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) 2012,2013,2014, 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_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m128i          vfitab;
 104     __m128i          ifour       = _mm_set1_epi32(4);
 105     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 106     real             *vftab;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm256_set1_pd(fr->ic->k_rf);
 126     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 127     crf              = _mm256_set1_pd(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 138     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 139     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 140     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 174
 175         fix0             = _mm256_setzero_pd();
 176         fiy0             = _mm256_setzero_pd();
 177         fiz0             = _mm256_setzero_pd();
 178         fix1             = _mm256_setzero_pd();
 179         fiy1             = _mm256_setzero_pd();
 180         fiz1             = _mm256_setzero_pd();
 181         fix2             = _mm256_setzero_pd();
 182         fiy2             = _mm256_setzero_pd();
 183         fiz2             = _mm256_setzero_pd();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm256_setzero_pd();
 187         vvdwsum          = _mm256_setzero_pd();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 206                                                  &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm256_sub_pd(ix0,jx0);
 210             dy00             = _mm256_sub_pd(iy0,jy0);
 211             dz00             = _mm256_sub_pd(iz0,jz0);
 212             dx10             = _mm256_sub_pd(ix1,jx0);
 213             dy10             = _mm256_sub_pd(iy1,jy0);
 214             dz10             = _mm256_sub_pd(iz1,jz0);
 215             dx20             = _mm256_sub_pd(ix2,jx0);
 216             dy20             = _mm256_sub_pd(iy2,jy0);
 217             dz20             = _mm256_sub_pd(iz2,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 221             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 222             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 223
 224             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 225             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 226             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 227
 228             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 229             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 230             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 234                                                                  charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             fjx0             = _mm256_setzero_pd();
 241             fjy0             = _mm256_setzero_pd();
 242             fjz0             = _mm256_setzero_pd();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             r00              = _mm256_mul_pd(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm256_mul_pd(iq0,jq0);
 252             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 253                                             vdwioffsetptr0+vdwjidx0B,
 254                                             vdwioffsetptr0+vdwjidx0C,
 255                                             vdwioffsetptr0+vdwjidx0D,
 256                                             &c6_00,&c12_00);
 257
 258             /* Calculate table index by multiplying r with table scale and truncate to integer */
 259             rt               = _mm256_mul_pd(r00,vftabscale);
 260             vfitab           = _mm256_cvttpd_epi32(rt);
 261             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 262             vfitab           = _mm_slli_epi32(vfitab,3);
 263
 264             /* REACTION-FIELD ELECTROSTATICS */
 265             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 266             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 267
 268             /* CUBIC SPLINE TABLE DISPERSION */
 269             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 270             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 271             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 272             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 273             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 274             Heps             = _mm256_mul_pd(vfeps,H);
 275             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 276             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 277             vvdw6            = _mm256_mul_pd(c6_00,VV);
 278             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 279             fvdw6            = _mm256_mul_pd(c6_00,FF);
 280
 281             /* CUBIC SPLINE TABLE REPULSION */
 282             vfitab           = _mm_add_epi32(vfitab,ifour);
 283             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 284             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 285             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 286             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 287             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 288             Heps             = _mm256_mul_pd(vfeps,H);
 289             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 290             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 291             vvdw12           = _mm256_mul_pd(c12_00,VV);
