwjidx0C        = 2*vdwtype[jnrC+0];
 818             vdwjidx0D        = 2*vdwtype[jnrD+0];
 819
 820             fjx0             = _mm256_setzero_pd();
 821             fjy0             = _mm256_setzero_pd();
 822             fjz0             = _mm256_setzero_pd();
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             r00              = _mm256_mul_pd(rsq00,rinv00);
 829
 830             /* Compute parameters for interactions between i and j atoms */
 831             qq00             = _mm256_mul_pd(iq0,jq0);
 832             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 833                                             vdwioffsetptr0+vdwjidx0B,
 834                                             vdwioffsetptr0+vdwjidx0C,
 835                                             vdwioffsetptr0+vdwjidx0D,
 836                                             &c6_00,&c12_00);
 837
 838             /* Calculate table index by multiplying r with table scale and truncate to integer */
 839             rt               = _mm256_mul_pd(r00,vftabscale);
 840             vfitab           = _mm256_cvttpd_epi32(rt);
 841             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 842             vfitab           = _mm_slli_epi32(vfitab,3);
 843
 844             /* REACTION-FIELD ELECTROSTATICS */
 845             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 846
 847             /* CUBIC SPLINE TABLE DISPERSION */
 848             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 849             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 850             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 851             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 852             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 853             Heps             = _mm256_mul_pd(vfeps,H);
 854             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 855             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 856             fvdw6            = _mm256_mul_pd(c6_00,FF);
 857
 858             /* CUBIC SPLINE TABLE REPULSION */
 859             vfitab           = _mm_add_epi32(vfitab,ifour);
 860             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 861             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 862             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 863             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 864             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 865             Heps             = _mm256_mul_pd(vfeps,H);
 866             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 867             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 868             fvdw12           = _mm256_mul_pd(c12_00,FF);
 869             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 870
 871             fscal            = _mm256_add_pd(felec,fvdw);
 872
 873             /* Calculate temporary vectorial force */
 874             tx               = _mm256_mul_pd(fscal,dx00);
 875             ty               = _mm256_mul_pd(fscal,dy00);
 876             tz               = _mm256_mul_pd(fscal,dz00);
 877
 878             /* Update vectorial force */
 879             fix0             = _mm256_add_pd(fix0,tx);
 880             fiy0             = _mm256_add_pd(fiy0,ty);
 881             fiz0             = _mm256_add_pd(fiz0,tz);
 882
 883             fjx0             = _mm256_add_pd(fjx0,tx);
 884             fjy0             = _mm256_add_pd(fjy0,ty);
 885             fjz0             = _mm256_add_pd(fjz0,tz);
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq10             = _mm256_mul_pd(iq1,jq0);
 893
 894             /* REACTION-FIELD ELECTROSTATICS */
 895             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 896
 897             fscal            = felec;
 898
 899             /* Calculate temporary vectorial force */
 900             tx               = _mm256_mul_pd(fscal,dx10);
 901             ty               = _mm256_mul_pd(fscal,dy10);
 902             tz               = _mm256_mul_pd(fscal,dz10);
 903
 904             /* Update vectorial force */
 905             fix1             = _mm256_add_pd(fix1,tx);
 906             fiy1             = _mm256_add_pd(fiy1,ty);
 907             fiz1             = _mm256_add_pd(fiz1,tz);
 908
 909             fjx0             = _mm256_add_pd(fjx0,tx);
 910             fjy0             = _mm256_add_pd(fjy0,ty);
 911             fjz0             = _mm256_add_pd(fjz0,tz);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             /* Compute parameters for interactions between i and j atoms */
 918             qq20             = _mm256_mul_pd(iq2,jq0);
 919
 920             /* REACTION-FIELD ELECTROSTATICS */
 921             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 922
 923             fscal            = felec;
 924
 925             /* Calculate temporary vectorial force */
 926             tx               = _mm256_mul_pd(fscal,dx20);
 927             ty               = _mm256_mul_pd(fscal,dy20);
 928             tz               = _mm256_mul_pd(fscal,dz20);
 929
 930             /* Update vectorial force */
 931             fix2             = _mm256_add_pd(fix2,tx);