 292             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 293             fvdw12           = _mm256_mul_pd(c12_00,FF);
 294             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 295             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm256_add_pd(velecsum,velec);
 299             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 300
 301             fscal            = _mm256_add_pd(felec,fvdw);
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm256_mul_pd(fscal,dx00);
 305             ty               = _mm256_mul_pd(fscal,dy00);
 306             tz               = _mm256_mul_pd(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm256_add_pd(fix0,tx);
 310             fiy0             = _mm256_add_pd(fiy0,ty);
 311             fiz0             = _mm256_add_pd(fiz0,tz);
 312
 313             fjx0             = _mm256_add_pd(fjx0,tx);
 314             fjy0             = _mm256_add_pd(fjy0,ty);
 315             fjz0             = _mm256_add_pd(fjz0,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm256_mul_pd(iq1,jq0);
 323
 324             /* REACTION-FIELD ELECTROSTATICS */
 325             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 326             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm256_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm256_mul_pd(fscal,dx10);
 335             ty               = _mm256_mul_pd(fscal,dy10);
 336             tz               = _mm256_mul_pd(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm256_add_pd(fix1,tx);
 340             fiy1             = _mm256_add_pd(fiy1,ty);
 341             fiz1             = _mm256_add_pd(fiz1,tz);
 342
 343             fjx0             = _mm256_add_pd(fjx0,tx);
 344             fjy0             = _mm256_add_pd(fjy0,ty);
 345             fjz0             = _mm256_add_pd(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm256_mul_pd(iq2,jq0);
 353
 354             /* REACTION-FIELD ELECTROSTATICS */
 355             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 356             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm256_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm256_mul_pd(fscal,dx20);
 365             ty               = _mm256_mul_pd(fscal,dy20);
 366             tz               = _mm256_mul_pd(fscal,dz20);
 367
 368             /* Update vectorial force */
 369             fix2             = _mm256_add_pd(fix2,tx);
 370             fiy2             = _mm256_add_pd(fiy2,ty);
 371             fiz2             = _mm256_add_pd(fiz2,tz);
 372
 373             fjx0             = _mm256_add_pd(fjx0,tx);
 374             fjy0             = _mm256_add_pd(fjy0,ty);
 375             fjz0             = _mm256_add_pd(fjz0,tz);
 376
 377             fjptrA             = f+j_coord_offsetA;
 378             fjptrB             = f+j_coord_offsetB;
 379             fjptrC             = f+j_coord_offsetC;
 380             fjptrD             = f+j_coord_offsetD;
 381
 382             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 383
 384             /* Inner loop uses 134 flops */
 385         }
 386
 387         if(jidx<j_index_end)
 388         {
 389
 390             /* Get j neighbor index, and coordinate index */
 391             jnrlistA         = jjnr[jidx];
 392             jnrlistB         = jjnr[jidx+1];
 393             jnrlistC         = jjnr[jidx+2];
 394             jnrlistD         = jjnr[jidx+3];
 395             /* Sign of each element will be negative for non-real atoms.
 396              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 397              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 398              */
 399             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 400
 401             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 402             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 403             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 404
 405             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 406             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 407             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 408             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 409             j_coord_offsetA  = DIM*jnrA;
 410             j_coord_offsetB  = DIM*jnrB;
 411             j_coord_offsetC  = DIM*jnrC;
 412             j_coord_offsetD  = DIM*jnrD;
 413
 414             /* load j atom coordinates */
 415             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 416                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 417                                                  &jx0,&jy0,&jz0);
 418
 419             /* Calculate displacement vector */
 420             dx00             = _mm256_sub_pd(ix0,jx0);
 421             dy00             = _mm256_sub_pd(iy0,jy0);
 422             dz00             = _mm256_sub_pd(iz0,jz0);
 423             dx10             = _mm256_sub_pd(ix1,jx0);
 424             dy10             = _mm256_sub_pd(iy1,jy0);
 425             dz10             = _mm256_sub_pd(iz1,jz0);
 426             dx20             = _mm256_sub_pd(ix2,jx0);
 427             dy20             = _mm256_sub_pd(iy2,jy0);
 428             dz20             = _mm256_sub_pd(iz2,jz0);
 429
 430             /* Calculate squared distance and things based on it */
 431             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 432             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 433             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 434
 435             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 436             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 437             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 438
 439             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 440             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 441             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 442
 443             /* Load parameters for j particles */
 444             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 445                                                                  charge+jnrC+0,charge+jnrD+0);
 446             vdwjidx0A        = 2*vdwtype[jnrA+0];
 447             vdwjidx0B        = 2*vdwtype[jnrB+0];
 448             vdwjidx0C        = 2*vdwtype[jnrC+0];
 449             vdwjidx0D        = 2*vdwtype[jnrD+0];
 450
 451             fjx0             = _mm256_setzero_pd();
 452             fjy0             = _mm256_setzero_pd();
 453             fjz0             = _mm256_setzero_pd();
 454
 455             /**************************
 456              * CALCULATE INTERACTIONS *
 457              **************************/
 458
 459             r00              = _mm256_mul_pd(rsq00,rinv00);
 460             r00              = _mm256_andnot_pd(dummy_mask,r00);
 461
 462             /* Compute parameters for interactions between i and j atoms */
 463             qq00             = _mm256_mul_pd(iq0,jq0);
 464             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 465                                             vdwioffsetptr0+vdwjidx0B,
 466                                             vdwioffsetptr0+vdwjidx0C,
 467                                             vdwioffsetptr0+vdwjidx0D,
 468                                             &c6_00,&c12_00);
 469
 470             /* Calculate table index by multiplying r with table scale and truncate to integer */
 471             rt               = _mm256_mul_pd(r00,vftabscale);
 472             vfitab           = _mm256_cvttpd_epi32(rt);
 473             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 474             vfitab           = _mm_slli_epi32(vfitab,3);
 475
 476             /* REACTION-FIELD ELECTROSTATICS */
 477             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 478             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 479
 480             /* CUBIC SPLINE TABLE DISPERSION */
 481             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 482             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 483             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 484             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 485             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 486             Heps             = _mm256_mul_pd(vfeps,H);
 487             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 488             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 489             vvdw6            = _mm256_mul_pd(c6_00,VV);
 490             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 491             fvdw6            = _mm256_mul_pd(c6_00,FF);
 492
 493             /* CUBIC SPLINE TABLE REPULSION */
 494             vfitab           = _mm_add_epi32(vfitab,ifour);
 495             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 496             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 497             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 498             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 499             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 500             Heps             = _mm256_mul_pd(vfeps,H);
 501             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 502             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 503             vvdw12           = _mm256_mul_pd(c12_00,VV);
 504             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 505             fvdw12           = _mm256_mul_pd(c12_00,FF);
 506             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 507             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velec            = _mm256_andnot_pd(dummy_mask,velec);
 511             velecsum         = _mm256_add_pd(velecsum,velec);
 512             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 513             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 514
 515             fscal            = _mm256_add_pd(felec,fvdw);
 516
 517             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 518
 519             /* Calculate temporary vectorial force */
 520             tx               = _mm256_mul_pd(fscal,dx00);
 521             ty               = _mm256_mul_pd(fscal,dy00);
 522             tz               = _mm256_mul_pd(fscal,dz00);
 523
 524             /* Update vectorial force */
 525             fix0             = _mm256_add_pd(fix0,tx);
 526             fiy0             = _mm256_add_pd(fiy0,ty);
 527             fiz0             = _mm256_add_pd(fiz0,tz);
 528
 529             fjx0             = _mm256_add_pd(fjx0,tx);
 530             fjy0             = _mm256_add_pd(fjy0,ty);
 531             fjz0             = _mm256_add_pd(fjz0,tz);
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq10             = _mm256_mul_pd(iq1,jq0);
 539