 932             fiy2             = _mm256_add_pd(fiy2,ty);
 933             fiz2             = _mm256_add_pd(fiz2,tz);
 934
 935             fjx0             = _mm256_add_pd(fjx0,tx);
 936             fjy0             = _mm256_add_pd(fjy0,ty);
 937             fjz0             = _mm256_add_pd(fjz0,tz);
 938
 939             fjptrA             = f+j_coord_offsetA;
 940             fjptrB             = f+j_coord_offsetB;
 941             fjptrC             = f+j_coord_offsetC;
 942             fjptrD             = f+j_coord_offsetD;
 943
 944             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 945
 946             /* Inner loop uses 111 flops */
 947         }
 948
 949         if(jidx<j_index_end)
 950         {
 951
 952             /* Get j neighbor index, and coordinate index */
 953             jnrlistA         = jjnr[jidx];
 954             jnrlistB         = jjnr[jidx+1];
 955             jnrlistC         = jjnr[jidx+2];
 956             jnrlistD         = jjnr[jidx+3];
 957             /* Sign of each element will be negative for non-real atoms.
 958              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 959              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 960              */
 961             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 962
 963             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 964             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 965             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 966
 967             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 968             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 969             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 970             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 971             j_coord_offsetA  = DIM*jnrA;
 972             j_coord_offsetB  = DIM*jnrB;
 973             j_coord_offsetC  = DIM*jnrC;
 974             j_coord_offsetD  = DIM*jnrD;
 975
 976             /* load j atom coordinates */
 977             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 978                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 979                                                  &jx0,&jy0,&jz0);
 980
 981             /* Calculate displacement vector */
 982             dx00             = _mm256_sub_pd(ix0,jx0);
 983             dy00             = _mm256_sub_pd(iy0,jy0);
 984             dz00             = _mm256_sub_pd(iz0,jz0);
 985             dx10             = _mm256_sub_pd(ix1,jx0);
 986             dy10             = _mm256_sub_pd(iy1,jy0);
 987             dz10             = _mm256_sub_pd(iz1,jz0);
 988             dx20             = _mm256_sub_pd(ix2,jx0);
 989             dy20             = _mm256_sub_pd(iy2,jy0);
 990             dz20             = _mm256_sub_pd(iz2,jz0);
 991
 992             /* Calculate squared distance and things based on it */
 993             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 994             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 995             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 996
 997             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 998             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 999             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1000
1001             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1002             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1003             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1004
1005             /* Load parameters for j particles */
1006             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1007                                                                  charge+jnrC+0,charge+jnrD+0);
1008             vdwjidx0A        = 2*vdwtype[jnrA+0];
1009             vdwjidx0B        = 2*vdwtype[jnrB+0];
1010             vdwjidx0C        = 2*vdwtype[jnrC+0];
1011             vdwjidx0D        = 2*vdwtype[jnrD+0];
1012
1013             fjx0             = _mm256_setzero_pd();
1014             fjy0             = _mm256_setzero_pd();
1015             fjz0             = _mm256_setzero_pd();
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             r00              = _mm256_mul_pd(rsq00,rinv00);
1022             r00              = _mm256_andnot_pd(dummy_mask,r00);
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq00             = _mm256_mul_pd(iq0,jq0);
1026             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1027                                             vdwioffsetptr0+vdwjidx0B,
1028                                             vdwioffsetptr0+vdwjidx0C,
1029                                             vdwioffsetptr0+vdwjidx0D,
1030                                             &c6_00,&c12_00);
1031
1032             /* Calculate table index by multiplying r with table scale and truncate to integer */
1033             rt               = _mm256_mul_pd(r00,vftabscale);
1034             vfitab           = _mm256_cvttpd_epi32(rt);
1035             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1036             vfitab           = _mm_slli_epi32(vfitab,3);
1037