 540             /* REACTION-FIELD ELECTROSTATICS */
 541             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 542             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm256_andnot_pd(dummy_mask,velec);
 546             velecsum         = _mm256_add_pd(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_pd(fscal,dx10);
 554             ty               = _mm256_mul_pd(fscal,dy10);
 555             tz               = _mm256_mul_pd(fscal,dz10);
 556
 557             /* Update vectorial force */
 558             fix1             = _mm256_add_pd(fix1,tx);
 559             fiy1             = _mm256_add_pd(fiy1,ty);
 560             fiz1             = _mm256_add_pd(fiz1,tz);
 561
 562             fjx0             = _mm256_add_pd(fjx0,tx);
 563             fjy0             = _mm256_add_pd(fjy0,ty);
 564             fjz0             = _mm256_add_pd(fjz0,tz);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq20             = _mm256_mul_pd(iq2,jq0);
 572
 573             /* REACTION-FIELD ELECTROSTATICS */
 574             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 575             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_andnot_pd(dummy_mask,velec);
 579             velecsum         = _mm256_add_pd(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_pd(fscal,dx20);
 587             ty               = _mm256_mul_pd(fscal,dy20);
 588             tz               = _mm256_mul_pd(fscal,dz20);
 589
 590             /* Update vectorial force */
 591             fix2             = _mm256_add_pd(fix2,tx);
 592             fiy2             = _mm256_add_pd(fiy2,ty);
 593             fiz2             = _mm256_add_pd(fiz2,tz);
 594
 595             fjx0             = _mm256_add_pd(fjx0,tx);
 596             fjy0             = _mm256_add_pd(fjy0,ty);
 597             fjz0             = _mm256_add_pd(fjz0,tz);
 598
 599             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 600             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 601             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 602             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 603
 604             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 605
 606             /* Inner loop uses 135 flops */
 607         }
 608
 609         /* End of innermost loop */
 610
 611         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 612                                                  f+i_coord_offset,fshift+i_shift_offset);
 613
 614         ggid                        = gid[iidx];
 615         /* Update potential energies */
 616         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 617         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 618
 619         /* Increment number of inner iterations */
 620         inneriter                  += j_index_end - j_index_start;
 621
 622         /* Outer loop uses 20 flops */
 623     }
 624
 625     /* Increment number of outer iterations */
 626     outeriter        += nri;
 627
 628     /* Update outer/inner flops */
 629
 630     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*135);
 631 }
 632 /*
 633  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_double
 634  * Electrostatics interaction: ReactionField
 635  * VdW interaction:            CubicSplineTable
 636  * Geometry:                   Water3-Particle
 637  * Calculate force/pot:        Force
 638  */
 639 void
 640 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_256_double
 641                     (t_nblist                    * gmx_restrict       nlist,
 642                      rvec                        * gmx_restrict          xx,
 643                      rvec                        * gmx_restrict          ff,
 644                      t_forcerec                  * gmx_restrict          fr,
 645                      t_mdatoms                   * gmx_restrict     mdatoms,
 646                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 647                      t_nrnb                      * gmx_restrict        nrnb)
 648 {
 649     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 650      * just 0 for non-waters.
 651      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 652      * jnr indices corresponding to data put in the four positions in the SIMD register.
 653      */
 654     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 655     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 656     int              jnrA,jnrB,jnrC,jnrD;
 657     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 658     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 659     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 660     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 661     real             rcutoff_scalar;
 662     real             *shiftvec,*fshift,*x,*f;
 663     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 664     real             scratch[4*DIM];
 665     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 666     real *           vdwioffsetptr0;
 667     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 668     real *           vdwioffsetptr1;
 669     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 670     real *           vdwioffsetptr2;
 671     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 672     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 673     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 674     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 675     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 676     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 677     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 678     real             *charge;
 679     int              nvdwtype;
 680     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 681     int              *vdwtype;
 682     real             *vdwparam;
 683     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 684     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 685     __m128i          vfitab;
 686     __m128i          ifour       = _mm_set1_epi32(4);
 687     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 688     real             *vftab;
 689     __m256d          dummy_mask,cutoff_mask;
 690     __m128           tmpmask0,tmpmask1;
 691     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 692     __m256d          one     = _mm256_set1_pd(1.0);
 693     __m256d          two     = _mm256_set1_pd(2.0);
 694     x                = xx[0];
 695     f                = ff[0];
 696
 697     nri              = nlist->nri;
 698     iinr             = nlist->iinr;
 699     jindex           = nlist->jindex;
 700     jjnr             = nlist->jjnr;
 701     shiftidx         = nlist->shift;
 702     gid              = nlist->gid;
 703     shiftvec         = fr->shift_vec[0];
 704     fshift           = fr->fshift[0];
 705     facel            = _mm256_set1_pd(fr->epsfac);
 706     charge           = mdatoms->chargeA;
 707     krf              = _mm256_set1_pd(fr->ic->k_rf);
 708     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 709     crf              = _mm256_set1_pd(fr->ic->c_rf);
 710     nvdwtype         = fr->ntype;
 711     vdwparam         = fr->nbfp;
 712     vdwtype          = mdatoms->typeA;
 713
 714     vftab            = kernel_data->table_vdw->data;
 715     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 716
 717     /* Setup water-specific parameters */
 718     inr              = nlist->iinr[0];
 719     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 720     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 721     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 722     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 723
 724     /* Avoid stupid compiler warnings */
 725     jnrA = jnrB = jnrC = jnrD = 0;
 726     j_coord_offsetA = 0;
 727     j_coord_offsetB = 0;
 728     j_coord_offsetC = 0;
 729     j_coord_offsetD = 0;
 730
 731     outeriter        = 0;
 732     inneriter        = 0;
 733
 734     for(iidx=0;iidx<4*DIM;iidx++)
 735     {
 736         scratch[iidx] = 0.0;
 737     }
 738
 739     /* Start outer loop over neighborlists */
 740     for(iidx=0; iidx<nri; iidx++)
 741     {
 742         /* Load shift vector for this list */
 743         i_shift_offset   = DIM*shiftidx[iidx];
 744
 745         /* Load limits for loop over neighbors */
 746         j_index_start    = jindex[iidx];
 747         j_index_end      = jindex[iidx+1];
 748
 749         /* Get outer coordinate index */
 750         inr              = iinr[iidx];
 751         i_coord_offset   = DIM*inr;
 752
 753         /* Load i particle coords and add shift vector */
 754         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 755                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 756
 757         fix0             = _mm256_setzero_pd();
 758         fiy0             = _mm256_setzero_pd();
 759         fiz0             = _mm256_setzero_pd();
 760         fix1             = _mm256_setzero_pd();
 761         fiy1             = _mm256_setzero_pd();
 762         fiz1             = _mm256_setzero_pd();
 763         fix2             = _mm256_setzero_pd();
 764         fiy2             = _mm256_setzero_pd();
 765         fiz2             = _mm256_setzero_pd();
 766
 767         /* Start inner kernel loop */
 768         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 769         {
 770
 771             /* Get j neighbor index, and coordinate index */
 772             jnrA             = jjnr[jidx];
 773             jnrB             = jjnr[jidx+1];
 774             jnrC             = jjnr[jidx+2];
 775             jnrD             = jjnr[jidx+3];
 776             j_coord_offsetA  = DIM*jnrA;
 777             j_coord_offsetB  = DIM*jnrB;
 778             j_coord_offsetC  = DIM*jnrC;
 779             j_coord_offsetD  = DIM*jnrD;
 780
 781             /* load j atom coordinates */
 782             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 783                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 784                                                  &jx0,&jy0,&jz0);
 785
 786             /* Calculate displacement vector */
 787             dx00             = _mm256_sub_pd(ix0,jx0);
 788             dy00             = _mm256_sub_pd(iy0,jy0);
 789             dz00             = _mm256_sub_pd(iz0,jz0);
 790             dx10             = _mm256_sub_pd(ix1,jx0);
 791             dy10             = _mm256_sub_pd(iy1,jy0);
 792             dz10             = _mm256_sub_pd(iz1,jz0);
 793             dx20             = _mm256_sub_pd(ix2,jx0);
 794             dy20             = _mm256_sub_pd(iy2,jy0);
 795             dz20             = _mm256_sub_pd(iz2,jz0);
 796
 797             /* Calculate squared distance and things based on it */
 798             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 799             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 800             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 801