1038             /* REACTION-FIELD ELECTROSTATICS */
1039             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1040
1041             /* CUBIC SPLINE TABLE DISPERSION */
1042             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1043             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1044             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1045             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1046             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1047             Heps             = _mm256_mul_pd(vfeps,H);
1048             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1049             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1050             fvdw6            = _mm256_mul_pd(c6_00,FF);
1051
1052             /* CUBIC SPLINE TABLE REPULSION */
1053             vfitab           = _mm_add_epi32(vfitab,ifour);
1054             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1055             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1056             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1057             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1058             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1059             Heps             = _mm256_mul_pd(vfeps,H);
1060             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1061             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1062             fvdw12           = _mm256_mul_pd(c12_00,FF);
1063             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1064
1065             fscal            = _mm256_add_pd(felec,fvdw);
1066
1067             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm256_mul_pd(fscal,dx00);
1071             ty               = _mm256_mul_pd(fscal,dy00);
1072             tz               = _mm256_mul_pd(fscal,dz00);
1073
1074             /* Update vectorial force */
1075             fix0             = _mm256_add_pd(fix0,tx);
1076             fiy0             = _mm256_add_pd(fiy0,ty);
1077             fiz0             = _mm256_add_pd(fiz0,tz);
1078
1079             fjx0             = _mm256_add_pd(fjx0,tx);
1080             fjy0             = _mm256_add_pd(fjy0,ty);
1081             fjz0             = _mm256_add_pd(fjz0,tz);
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             /* Compute parameters for interactions between i and j atoms */
1088             qq10             = _mm256_mul_pd(iq1,jq0);
1089
1090             /* REACTION-FIELD ELECTROSTATICS */
1091             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1092
1093             fscal            = felec;
1094
1095             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm256_mul_pd(fscal,dx10);
1099             ty               = _mm256_mul_pd(fscal,dy10);
1100             tz               = _mm256_mul_pd(fscal,dz10);
1101
1102             /* Update vectorial force */
1103             fix1             = _mm256_add_pd(fix1,tx);
1104             fiy1             = _mm256_add_pd(fiy1,ty);
1105             fiz1             = _mm256_add_pd(fiz1,tz);
1106
1107             fjx0             = _mm256_add_pd(fjx0,tx);
1108             fjy0             = _mm256_add_pd(fjy0,ty);
1109             fjz0             = _mm256_add_pd(fjz0,tz);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq20             = _mm256_mul_pd(iq2,jq0);
1117
1118             /* REACTION-FIELD ELECTROSTATICS */
1119             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1120
1121             fscal            = felec;
1122
1123             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm256_mul_pd(fscal,dx20);
1127             ty               = _mm256_mul_pd(fscal,dy20);
1128             tz               = _mm256_mul_pd(fscal,dz20);
1129
1130             /* Update vectorial force */
1131             fix2             = _mm256_add_pd(fix2,tx);
1132             fiy2             = _mm256_add_pd(fiy2,ty);
1133             fiz2             = _mm256_add_pd(fiz2,tz);
1134
1135             fjx0             = _mm256_add_pd(fjx0,tx);
1136             fjy0             = _mm256_add_pd(fjy0,ty);
1137             fjz0             = _mm256_add_pd(fjz0,tz);
1138
1139             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1140             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1141             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1142             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1143
1144             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1145
1146             /* Inner loop uses 112 flops */
1147         }
1148
1149         /* End of innermost loop */
1150
1151         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1152                                                  f+i_coord_offset,fshift+i_shift_offset);
1153
1154         /* Increment number of inner iterations */
1155         inneriter                  += j_index_end - j_index_start;
1156
1157         /* Outer loop uses 18 flops */
1158     }
1159
1160     /* Increment number of outer iterations */
1161     outeriter        += nri;
1162
1163     /* Update outer/inner flops */
1164
1165     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*112);
1166 }