 802             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 803             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 804             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 805
 806             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 807             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 808             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 809
 810             /* Load parameters for j particles */
 811             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 812                                                                  charge+jnrC+0,charge+jnrD+0);
 813             vdwjidx0A        = 2*vdwtype[jnrA+0];
 814             vdwjidx0B        = 2*vdwtype[jnrB+0];
 815             vdwjidx0C        = 2*vdwtype[jnrC+0];
 816             vdwjidx0D        = 2*vdwtype[jnrD+0];
 817
 818             fjx0             = _mm256_setzero_pd();
 819             fjy0             = _mm256_setzero_pd();
 820             fjz0             = _mm256_setzero_pd();
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             r00              = _mm256_mul_pd(rsq00,rinv00);
 827
 828             /* Compute parameters for interactions between i and j atoms */
 829             qq00             = _mm256_mul_pd(iq0,jq0);
 830             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 831                                             vdwioffsetptr0+vdwjidx0B,
 832                                             vdwioffsetptr0+vdwjidx0C,
 833                                             vdwioffsetptr0+vdwjidx0D,
 834                                             &c6_00,&c12_00);
 835
 836             /* Calculate table index by multiplying r with table scale and truncate to integer */
 837             rt               = _mm256_mul_pd(r00,vftabscale);
 838             vfitab           = _mm256_cvttpd_epi32(rt);
 839             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 840             vfitab           = _mm_slli_epi32(vfitab,3);
 841
 842             /* REACTION-FIELD ELECTROSTATICS */
 843             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 844
 845             /* CUBIC SPLINE TABLE DISPERSION */
 846             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 847             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 848             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 849             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 850             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 851             Heps             = _mm256_mul_pd(vfeps,H);
 852             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 853             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 854             fvdw6            = _mm256_mul_pd(c6_00,FF);
 855
 856             /* CUBIC SPLINE TABLE REPULSION */
 857             vfitab           = _mm_add_epi32(vfitab,ifour);
 858             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 859             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 860             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 861             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 862             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 863             Heps             = _mm256_mul_pd(vfeps,H);
 864             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 865             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 866             fvdw12           = _mm256_mul_pd(c12_00,FF);
 867             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 868
 869             fscal            = _mm256_add_pd(felec,fvdw);
 870
 871             /* Calculate temporary vectorial force */
 872             tx               = _mm256_mul_pd(fscal,dx00);
 873             ty               = _mm256_mul_pd(fscal,dy00);
 874             tz               = _mm256_mul_pd(fscal,dz00);
 875
 876             /* Update vectorial force */
 877             fix0             = _mm256_add_pd(fix0,tx);
 878             fiy0             = _mm256_add_pd(fiy0,ty);
 879             fiz0             = _mm256_add_pd(fiz0,tz);
 880
 881             fjx0             = _mm256_add_pd(fjx0,tx);
 882             fjy0             = _mm256_add_pd(fjy0,ty);
 883             fjz0             = _mm256_add_pd(fjz0,tz);
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             /* Compute parameters for interactions between i and j atoms */
 890             qq10             = _mm256_mul_pd(iq1,jq0);
 891
 892             /* REACTION-FIELD ELECTROSTATICS */
 893             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 894
 895             fscal            = felec;
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm256_mul_pd(fscal,dx10);
 899             ty               = _mm256_mul_pd(fscal,dy10);
 900             tz               = _mm256_mul_pd(fscal,dz10);
 901
 902             /* Update vectorial force */
 903             fix1             = _mm256_add_pd(fix1,tx);
 904             fiy1             = _mm256_add_pd(fiy1,ty);
 905             fiz1             = _mm256_add_pd(fiz1,tz);
 906
 907             fjx0             = _mm256_add_pd(fjx0,tx);
 908             fjy0             = _mm256_add_pd(fjy0,ty);
 909             fjz0             = _mm256_add_pd(fjz0,tz);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq20             = _mm256_mul_pd(iq2,jq0);
 917
 918             /* REACTION-FIELD ELECTROSTATICS */
 919             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 920
 921             fscal            = felec;
 922
 923             /* Calculate temporary vectorial force */
 924             tx               = _mm256_mul_pd(fscal,dx20);
 925             ty               = _mm256_mul_pd(fscal,dy20);
 926             tz               = _mm256_mul_pd(fscal,dz20);
 927
 928             /* Update vectorial force */
 929             fix2             = _mm256_add_pd(fix2,tx);
 930             fiy2             = _mm256_add_pd(fiy2,ty);
 931             fiz2             = _mm256_add_pd(fiz2,tz);
 932
 933             fjx0             = _mm256_add_pd(fjx0,tx);
 934             fjy0             = _mm256_add_pd(fjy0,ty);
 935             fjz0             = _mm256_add_pd(fjz0,tz);
 936
 937             fjptrA             = f+j_coord_offsetA;
 938             fjptrB             = f+j_coord_offsetB;
 939             fjptrC             = f+j_coord_offsetC;
 940             fjptrD             = f+j_coord_offsetD;
 941
 942             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 943
 944             /* Inner loop uses 111 flops */
 945         }
 946
 947         if(jidx<j_index_end)
 948         {
 949
 950             /* Get j neighbor index, and coordinate index */
 951             jnrlistA         = jjnr[jidx];
 952             jnrlistB         = jjnr[jidx+1];
 953             jnrlistC         = jjnr[jidx+2];
 954             jnrlistD         = jjnr[jidx+3];
 955             /* Sign of each element will be negative for non-real atoms.
 956              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 957              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 958              */
 959             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 960
 961             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 962             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 963             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 964
 965             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 966             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 967             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 968             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 969             j_coord_offsetA  = DIM*jnrA;
 970             j_coord_offsetB  = DIM*jnrB;
 971             j_coord_offsetC  = DIM*jnrC;
 972             j_coord_offsetD  = DIM*jnrD;
 973
 974             /* load j atom coordinates */
 975             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 976                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 977                                                  &jx0,&jy0,&jz0);
 978
 979             /* Calculate displacement vector */
 980             dx00             = _mm256_sub_pd(ix0,jx0);
 981             dy00             = _mm256_sub_pd(iy0,jy0);
 982             dz00             = _mm256_sub_pd(iz0,jz0);
 983             dx10             = _mm256_sub_pd(ix1,jx0);
 984             dy10             = _mm256_sub_pd(iy1,jy0);
 985             dz10             = _mm256_sub_pd(iz1,jz0);
 986             dx20             = _mm256_sub_pd(ix2,jx0);
 987             dy20             = _mm256_sub_pd(iy2,jy0);
 988             dz20             = _mm256_sub_pd(iz2,jz0);
 989
 990             /* Calculate squared distance and things based on it */
 991             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 992             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 993             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 994
 995             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 996             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 997             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 998
 999             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1000             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1001             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1002
1003             /* Load parameters for j particles */
1004             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1005                                                                  charge+jnrC+0,charge+jnrD+0);
1006             vdwjidx0A        = 2*vdwtype[jnrA+0];
1007             vdwjidx0B        = 2*vdwtype[jnrB+0];
1008             vdwjidx0C        = 2*vdwtype[jnrC+0];
1009             vdwjidx0D        = 2*vdwtype[jnrD+0];
1010
1011             fjx0             = _mm256_setzero_pd();
1012             fjy0             = _mm256_setzero_pd();
1013             fjz0             = _mm256_setzero_pd();
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             r00              = _mm256_mul_pd(rsq00,rinv00);
1020             r00              = _mm256_andnot_pd(dummy_mask,r00);
1021
1022             /* Compute parameters for interactions between i and j atoms */
1023             qq00             = _mm256_mul_pd(iq0,jq0);
1024             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1025                                             vdwioffsetptr0+vdwjidx0B,
1026                                             vdwioffsetptr0+vdwjidx0C,
1027                                             vdwioffsetptr0+vdwjidx0D,
1028                                             &c6_00,&c12_00);
1029
1030             /* Calculate table index by multiplying r with table scale and truncate to integer */
1031             rt               = _mm256_mul_pd(r00,vftabscale);
1032             vfitab           = _mm256_cvttpd_epi32(rt);
1033             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1034             vfitab           = _mm_slli_epi32(vfitab,3);
1035
1036             /* REACTION-FIELD ELECTROSTATICS */
1037             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1038
1039             /* CUBIC SPLINE TABLE DISPERSION */
1040             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1041             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1042             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1043             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1044             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1045             Heps             = _mm256_mul_pd(vfeps,H);
1046             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1047             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1048             fvdw6            = _mm256_mul_pd(c6_00,FF);
1049
1050             /* CUBIC SPLINE TABLE REPULSION */
1051             vfitab           = _mm_add_epi32(vfitab,ifour);
1052             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1053             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1054             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1055             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1056             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1057             Heps             = _mm256_mul_pd(vfeps,H);
1058             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1059             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1060             fvdw12           = _mm256_mul_pd(c12_00,FF);
1061             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1062
1063             fscal            = _mm256_add_pd(felec,fvdw);
1064
1065             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm256_mul_pd(fscal,dx00);
1069             ty               = _mm256_mul_pd(fscal,dy00);
1070             tz               = _mm256_mul_pd(fscal,dz00);
1071
1072             /* Update vectorial force */
1073             fix0             = _mm256_add_pd(fix0,tx);
1074             fiy0             = _mm256_add_pd(fiy0,ty);
1075             fiz0             = _mm256_add_pd(fiz0,tz);
1076
1077             fjx0             = _mm256_add_pd(fjx0,tx);
1078             fjy0             = _mm256_add_pd(fjy0,ty);
1079             fjz0             = _mm256_add_pd(fjz0,tz);
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq10             = _mm256_mul_pd(iq1,jq0);
1087
1088             /* REACTION-FIELD ELECTROSTATICS */
1089             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1090
1091             fscal            = felec;
1092
1093             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm256_mul_pd(fscal,dx10);
1097             ty               = _mm256_mul_pd(fscal,dy10);
1098             tz               = _mm256_mul_pd(fscal,dz10);
1099
1100             /* Update vectorial force */
1101             fix1             = _mm256_add_pd(fix1,tx);
1102             fiy1             = _mm256_add_pd(fiy1,ty);
1103             fiz1             = _mm256_add_pd(fiz1,tz);
1104
1105             fjx0             = _mm256_add_pd(fjx0,tx);
1106             fjy0             = _mm256_add_pd(fjy0,ty);
1107             fjz0             = _mm256_add_pd(fjz0,tz);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             /* Compute parameters for interactions between i and j atoms */
1114             qq20             = _mm256_mul_pd(iq2,jq0);
1115
1116             /* REACTION-FIELD ELECTROSTATICS */
1117             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1118
1119             fscal            = felec;
1120
1121             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm256_mul_pd(fscal,dx20);
1125             ty               = _mm256_mul_pd(fscal,dy20);
1126             tz               = _mm256_mul_pd(fscal,dz20);
1127
1128             /* Update vectorial force */
1129             fix2             = _mm256_add_pd(fix2,tx);
1130             fiy2             = _mm256_add_pd(fiy2,ty);
1131             fiz2             = _mm256_add_pd(fiz2,tz);
1132
1133             fjx0             = _mm256_add_pd(fjx0,tx);
1134             fjy0             = _mm256_add_pd(fjy0,ty);
1135             fjz0             = _mm256_add_pd(fjz0,tz);
1136
1137             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1138             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1139             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1140             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1141
1142             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1143
1144             /* Inner loop uses 112 flops */
1145         }
1146
1147         /* End of innermost loop */
1148
1149         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1150                                                  f+i_coord_offset,fshift+i_shift_offset);
1151
1152         /* Increment number of inner iterations */
1153         inneriter                  += j_index_end - j_index_start;
1154
1155         /* Outer loop uses 18 flops */
1156     }
1157
1158     /* Increment number of outer iterations */
1159     outeriter        += nri;
1160
1161     /* Update outer/inner flops */
1162
1163     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*112);
1164 }